Research ArticleRelationship between Hyperuricemia andHaar-Like Features on Tongue Images

Yan Cui12 Shizhong Liao1 Hongwu Wang3 Hongyu Liu4 Wenhua Wang4 and Liqun Yin2

1School of Computer Science and Technology Tianjin University 92 Weijin Road Nankai District Tianjin 300072 China2Department of Common Required Courses Tianjin University of Traditional Chinese Medicine 312 Anshanxi RoadNankai District Tianjin 300193 China3College of Traditional Chinese Medicine Tianjin University of Traditional Chinese Medicine 312 Anshanxi RoadNankai District Tianjin 300193 China4Wuqing Chinese Medicine Hospital 10 Jichang Road Wuqing District Tianjin 301700 China

Correspondence should be addressed to HongwuWang cms tjutcm126com

Received 19 September 2014 Accepted 24 December 2014

Academic Editor Tao Huang

Copyright copy 2015 Yan Cui et al This is an open access article distributed under the Creative Commons Attribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Objective To investigate differences in tongue images of subjects with and without hyperuricemia Materials and Methods Thispopulation-based case-control study was performed in 2012-2013 We collected data from 46 case subjects with hyperuricemiaand 46 control subjects including results of biochemical examinations and tongue images Symmetrical Haar-like features basedon integral images were extracted from tongue images T-tests were performed to determine the ability of extracted features todistinguish between the case and control groups We first selected features using the common criterion 119875 lt 005 then conductedfurther examination of feature characteristics and feature selection usingmeans and standard deviations of distributions in the caseand control groups Results A total of 115683 features were selected using the criterion 119875 lt 005 The maximum area under thereceiver operating characteristic curve (AUC) of these featureswas 0877The sensitivity of the featurewith themaximumAUCvaluewas 0800 and specificity was 0826 when the Youden index was maximized Features that performed well were concentrated in thetongue root regionConclusions SymmetricalHaar-like features enabled discrimination of subjects with andwithout hyperuricemiain our sample The locations of these discriminative features were in agreement with the interpretation of tongue appearance intraditional Chinese and Western medicine

1 Introduction

Hyperuricemia is a metabolic disorder in which the bodyproduces excessive uric acid and fails to excrete it Excessdietary purines (eg from meat and certain seafood) play asignificant role in hyperuricemia and contribute to gout [1]More precisely hypoxanthine is considered to be an impor-tant factor contributing to hyperuricemia [2] Decreased uricacid excretion is most commonly attributed to genetic factorsand medications [3 4] Although the mechanism remainsunknown many studies have found relationships betweenhyperuricemia or urinary abnormalities and impaired kidneyfunction [5 6]Thus impaired kidney function is consideredto be a risk factor for hyperuricemia [7] In turn hyper-uricemia is considered to be a risk factor for severe diseases

that can impact quality of life and lead to disability and evendeath including coronary heart disease hypertension strokeand insulin resistance [8ndash11]

With rapid economic development daily diet and health-care in China have improved The prevalence of hyper-uricemia has increased with dietary purine content accord-ing to a meta-analysis conducted in 2011 it was 216among males and 86 among females in China [12] Forcomparison the prevalence of hyperuricemia in the UnitedStates was only 127 in 2010 [13] The high prevalence ofhyperuricemia renders its accurate diagnosis critical

Serum uric acid (SUA) concentration analysis is thegold standard for hyperuricemia diagnosis However thismethod necessitates invasive blood sample collection andbiochemical examination which are time consuming and

Hindawi Publishing CorporationBioMed Research InternationalVolume 2015 Article ID 363216 10 pageshttpdxdoiorg1011552015363216

2 BioMed Research International

(a) (b) (c)

Figure 1 Image processing procedure (a) Original image acquired by the tongue analyzer (b) Rectangular area containing the tonguesegmented from the original image (c) Final image scaled to 120 times 100 pixels

laborious and risk patient injury The development of arapid simple noninvasive method would thus improve thediagnostic procedure for hyperuricemia

Tongue images have been applied as inexpensive andnoninvasive means of diagnosing several diseases such asstroke and appendicitis [14ndash16] Wang et al [17] statisticallyanalyzed features extracted from tongue images defining 12image classes Other statistical methods such as Bayesiannetworks and a bagging tree algorithm have been applied totongue image analysis [15 16] However these studies didnot employ case-control designs that would have avoidedbias introduced by age and sex differences in tongue fea-tures Jung et al [18] performed a case-control study toexamine differences in color distribution on tongue imagesbetween subjects with and without sleep disorders but thediagnostic criteria used in this study were based on thephysicianrsquos judgment rather than biochemical examinationWestern medical studies have found that tongue appearance(coloration and coating) is related to kidney diseases orconditions such as renal adenocarcinoma tongue metastasisand kidney transplantation [19 20] Traditional Chinesemedicine (TCM) studies have also found that tongue imagecharacteristics can reflect renal deficiency [21 22]

The present case-control study was performed to identifytongue image features useful for the diagnosis of hyper-uricemia A series of symmetrical Haar-like features whichhave been applied successfully to face detection [23] wereextracted from tongue images from subjects with andwithouthyperuricemia (diagnoses were confirmed biochemically)We sought to identify independently useful and readilyinterpretable Haar-like features for the diagnosis of hyper-uricemia

2 Subjects and Methods

21 Subjects and Examination BetweenAugust 2011 and June2012 outpatients from Wuqing Chinese Medicine Hospitala medical examination center of teaching hospital affiliatedwith the Tianjin University of Traditional Chinese Medicine

(TJUTCM) participated in this study All participants pro-vided informed consent and this study was approved by themedical ethics committee of TJUTCM Adults from all agegroups were included to avoid bias introduced by uneven agedistribution Based on data from medical records accessedthrough the hospitalrsquos health information system subjectswith diseases impacting the appearance of the tongue suchas hypertension diabetes and cancer were excluded Thosewith dyed and scraped tongue fur as determined by outpa-tient interviews were also excluded

The diagnostic criteria for hyperuricemia in malesand females were SUA gt416 120583molL (7mgdL) and SUAgt357 120583molL (6mgdL) respectively [24] An image of eachsubjectrsquos tongue was acquired using a YM-III tongue imageanalyzer (httptjtzymcomtonguehtml)

Case and control subjects were matched 1 1 by age(within 1 year) and sex to exclude the impacts of thesecovariates and improve the value of the empirical data [2526] Two-tailed t-tests for samples with equal and unequalvariance were used to confirm similarity in age and differencein SUA value respectively between case and control subjects

22 Image Processing and Feature Selection Tongue imageswere processed as shown in Figure 1 The original imageacquired by the tongue analyzer which depicts the subjectrsquosentire face (Figure 1(a)) was first segmented to include onlythe rectangular area depicting the tongue (Figure 1(b)) Eachimage was then scaled to 120 times 100 pixels (Figure 1(c))to enable efficient feature extraction while retaining colorinformation

Several feature types can be used in image analysisFeatures based on statistical analysis of color represent globaldifferences (expressed as means and standard deviations)among images and are the most intuitive feature type [15]but they cannot describe differences among areas in a singleimage The use of pixel analysis to define image features hasa high computational cost and does not provide high-levelinformation about the images [23] Moreover the number ofpixels is much greater than the number of images in mostsituations and adjacent pixels are often closely correlated

BioMed Research International 3

(a) (b)

Figure 2 Upper-lower and left-right area arrangements The sum of pixels in the lower (a) or right (b) rectangle is subtracted from the sumof pixels in the upper or left rectangle respectively

One Haar-like featureH

W

Top-left corner of tongue image

X

Y

V

Figure 3 Haar-like feature parameters119883 and119884 describe the position of the top-left corner of the feature119867 represents the height of a feature119882 and 119881 represent the widths of the left and right rectangles respectively

these characteristics complicate statistical analysis For thisreason we used Haar-like features [23] which fall betweenthe pixel and global levels in the present study These fea-tures enable examination of color differences between areaspartially solving the problem of correlations among pixelsHowever the number of such features is large exceeding160000 in a 24 times 24-pixel image and the computational costof Haar-like feature extraction remains large [23]

We first sought to reduce the number of Haar-likefeatures in the tongue images which exceeded our computingcapabilities Considering that observation of the tongue isbased on color we first selected features in the red greenand blue color plains ignoring plains in other color spaces(ie Lab) We then employed directional selection whichinvolved the delineation of two adjacent rectangles on eachimage Figure 2 shows two approaches to such selection thesumof pixels in the lower or right rectanglemay be subtracted

from that in the upper or left rectangle respectively Giventhat the human body is characterized predominantly bybilateral symmetry we subtracted the sum of pixels in theright from that in the left rectangle to select Haar-likefeatures Finally we applied scale selection based on the fiveparameters of the left-right feature (Figure 3) Given thatfacial pixels near image boundaries were meaningless forthis study we determined the values of 119884 and 119883 (whichdescribe the position of the top left corner of a feature)as 11 12 110 and 11 12 90 respectively Giventhat overly narrow or short features provide inadequateinformation we determined the values of 119882 (width of leftrectangle) and 119867 (feature height) as 10 15 lfloor(100 minus119883)2rfloor and 10 15 100 minus 119884 respectively Consideringthe number of pixels contained in the two rectangles we set119882 = 119881 (width of right rectangle) Integral image scanning[20] was applied to reduce the computational cost of feature

4 BioMed Research International

(a) (b)

Figure 4 Three features on the red color plain examined in this study (a) contains 6800 pixels (b) contains 600 pixels

extraction This technique requires a single image scan andvalues of all features can be computed within several secondsBecause feature positions and scales are determined by119883 119884119882 and 119867 features are identified by these values using theformat ldquofeature (119883119884119882119867)rdquo in this text Each color plaincontained 195840 features (total = 587520 features in threeplains)

23 Statistical Analysis At this stage of processing thenumber of selected Haar-like features far exceeds the numberof subjects and correlation among features remains strongdue to overlap preventing direct application in classificatorymodels Gorkani and Picard [27] found that human eyesdistinguish images using high-level textural features In thisstudy we thus assumed that the diagnosis of hyperuricemiawould be based on instantaneous extraction of a feature froma tongue image in a single glance We also assumed that allglances would be independent We used Studentrsquos 119905-tests toexamine the null hypothesis that 120583

1= 1205832 where 120583

1and 120583

2

represent themean values of one Haar-like feature in samplesfrom the case and control groups respectively To speed upthe calculation we divided these data into four almost equalparts and ran tests on a personal computer (Lenovo M8000tQuad Core Q6600 CPU 8GB RAM)The statistical softwareused was R 2152 [28]

Given recent suspicion of the discriminatory value of 119875 lt005 [29] and the small deviation in mean values betweenfeatures associated and not associated with hyperuricemia incomparison with their standard deviations we investigateddata dispersion using the following formula

119889119891 =210038161003816100381610038161205831 minus 12058321003816100381610038161003816

1205901minus 1205902

(1)

where 1205831and 120590

1are the mean value and standard deviation

respectively of a feature associated with hyperuricemia1205832and 120590

2are the corresponding values for a feature not

associated with hyperuricemia

We selected 50 features with smallest 119875 and largest 119889119891values to serve as single classifiers in this study We thentested the ability of these features to correctly classify caseand control subjects Receiver operating characteristic (ROC)analysis was performed and areas under the ROC curve(AUCs) were calculated For features with the smallest 119875values largest 119889119891 values and largest areas we considered thata classifier would perform best when its Youden index wasmaximizedWe also determined the sensitivity and specificityof these classifiers

3 Results

31 Participant Characteristics Blood samples from 13321437 eligible participants were analyzed to determine SUAconcentrationThemean age of this populationwas 438plusmn119(range 19ndash87) years and the prevalence of hyperuricemiawas 901 (1201332 130 in men and 315 in women)Application of the exclusion criteria left a sample of 83subjects with and 211 without hyperuricemia The final case-control-matched sample comprised 46 subjects (36 men and10 women) per group Mean age did not differ significantlybetween the case and control groups (462 plusmn 131 versus452 plusmn 131 years 119875 = 0763) but SUA concentration did(324 plusmn 818 versus 182 plusmn 327 120583molL 119875 = 418 times 10minus18)

32 Haar-Like Features Useful for Hyperuricemia DiagnosisInitial screening of features using the criterion 119875 lt 005yielded a sample of 239035 features Selection based on colorplains reduced the number of features to several fractions ofthe original yielding 97124 features on the red plain 26228features on the green plain and 115683 features on the blueplain that were useful for the diagnosis of hyperuricemia (all119875 lt 005) The largest features on the red green and bluecolor plains all of whichwere in the centers of tongue imageswere feature (11 21 40 85) (119875 = 172 times 10minus2 Figure 4(a))feature (12 11 35 85) (119875 = 311 times 10minus2 Figure 5(a)) andfeature (11 21 40 85) (119875 = 119 times 10minus3 Figure 6(a))

BioMed Research International 5

(a) (b) (c)

Figure 5 Three features on the green color plain examined in this study Panel (a) contains 5950 pixels (b) contains 300 pixels and (c)contains 200 pixels

(a) (b) (c)

Figure 6 Three features on the blue color plain examined in this study Panel (a) contains 6800 pixels (b) contains 1200 pixels and (c)contains 1200 pixels

respectively (Table 1)The smallest119875 values were obtained forfeature (30 23 10 30) (119875 = 209 times 10minus7 Figure 4(b)) feature(33 24 10 15) (119875 = 116 times 10minus4 Figure 5(b)) and feature(29 11 20 30) (119875 = 485 times 10minus11 Figure 6(b)) on the redgreen and blue color plains respectively (Table 1) Featureswith the largest 119889119891 values in the red green and blue plainswere feature (30 23 10 30) (119889119891 = 119 Figure 4(b)) feature(38 24 10 10) (119889119891 = 848 times 10minus1 Figure 5(c)) and feature(30 11 20 30) (119889119891 = 158 Figure 6(c)) respectively (Table 1)Feature (30 23 10 30) in the red plain had both the smallest119875 value and largest 119889119891 value

33 Single Classifier Performance The ROC curves of thetwo features in the red plain are shown in Figure 7 The

Table 1 Parameters of eight Haar-like features potentially useful forhyperuricemia diagnosis

Color plain 119875 119889119891 119883 119884 119882 119867

Red 172119864 minus 02 513119864 minus 01 11 21 40 85Red 209119864 minus 07 119 30 23 10 30Green 311119864 minus 02 459119864 minus 02 12 11 35 85Green 116119864 minus 04 852119864 minus 01 33 24 10 15Green 107119864 minus 04 848119864 minus 01 38 24 10 10Blue 119119864 minus 03 719119864 minus 01 11 21 40 85Blue 485119864 minus 11 158 29 11 20 30Blue 450119864 minus 11 158 30 11 20 30119883 and 119884 describe the position of the top-left corner of the feature 119867represents the height of a feature119882 and 119881 represent the widths of the leftand right rectangles respectively

6 BioMed Research International

False positive rate

True

pos

itive

rate

00 02 04 06 08 10

00

02

04

06

08

10

AUC 0654 AUC 081

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

Feature (11 21 40 85) Feature (30 23 10 30)

Figure 7 Receiver operating characteristic curves of features in the red plain

AUC 0612 AUC 0726 AUC 072

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

Feature (12 11 35 85) Feature (38 24 10 10) Feature (33 24 10 15)

Figure 8 Receiver operating characteristic curves of features in the green plain

AUC of feature (11 21 40 85) was 0654 preventing properanalysis of sensitivity and specificity The AUC of feature(30 23 10 30) was 0810 maximization of the Youden indexyielded sensitivity and specificity values of 0844 and 0652respectively Features in the green and blue color plainsshowed similar characteristics In the green plain the AUCvalue of feature (12 11 35 85) was only 0612 and thoseof feature (38 24 10 10) and feature (33 24 10 15) were0726 and 0720 respectively (Figure 8) Maximization of theYouden index yielded sensitivity values of 0500 and 0844and specificity values of 0391 and 0933 for the latter twofeatures respectively These two features were also deemed

to be inapplicable because of their low sensitivity values Inthe blue plain the AUC of feature (11 21 40 85) was 0704and those of feature (30 11 20 30) and feature (29 11 20 30)were 0877 and 0875 respectively (Figure 9) Sensitivity andspecificity values for feature (29 11 20 30) were 0800 and0804 respectively when the Youden index was maximizedThis feature achieved the best performance (sensitivity 0800specificity 0826) when the maximum value of the Youdenindex was 0626

34 Cumulative Feature Cumulative features in the redgreen and blue color plains (each composed of 50 features)

BioMed Research International 7

AUC 0704 AUC 0877 AUC 0875

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10Feature (11 21 40 85) Feature (30 11 20 30) Feature (29 11 20 30)

Figure 9 Receiver operating characteristic curves of features in the blue plain

0

0

20

20

40

40

60

60

80

80

100

100

120

P df

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 10 Cumulative image of 50 Haar-like features on the red plain Note the cumulative featurersquos centralization around the tongue root

are shown in Figures 10 11 and 12 respectively All ofthese features were centralized around the tongue rootvalidating our hypothesis The red cumulative feature has acircular distribution the green cumulative feature is moreconcentrated than the red feature and the blue cumulativefeature shows vertical symmetry

4 Discussion

Feature extraction is among the most important issues inimage processing These feature classes are based on perfect

segmentation of a tongue image from the background whichis difficult for the human eye [14ndash16] The extraction ofHaar-like features does not require segmentation [20] greatlysimplifying image preprocessing In this study we examined arectangular area including the tongue rather than attemptingto performmore precise segmentation as in previous studies

The use of Haar-like feature extraction from imagesis superior to extraction based solely on color because itallows the identification of local characteristics [19 30 31]Other studies have focused on color differences of the entiretongue [17 18] using global features such as means and

8 BioMed Research International

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixelsy

-pix

els

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 11 Cumulative image of 50 Haar-like features in the green plain

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 12 Cumulative image of 50 Haar-like features in the blue plain

standard deviations of color value These features prohibitdetailed medical interpretation because they do not considerdifferences among parts of the tongue In a previous study theexamination of tongue portions resulted in the identificationof some features that were located outside of the tongue [16]In our study we scanned the entire tongue image and foundthat all meaningful features (those with the smallest 119875 valuesand largest 119889119891 values) were located within the tongue area Incontrast to those of previous studies our results indicate that

tongue image preprocessing does not require perfect tonguesegmentation

Tongue image preprocessing using Haar-like features is anew method that not only resolves the segmentation issuebut also provides a novel means of interpreting tongueimages A face detection study using Haar-like featuresprovided the intuitive explanation that the most decisivefeatures include the eyes and nose [23] In our study wefound that the most decisive features for the diagnosis of

BioMed Research International 9

hyperuricemia are centralized on the tongue root A previousstudy described the results of tongue image analysis for thediagnosis of metastatic cancer [19] but applicable quantita-tive image analysis was not available at the time the studywas performed and the study also lacked a control groupAnother study focused on elderly subjects [20] In our studywe calculated quantitative feature values (119875 and119889119891) to expressdifferences between subjects with andwithout hyperuricemiausing a case-control design and including subjects from allage groups

The features identified in this study can be interpretedwithin the framework of TCM because they are based onpixels All features that performed well in this study werecentralized around the tongue root the area consideredto reflect kidney disease in TCM This study provideddirect evidence of the relationship between changes in thetongue root and kidney disease The kidney filters bloodand excretes metabolic waste products including uric acidIn the human body 70 of urate is disposed of via thekidneys [32] Hyperuricemia is not only related to severaldiseases but is also a risk factor for kidney injury [33ndash35] The diagnosis of hyperuricemia thus provides earlywarning of kidney injury However the determination ofserum urea nitrogen creatinine carbon dioxide and uricacid concentrations requires time-consuming biochemicalexamination An intuitive inexpensive and noninvasivemethod of hyperuricemia would thus be of benefit TCMprovides examination tools fulfilling these requirements Ourstudy provided direct evidence supporting the TCMmethodof diagnosing hyperuricemia based on tongue features

However this study has several limitations First theROCanalysiswas performedusing the same sample In futurestudies a test dataset will be collected to confirm the findingsof this study Second given that the use of tongue imagesis a complementary and alternative diagnostic method themethod described in this study should be combined withother available variables associated with hyperuricemia suchas body mass index and alcohol intake We plan to takethis approach in a further study Third because the samplewas carefully selected and patients with underlying diseasesassociated with hyperuricemia were excluded the use oftongue images for the diagnosis of hyperuricemia should berestricted

5 Conclusions

Haar-like features extracted from tongue images differed sig-nificantly between subjects with and without hyperuricemiaThe locations of these features are consistent with interpre-tations of tongue appearance in TCM andWestern medicineindicating the existence of a relationship between tongue rootcolor and hyperuricemia in our sample

Conflict of Interests

The authors declare that they have no conflict of interestsrelated to this work

Authorsrsquo Contribution

Yan Cui completed the statistical analysis and wrote thepaper Shizhong Liao performed critical revision of the paperand HongwuWang performed the mathematical models andmethods of this studyHongyuLiuWenhuaWang andLiqunYin validated health information for all subjects

Acknowledgment

This work was supported in part by the National BasicResearch Program of China (973 Program Grant no2011CB505406)

References

[1] H K Choi K Atkinson E W Karlson W Willett and GCurhan ldquoPurine-rich foods daily and protein intake and therisk of gout in menrdquoThe New England Journal of Medicine vol350 no 11 pp 1093ndash1103 2004

[2] K Kaneko Y Aoyagi T Fukuuchi K Inazawa andNYamaokaldquoTotal purine and purine base content of common foodstuffsfor facilitating nutritional therapy for gout and hyperuricemiardquoBiological and Pharmaceutical Bulletin vol 37 no 5 pp 709ndash721 2014

[3] J T Scott ldquoDrug-induced goutrdquo Baillierersquos Clinical Rheumatol-ogy vol 5 no 1 pp 39ndash60 1991

[4] A Brandstater S Kiechl B Kollerits et al ldquoSex-specificassociation of the putative fructose transporter SLC2A9 variantswith uric acid levels is modified by BMIrdquo Diabetes Care vol 31no 8 pp 1662ndash1667 2008

[5] L-H Lai S-Y Chou F-Y Wu J J-H Chen and H-W KuoldquoRenal dysfunction and hyperuricemia with low blood leadlevels and ethnicity in community-based studyrdquo Science of theTotal Environment vol 401 no 1-3 pp 39ndash43 2008

[6] Q Liu Z Li H Wang et al ldquoHigh prevalence and associatedrisk factors for impaired renal function and urinary abnormal-ities in a rural adult population from Southern Chinardquo PLoSONE vol 7 no 10 Article ID e47100 2012

[7] W-G Fang X-M Huang Y Wang et al ldquoA cross-sectionalstudy of hyperuricemia in state-employees in Beijing preva-lence and risk factorsrdquo National Medical Journal of China vol86 no 25 pp 1764ndash1768 2006

[8] Y S Kim J P Guevara K M Kim H K Choi D F Heitjanand D A Albert ldquoHyperuricemia and coronary heart diseasea systematic review and meta-analysisrdquo Arthritis Care andResearch vol 62 no 2 pp 170ndash180 2010

[9] P C Grayson S Y Kim M LaValley and H K ChoildquoHyperuricemia and incident hypertension a systematic reviewand meta-analysisrdquo Arthritis Care and Research vol 63 no 1pp 102ndash110 2011

[10] M Li W Hou X Zhang L Hu and Z Tang ldquoHyperuricemiaand risk of stroke a systematic review and meta-analysis ofprospective studiesrdquoAtherosclerosis vol 232 no 2 pp 265ndash2702014

[11] M R Carnethon S P Fortmann L Palaniappan B B Dun-can M I Schmidt and L E Chambless ldquoRisk factors forprogression to incident hyperinsulinemia the atherosclerosisrisk in communities study 1987ndash1998rdquo American Journal ofEpidemiology vol 158 no 11 pp 1058ndash1067 2003

10 BioMed Research International

[12] B Liu T Wang H Zhao et al ldquoThe prevalence of hyper-uricemia in china a meta-analysisrdquo BMC Public Health vol 11no 1 article 832 2011

[13] E Krishnan ldquoInteraction of inflammation hyperuricemia andthe prevalence of hypertension among adults free of metabolicsyndromeNHANES 2009ndash2010rdquo Journal of the AmericanHeartAssociation vol 3 no 2 Article ID e000157 2014

[14] B Zhang XWang J You andD Zhang ldquoTongue color analysisfor medical applicationrdquo Evidence-Based Complementary andAlternativeMedicine vol 2013 Article ID 264742 11 pages 2013

[15] B Pang D Zhang and K Wang ldquoTongue image analysis forappendicitis diagnosisrdquo Information Sciences vol 175 no 3 pp160ndash176 2005

[16] J Kim J Son S Jang et al ldquoAvailability of tongue diagno-sis system for assessing tongue coating thickness in patientswith functional dyspepsiardquoEvidence-BasedComplementary andAlternativeMedicine vol 2013 Article ID 348272 6 pages 2013

[17] XWang B Zhang Z Yang HWang andD Zhang ldquoStatisticalanalysis of tongue images for feature extraction and diagnos-ticsrdquo IEEE Transactions on Image Processing vol 22 no 12 pp5336ndash5347 2013

[18] C J Jung J H Nam Y J Jeon and K H Kim ldquoColor distribu-tion differences in the tongue in sleep disorderrdquo Evidence-BasedComplementary and Alternative Medicine vol 2014 Article ID323645 8 pages 2014

[19] A H Friedlander and R Singer ldquoRenal adenocarcinoma ofthe kidney with metastasis to the tonguerdquo The Journal of theAmerican Dental Association vol 97 no 6 pp 989ndash991 1978

[20] K A Amir R K Bobba B Clarke et al ldquoTongue discolorationin an elderly kidney transplant recipient treatment-relatedadverse eventrdquoThe American Journal Geriatric Pharmacother-apy vol 4 no 3 pp 260ndash263 2006

[21] Q Tang J Chen and J Li ldquoAnalysis of the tongue characteristicsof liver depression and kidney insufficiency pid patientsrdquoChinaModern Medicine vol 15 pp 82ndash83 2010

[22] M Yang C-D Li W-N Liang and H Li ldquoCorrelationalstudy on pathological changes of liver qi depression and kidneydeficiency tongue coating exfoliative cells and sex hormonesin perimenopausal syndromerdquo China Journal of TraditionalChinese Medicine and Pharmacy vol 9 pp 1984ndash1986 2011

[23] P Viola and M J Jones ldquoRobust real-time face detectionrdquoInternational Journal of Computer Vision vol 57 no 2 pp 137ndash154 2004

[24] F Valbusa L Bertolini S Bonapace et al ldquoRelation of elevatedserum uric acid levels to incidence of atrial fibrillation inpatients with type 2 diabetes mellitusrdquoThe American Journal ofCardiology vol 112 no 4 pp 499ndash504 2013

[25] D A Grimes and K F Schulz ldquoAn overview of clinical researchthe lay of the landrdquo The Lancet vol 359 no 9300 pp 57ndash612002

[26] D A Grimes and K F Schulz ldquoCompared to what Findingcontrols for case-control studiesrdquoTheLancet vol 365 no 9468pp 1429ndash1433 2005

[27] M M Gorkani and R W Picard ldquo Texture orientation forsorting photos lsquoat a glancersquordquo inProceedings of the 12th IAPR Inter-national Conference on Pattern Recognition vol 1mdashConferenceA Computer Vision amp Image Processing pp 459ndash464 IEEEOctober 1994

[28] R Core Team R A Language and Environment for StatisticalComputing 2012

[29] R Nuzzo ldquoStatistical errorsrdquoNature vol 506 no 7487 pp 150ndash152 2014

[30] R Kanawong T Obafemi-Ajayi J Yu D Xu S Li and YDuan ldquoZheng classification in traditional chinese medicinebased on modified specular-free tongue imagesrdquo in Proceed-ings of the IEEE International Conference on Bioinformaticsand Biomedicine Workshops (BIBMW rsquo12) pp 288ndash294 IEEEOctober 2012

[31] R Kanawong T Obafemi-Ajayi T Ma D Xu S Li andY Duan ldquoAutomated tongue feature extraction for ZHENGclassification in traditional Chinese medicinerdquo Evidence-BasedComplementary and Alternative Medicine vol 2012 Article ID912852 14 pages 2012

[32] V Vitart I Rudan C Hayward et al ldquoSLC2A9 is a newly iden-tified urate transporter influencing serum urate concentrationurate excretion and goutrdquo Nature Genetics vol 40 no 4 pp437ndash442 2008

[33] M Tomita SMizunoH Yamanaka et al ldquoDoes hyperuricemiaaffect mortality A prospective cohort study of Japanese maleworkersrdquo Journal of epidemiologyJapan Epidemiological Associ-ation vol 10 no 6 pp 403ndash409 2000

[34] J Fang andMH Alderman ldquoSerumuric acid and cardiovascu-lar mortality the nhanes i epidemiologic follow-up study 1971ndash1992rdquo Journal of the American Medical Association vol 283 no18 pp 2404ndash2410 2000

[35] O Toprak M Cirit E Esi N Postaci M Yesil and S BayataldquoHyperuricemia as a risk factor for contrast-induced nephropa-thy in patientswith chronic kidney diseaserdquoCatheterization andCardiovascular Interventions vol 67 no 2 pp 227ndash235 2006

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

2 BioMed Research International

(a) (b) (c)

Figure 1 Image processing procedure (a) Original image acquired by the tongue analyzer (b) Rectangular area containing the tonguesegmented from the original image (c) Final image scaled to 120 times 100 pixels

laborious and risk patient injury The development of arapid simple noninvasive method would thus improve thediagnostic procedure for hyperuricemia

Tongue images have been applied as inexpensive andnoninvasive means of diagnosing several diseases such asstroke and appendicitis [14ndash16] Wang et al [17] statisticallyanalyzed features extracted from tongue images defining 12image classes Other statistical methods such as Bayesiannetworks and a bagging tree algorithm have been applied totongue image analysis [15 16] However these studies didnot employ case-control designs that would have avoidedbias introduced by age and sex differences in tongue fea-tures Jung et al [18] performed a case-control study toexamine differences in color distribution on tongue imagesbetween subjects with and without sleep disorders but thediagnostic criteria used in this study were based on thephysicianrsquos judgment rather than biochemical examinationWestern medical studies have found that tongue appearance(coloration and coating) is related to kidney diseases orconditions such as renal adenocarcinoma tongue metastasisand kidney transplantation [19 20] Traditional Chinesemedicine (TCM) studies have also found that tongue imagecharacteristics can reflect renal deficiency [21 22]

The present case-control study was performed to identifytongue image features useful for the diagnosis of hyper-uricemia A series of symmetrical Haar-like features whichhave been applied successfully to face detection [23] wereextracted from tongue images from subjects with andwithouthyperuricemia (diagnoses were confirmed biochemically)We sought to identify independently useful and readilyinterpretable Haar-like features for the diagnosis of hyper-uricemia

2 Subjects and Methods

21 Subjects and Examination BetweenAugust 2011 and June2012 outpatients from Wuqing Chinese Medicine Hospitala medical examination center of teaching hospital affiliatedwith the Tianjin University of Traditional Chinese Medicine

(TJUTCM) participated in this study All participants pro-vided informed consent and this study was approved by themedical ethics committee of TJUTCM Adults from all agegroups were included to avoid bias introduced by uneven agedistribution Based on data from medical records accessedthrough the hospitalrsquos health information system subjectswith diseases impacting the appearance of the tongue suchas hypertension diabetes and cancer were excluded Thosewith dyed and scraped tongue fur as determined by outpa-tient interviews were also excluded

The diagnostic criteria for hyperuricemia in malesand females were SUA gt416 120583molL (7mgdL) and SUAgt357 120583molL (6mgdL) respectively [24] An image of eachsubjectrsquos tongue was acquired using a YM-III tongue imageanalyzer (httptjtzymcomtonguehtml)

Case and control subjects were matched 1 1 by age(within 1 year) and sex to exclude the impacts of thesecovariates and improve the value of the empirical data [2526] Two-tailed t-tests for samples with equal and unequalvariance were used to confirm similarity in age and differencein SUA value respectively between case and control subjects

22 Image Processing and Feature Selection Tongue imageswere processed as shown in Figure 1 The original imageacquired by the tongue analyzer which depicts the subjectrsquosentire face (Figure 1(a)) was first segmented to include onlythe rectangular area depicting the tongue (Figure 1(b)) Eachimage was then scaled to 120 times 100 pixels (Figure 1(c))to enable efficient feature extraction while retaining colorinformation

Several feature types can be used in image analysisFeatures based on statistical analysis of color represent globaldifferences (expressed as means and standard deviations)among images and are the most intuitive feature type [15]but they cannot describe differences among areas in a singleimage The use of pixel analysis to define image features hasa high computational cost and does not provide high-levelinformation about the images [23] Moreover the number ofpixels is much greater than the number of images in mostsituations and adjacent pixels are often closely correlated

BioMed Research International 3

(a) (b)

Figure 2 Upper-lower and left-right area arrangements The sum of pixels in the lower (a) or right (b) rectangle is subtracted from the sumof pixels in the upper or left rectangle respectively

One Haar-like featureH

W

Top-left corner of tongue image

X

Y

V

Figure 3 Haar-like feature parameters119883 and119884 describe the position of the top-left corner of the feature119867 represents the height of a feature119882 and 119881 represent the widths of the left and right rectangles respectively

these characteristics complicate statistical analysis For thisreason we used Haar-like features [23] which fall betweenthe pixel and global levels in the present study These fea-tures enable examination of color differences between areaspartially solving the problem of correlations among pixelsHowever the number of such features is large exceeding160000 in a 24 times 24-pixel image and the computational costof Haar-like feature extraction remains large [23]

We first sought to reduce the number of Haar-likefeatures in the tongue images which exceeded our computingcapabilities Considering that observation of the tongue isbased on color we first selected features in the red greenand blue color plains ignoring plains in other color spaces(ie Lab) We then employed directional selection whichinvolved the delineation of two adjacent rectangles on eachimage Figure 2 shows two approaches to such selection thesumof pixels in the lower or right rectanglemay be subtracted

from that in the upper or left rectangle respectively Giventhat the human body is characterized predominantly bybilateral symmetry we subtracted the sum of pixels in theright from that in the left rectangle to select Haar-likefeatures Finally we applied scale selection based on the fiveparameters of the left-right feature (Figure 3) Given thatfacial pixels near image boundaries were meaningless forthis study we determined the values of 119884 and 119883 (whichdescribe the position of the top left corner of a feature)as 11 12 110 and 11 12 90 respectively Giventhat overly narrow or short features provide inadequateinformation we determined the values of 119882 (width of leftrectangle) and 119867 (feature height) as 10 15 lfloor(100 minus119883)2rfloor and 10 15 100 minus 119884 respectively Consideringthe number of pixels contained in the two rectangles we set119882 = 119881 (width of right rectangle) Integral image scanning[20] was applied to reduce the computational cost of feature

4 BioMed Research International

(a) (b)

Figure 4 Three features on the red color plain examined in this study (a) contains 6800 pixels (b) contains 600 pixels

extraction This technique requires a single image scan andvalues of all features can be computed within several secondsBecause feature positions and scales are determined by119883 119884119882 and 119867 features are identified by these values using theformat ldquofeature (119883119884119882119867)rdquo in this text Each color plaincontained 195840 features (total = 587520 features in threeplains)

23 Statistical Analysis At this stage of processing thenumber of selected Haar-like features far exceeds the numberof subjects and correlation among features remains strongdue to overlap preventing direct application in classificatorymodels Gorkani and Picard [27] found that human eyesdistinguish images using high-level textural features In thisstudy we thus assumed that the diagnosis of hyperuricemiawould be based on instantaneous extraction of a feature froma tongue image in a single glance We also assumed that allglances would be independent We used Studentrsquos 119905-tests toexamine the null hypothesis that 120583

1= 1205832 where 120583

1and 120583

2

represent themean values of one Haar-like feature in samplesfrom the case and control groups respectively To speed upthe calculation we divided these data into four almost equalparts and ran tests on a personal computer (Lenovo M8000tQuad Core Q6600 CPU 8GB RAM)The statistical softwareused was R 2152 [28]

Given recent suspicion of the discriminatory value of 119875 lt005 [29] and the small deviation in mean values betweenfeatures associated and not associated with hyperuricemia incomparison with their standard deviations we investigateddata dispersion using the following formula

119889119891 =210038161003816100381610038161205831 minus 12058321003816100381610038161003816

1205901minus 1205902

(1)

where 1205831and 120590

1are the mean value and standard deviation

respectively of a feature associated with hyperuricemia1205832and 120590

2are the corresponding values for a feature not

associated with hyperuricemia

We selected 50 features with smallest 119875 and largest 119889119891values to serve as single classifiers in this study We thentested the ability of these features to correctly classify caseand control subjects Receiver operating characteristic (ROC)analysis was performed and areas under the ROC curve(AUCs) were calculated For features with the smallest 119875values largest 119889119891 values and largest areas we considered thata classifier would perform best when its Youden index wasmaximizedWe also determined the sensitivity and specificityof these classifiers

3 Results

31 Participant Characteristics Blood samples from 13321437 eligible participants were analyzed to determine SUAconcentrationThemean age of this populationwas 438plusmn119(range 19ndash87) years and the prevalence of hyperuricemiawas 901 (1201332 130 in men and 315 in women)Application of the exclusion criteria left a sample of 83subjects with and 211 without hyperuricemia The final case-control-matched sample comprised 46 subjects (36 men and10 women) per group Mean age did not differ significantlybetween the case and control groups (462 plusmn 131 versus452 plusmn 131 years 119875 = 0763) but SUA concentration did(324 plusmn 818 versus 182 plusmn 327 120583molL 119875 = 418 times 10minus18)

32 Haar-Like Features Useful for Hyperuricemia DiagnosisInitial screening of features using the criterion 119875 lt 005yielded a sample of 239035 features Selection based on colorplains reduced the number of features to several fractions ofthe original yielding 97124 features on the red plain 26228features on the green plain and 115683 features on the blueplain that were useful for the diagnosis of hyperuricemia (all119875 lt 005) The largest features on the red green and bluecolor plains all of whichwere in the centers of tongue imageswere feature (11 21 40 85) (119875 = 172 times 10minus2 Figure 4(a))feature (12 11 35 85) (119875 = 311 times 10minus2 Figure 5(a)) andfeature (11 21 40 85) (119875 = 119 times 10minus3 Figure 6(a))

BioMed Research International 5

(a) (b) (c)

Figure 5 Three features on the green color plain examined in this study Panel (a) contains 5950 pixels (b) contains 300 pixels and (c)contains 200 pixels

(a) (b) (c)

Figure 6 Three features on the blue color plain examined in this study Panel (a) contains 6800 pixels (b) contains 1200 pixels and (c)contains 1200 pixels

respectively (Table 1)The smallest119875 values were obtained forfeature (30 23 10 30) (119875 = 209 times 10minus7 Figure 4(b)) feature(33 24 10 15) (119875 = 116 times 10minus4 Figure 5(b)) and feature(29 11 20 30) (119875 = 485 times 10minus11 Figure 6(b)) on the redgreen and blue color plains respectively (Table 1) Featureswith the largest 119889119891 values in the red green and blue plainswere feature (30 23 10 30) (119889119891 = 119 Figure 4(b)) feature(38 24 10 10) (119889119891 = 848 times 10minus1 Figure 5(c)) and feature(30 11 20 30) (119889119891 = 158 Figure 6(c)) respectively (Table 1)Feature (30 23 10 30) in the red plain had both the smallest119875 value and largest 119889119891 value

33 Single Classifier Performance The ROC curves of thetwo features in the red plain are shown in Figure 7 The

Table 1 Parameters of eight Haar-like features potentially useful forhyperuricemia diagnosis

Color plain 119875 119889119891 119883 119884 119882 119867

Red 172119864 minus 02 513119864 minus 01 11 21 40 85Red 209119864 minus 07 119 30 23 10 30Green 311119864 minus 02 459119864 minus 02 12 11 35 85Green 116119864 minus 04 852119864 minus 01 33 24 10 15Green 107119864 minus 04 848119864 minus 01 38 24 10 10Blue 119119864 minus 03 719119864 minus 01 11 21 40 85Blue 485119864 minus 11 158 29 11 20 30Blue 450119864 minus 11 158 30 11 20 30119883 and 119884 describe the position of the top-left corner of the feature 119867represents the height of a feature119882 and 119881 represent the widths of the leftand right rectangles respectively

6 BioMed Research International

False positive rate

True

pos

itive

rate

00 02 04 06 08 10

00

02

04

06

08

10

AUC 0654 AUC 081

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

Feature (11 21 40 85) Feature (30 23 10 30)

Figure 7 Receiver operating characteristic curves of features in the red plain

AUC 0612 AUC 0726 AUC 072

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

Feature (12 11 35 85) Feature (38 24 10 10) Feature (33 24 10 15)

Figure 8 Receiver operating characteristic curves of features in the green plain

AUC of feature (11 21 40 85) was 0654 preventing properanalysis of sensitivity and specificity The AUC of feature(30 23 10 30) was 0810 maximization of the Youden indexyielded sensitivity and specificity values of 0844 and 0652respectively Features in the green and blue color plainsshowed similar characteristics In the green plain the AUCvalue of feature (12 11 35 85) was only 0612 and thoseof feature (38 24 10 10) and feature (33 24 10 15) were0726 and 0720 respectively (Figure 8) Maximization of theYouden index yielded sensitivity values of 0500 and 0844and specificity values of 0391 and 0933 for the latter twofeatures respectively These two features were also deemed

to be inapplicable because of their low sensitivity values Inthe blue plain the AUC of feature (11 21 40 85) was 0704and those of feature (30 11 20 30) and feature (29 11 20 30)were 0877 and 0875 respectively (Figure 9) Sensitivity andspecificity values for feature (29 11 20 30) were 0800 and0804 respectively when the Youden index was maximizedThis feature achieved the best performance (sensitivity 0800specificity 0826) when the maximum value of the Youdenindex was 0626

34 Cumulative Feature Cumulative features in the redgreen and blue color plains (each composed of 50 features)

BioMed Research International 7

AUC 0704 AUC 0877 AUC 0875

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10Feature (11 21 40 85) Feature (30 11 20 30) Feature (29 11 20 30)

Figure 9 Receiver operating characteristic curves of features in the blue plain

0

0

20

20

40

40

60

60

80

80

100

100

120

P df

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 10 Cumulative image of 50 Haar-like features on the red plain Note the cumulative featurersquos centralization around the tongue root

are shown in Figures 10 11 and 12 respectively All ofthese features were centralized around the tongue rootvalidating our hypothesis The red cumulative feature has acircular distribution the green cumulative feature is moreconcentrated than the red feature and the blue cumulativefeature shows vertical symmetry

4 Discussion

Feature extraction is among the most important issues inimage processing These feature classes are based on perfect

segmentation of a tongue image from the background whichis difficult for the human eye [14ndash16] The extraction ofHaar-like features does not require segmentation [20] greatlysimplifying image preprocessing In this study we examined arectangular area including the tongue rather than attemptingto performmore precise segmentation as in previous studies

The use of Haar-like feature extraction from imagesis superior to extraction based solely on color because itallows the identification of local characteristics [19 30 31]Other studies have focused on color differences of the entiretongue [17 18] using global features such as means and

8 BioMed Research International

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixelsy

-pix

els

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 11 Cumulative image of 50 Haar-like features in the green plain

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 12 Cumulative image of 50 Haar-like features in the blue plain

standard deviations of color value These features prohibitdetailed medical interpretation because they do not considerdifferences among parts of the tongue In a previous study theexamination of tongue portions resulted in the identificationof some features that were located outside of the tongue [16]In our study we scanned the entire tongue image and foundthat all meaningful features (those with the smallest 119875 valuesand largest 119889119891 values) were located within the tongue area Incontrast to those of previous studies our results indicate that

tongue image preprocessing does not require perfect tonguesegmentation

Tongue image preprocessing using Haar-like features is anew method that not only resolves the segmentation issuebut also provides a novel means of interpreting tongueimages A face detection study using Haar-like featuresprovided the intuitive explanation that the most decisivefeatures include the eyes and nose [23] In our study wefound that the most decisive features for the diagnosis of

BioMed Research International 9

hyperuricemia are centralized on the tongue root A previousstudy described the results of tongue image analysis for thediagnosis of metastatic cancer [19] but applicable quantita-tive image analysis was not available at the time the studywas performed and the study also lacked a control groupAnother study focused on elderly subjects [20] In our studywe calculated quantitative feature values (119875 and119889119891) to expressdifferences between subjects with andwithout hyperuricemiausing a case-control design and including subjects from allage groups

The features identified in this study can be interpretedwithin the framework of TCM because they are based onpixels All features that performed well in this study werecentralized around the tongue root the area consideredto reflect kidney disease in TCM This study provideddirect evidence of the relationship between changes in thetongue root and kidney disease The kidney filters bloodand excretes metabolic waste products including uric acidIn the human body 70 of urate is disposed of via thekidneys [32] Hyperuricemia is not only related to severaldiseases but is also a risk factor for kidney injury [33ndash35] The diagnosis of hyperuricemia thus provides earlywarning of kidney injury However the determination ofserum urea nitrogen creatinine carbon dioxide and uricacid concentrations requires time-consuming biochemicalexamination An intuitive inexpensive and noninvasivemethod of hyperuricemia would thus be of benefit TCMprovides examination tools fulfilling these requirements Ourstudy provided direct evidence supporting the TCMmethodof diagnosing hyperuricemia based on tongue features

However this study has several limitations First theROCanalysiswas performedusing the same sample In futurestudies a test dataset will be collected to confirm the findingsof this study Second given that the use of tongue imagesis a complementary and alternative diagnostic method themethod described in this study should be combined withother available variables associated with hyperuricemia suchas body mass index and alcohol intake We plan to takethis approach in a further study Third because the samplewas carefully selected and patients with underlying diseasesassociated with hyperuricemia were excluded the use oftongue images for the diagnosis of hyperuricemia should berestricted

5 Conclusions

Haar-like features extracted from tongue images differed sig-nificantly between subjects with and without hyperuricemiaThe locations of these features are consistent with interpre-tations of tongue appearance in TCM andWestern medicineindicating the existence of a relationship between tongue rootcolor and hyperuricemia in our sample

Conflict of Interests

The authors declare that they have no conflict of interestsrelated to this work

Authorsrsquo Contribution

Yan Cui completed the statistical analysis and wrote thepaper Shizhong Liao performed critical revision of the paperand HongwuWang performed the mathematical models andmethods of this studyHongyuLiuWenhuaWang andLiqunYin validated health information for all subjects

Acknowledgment

This work was supported in part by the National BasicResearch Program of China (973 Program Grant no2011CB505406)

References

[1] H K Choi K Atkinson E W Karlson W Willett and GCurhan ldquoPurine-rich foods daily and protein intake and therisk of gout in menrdquoThe New England Journal of Medicine vol350 no 11 pp 1093ndash1103 2004

[2] K Kaneko Y Aoyagi T Fukuuchi K Inazawa andNYamaokaldquoTotal purine and purine base content of common foodstuffsfor facilitating nutritional therapy for gout and hyperuricemiardquoBiological and Pharmaceutical Bulletin vol 37 no 5 pp 709ndash721 2014

[3] J T Scott ldquoDrug-induced goutrdquo Baillierersquos Clinical Rheumatol-ogy vol 5 no 1 pp 39ndash60 1991

[4] A Brandstater S Kiechl B Kollerits et al ldquoSex-specificassociation of the putative fructose transporter SLC2A9 variantswith uric acid levels is modified by BMIrdquo Diabetes Care vol 31no 8 pp 1662ndash1667 2008

[5] L-H Lai S-Y Chou F-Y Wu J J-H Chen and H-W KuoldquoRenal dysfunction and hyperuricemia with low blood leadlevels and ethnicity in community-based studyrdquo Science of theTotal Environment vol 401 no 1-3 pp 39ndash43 2008

[6] Q Liu Z Li H Wang et al ldquoHigh prevalence and associatedrisk factors for impaired renal function and urinary abnormal-ities in a rural adult population from Southern Chinardquo PLoSONE vol 7 no 10 Article ID e47100 2012

[7] W-G Fang X-M Huang Y Wang et al ldquoA cross-sectionalstudy of hyperuricemia in state-employees in Beijing preva-lence and risk factorsrdquo National Medical Journal of China vol86 no 25 pp 1764ndash1768 2006

[8] Y S Kim J P Guevara K M Kim H K Choi D F Heitjanand D A Albert ldquoHyperuricemia and coronary heart diseasea systematic review and meta-analysisrdquo Arthritis Care andResearch vol 62 no 2 pp 170ndash180 2010

[9] P C Grayson S Y Kim M LaValley and H K ChoildquoHyperuricemia and incident hypertension a systematic reviewand meta-analysisrdquo Arthritis Care and Research vol 63 no 1pp 102ndash110 2011

[10] M Li W Hou X Zhang L Hu and Z Tang ldquoHyperuricemiaand risk of stroke a systematic review and meta-analysis ofprospective studiesrdquoAtherosclerosis vol 232 no 2 pp 265ndash2702014

[11] M R Carnethon S P Fortmann L Palaniappan B B Dun-can M I Schmidt and L E Chambless ldquoRisk factors forprogression to incident hyperinsulinemia the atherosclerosisrisk in communities study 1987ndash1998rdquo American Journal ofEpidemiology vol 158 no 11 pp 1058ndash1067 2003

10 BioMed Research International

[12] B Liu T Wang H Zhao et al ldquoThe prevalence of hyper-uricemia in china a meta-analysisrdquo BMC Public Health vol 11no 1 article 832 2011

[13] E Krishnan ldquoInteraction of inflammation hyperuricemia andthe prevalence of hypertension among adults free of metabolicsyndromeNHANES 2009ndash2010rdquo Journal of the AmericanHeartAssociation vol 3 no 2 Article ID e000157 2014

[14] B Zhang XWang J You andD Zhang ldquoTongue color analysisfor medical applicationrdquo Evidence-Based Complementary andAlternativeMedicine vol 2013 Article ID 264742 11 pages 2013

[15] B Pang D Zhang and K Wang ldquoTongue image analysis forappendicitis diagnosisrdquo Information Sciences vol 175 no 3 pp160ndash176 2005

[16] J Kim J Son S Jang et al ldquoAvailability of tongue diagno-sis system for assessing tongue coating thickness in patientswith functional dyspepsiardquoEvidence-BasedComplementary andAlternativeMedicine vol 2013 Article ID 348272 6 pages 2013

[17] XWang B Zhang Z Yang HWang andD Zhang ldquoStatisticalanalysis of tongue images for feature extraction and diagnos-ticsrdquo IEEE Transactions on Image Processing vol 22 no 12 pp5336ndash5347 2013

[18] C J Jung J H Nam Y J Jeon and K H Kim ldquoColor distribu-tion differences in the tongue in sleep disorderrdquo Evidence-BasedComplementary and Alternative Medicine vol 2014 Article ID323645 8 pages 2014

[19] A H Friedlander and R Singer ldquoRenal adenocarcinoma ofthe kidney with metastasis to the tonguerdquo The Journal of theAmerican Dental Association vol 97 no 6 pp 989ndash991 1978

[20] K A Amir R K Bobba B Clarke et al ldquoTongue discolorationin an elderly kidney transplant recipient treatment-relatedadverse eventrdquoThe American Journal Geriatric Pharmacother-apy vol 4 no 3 pp 260ndash263 2006

[21] Q Tang J Chen and J Li ldquoAnalysis of the tongue characteristicsof liver depression and kidney insufficiency pid patientsrdquoChinaModern Medicine vol 15 pp 82ndash83 2010

[22] M Yang C-D Li W-N Liang and H Li ldquoCorrelationalstudy on pathological changes of liver qi depression and kidneydeficiency tongue coating exfoliative cells and sex hormonesin perimenopausal syndromerdquo China Journal of TraditionalChinese Medicine and Pharmacy vol 9 pp 1984ndash1986 2011

[23] P Viola and M J Jones ldquoRobust real-time face detectionrdquoInternational Journal of Computer Vision vol 57 no 2 pp 137ndash154 2004

[24] F Valbusa L Bertolini S Bonapace et al ldquoRelation of elevatedserum uric acid levels to incidence of atrial fibrillation inpatients with type 2 diabetes mellitusrdquoThe American Journal ofCardiology vol 112 no 4 pp 499ndash504 2013

[25] D A Grimes and K F Schulz ldquoAn overview of clinical researchthe lay of the landrdquo The Lancet vol 359 no 9300 pp 57ndash612002

[26] D A Grimes and K F Schulz ldquoCompared to what Findingcontrols for case-control studiesrdquoTheLancet vol 365 no 9468pp 1429ndash1433 2005

[27] M M Gorkani and R W Picard ldquo Texture orientation forsorting photos lsquoat a glancersquordquo inProceedings of the 12th IAPR Inter-national Conference on Pattern Recognition vol 1mdashConferenceA Computer Vision amp Image Processing pp 459ndash464 IEEEOctober 1994

[28] R Core Team R A Language and Environment for StatisticalComputing 2012

[29] R Nuzzo ldquoStatistical errorsrdquoNature vol 506 no 7487 pp 150ndash152 2014

[30] R Kanawong T Obafemi-Ajayi J Yu D Xu S Li and YDuan ldquoZheng classification in traditional chinese medicinebased on modified specular-free tongue imagesrdquo in Proceed-ings of the IEEE International Conference on Bioinformaticsand Biomedicine Workshops (BIBMW rsquo12) pp 288ndash294 IEEEOctober 2012

[31] R Kanawong T Obafemi-Ajayi T Ma D Xu S Li andY Duan ldquoAutomated tongue feature extraction for ZHENGclassification in traditional Chinese medicinerdquo Evidence-BasedComplementary and Alternative Medicine vol 2012 Article ID912852 14 pages 2012

[32] V Vitart I Rudan C Hayward et al ldquoSLC2A9 is a newly iden-tified urate transporter influencing serum urate concentrationurate excretion and goutrdquo Nature Genetics vol 40 no 4 pp437ndash442 2008

[33] M Tomita SMizunoH Yamanaka et al ldquoDoes hyperuricemiaaffect mortality A prospective cohort study of Japanese maleworkersrdquo Journal of epidemiologyJapan Epidemiological Associ-ation vol 10 no 6 pp 403ndash409 2000

[34] J Fang andMH Alderman ldquoSerumuric acid and cardiovascu-lar mortality the nhanes i epidemiologic follow-up study 1971ndash1992rdquo Journal of the American Medical Association vol 283 no18 pp 2404ndash2410 2000

[35] O Toprak M Cirit E Esi N Postaci M Yesil and S BayataldquoHyperuricemia as a risk factor for contrast-induced nephropa-thy in patientswith chronic kidney diseaserdquoCatheterization andCardiovascular Interventions vol 67 no 2 pp 227ndash235 2006

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

BioMed Research International 3

(a) (b)

Figure 2 Upper-lower and left-right area arrangements The sum of pixels in the lower (a) or right (b) rectangle is subtracted from the sumof pixels in the upper or left rectangle respectively

One Haar-like featureH

W

Top-left corner of tongue image

X

Y

V

Figure 3 Haar-like feature parameters119883 and119884 describe the position of the top-left corner of the feature119867 represents the height of a feature119882 and 119881 represent the widths of the left and right rectangles respectively

these characteristics complicate statistical analysis For thisreason we used Haar-like features [23] which fall betweenthe pixel and global levels in the present study These fea-tures enable examination of color differences between areaspartially solving the problem of correlations among pixelsHowever the number of such features is large exceeding160000 in a 24 times 24-pixel image and the computational costof Haar-like feature extraction remains large [23]

We first sought to reduce the number of Haar-likefeatures in the tongue images which exceeded our computingcapabilities Considering that observation of the tongue isbased on color we first selected features in the red greenand blue color plains ignoring plains in other color spaces(ie Lab) We then employed directional selection whichinvolved the delineation of two adjacent rectangles on eachimage Figure 2 shows two approaches to such selection thesumof pixels in the lower or right rectanglemay be subtracted

from that in the upper or left rectangle respectively Giventhat the human body is characterized predominantly bybilateral symmetry we subtracted the sum of pixels in theright from that in the left rectangle to select Haar-likefeatures Finally we applied scale selection based on the fiveparameters of the left-right feature (Figure 3) Given thatfacial pixels near image boundaries were meaningless forthis study we determined the values of 119884 and 119883 (whichdescribe the position of the top left corner of a feature)as 11 12 110 and 11 12 90 respectively Giventhat overly narrow or short features provide inadequateinformation we determined the values of 119882 (width of leftrectangle) and 119867 (feature height) as 10 15 lfloor(100 minus119883)2rfloor and 10 15 100 minus 119884 respectively Consideringthe number of pixels contained in the two rectangles we set119882 = 119881 (width of right rectangle) Integral image scanning[20] was applied to reduce the computational cost of feature

4 BioMed Research International

(a) (b)

Figure 4 Three features on the red color plain examined in this study (a) contains 6800 pixels (b) contains 600 pixels

extraction This technique requires a single image scan andvalues of all features can be computed within several secondsBecause feature positions and scales are determined by119883 119884119882 and 119867 features are identified by these values using theformat ldquofeature (119883119884119882119867)rdquo in this text Each color plaincontained 195840 features (total = 587520 features in threeplains)

23 Statistical Analysis At this stage of processing thenumber of selected Haar-like features far exceeds the numberof subjects and correlation among features remains strongdue to overlap preventing direct application in classificatorymodels Gorkani and Picard [27] found that human eyesdistinguish images using high-level textural features In thisstudy we thus assumed that the diagnosis of hyperuricemiawould be based on instantaneous extraction of a feature froma tongue image in a single glance We also assumed that allglances would be independent We used Studentrsquos 119905-tests toexamine the null hypothesis that 120583

1= 1205832 where 120583

1and 120583

2

represent themean values of one Haar-like feature in samplesfrom the case and control groups respectively To speed upthe calculation we divided these data into four almost equalparts and ran tests on a personal computer (Lenovo M8000tQuad Core Q6600 CPU 8GB RAM)The statistical softwareused was R 2152 [28]

Given recent suspicion of the discriminatory value of 119875 lt005 [29] and the small deviation in mean values betweenfeatures associated and not associated with hyperuricemia incomparison with their standard deviations we investigateddata dispersion using the following formula

119889119891 =210038161003816100381610038161205831 minus 12058321003816100381610038161003816

1205901minus 1205902

(1)

where 1205831and 120590

1are the mean value and standard deviation

respectively of a feature associated with hyperuricemia1205832and 120590

2are the corresponding values for a feature not

associated with hyperuricemia

We selected 50 features with smallest 119875 and largest 119889119891values to serve as single classifiers in this study We thentested the ability of these features to correctly classify caseand control subjects Receiver operating characteristic (ROC)analysis was performed and areas under the ROC curve(AUCs) were calculated For features with the smallest 119875values largest 119889119891 values and largest areas we considered thata classifier would perform best when its Youden index wasmaximizedWe also determined the sensitivity and specificityof these classifiers

3 Results

31 Participant Characteristics Blood samples from 13321437 eligible participants were analyzed to determine SUAconcentrationThemean age of this populationwas 438plusmn119(range 19ndash87) years and the prevalence of hyperuricemiawas 901 (1201332 130 in men and 315 in women)Application of the exclusion criteria left a sample of 83subjects with and 211 without hyperuricemia The final case-control-matched sample comprised 46 subjects (36 men and10 women) per group Mean age did not differ significantlybetween the case and control groups (462 plusmn 131 versus452 plusmn 131 years 119875 = 0763) but SUA concentration did(324 plusmn 818 versus 182 plusmn 327 120583molL 119875 = 418 times 10minus18)

32 Haar-Like Features Useful for Hyperuricemia DiagnosisInitial screening of features using the criterion 119875 lt 005yielded a sample of 239035 features Selection based on colorplains reduced the number of features to several fractions ofthe original yielding 97124 features on the red plain 26228features on the green plain and 115683 features on the blueplain that were useful for the diagnosis of hyperuricemia (all119875 lt 005) The largest features on the red green and bluecolor plains all of whichwere in the centers of tongue imageswere feature (11 21 40 85) (119875 = 172 times 10minus2 Figure 4(a))feature (12 11 35 85) (119875 = 311 times 10minus2 Figure 5(a)) andfeature (11 21 40 85) (119875 = 119 times 10minus3 Figure 6(a))

BioMed Research International 5

(a) (b) (c)

Figure 5 Three features on the green color plain examined in this study Panel (a) contains 5950 pixels (b) contains 300 pixels and (c)contains 200 pixels

(a) (b) (c)

Figure 6 Three features on the blue color plain examined in this study Panel (a) contains 6800 pixels (b) contains 1200 pixels and (c)contains 1200 pixels

respectively (Table 1)The smallest119875 values were obtained forfeature (30 23 10 30) (119875 = 209 times 10minus7 Figure 4(b)) feature(33 24 10 15) (119875 = 116 times 10minus4 Figure 5(b)) and feature(29 11 20 30) (119875 = 485 times 10minus11 Figure 6(b)) on the redgreen and blue color plains respectively (Table 1) Featureswith the largest 119889119891 values in the red green and blue plainswere feature (30 23 10 30) (119889119891 = 119 Figure 4(b)) feature(38 24 10 10) (119889119891 = 848 times 10minus1 Figure 5(c)) and feature(30 11 20 30) (119889119891 = 158 Figure 6(c)) respectively (Table 1)Feature (30 23 10 30) in the red plain had both the smallest119875 value and largest 119889119891 value

33 Single Classifier Performance The ROC curves of thetwo features in the red plain are shown in Figure 7 The

Table 1 Parameters of eight Haar-like features potentially useful forhyperuricemia diagnosis

Color plain 119875 119889119891 119883 119884 119882 119867

Red 172119864 minus 02 513119864 minus 01 11 21 40 85Red 209119864 minus 07 119 30 23 10 30Green 311119864 minus 02 459119864 minus 02 12 11 35 85Green 116119864 minus 04 852119864 minus 01 33 24 10 15Green 107119864 minus 04 848119864 minus 01 38 24 10 10Blue 119119864 minus 03 719119864 minus 01 11 21 40 85Blue 485119864 minus 11 158 29 11 20 30Blue 450119864 minus 11 158 30 11 20 30119883 and 119884 describe the position of the top-left corner of the feature 119867represents the height of a feature119882 and 119881 represent the widths of the leftand right rectangles respectively

6 BioMed Research International

False positive rate

True

pos

itive

rate

00 02 04 06 08 10

00

02

04

06

08

10

AUC 0654 AUC 081

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

Feature (11 21 40 85) Feature (30 23 10 30)

Figure 7 Receiver operating characteristic curves of features in the red plain

AUC 0612 AUC 0726 AUC 072

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

Feature (12 11 35 85) Feature (38 24 10 10) Feature (33 24 10 15)

Figure 8 Receiver operating characteristic curves of features in the green plain

AUC of feature (11 21 40 85) was 0654 preventing properanalysis of sensitivity and specificity The AUC of feature(30 23 10 30) was 0810 maximization of the Youden indexyielded sensitivity and specificity values of 0844 and 0652respectively Features in the green and blue color plainsshowed similar characteristics In the green plain the AUCvalue of feature (12 11 35 85) was only 0612 and thoseof feature (38 24 10 10) and feature (33 24 10 15) were0726 and 0720 respectively (Figure 8) Maximization of theYouden index yielded sensitivity values of 0500 and 0844and specificity values of 0391 and 0933 for the latter twofeatures respectively These two features were also deemed

to be inapplicable because of their low sensitivity values Inthe blue plain the AUC of feature (11 21 40 85) was 0704and those of feature (30 11 20 30) and feature (29 11 20 30)were 0877 and 0875 respectively (Figure 9) Sensitivity andspecificity values for feature (29 11 20 30) were 0800 and0804 respectively when the Youden index was maximizedThis feature achieved the best performance (sensitivity 0800specificity 0826) when the maximum value of the Youdenindex was 0626

34 Cumulative Feature Cumulative features in the redgreen and blue color plains (each composed of 50 features)

BioMed Research International 7

AUC 0704 AUC 0877 AUC 0875

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10Feature (11 21 40 85) Feature (30 11 20 30) Feature (29 11 20 30)

Figure 9 Receiver operating characteristic curves of features in the blue plain

0

0

20

20

40

40

60

60

80

80

100

100

120

P df

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 10 Cumulative image of 50 Haar-like features on the red plain Note the cumulative featurersquos centralization around the tongue root

are shown in Figures 10 11 and 12 respectively All ofthese features were centralized around the tongue rootvalidating our hypothesis The red cumulative feature has acircular distribution the green cumulative feature is moreconcentrated than the red feature and the blue cumulativefeature shows vertical symmetry

4 Discussion

Feature extraction is among the most important issues inimage processing These feature classes are based on perfect

segmentation of a tongue image from the background whichis difficult for the human eye [14ndash16] The extraction ofHaar-like features does not require segmentation [20] greatlysimplifying image preprocessing In this study we examined arectangular area including the tongue rather than attemptingto performmore precise segmentation as in previous studies

The use of Haar-like feature extraction from imagesis superior to extraction based solely on color because itallows the identification of local characteristics [19 30 31]Other studies have focused on color differences of the entiretongue [17 18] using global features such as means and

8 BioMed Research International

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixelsy

-pix

els

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 11 Cumulative image of 50 Haar-like features in the green plain

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 12 Cumulative image of 50 Haar-like features in the blue plain

standard deviations of color value These features prohibitdetailed medical interpretation because they do not considerdifferences among parts of the tongue In a previous study theexamination of tongue portions resulted in the identificationof some features that were located outside of the tongue [16]In our study we scanned the entire tongue image and foundthat all meaningful features (those with the smallest 119875 valuesand largest 119889119891 values) were located within the tongue area Incontrast to those of previous studies our results indicate that

tongue image preprocessing does not require perfect tonguesegmentation

Tongue image preprocessing using Haar-like features is anew method that not only resolves the segmentation issuebut also provides a novel means of interpreting tongueimages A face detection study using Haar-like featuresprovided the intuitive explanation that the most decisivefeatures include the eyes and nose [23] In our study wefound that the most decisive features for the diagnosis of

BioMed Research International 9

hyperuricemia are centralized on the tongue root A previousstudy described the results of tongue image analysis for thediagnosis of metastatic cancer [19] but applicable quantita-tive image analysis was not available at the time the studywas performed and the study also lacked a control groupAnother study focused on elderly subjects [20] In our studywe calculated quantitative feature values (119875 and119889119891) to expressdifferences between subjects with andwithout hyperuricemiausing a case-control design and including subjects from allage groups

The features identified in this study can be interpretedwithin the framework of TCM because they are based onpixels All features that performed well in this study werecentralized around the tongue root the area consideredto reflect kidney disease in TCM This study provideddirect evidence of the relationship between changes in thetongue root and kidney disease The kidney filters bloodand excretes metabolic waste products including uric acidIn the human body 70 of urate is disposed of via thekidneys [32] Hyperuricemia is not only related to severaldiseases but is also a risk factor for kidney injury [33ndash35] The diagnosis of hyperuricemia thus provides earlywarning of kidney injury However the determination ofserum urea nitrogen creatinine carbon dioxide and uricacid concentrations requires time-consuming biochemicalexamination An intuitive inexpensive and noninvasivemethod of hyperuricemia would thus be of benefit TCMprovides examination tools fulfilling these requirements Ourstudy provided direct evidence supporting the TCMmethodof diagnosing hyperuricemia based on tongue features

However this study has several limitations First theROCanalysiswas performedusing the same sample In futurestudies a test dataset will be collected to confirm the findingsof this study Second given that the use of tongue imagesis a complementary and alternative diagnostic method themethod described in this study should be combined withother available variables associated with hyperuricemia suchas body mass index and alcohol intake We plan to takethis approach in a further study Third because the samplewas carefully selected and patients with underlying diseasesassociated with hyperuricemia were excluded the use oftongue images for the diagnosis of hyperuricemia should berestricted

5 Conclusions

Haar-like features extracted from tongue images differed sig-nificantly between subjects with and without hyperuricemiaThe locations of these features are consistent with interpre-tations of tongue appearance in TCM andWestern medicineindicating the existence of a relationship between tongue rootcolor and hyperuricemia in our sample

Conflict of Interests

The authors declare that they have no conflict of interestsrelated to this work

Authorsrsquo Contribution

Yan Cui completed the statistical analysis and wrote thepaper Shizhong Liao performed critical revision of the paperand HongwuWang performed the mathematical models andmethods of this studyHongyuLiuWenhuaWang andLiqunYin validated health information for all subjects

Acknowledgment

This work was supported in part by the National BasicResearch Program of China (973 Program Grant no2011CB505406)

References

[1] H K Choi K Atkinson E W Karlson W Willett and GCurhan ldquoPurine-rich foods daily and protein intake and therisk of gout in menrdquoThe New England Journal of Medicine vol350 no 11 pp 1093ndash1103 2004

[2] K Kaneko Y Aoyagi T Fukuuchi K Inazawa andNYamaokaldquoTotal purine and purine base content of common foodstuffsfor facilitating nutritional therapy for gout and hyperuricemiardquoBiological and Pharmaceutical Bulletin vol 37 no 5 pp 709ndash721 2014

[3] J T Scott ldquoDrug-induced goutrdquo Baillierersquos Clinical Rheumatol-ogy vol 5 no 1 pp 39ndash60 1991

[4] A Brandstater S Kiechl B Kollerits et al ldquoSex-specificassociation of the putative fructose transporter SLC2A9 variantswith uric acid levels is modified by BMIrdquo Diabetes Care vol 31no 8 pp 1662ndash1667 2008

[5] L-H Lai S-Y Chou F-Y Wu J J-H Chen and H-W KuoldquoRenal dysfunction and hyperuricemia with low blood leadlevels and ethnicity in community-based studyrdquo Science of theTotal Environment vol 401 no 1-3 pp 39ndash43 2008

[6] Q Liu Z Li H Wang et al ldquoHigh prevalence and associatedrisk factors for impaired renal function and urinary abnormal-ities in a rural adult population from Southern Chinardquo PLoSONE vol 7 no 10 Article ID e47100 2012

[7] W-G Fang X-M Huang Y Wang et al ldquoA cross-sectionalstudy of hyperuricemia in state-employees in Beijing preva-lence and risk factorsrdquo National Medical Journal of China vol86 no 25 pp 1764ndash1768 2006

[8] Y S Kim J P Guevara K M Kim H K Choi D F Heitjanand D A Albert ldquoHyperuricemia and coronary heart diseasea systematic review and meta-analysisrdquo Arthritis Care andResearch vol 62 no 2 pp 170ndash180 2010

[9] P C Grayson S Y Kim M LaValley and H K ChoildquoHyperuricemia and incident hypertension a systematic reviewand meta-analysisrdquo Arthritis Care and Research vol 63 no 1pp 102ndash110 2011

[10] M Li W Hou X Zhang L Hu and Z Tang ldquoHyperuricemiaand risk of stroke a systematic review and meta-analysis ofprospective studiesrdquoAtherosclerosis vol 232 no 2 pp 265ndash2702014

[11] M R Carnethon S P Fortmann L Palaniappan B B Dun-can M I Schmidt and L E Chambless ldquoRisk factors forprogression to incident hyperinsulinemia the atherosclerosisrisk in communities study 1987ndash1998rdquo American Journal ofEpidemiology vol 158 no 11 pp 1058ndash1067 2003

10 BioMed Research International

[12] B Liu T Wang H Zhao et al ldquoThe prevalence of hyper-uricemia in china a meta-analysisrdquo BMC Public Health vol 11no 1 article 832 2011

[13] E Krishnan ldquoInteraction of inflammation hyperuricemia andthe prevalence of hypertension among adults free of metabolicsyndromeNHANES 2009ndash2010rdquo Journal of the AmericanHeartAssociation vol 3 no 2 Article ID e000157 2014

[14] B Zhang XWang J You andD Zhang ldquoTongue color analysisfor medical applicationrdquo Evidence-Based Complementary andAlternativeMedicine vol 2013 Article ID 264742 11 pages 2013

[15] B Pang D Zhang and K Wang ldquoTongue image analysis forappendicitis diagnosisrdquo Information Sciences vol 175 no 3 pp160ndash176 2005

[16] J Kim J Son S Jang et al ldquoAvailability of tongue diagno-sis system for assessing tongue coating thickness in patientswith functional dyspepsiardquoEvidence-BasedComplementary andAlternativeMedicine vol 2013 Article ID 348272 6 pages 2013

[17] XWang B Zhang Z Yang HWang andD Zhang ldquoStatisticalanalysis of tongue images for feature extraction and diagnos-ticsrdquo IEEE Transactions on Image Processing vol 22 no 12 pp5336ndash5347 2013

[18] C J Jung J H Nam Y J Jeon and K H Kim ldquoColor distribu-tion differences in the tongue in sleep disorderrdquo Evidence-BasedComplementary and Alternative Medicine vol 2014 Article ID323645 8 pages 2014

[19] A H Friedlander and R Singer ldquoRenal adenocarcinoma ofthe kidney with metastasis to the tonguerdquo The Journal of theAmerican Dental Association vol 97 no 6 pp 989ndash991 1978

[20] K A Amir R K Bobba B Clarke et al ldquoTongue discolorationin an elderly kidney transplant recipient treatment-relatedadverse eventrdquoThe American Journal Geriatric Pharmacother-apy vol 4 no 3 pp 260ndash263 2006

[21] Q Tang J Chen and J Li ldquoAnalysis of the tongue characteristicsof liver depression and kidney insufficiency pid patientsrdquoChinaModern Medicine vol 15 pp 82ndash83 2010

[22] M Yang C-D Li W-N Liang and H Li ldquoCorrelationalstudy on pathological changes of liver qi depression and kidneydeficiency tongue coating exfoliative cells and sex hormonesin perimenopausal syndromerdquo China Journal of TraditionalChinese Medicine and Pharmacy vol 9 pp 1984ndash1986 2011

[23] P Viola and M J Jones ldquoRobust real-time face detectionrdquoInternational Journal of Computer Vision vol 57 no 2 pp 137ndash154 2004

[24] F Valbusa L Bertolini S Bonapace et al ldquoRelation of elevatedserum uric acid levels to incidence of atrial fibrillation inpatients with type 2 diabetes mellitusrdquoThe American Journal ofCardiology vol 112 no 4 pp 499ndash504 2013

[25] D A Grimes and K F Schulz ldquoAn overview of clinical researchthe lay of the landrdquo The Lancet vol 359 no 9300 pp 57ndash612002

[26] D A Grimes and K F Schulz ldquoCompared to what Findingcontrols for case-control studiesrdquoTheLancet vol 365 no 9468pp 1429ndash1433 2005

[27] M M Gorkani and R W Picard ldquo Texture orientation forsorting photos lsquoat a glancersquordquo inProceedings of the 12th IAPR Inter-national Conference on Pattern Recognition vol 1mdashConferenceA Computer Vision amp Image Processing pp 459ndash464 IEEEOctober 1994

[28] R Core Team R A Language and Environment for StatisticalComputing 2012

[29] R Nuzzo ldquoStatistical errorsrdquoNature vol 506 no 7487 pp 150ndash152 2014

[30] R Kanawong T Obafemi-Ajayi J Yu D Xu S Li and YDuan ldquoZheng classification in traditional chinese medicinebased on modified specular-free tongue imagesrdquo in Proceed-ings of the IEEE International Conference on Bioinformaticsand Biomedicine Workshops (BIBMW rsquo12) pp 288ndash294 IEEEOctober 2012

[31] R Kanawong T Obafemi-Ajayi T Ma D Xu S Li andY Duan ldquoAutomated tongue feature extraction for ZHENGclassification in traditional Chinese medicinerdquo Evidence-BasedComplementary and Alternative Medicine vol 2012 Article ID912852 14 pages 2012

[32] V Vitart I Rudan C Hayward et al ldquoSLC2A9 is a newly iden-tified urate transporter influencing serum urate concentrationurate excretion and goutrdquo Nature Genetics vol 40 no 4 pp437ndash442 2008

[33] M Tomita SMizunoH Yamanaka et al ldquoDoes hyperuricemiaaffect mortality A prospective cohort study of Japanese maleworkersrdquo Journal of epidemiologyJapan Epidemiological Associ-ation vol 10 no 6 pp 403ndash409 2000

[34] J Fang andMH Alderman ldquoSerumuric acid and cardiovascu-lar mortality the nhanes i epidemiologic follow-up study 1971ndash1992rdquo Journal of the American Medical Association vol 283 no18 pp 2404ndash2410 2000

[35] O Toprak M Cirit E Esi N Postaci M Yesil and S BayataldquoHyperuricemia as a risk factor for contrast-induced nephropa-thy in patientswith chronic kidney diseaserdquoCatheterization andCardiovascular Interventions vol 67 no 2 pp 227ndash235 2006

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

4 BioMed Research International

(a) (b)

Figure 4 Three features on the red color plain examined in this study (a) contains 6800 pixels (b) contains 600 pixels

extraction This technique requires a single image scan andvalues of all features can be computed within several secondsBecause feature positions and scales are determined by119883 119884119882 and 119867 features are identified by these values using theformat ldquofeature (119883119884119882119867)rdquo in this text Each color plaincontained 195840 features (total = 587520 features in threeplains)

23 Statistical Analysis At this stage of processing thenumber of selected Haar-like features far exceeds the numberof subjects and correlation among features remains strongdue to overlap preventing direct application in classificatorymodels Gorkani and Picard [27] found that human eyesdistinguish images using high-level textural features In thisstudy we thus assumed that the diagnosis of hyperuricemiawould be based on instantaneous extraction of a feature froma tongue image in a single glance We also assumed that allglances would be independent We used Studentrsquos 119905-tests toexamine the null hypothesis that 120583

1= 1205832 where 120583

1and 120583

2

represent themean values of one Haar-like feature in samplesfrom the case and control groups respectively To speed upthe calculation we divided these data into four almost equalparts and ran tests on a personal computer (Lenovo M8000tQuad Core Q6600 CPU 8GB RAM)The statistical softwareused was R 2152 [28]

Given recent suspicion of the discriminatory value of 119875 lt005 [29] and the small deviation in mean values betweenfeatures associated and not associated with hyperuricemia incomparison with their standard deviations we investigateddata dispersion using the following formula

119889119891 =210038161003816100381610038161205831 minus 12058321003816100381610038161003816

1205901minus 1205902

(1)

where 1205831and 120590

1are the mean value and standard deviation

respectively of a feature associated with hyperuricemia1205832and 120590

2are the corresponding values for a feature not

associated with hyperuricemia

We selected 50 features with smallest 119875 and largest 119889119891values to serve as single classifiers in this study We thentested the ability of these features to correctly classify caseand control subjects Receiver operating characteristic (ROC)analysis was performed and areas under the ROC curve(AUCs) were calculated For features with the smallest 119875values largest 119889119891 values and largest areas we considered thata classifier would perform best when its Youden index wasmaximizedWe also determined the sensitivity and specificityof these classifiers

3 Results

31 Participant Characteristics Blood samples from 13321437 eligible participants were analyzed to determine SUAconcentrationThemean age of this populationwas 438plusmn119(range 19ndash87) years and the prevalence of hyperuricemiawas 901 (1201332 130 in men and 315 in women)Application of the exclusion criteria left a sample of 83subjects with and 211 without hyperuricemia The final case-control-matched sample comprised 46 subjects (36 men and10 women) per group Mean age did not differ significantlybetween the case and control groups (462 plusmn 131 versus452 plusmn 131 years 119875 = 0763) but SUA concentration did(324 plusmn 818 versus 182 plusmn 327 120583molL 119875 = 418 times 10minus18)

32 Haar-Like Features Useful for Hyperuricemia DiagnosisInitial screening of features using the criterion 119875 lt 005yielded a sample of 239035 features Selection based on colorplains reduced the number of features to several fractions ofthe original yielding 97124 features on the red plain 26228features on the green plain and 115683 features on the blueplain that were useful for the diagnosis of hyperuricemia (all119875 lt 005) The largest features on the red green and bluecolor plains all of whichwere in the centers of tongue imageswere feature (11 21 40 85) (119875 = 172 times 10minus2 Figure 4(a))feature (12 11 35 85) (119875 = 311 times 10minus2 Figure 5(a)) andfeature (11 21 40 85) (119875 = 119 times 10minus3 Figure 6(a))

BioMed Research International 5

(a) (b) (c)

Figure 5 Three features on the green color plain examined in this study Panel (a) contains 5950 pixels (b) contains 300 pixels and (c)contains 200 pixels

(a) (b) (c)

Figure 6 Three features on the blue color plain examined in this study Panel (a) contains 6800 pixels (b) contains 1200 pixels and (c)contains 1200 pixels

respectively (Table 1)The smallest119875 values were obtained forfeature (30 23 10 30) (119875 = 209 times 10minus7 Figure 4(b)) feature(33 24 10 15) (119875 = 116 times 10minus4 Figure 5(b)) and feature(29 11 20 30) (119875 = 485 times 10minus11 Figure 6(b)) on the redgreen and blue color plains respectively (Table 1) Featureswith the largest 119889119891 values in the red green and blue plainswere feature (30 23 10 30) (119889119891 = 119 Figure 4(b)) feature(38 24 10 10) (119889119891 = 848 times 10minus1 Figure 5(c)) and feature(30 11 20 30) (119889119891 = 158 Figure 6(c)) respectively (Table 1)Feature (30 23 10 30) in the red plain had both the smallest119875 value and largest 119889119891 value

33 Single Classifier Performance The ROC curves of thetwo features in the red plain are shown in Figure 7 The

Table 1 Parameters of eight Haar-like features potentially useful forhyperuricemia diagnosis

Color plain 119875 119889119891 119883 119884 119882 119867

Red 172119864 minus 02 513119864 minus 01 11 21 40 85Red 209119864 minus 07 119 30 23 10 30Green 311119864 minus 02 459119864 minus 02 12 11 35 85Green 116119864 minus 04 852119864 minus 01 33 24 10 15Green 107119864 minus 04 848119864 minus 01 38 24 10 10Blue 119119864 minus 03 719119864 minus 01 11 21 40 85Blue 485119864 minus 11 158 29 11 20 30Blue 450119864 minus 11 158 30 11 20 30119883 and 119884 describe the position of the top-left corner of the feature 119867represents the height of a feature119882 and 119881 represent the widths of the leftand right rectangles respectively

6 BioMed Research International

False positive rate

True

pos

itive

rate

00 02 04 06 08 10

00

02

04

06

08

10

AUC 0654 AUC 081

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

Feature (11 21 40 85) Feature (30 23 10 30)

Figure 7 Receiver operating characteristic curves of features in the red plain

AUC 0612 AUC 0726 AUC 072

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

Feature (12 11 35 85) Feature (38 24 10 10) Feature (33 24 10 15)

Figure 8 Receiver operating characteristic curves of features in the green plain

AUC of feature (11 21 40 85) was 0654 preventing properanalysis of sensitivity and specificity The AUC of feature(30 23 10 30) was 0810 maximization of the Youden indexyielded sensitivity and specificity values of 0844 and 0652respectively Features in the green and blue color plainsshowed similar characteristics In the green plain the AUCvalue of feature (12 11 35 85) was only 0612 and thoseof feature (38 24 10 10) and feature (33 24 10 15) were0726 and 0720 respectively (Figure 8) Maximization of theYouden index yielded sensitivity values of 0500 and 0844and specificity values of 0391 and 0933 for the latter twofeatures respectively These two features were also deemed

to be inapplicable because of their low sensitivity values Inthe blue plain the AUC of feature (11 21 40 85) was 0704and those of feature (30 11 20 30) and feature (29 11 20 30)were 0877 and 0875 respectively (Figure 9) Sensitivity andspecificity values for feature (29 11 20 30) were 0800 and0804 respectively when the Youden index was maximizedThis feature achieved the best performance (sensitivity 0800specificity 0826) when the maximum value of the Youdenindex was 0626

34 Cumulative Feature Cumulative features in the redgreen and blue color plains (each composed of 50 features)

BioMed Research International 7

AUC 0704 AUC 0877 AUC 0875

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10Feature (11 21 40 85) Feature (30 11 20 30) Feature (29 11 20 30)

Figure 9 Receiver operating characteristic curves of features in the blue plain

0

0

20

20

40

40

60

60

80

80

100

100

120

P df

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 10 Cumulative image of 50 Haar-like features on the red plain Note the cumulative featurersquos centralization around the tongue root

are shown in Figures 10 11 and 12 respectively All ofthese features were centralized around the tongue rootvalidating our hypothesis The red cumulative feature has acircular distribution the green cumulative feature is moreconcentrated than the red feature and the blue cumulativefeature shows vertical symmetry

4 Discussion

Feature extraction is among the most important issues inimage processing These feature classes are based on perfect

segmentation of a tongue image from the background whichis difficult for the human eye [14ndash16] The extraction ofHaar-like features does not require segmentation [20] greatlysimplifying image preprocessing In this study we examined arectangular area including the tongue rather than attemptingto performmore precise segmentation as in previous studies

The use of Haar-like feature extraction from imagesis superior to extraction based solely on color because itallows the identification of local characteristics [19 30 31]Other studies have focused on color differences of the entiretongue [17 18] using global features such as means and

8 BioMed Research International

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixelsy

-pix

els

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 11 Cumulative image of 50 Haar-like features in the green plain

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 12 Cumulative image of 50 Haar-like features in the blue plain

standard deviations of color value These features prohibitdetailed medical interpretation because they do not considerdifferences among parts of the tongue In a previous study theexamination of tongue portions resulted in the identificationof some features that were located outside of the tongue [16]In our study we scanned the entire tongue image and foundthat all meaningful features (those with the smallest 119875 valuesand largest 119889119891 values) were located within the tongue area Incontrast to those of previous studies our results indicate that

tongue image preprocessing does not require perfect tonguesegmentation

Tongue image preprocessing using Haar-like features is anew method that not only resolves the segmentation issuebut also provides a novel means of interpreting tongueimages A face detection study using Haar-like featuresprovided the intuitive explanation that the most decisivefeatures include the eyes and nose [23] In our study wefound that the most decisive features for the diagnosis of

BioMed Research International 9

hyperuricemia are centralized on the tongue root A previousstudy described the results of tongue image analysis for thediagnosis of metastatic cancer [19] but applicable quantita-tive image analysis was not available at the time the studywas performed and the study also lacked a control groupAnother study focused on elderly subjects [20] In our studywe calculated quantitative feature values (119875 and119889119891) to expressdifferences between subjects with andwithout hyperuricemiausing a case-control design and including subjects from allage groups

The features identified in this study can be interpretedwithin the framework of TCM because they are based onpixels All features that performed well in this study werecentralized around the tongue root the area consideredto reflect kidney disease in TCM This study provideddirect evidence of the relationship between changes in thetongue root and kidney disease The kidney filters bloodand excretes metabolic waste products including uric acidIn the human body 70 of urate is disposed of via thekidneys [32] Hyperuricemia is not only related to severaldiseases but is also a risk factor for kidney injury [33ndash35] The diagnosis of hyperuricemia thus provides earlywarning of kidney injury However the determination ofserum urea nitrogen creatinine carbon dioxide and uricacid concentrations requires time-consuming biochemicalexamination An intuitive inexpensive and noninvasivemethod of hyperuricemia would thus be of benefit TCMprovides examination tools fulfilling these requirements Ourstudy provided direct evidence supporting the TCMmethodof diagnosing hyperuricemia based on tongue features

However this study has several limitations First theROCanalysiswas performedusing the same sample In futurestudies a test dataset will be collected to confirm the findingsof this study Second given that the use of tongue imagesis a complementary and alternative diagnostic method themethod described in this study should be combined withother available variables associated with hyperuricemia suchas body mass index and alcohol intake We plan to takethis approach in a further study Third because the samplewas carefully selected and patients with underlying diseasesassociated with hyperuricemia were excluded the use oftongue images for the diagnosis of hyperuricemia should berestricted

5 Conclusions

Haar-like features extracted from tongue images differed sig-nificantly between subjects with and without hyperuricemiaThe locations of these features are consistent with interpre-tations of tongue appearance in TCM andWestern medicineindicating the existence of a relationship between tongue rootcolor and hyperuricemia in our sample

Conflict of Interests

The authors declare that they have no conflict of interestsrelated to this work

Authorsrsquo Contribution

Yan Cui completed the statistical analysis and wrote thepaper Shizhong Liao performed critical revision of the paperand HongwuWang performed the mathematical models andmethods of this studyHongyuLiuWenhuaWang andLiqunYin validated health information for all subjects

Acknowledgment

This work was supported in part by the National BasicResearch Program of China (973 Program Grant no2011CB505406)

References

[1] H K Choi K Atkinson E W Karlson W Willett and GCurhan ldquoPurine-rich foods daily and protein intake and therisk of gout in menrdquoThe New England Journal of Medicine vol350 no 11 pp 1093ndash1103 2004

[2] K Kaneko Y Aoyagi T Fukuuchi K Inazawa andNYamaokaldquoTotal purine and purine base content of common foodstuffsfor facilitating nutritional therapy for gout and hyperuricemiardquoBiological and Pharmaceutical Bulletin vol 37 no 5 pp 709ndash721 2014

[3] J T Scott ldquoDrug-induced goutrdquo Baillierersquos Clinical Rheumatol-ogy vol 5 no 1 pp 39ndash60 1991

[4] A Brandstater S Kiechl B Kollerits et al ldquoSex-specificassociation of the putative fructose transporter SLC2A9 variantswith uric acid levels is modified by BMIrdquo Diabetes Care vol 31no 8 pp 1662ndash1667 2008

[5] L-H Lai S-Y Chou F-Y Wu J J-H Chen and H-W KuoldquoRenal dysfunction and hyperuricemia with low blood leadlevels and ethnicity in community-based studyrdquo Science of theTotal Environment vol 401 no 1-3 pp 39ndash43 2008

[6] Q Liu Z Li H Wang et al ldquoHigh prevalence and associatedrisk factors for impaired renal function and urinary abnormal-ities in a rural adult population from Southern Chinardquo PLoSONE vol 7 no 10 Article ID e47100 2012

[7] W-G Fang X-M Huang Y Wang et al ldquoA cross-sectionalstudy of hyperuricemia in state-employees in Beijing preva-lence and risk factorsrdquo National Medical Journal of China vol86 no 25 pp 1764ndash1768 2006

[8] Y S Kim J P Guevara K M Kim H K Choi D F Heitjanand D A Albert ldquoHyperuricemia and coronary heart diseasea systematic review and meta-analysisrdquo Arthritis Care andResearch vol 62 no 2 pp 170ndash180 2010

[9] P C Grayson S Y Kim M LaValley and H K ChoildquoHyperuricemia and incident hypertension a systematic reviewand meta-analysisrdquo Arthritis Care and Research vol 63 no 1pp 102ndash110 2011

[10] M Li W Hou X Zhang L Hu and Z Tang ldquoHyperuricemiaand risk of stroke a systematic review and meta-analysis ofprospective studiesrdquoAtherosclerosis vol 232 no 2 pp 265ndash2702014

[11] M R Carnethon S P Fortmann L Palaniappan B B Dun-can M I Schmidt and L E Chambless ldquoRisk factors forprogression to incident hyperinsulinemia the atherosclerosisrisk in communities study 1987ndash1998rdquo American Journal ofEpidemiology vol 158 no 11 pp 1058ndash1067 2003

10 BioMed Research International

[12] B Liu T Wang H Zhao et al ldquoThe prevalence of hyper-uricemia in china a meta-analysisrdquo BMC Public Health vol 11no 1 article 832 2011

[13] E Krishnan ldquoInteraction of inflammation hyperuricemia andthe prevalence of hypertension among adults free of metabolicsyndromeNHANES 2009ndash2010rdquo Journal of the AmericanHeartAssociation vol 3 no 2 Article ID e000157 2014

[14] B Zhang XWang J You andD Zhang ldquoTongue color analysisfor medical applicationrdquo Evidence-Based Complementary andAlternativeMedicine vol 2013 Article ID 264742 11 pages 2013

[15] B Pang D Zhang and K Wang ldquoTongue image analysis forappendicitis diagnosisrdquo Information Sciences vol 175 no 3 pp160ndash176 2005

[16] J Kim J Son S Jang et al ldquoAvailability of tongue diagno-sis system for assessing tongue coating thickness in patientswith functional dyspepsiardquoEvidence-BasedComplementary andAlternativeMedicine vol 2013 Article ID 348272 6 pages 2013

[17] XWang B Zhang Z Yang HWang andD Zhang ldquoStatisticalanalysis of tongue images for feature extraction and diagnos-ticsrdquo IEEE Transactions on Image Processing vol 22 no 12 pp5336ndash5347 2013

[18] C J Jung J H Nam Y J Jeon and K H Kim ldquoColor distribu-tion differences in the tongue in sleep disorderrdquo Evidence-BasedComplementary and Alternative Medicine vol 2014 Article ID323645 8 pages 2014

[19] A H Friedlander and R Singer ldquoRenal adenocarcinoma ofthe kidney with metastasis to the tonguerdquo The Journal of theAmerican Dental Association vol 97 no 6 pp 989ndash991 1978

[20] K A Amir R K Bobba B Clarke et al ldquoTongue discolorationin an elderly kidney transplant recipient treatment-relatedadverse eventrdquoThe American Journal Geriatric Pharmacother-apy vol 4 no 3 pp 260ndash263 2006

[21] Q Tang J Chen and J Li ldquoAnalysis of the tongue characteristicsof liver depression and kidney insufficiency pid patientsrdquoChinaModern Medicine vol 15 pp 82ndash83 2010

[22] M Yang C-D Li W-N Liang and H Li ldquoCorrelationalstudy on pathological changes of liver qi depression and kidneydeficiency tongue coating exfoliative cells and sex hormonesin perimenopausal syndromerdquo China Journal of TraditionalChinese Medicine and Pharmacy vol 9 pp 1984ndash1986 2011

[23] P Viola and M J Jones ldquoRobust real-time face detectionrdquoInternational Journal of Computer Vision vol 57 no 2 pp 137ndash154 2004

[24] F Valbusa L Bertolini S Bonapace et al ldquoRelation of elevatedserum uric acid levels to incidence of atrial fibrillation inpatients with type 2 diabetes mellitusrdquoThe American Journal ofCardiology vol 112 no 4 pp 499ndash504 2013

[25] D A Grimes and K F Schulz ldquoAn overview of clinical researchthe lay of the landrdquo The Lancet vol 359 no 9300 pp 57ndash612002

[26] D A Grimes and K F Schulz ldquoCompared to what Findingcontrols for case-control studiesrdquoTheLancet vol 365 no 9468pp 1429ndash1433 2005

[27] M M Gorkani and R W Picard ldquo Texture orientation forsorting photos lsquoat a glancersquordquo inProceedings of the 12th IAPR Inter-national Conference on Pattern Recognition vol 1mdashConferenceA Computer Vision amp Image Processing pp 459ndash464 IEEEOctober 1994

[28] R Core Team R A Language and Environment for StatisticalComputing 2012

[29] R Nuzzo ldquoStatistical errorsrdquoNature vol 506 no 7487 pp 150ndash152 2014

[30] R Kanawong T Obafemi-Ajayi J Yu D Xu S Li and YDuan ldquoZheng classification in traditional chinese medicinebased on modified specular-free tongue imagesrdquo in Proceed-ings of the IEEE International Conference on Bioinformaticsand Biomedicine Workshops (BIBMW rsquo12) pp 288ndash294 IEEEOctober 2012

[31] R Kanawong T Obafemi-Ajayi T Ma D Xu S Li andY Duan ldquoAutomated tongue feature extraction for ZHENGclassification in traditional Chinese medicinerdquo Evidence-BasedComplementary and Alternative Medicine vol 2012 Article ID912852 14 pages 2012

[32] V Vitart I Rudan C Hayward et al ldquoSLC2A9 is a newly iden-tified urate transporter influencing serum urate concentrationurate excretion and goutrdquo Nature Genetics vol 40 no 4 pp437ndash442 2008

[33] M Tomita SMizunoH Yamanaka et al ldquoDoes hyperuricemiaaffect mortality A prospective cohort study of Japanese maleworkersrdquo Journal of epidemiologyJapan Epidemiological Associ-ation vol 10 no 6 pp 403ndash409 2000

[34] J Fang andMH Alderman ldquoSerumuric acid and cardiovascu-lar mortality the nhanes i epidemiologic follow-up study 1971ndash1992rdquo Journal of the American Medical Association vol 283 no18 pp 2404ndash2410 2000

[35] O Toprak M Cirit E Esi N Postaci M Yesil and S BayataldquoHyperuricemia as a risk factor for contrast-induced nephropa-thy in patientswith chronic kidney diseaserdquoCatheterization andCardiovascular Interventions vol 67 no 2 pp 227ndash235 2006

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

BioMed Research International 5

(a) (b) (c)

Figure 5 Three features on the green color plain examined in this study Panel (a) contains 5950 pixels (b) contains 300 pixels and (c)contains 200 pixels

(a) (b) (c)

Figure 6 Three features on the blue color plain examined in this study Panel (a) contains 6800 pixels (b) contains 1200 pixels and (c)contains 1200 pixels

respectively (Table 1)The smallest119875 values were obtained forfeature (30 23 10 30) (119875 = 209 times 10minus7 Figure 4(b)) feature(33 24 10 15) (119875 = 116 times 10minus4 Figure 5(b)) and feature(29 11 20 30) (119875 = 485 times 10minus11 Figure 6(b)) on the redgreen and blue color plains respectively (Table 1) Featureswith the largest 119889119891 values in the red green and blue plainswere feature (30 23 10 30) (119889119891 = 119 Figure 4(b)) feature(38 24 10 10) (119889119891 = 848 times 10minus1 Figure 5(c)) and feature(30 11 20 30) (119889119891 = 158 Figure 6(c)) respectively (Table 1)Feature (30 23 10 30) in the red plain had both the smallest119875 value and largest 119889119891 value

33 Single Classifier Performance The ROC curves of thetwo features in the red plain are shown in Figure 7 The

Table 1 Parameters of eight Haar-like features potentially useful forhyperuricemia diagnosis

Color plain 119875 119889119891 119883 119884 119882 119867

Red 172119864 minus 02 513119864 minus 01 11 21 40 85Red 209119864 minus 07 119 30 23 10 30Green 311119864 minus 02 459119864 minus 02 12 11 35 85Green 116119864 minus 04 852119864 minus 01 33 24 10 15Green 107119864 minus 04 848119864 minus 01 38 24 10 10Blue 119119864 minus 03 719119864 minus 01 11 21 40 85Blue 485119864 minus 11 158 29 11 20 30Blue 450119864 minus 11 158 30 11 20 30119883 and 119884 describe the position of the top-left corner of the feature 119867represents the height of a feature119882 and 119881 represent the widths of the leftand right rectangles respectively

6 BioMed Research International

False positive rate

True

pos

itive

rate

00 02 04 06 08 10

00

02

04

06

08

10

AUC 0654 AUC 081

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

Feature (11 21 40 85) Feature (30 23 10 30)

Figure 7 Receiver operating characteristic curves of features in the red plain

AUC 0612 AUC 0726 AUC 072

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

Feature (12 11 35 85) Feature (38 24 10 10) Feature (33 24 10 15)

Figure 8 Receiver operating characteristic curves of features in the green plain

AUC of feature (11 21 40 85) was 0654 preventing properanalysis of sensitivity and specificity The AUC of feature(30 23 10 30) was 0810 maximization of the Youden indexyielded sensitivity and specificity values of 0844 and 0652respectively Features in the green and blue color plainsshowed similar characteristics In the green plain the AUCvalue of feature (12 11 35 85) was only 0612 and thoseof feature (38 24 10 10) and feature (33 24 10 15) were0726 and 0720 respectively (Figure 8) Maximization of theYouden index yielded sensitivity values of 0500 and 0844and specificity values of 0391 and 0933 for the latter twofeatures respectively These two features were also deemed

to be inapplicable because of their low sensitivity values Inthe blue plain the AUC of feature (11 21 40 85) was 0704and those of feature (30 11 20 30) and feature (29 11 20 30)were 0877 and 0875 respectively (Figure 9) Sensitivity andspecificity values for feature (29 11 20 30) were 0800 and0804 respectively when the Youden index was maximizedThis feature achieved the best performance (sensitivity 0800specificity 0826) when the maximum value of the Youdenindex was 0626

34 Cumulative Feature Cumulative features in the redgreen and blue color plains (each composed of 50 features)

BioMed Research International 7

AUC 0704 AUC 0877 AUC 0875

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10Feature (11 21 40 85) Feature (30 11 20 30) Feature (29 11 20 30)

Figure 9 Receiver operating characteristic curves of features in the blue plain

0

0

20

20

40

40

60

60

80

80

100

100

120

P df

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 10 Cumulative image of 50 Haar-like features on the red plain Note the cumulative featurersquos centralization around the tongue root

are shown in Figures 10 11 and 12 respectively All ofthese features were centralized around the tongue rootvalidating our hypothesis The red cumulative feature has acircular distribution the green cumulative feature is moreconcentrated than the red feature and the blue cumulativefeature shows vertical symmetry

4 Discussion

Feature extraction is among the most important issues inimage processing These feature classes are based on perfect

segmentation of a tongue image from the background whichis difficult for the human eye [14ndash16] The extraction ofHaar-like features does not require segmentation [20] greatlysimplifying image preprocessing In this study we examined arectangular area including the tongue rather than attemptingto performmore precise segmentation as in previous studies

The use of Haar-like feature extraction from imagesis superior to extraction based solely on color because itallows the identification of local characteristics [19 30 31]Other studies have focused on color differences of the entiretongue [17 18] using global features such as means and

8 BioMed Research International

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixelsy

-pix

els

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 11 Cumulative image of 50 Haar-like features in the green plain

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 12 Cumulative image of 50 Haar-like features in the blue plain

standard deviations of color value These features prohibitdetailed medical interpretation because they do not considerdifferences among parts of the tongue In a previous study theexamination of tongue portions resulted in the identificationof some features that were located outside of the tongue [16]In our study we scanned the entire tongue image and foundthat all meaningful features (those with the smallest 119875 valuesand largest 119889119891 values) were located within the tongue area Incontrast to those of previous studies our results indicate that

tongue image preprocessing does not require perfect tonguesegmentation

Tongue image preprocessing using Haar-like features is anew method that not only resolves the segmentation issuebut also provides a novel means of interpreting tongueimages A face detection study using Haar-like featuresprovided the intuitive explanation that the most decisivefeatures include the eyes and nose [23] In our study wefound that the most decisive features for the diagnosis of

BioMed Research International 9

hyperuricemia are centralized on the tongue root A previousstudy described the results of tongue image analysis for thediagnosis of metastatic cancer [19] but applicable quantita-tive image analysis was not available at the time the studywas performed and the study also lacked a control groupAnother study focused on elderly subjects [20] In our studywe calculated quantitative feature values (119875 and119889119891) to expressdifferences between subjects with andwithout hyperuricemiausing a case-control design and including subjects from allage groups

The features identified in this study can be interpretedwithin the framework of TCM because they are based onpixels All features that performed well in this study werecentralized around the tongue root the area consideredto reflect kidney disease in TCM This study provideddirect evidence of the relationship between changes in thetongue root and kidney disease The kidney filters bloodand excretes metabolic waste products including uric acidIn the human body 70 of urate is disposed of via thekidneys [32] Hyperuricemia is not only related to severaldiseases but is also a risk factor for kidney injury [33ndash35] The diagnosis of hyperuricemia thus provides earlywarning of kidney injury However the determination ofserum urea nitrogen creatinine carbon dioxide and uricacid concentrations requires time-consuming biochemicalexamination An intuitive inexpensive and noninvasivemethod of hyperuricemia would thus be of benefit TCMprovides examination tools fulfilling these requirements Ourstudy provided direct evidence supporting the TCMmethodof diagnosing hyperuricemia based on tongue features

However this study has several limitations First theROCanalysiswas performedusing the same sample In futurestudies a test dataset will be collected to confirm the findingsof this study Second given that the use of tongue imagesis a complementary and alternative diagnostic method themethod described in this study should be combined withother available variables associated with hyperuricemia suchas body mass index and alcohol intake We plan to takethis approach in a further study Third because the samplewas carefully selected and patients with underlying diseasesassociated with hyperuricemia were excluded the use oftongue images for the diagnosis of hyperuricemia should berestricted

5 Conclusions

Haar-like features extracted from tongue images differed sig-nificantly between subjects with and without hyperuricemiaThe locations of these features are consistent with interpre-tations of tongue appearance in TCM andWestern medicineindicating the existence of a relationship between tongue rootcolor and hyperuricemia in our sample

Conflict of Interests

The authors declare that they have no conflict of interestsrelated to this work

Authorsrsquo Contribution

Yan Cui completed the statistical analysis and wrote thepaper Shizhong Liao performed critical revision of the paperand HongwuWang performed the mathematical models andmethods of this studyHongyuLiuWenhuaWang andLiqunYin validated health information for all subjects

Acknowledgment

This work was supported in part by the National BasicResearch Program of China (973 Program Grant no2011CB505406)

References

[1] H K Choi K Atkinson E W Karlson W Willett and GCurhan ldquoPurine-rich foods daily and protein intake and therisk of gout in menrdquoThe New England Journal of Medicine vol350 no 11 pp 1093ndash1103 2004

[2] K Kaneko Y Aoyagi T Fukuuchi K Inazawa andNYamaokaldquoTotal purine and purine base content of common foodstuffsfor facilitating nutritional therapy for gout and hyperuricemiardquoBiological and Pharmaceutical Bulletin vol 37 no 5 pp 709ndash721 2014

[3] J T Scott ldquoDrug-induced goutrdquo Baillierersquos Clinical Rheumatol-ogy vol 5 no 1 pp 39ndash60 1991

[4] A Brandstater S Kiechl B Kollerits et al ldquoSex-specificassociation of the putative fructose transporter SLC2A9 variantswith uric acid levels is modified by BMIrdquo Diabetes Care vol 31no 8 pp 1662ndash1667 2008

[5] L-H Lai S-Y Chou F-Y Wu J J-H Chen and H-W KuoldquoRenal dysfunction and hyperuricemia with low blood leadlevels and ethnicity in community-based studyrdquo Science of theTotal Environment vol 401 no 1-3 pp 39ndash43 2008

[6] Q Liu Z Li H Wang et al ldquoHigh prevalence and associatedrisk factors for impaired renal function and urinary abnormal-ities in a rural adult population from Southern Chinardquo PLoSONE vol 7 no 10 Article ID e47100 2012

[7] W-G Fang X-M Huang Y Wang et al ldquoA cross-sectionalstudy of hyperuricemia in state-employees in Beijing preva-lence and risk factorsrdquo National Medical Journal of China vol86 no 25 pp 1764ndash1768 2006

[8] Y S Kim J P Guevara K M Kim H K Choi D F Heitjanand D A Albert ldquoHyperuricemia and coronary heart diseasea systematic review and meta-analysisrdquo Arthritis Care andResearch vol 62 no 2 pp 170ndash180 2010

[9] P C Grayson S Y Kim M LaValley and H K ChoildquoHyperuricemia and incident hypertension a systematic reviewand meta-analysisrdquo Arthritis Care and Research vol 63 no 1pp 102ndash110 2011

[10] M Li W Hou X Zhang L Hu and Z Tang ldquoHyperuricemiaand risk of stroke a systematic review and meta-analysis ofprospective studiesrdquoAtherosclerosis vol 232 no 2 pp 265ndash2702014

[11] M R Carnethon S P Fortmann L Palaniappan B B Dun-can M I Schmidt and L E Chambless ldquoRisk factors forprogression to incident hyperinsulinemia the atherosclerosisrisk in communities study 1987ndash1998rdquo American Journal ofEpidemiology vol 158 no 11 pp 1058ndash1067 2003

10 BioMed Research International

[12] B Liu T Wang H Zhao et al ldquoThe prevalence of hyper-uricemia in china a meta-analysisrdquo BMC Public Health vol 11no 1 article 832 2011

[13] E Krishnan ldquoInteraction of inflammation hyperuricemia andthe prevalence of hypertension among adults free of metabolicsyndromeNHANES 2009ndash2010rdquo Journal of the AmericanHeartAssociation vol 3 no 2 Article ID e000157 2014

[14] B Zhang XWang J You andD Zhang ldquoTongue color analysisfor medical applicationrdquo Evidence-Based Complementary andAlternativeMedicine vol 2013 Article ID 264742 11 pages 2013

[15] B Pang D Zhang and K Wang ldquoTongue image analysis forappendicitis diagnosisrdquo Information Sciences vol 175 no 3 pp160ndash176 2005

[16] J Kim J Son S Jang et al ldquoAvailability of tongue diagno-sis system for assessing tongue coating thickness in patientswith functional dyspepsiardquoEvidence-BasedComplementary andAlternativeMedicine vol 2013 Article ID 348272 6 pages 2013

[17] XWang B Zhang Z Yang HWang andD Zhang ldquoStatisticalanalysis of tongue images for feature extraction and diagnos-ticsrdquo IEEE Transactions on Image Processing vol 22 no 12 pp5336ndash5347 2013

[18] C J Jung J H Nam Y J Jeon and K H Kim ldquoColor distribu-tion differences in the tongue in sleep disorderrdquo Evidence-BasedComplementary and Alternative Medicine vol 2014 Article ID323645 8 pages 2014

[19] A H Friedlander and R Singer ldquoRenal adenocarcinoma ofthe kidney with metastasis to the tonguerdquo The Journal of theAmerican Dental Association vol 97 no 6 pp 989ndash991 1978

[20] K A Amir R K Bobba B Clarke et al ldquoTongue discolorationin an elderly kidney transplant recipient treatment-relatedadverse eventrdquoThe American Journal Geriatric Pharmacother-apy vol 4 no 3 pp 260ndash263 2006

[21] Q Tang J Chen and J Li ldquoAnalysis of the tongue characteristicsof liver depression and kidney insufficiency pid patientsrdquoChinaModern Medicine vol 15 pp 82ndash83 2010

[22] M Yang C-D Li W-N Liang and H Li ldquoCorrelationalstudy on pathological changes of liver qi depression and kidneydeficiency tongue coating exfoliative cells and sex hormonesin perimenopausal syndromerdquo China Journal of TraditionalChinese Medicine and Pharmacy vol 9 pp 1984ndash1986 2011

[23] P Viola and M J Jones ldquoRobust real-time face detectionrdquoInternational Journal of Computer Vision vol 57 no 2 pp 137ndash154 2004

[24] F Valbusa L Bertolini S Bonapace et al ldquoRelation of elevatedserum uric acid levels to incidence of atrial fibrillation inpatients with type 2 diabetes mellitusrdquoThe American Journal ofCardiology vol 112 no 4 pp 499ndash504 2013

[25] D A Grimes and K F Schulz ldquoAn overview of clinical researchthe lay of the landrdquo The Lancet vol 359 no 9300 pp 57ndash612002

[26] D A Grimes and K F Schulz ldquoCompared to what Findingcontrols for case-control studiesrdquoTheLancet vol 365 no 9468pp 1429ndash1433 2005

[27] M M Gorkani and R W Picard ldquo Texture orientation forsorting photos lsquoat a glancersquordquo inProceedings of the 12th IAPR Inter-national Conference on Pattern Recognition vol 1mdashConferenceA Computer Vision amp Image Processing pp 459ndash464 IEEEOctober 1994

[28] R Core Team R A Language and Environment for StatisticalComputing 2012

[29] R Nuzzo ldquoStatistical errorsrdquoNature vol 506 no 7487 pp 150ndash152 2014

[30] R Kanawong T Obafemi-Ajayi J Yu D Xu S Li and YDuan ldquoZheng classification in traditional chinese medicinebased on modified specular-free tongue imagesrdquo in Proceed-ings of the IEEE International Conference on Bioinformaticsand Biomedicine Workshops (BIBMW rsquo12) pp 288ndash294 IEEEOctober 2012

[31] R Kanawong T Obafemi-Ajayi T Ma D Xu S Li andY Duan ldquoAutomated tongue feature extraction for ZHENGclassification in traditional Chinese medicinerdquo Evidence-BasedComplementary and Alternative Medicine vol 2012 Article ID912852 14 pages 2012

[32] V Vitart I Rudan C Hayward et al ldquoSLC2A9 is a newly iden-tified urate transporter influencing serum urate concentrationurate excretion and goutrdquo Nature Genetics vol 40 no 4 pp437ndash442 2008

[33] M Tomita SMizunoH Yamanaka et al ldquoDoes hyperuricemiaaffect mortality A prospective cohort study of Japanese maleworkersrdquo Journal of epidemiologyJapan Epidemiological Associ-ation vol 10 no 6 pp 403ndash409 2000

[34] J Fang andMH Alderman ldquoSerumuric acid and cardiovascu-lar mortality the nhanes i epidemiologic follow-up study 1971ndash1992rdquo Journal of the American Medical Association vol 283 no18 pp 2404ndash2410 2000

[35] O Toprak M Cirit E Esi N Postaci M Yesil and S BayataldquoHyperuricemia as a risk factor for contrast-induced nephropa-thy in patientswith chronic kidney diseaserdquoCatheterization andCardiovascular Interventions vol 67 no 2 pp 227ndash235 2006

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

6 BioMed Research International

False positive rate

True

pos

itive

rate

00 02 04 06 08 10

00

02

04

06

08

10

AUC 0654 AUC 081

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

Feature (11 21 40 85) Feature (30 23 10 30)

Figure 7 Receiver operating characteristic curves of features in the red plain

AUC 0612 AUC 0726 AUC 072

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

Feature (12 11 35 85) Feature (38 24 10 10) Feature (33 24 10 15)

Figure 8 Receiver operating characteristic curves of features in the green plain

AUC of feature (11 21 40 85) was 0654 preventing properanalysis of sensitivity and specificity The AUC of feature(30 23 10 30) was 0810 maximization of the Youden indexyielded sensitivity and specificity values of 0844 and 0652respectively Features in the green and blue color plainsshowed similar characteristics In the green plain the AUCvalue of feature (12 11 35 85) was only 0612 and thoseof feature (38 24 10 10) and feature (33 24 10 15) were0726 and 0720 respectively (Figure 8) Maximization of theYouden index yielded sensitivity values of 0500 and 0844and specificity values of 0391 and 0933 for the latter twofeatures respectively These two features were also deemed

to be inapplicable because of their low sensitivity values Inthe blue plain the AUC of feature (11 21 40 85) was 0704and those of feature (30 11 20 30) and feature (29 11 20 30)were 0877 and 0875 respectively (Figure 9) Sensitivity andspecificity values for feature (29 11 20 30) were 0800 and0804 respectively when the Youden index was maximizedThis feature achieved the best performance (sensitivity 0800specificity 0826) when the maximum value of the Youdenindex was 0626

34 Cumulative Feature Cumulative features in the redgreen and blue color plains (each composed of 50 features)

BioMed Research International 7

AUC 0704 AUC 0877 AUC 0875

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10Feature (11 21 40 85) Feature (30 11 20 30) Feature (29 11 20 30)

Figure 9 Receiver operating characteristic curves of features in the blue plain

0

0

20

20

40

40

60

60

80

80

100

100

120

P df

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 10 Cumulative image of 50 Haar-like features on the red plain Note the cumulative featurersquos centralization around the tongue root

are shown in Figures 10 11 and 12 respectively All ofthese features were centralized around the tongue rootvalidating our hypothesis The red cumulative feature has acircular distribution the green cumulative feature is moreconcentrated than the red feature and the blue cumulativefeature shows vertical symmetry

4 Discussion

Feature extraction is among the most important issues inimage processing These feature classes are based on perfect

segmentation of a tongue image from the background whichis difficult for the human eye [14ndash16] The extraction ofHaar-like features does not require segmentation [20] greatlysimplifying image preprocessing In this study we examined arectangular area including the tongue rather than attemptingto performmore precise segmentation as in previous studies

The use of Haar-like feature extraction from imagesis superior to extraction based solely on color because itallows the identification of local characteristics [19 30 31]Other studies have focused on color differences of the entiretongue [17 18] using global features such as means and

8 BioMed Research International

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixelsy

-pix

els

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 11 Cumulative image of 50 Haar-like features in the green plain

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 12 Cumulative image of 50 Haar-like features in the blue plain

standard deviations of color value These features prohibitdetailed medical interpretation because they do not considerdifferences among parts of the tongue In a previous study theexamination of tongue portions resulted in the identificationof some features that were located outside of the tongue [16]In our study we scanned the entire tongue image and foundthat all meaningful features (those with the smallest 119875 valuesand largest 119889119891 values) were located within the tongue area Incontrast to those of previous studies our results indicate that

tongue image preprocessing does not require perfect tonguesegmentation

Tongue image preprocessing using Haar-like features is anew method that not only resolves the segmentation issuebut also provides a novel means of interpreting tongueimages A face detection study using Haar-like featuresprovided the intuitive explanation that the most decisivefeatures include the eyes and nose [23] In our study wefound that the most decisive features for the diagnosis of

BioMed Research International 9

hyperuricemia are centralized on the tongue root A previousstudy described the results of tongue image analysis for thediagnosis of metastatic cancer [19] but applicable quantita-tive image analysis was not available at the time the studywas performed and the study also lacked a control groupAnother study focused on elderly subjects [20] In our studywe calculated quantitative feature values (119875 and119889119891) to expressdifferences between subjects with andwithout hyperuricemiausing a case-control design and including subjects from allage groups

The features identified in this study can be interpretedwithin the framework of TCM because they are based onpixels All features that performed well in this study werecentralized around the tongue root the area consideredto reflect kidney disease in TCM This study provideddirect evidence of the relationship between changes in thetongue root and kidney disease The kidney filters bloodand excretes metabolic waste products including uric acidIn the human body 70 of urate is disposed of via thekidneys [32] Hyperuricemia is not only related to severaldiseases but is also a risk factor for kidney injury [33ndash35] The diagnosis of hyperuricemia thus provides earlywarning of kidney injury However the determination ofserum urea nitrogen creatinine carbon dioxide and uricacid concentrations requires time-consuming biochemicalexamination An intuitive inexpensive and noninvasivemethod of hyperuricemia would thus be of benefit TCMprovides examination tools fulfilling these requirements Ourstudy provided direct evidence supporting the TCMmethodof diagnosing hyperuricemia based on tongue features

However this study has several limitations First theROCanalysiswas performedusing the same sample In futurestudies a test dataset will be collected to confirm the findingsof this study Second given that the use of tongue imagesis a complementary and alternative diagnostic method themethod described in this study should be combined withother available variables associated with hyperuricemia suchas body mass index and alcohol intake We plan to takethis approach in a further study Third because the samplewas carefully selected and patients with underlying diseasesassociated with hyperuricemia were excluded the use oftongue images for the diagnosis of hyperuricemia should berestricted

5 Conclusions

Haar-like features extracted from tongue images differed sig-nificantly between subjects with and without hyperuricemiaThe locations of these features are consistent with interpre-tations of tongue appearance in TCM andWestern medicineindicating the existence of a relationship between tongue rootcolor and hyperuricemia in our sample

Conflict of Interests

The authors declare that they have no conflict of interestsrelated to this work

Authorsrsquo Contribution

Yan Cui completed the statistical analysis and wrote thepaper Shizhong Liao performed critical revision of the paperand HongwuWang performed the mathematical models andmethods of this studyHongyuLiuWenhuaWang andLiqunYin validated health information for all subjects

Acknowledgment

This work was supported in part by the National BasicResearch Program of China (973 Program Grant no2011CB505406)

References

[1] H K Choi K Atkinson E W Karlson W Willett and GCurhan ldquoPurine-rich foods daily and protein intake and therisk of gout in menrdquoThe New England Journal of Medicine vol350 no 11 pp 1093ndash1103 2004

[2] K Kaneko Y Aoyagi T Fukuuchi K Inazawa andNYamaokaldquoTotal purine and purine base content of common foodstuffsfor facilitating nutritional therapy for gout and hyperuricemiardquoBiological and Pharmaceutical Bulletin vol 37 no 5 pp 709ndash721 2014

[3] J T Scott ldquoDrug-induced goutrdquo Baillierersquos Clinical Rheumatol-ogy vol 5 no 1 pp 39ndash60 1991

[4] A Brandstater S Kiechl B Kollerits et al ldquoSex-specificassociation of the putative fructose transporter SLC2A9 variantswith uric acid levels is modified by BMIrdquo Diabetes Care vol 31no 8 pp 1662ndash1667 2008

[5] L-H Lai S-Y Chou F-Y Wu J J-H Chen and H-W KuoldquoRenal dysfunction and hyperuricemia with low blood leadlevels and ethnicity in community-based studyrdquo Science of theTotal Environment vol 401 no 1-3 pp 39ndash43 2008

[6] Q Liu Z Li H Wang et al ldquoHigh prevalence and associatedrisk factors for impaired renal function and urinary abnormal-ities in a rural adult population from Southern Chinardquo PLoSONE vol 7 no 10 Article ID e47100 2012

[7] W-G Fang X-M Huang Y Wang et al ldquoA cross-sectionalstudy of hyperuricemia in state-employees in Beijing preva-lence and risk factorsrdquo National Medical Journal of China vol86 no 25 pp 1764ndash1768 2006

[8] Y S Kim J P Guevara K M Kim H K Choi D F Heitjanand D A Albert ldquoHyperuricemia and coronary heart diseasea systematic review and meta-analysisrdquo Arthritis Care andResearch vol 62 no 2 pp 170ndash180 2010

[9] P C Grayson S Y Kim M LaValley and H K ChoildquoHyperuricemia and incident hypertension a systematic reviewand meta-analysisrdquo Arthritis Care and Research vol 63 no 1pp 102ndash110 2011

[10] M Li W Hou X Zhang L Hu and Z Tang ldquoHyperuricemiaand risk of stroke a systematic review and meta-analysis ofprospective studiesrdquoAtherosclerosis vol 232 no 2 pp 265ndash2702014

[11] M R Carnethon S P Fortmann L Palaniappan B B Dun-can M I Schmidt and L E Chambless ldquoRisk factors forprogression to incident hyperinsulinemia the atherosclerosisrisk in communities study 1987ndash1998rdquo American Journal ofEpidemiology vol 158 no 11 pp 1058ndash1067 2003

10 BioMed Research International

[12] B Liu T Wang H Zhao et al ldquoThe prevalence of hyper-uricemia in china a meta-analysisrdquo BMC Public Health vol 11no 1 article 832 2011

[13] E Krishnan ldquoInteraction of inflammation hyperuricemia andthe prevalence of hypertension among adults free of metabolicsyndromeNHANES 2009ndash2010rdquo Journal of the AmericanHeartAssociation vol 3 no 2 Article ID e000157 2014

[14] B Zhang XWang J You andD Zhang ldquoTongue color analysisfor medical applicationrdquo Evidence-Based Complementary andAlternativeMedicine vol 2013 Article ID 264742 11 pages 2013

[15] B Pang D Zhang and K Wang ldquoTongue image analysis forappendicitis diagnosisrdquo Information Sciences vol 175 no 3 pp160ndash176 2005

[16] J Kim J Son S Jang et al ldquoAvailability of tongue diagno-sis system for assessing tongue coating thickness in patientswith functional dyspepsiardquoEvidence-BasedComplementary andAlternativeMedicine vol 2013 Article ID 348272 6 pages 2013

[17] XWang B Zhang Z Yang HWang andD Zhang ldquoStatisticalanalysis of tongue images for feature extraction and diagnos-ticsrdquo IEEE Transactions on Image Processing vol 22 no 12 pp5336ndash5347 2013

[18] C J Jung J H Nam Y J Jeon and K H Kim ldquoColor distribu-tion differences in the tongue in sleep disorderrdquo Evidence-BasedComplementary and Alternative Medicine vol 2014 Article ID323645 8 pages 2014

[19] A H Friedlander and R Singer ldquoRenal adenocarcinoma ofthe kidney with metastasis to the tonguerdquo The Journal of theAmerican Dental Association vol 97 no 6 pp 989ndash991 1978

[20] K A Amir R K Bobba B Clarke et al ldquoTongue discolorationin an elderly kidney transplant recipient treatment-relatedadverse eventrdquoThe American Journal Geriatric Pharmacother-apy vol 4 no 3 pp 260ndash263 2006

[21] Q Tang J Chen and J Li ldquoAnalysis of the tongue characteristicsof liver depression and kidney insufficiency pid patientsrdquoChinaModern Medicine vol 15 pp 82ndash83 2010

[22] M Yang C-D Li W-N Liang and H Li ldquoCorrelationalstudy on pathological changes of liver qi depression and kidneydeficiency tongue coating exfoliative cells and sex hormonesin perimenopausal syndromerdquo China Journal of TraditionalChinese Medicine and Pharmacy vol 9 pp 1984ndash1986 2011

[23] P Viola and M J Jones ldquoRobust real-time face detectionrdquoInternational Journal of Computer Vision vol 57 no 2 pp 137ndash154 2004

[24] F Valbusa L Bertolini S Bonapace et al ldquoRelation of elevatedserum uric acid levels to incidence of atrial fibrillation inpatients with type 2 diabetes mellitusrdquoThe American Journal ofCardiology vol 112 no 4 pp 499ndash504 2013

[25] D A Grimes and K F Schulz ldquoAn overview of clinical researchthe lay of the landrdquo The Lancet vol 359 no 9300 pp 57ndash612002

[26] D A Grimes and K F Schulz ldquoCompared to what Findingcontrols for case-control studiesrdquoTheLancet vol 365 no 9468pp 1429ndash1433 2005

[27] M M Gorkani and R W Picard ldquo Texture orientation forsorting photos lsquoat a glancersquordquo inProceedings of the 12th IAPR Inter-national Conference on Pattern Recognition vol 1mdashConferenceA Computer Vision amp Image Processing pp 459ndash464 IEEEOctober 1994

[28] R Core Team R A Language and Environment for StatisticalComputing 2012

[29] R Nuzzo ldquoStatistical errorsrdquoNature vol 506 no 7487 pp 150ndash152 2014

[30] R Kanawong T Obafemi-Ajayi J Yu D Xu S Li and YDuan ldquoZheng classification in traditional chinese medicinebased on modified specular-free tongue imagesrdquo in Proceed-ings of the IEEE International Conference on Bioinformaticsand Biomedicine Workshops (BIBMW rsquo12) pp 288ndash294 IEEEOctober 2012

[31] R Kanawong T Obafemi-Ajayi T Ma D Xu S Li andY Duan ldquoAutomated tongue feature extraction for ZHENGclassification in traditional Chinese medicinerdquo Evidence-BasedComplementary and Alternative Medicine vol 2012 Article ID912852 14 pages 2012

[32] V Vitart I Rudan C Hayward et al ldquoSLC2A9 is a newly iden-tified urate transporter influencing serum urate concentrationurate excretion and goutrdquo Nature Genetics vol 40 no 4 pp437ndash442 2008

[33] M Tomita SMizunoH Yamanaka et al ldquoDoes hyperuricemiaaffect mortality A prospective cohort study of Japanese maleworkersrdquo Journal of epidemiologyJapan Epidemiological Associ-ation vol 10 no 6 pp 403ndash409 2000

[34] J Fang andMH Alderman ldquoSerumuric acid and cardiovascu-lar mortality the nhanes i epidemiologic follow-up study 1971ndash1992rdquo Journal of the American Medical Association vol 283 no18 pp 2404ndash2410 2000

[35] O Toprak M Cirit E Esi N Postaci M Yesil and S BayataldquoHyperuricemia as a risk factor for contrast-induced nephropa-thy in patientswith chronic kidney diseaserdquoCatheterization andCardiovascular Interventions vol 67 no 2 pp 227ndash235 2006

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

BioMed Research International 7

AUC 0704 AUC 0877 AUC 0875

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10

False positive rate00 02 04 06 08 10

True

pos

itive

rate

00

02

04

06

08

10Feature (11 21 40 85) Feature (30 11 20 30) Feature (29 11 20 30)

Figure 9 Receiver operating characteristic curves of features in the blue plain

0

0

20

20

40

40

60

60

80

80

100

100

120

P df

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 10 Cumulative image of 50 Haar-like features on the red plain Note the cumulative featurersquos centralization around the tongue root

are shown in Figures 10 11 and 12 respectively All ofthese features were centralized around the tongue rootvalidating our hypothesis The red cumulative feature has acircular distribution the green cumulative feature is moreconcentrated than the red feature and the blue cumulativefeature shows vertical symmetry

4 Discussion

Feature extraction is among the most important issues inimage processing These feature classes are based on perfect

segmentation of a tongue image from the background whichis difficult for the human eye [14ndash16] The extraction ofHaar-like features does not require segmentation [20] greatlysimplifying image preprocessing In this study we examined arectangular area including the tongue rather than attemptingto performmore precise segmentation as in previous studies

The use of Haar-like feature extraction from imagesis superior to extraction based solely on color because itallows the identification of local characteristics [19 30 31]Other studies have focused on color differences of the entiretongue [17 18] using global features such as means and

8 BioMed Research International

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixelsy

-pix

els

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 11 Cumulative image of 50 Haar-like features in the green plain

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 12 Cumulative image of 50 Haar-like features in the blue plain

standard deviations of color value These features prohibitdetailed medical interpretation because they do not considerdifferences among parts of the tongue In a previous study theexamination of tongue portions resulted in the identificationof some features that were located outside of the tongue [16]In our study we scanned the entire tongue image and foundthat all meaningful features (those with the smallest 119875 valuesand largest 119889119891 values) were located within the tongue area Incontrast to those of previous studies our results indicate that

tongue image preprocessing does not require perfect tonguesegmentation

Tongue image preprocessing using Haar-like features is anew method that not only resolves the segmentation issuebut also provides a novel means of interpreting tongueimages A face detection study using Haar-like featuresprovided the intuitive explanation that the most decisivefeatures include the eyes and nose [23] In our study wefound that the most decisive features for the diagnosis of

BioMed Research International 9

hyperuricemia are centralized on the tongue root A previousstudy described the results of tongue image analysis for thediagnosis of metastatic cancer [19] but applicable quantita-tive image analysis was not available at the time the studywas performed and the study also lacked a control groupAnother study focused on elderly subjects [20] In our studywe calculated quantitative feature values (119875 and119889119891) to expressdifferences between subjects with andwithout hyperuricemiausing a case-control design and including subjects from allage groups

The features identified in this study can be interpretedwithin the framework of TCM because they are based onpixels All features that performed well in this study werecentralized around the tongue root the area consideredto reflect kidney disease in TCM This study provideddirect evidence of the relationship between changes in thetongue root and kidney disease The kidney filters bloodand excretes metabolic waste products including uric acidIn the human body 70 of urate is disposed of via thekidneys [32] Hyperuricemia is not only related to severaldiseases but is also a risk factor for kidney injury [33ndash35] The diagnosis of hyperuricemia thus provides earlywarning of kidney injury However the determination ofserum urea nitrogen creatinine carbon dioxide and uricacid concentrations requires time-consuming biochemicalexamination An intuitive inexpensive and noninvasivemethod of hyperuricemia would thus be of benefit TCMprovides examination tools fulfilling these requirements Ourstudy provided direct evidence supporting the TCMmethodof diagnosing hyperuricemia based on tongue features

However this study has several limitations First theROCanalysiswas performedusing the same sample In futurestudies a test dataset will be collected to confirm the findingsof this study Second given that the use of tongue imagesis a complementary and alternative diagnostic method themethod described in this study should be combined withother available variables associated with hyperuricemia suchas body mass index and alcohol intake We plan to takethis approach in a further study Third because the samplewas carefully selected and patients with underlying diseasesassociated with hyperuricemia were excluded the use oftongue images for the diagnosis of hyperuricemia should berestricted

5 Conclusions

Haar-like features extracted from tongue images differed sig-nificantly between subjects with and without hyperuricemiaThe locations of these features are consistent with interpre-tations of tongue appearance in TCM andWestern medicineindicating the existence of a relationship between tongue rootcolor and hyperuricemia in our sample

Conflict of Interests

The authors declare that they have no conflict of interestsrelated to this work

Authorsrsquo Contribution

Yan Cui completed the statistical analysis and wrote thepaper Shizhong Liao performed critical revision of the paperand HongwuWang performed the mathematical models andmethods of this studyHongyuLiuWenhuaWang andLiqunYin validated health information for all subjects

Acknowledgment

This work was supported in part by the National BasicResearch Program of China (973 Program Grant no2011CB505406)

References

[1] H K Choi K Atkinson E W Karlson W Willett and GCurhan ldquoPurine-rich foods daily and protein intake and therisk of gout in menrdquoThe New England Journal of Medicine vol350 no 11 pp 1093ndash1103 2004

[2] K Kaneko Y Aoyagi T Fukuuchi K Inazawa andNYamaokaldquoTotal purine and purine base content of common foodstuffsfor facilitating nutritional therapy for gout and hyperuricemiardquoBiological and Pharmaceutical Bulletin vol 37 no 5 pp 709ndash721 2014

[3] J T Scott ldquoDrug-induced goutrdquo Baillierersquos Clinical Rheumatol-ogy vol 5 no 1 pp 39ndash60 1991

[4] A Brandstater S Kiechl B Kollerits et al ldquoSex-specificassociation of the putative fructose transporter SLC2A9 variantswith uric acid levels is modified by BMIrdquo Diabetes Care vol 31no 8 pp 1662ndash1667 2008

[5] L-H Lai S-Y Chou F-Y Wu J J-H Chen and H-W KuoldquoRenal dysfunction and hyperuricemia with low blood leadlevels and ethnicity in community-based studyrdquo Science of theTotal Environment vol 401 no 1-3 pp 39ndash43 2008

[6] Q Liu Z Li H Wang et al ldquoHigh prevalence and associatedrisk factors for impaired renal function and urinary abnormal-ities in a rural adult population from Southern Chinardquo PLoSONE vol 7 no 10 Article ID e47100 2012

[7] W-G Fang X-M Huang Y Wang et al ldquoA cross-sectionalstudy of hyperuricemia in state-employees in Beijing preva-lence and risk factorsrdquo National Medical Journal of China vol86 no 25 pp 1764ndash1768 2006

[8] Y S Kim J P Guevara K M Kim H K Choi D F Heitjanand D A Albert ldquoHyperuricemia and coronary heart diseasea systematic review and meta-analysisrdquo Arthritis Care andResearch vol 62 no 2 pp 170ndash180 2010

[9] P C Grayson S Y Kim M LaValley and H K ChoildquoHyperuricemia and incident hypertension a systematic reviewand meta-analysisrdquo Arthritis Care and Research vol 63 no 1pp 102ndash110 2011

[10] M Li W Hou X Zhang L Hu and Z Tang ldquoHyperuricemiaand risk of stroke a systematic review and meta-analysis ofprospective studiesrdquoAtherosclerosis vol 232 no 2 pp 265ndash2702014

[11] M R Carnethon S P Fortmann L Palaniappan B B Dun-can M I Schmidt and L E Chambless ldquoRisk factors forprogression to incident hyperinsulinemia the atherosclerosisrisk in communities study 1987ndash1998rdquo American Journal ofEpidemiology vol 158 no 11 pp 1058ndash1067 2003

10 BioMed Research International

[12] B Liu T Wang H Zhao et al ldquoThe prevalence of hyper-uricemia in china a meta-analysisrdquo BMC Public Health vol 11no 1 article 832 2011

[13] E Krishnan ldquoInteraction of inflammation hyperuricemia andthe prevalence of hypertension among adults free of metabolicsyndromeNHANES 2009ndash2010rdquo Journal of the AmericanHeartAssociation vol 3 no 2 Article ID e000157 2014

[14] B Zhang XWang J You andD Zhang ldquoTongue color analysisfor medical applicationrdquo Evidence-Based Complementary andAlternativeMedicine vol 2013 Article ID 264742 11 pages 2013

[15] B Pang D Zhang and K Wang ldquoTongue image analysis forappendicitis diagnosisrdquo Information Sciences vol 175 no 3 pp160ndash176 2005

[16] J Kim J Son S Jang et al ldquoAvailability of tongue diagno-sis system for assessing tongue coating thickness in patientswith functional dyspepsiardquoEvidence-BasedComplementary andAlternativeMedicine vol 2013 Article ID 348272 6 pages 2013

[17] XWang B Zhang Z Yang HWang andD Zhang ldquoStatisticalanalysis of tongue images for feature extraction and diagnos-ticsrdquo IEEE Transactions on Image Processing vol 22 no 12 pp5336ndash5347 2013

[18] C J Jung J H Nam Y J Jeon and K H Kim ldquoColor distribu-tion differences in the tongue in sleep disorderrdquo Evidence-BasedComplementary and Alternative Medicine vol 2014 Article ID323645 8 pages 2014

[19] A H Friedlander and R Singer ldquoRenal adenocarcinoma ofthe kidney with metastasis to the tonguerdquo The Journal of theAmerican Dental Association vol 97 no 6 pp 989ndash991 1978

[20] K A Amir R K Bobba B Clarke et al ldquoTongue discolorationin an elderly kidney transplant recipient treatment-relatedadverse eventrdquoThe American Journal Geriatric Pharmacother-apy vol 4 no 3 pp 260ndash263 2006

[21] Q Tang J Chen and J Li ldquoAnalysis of the tongue characteristicsof liver depression and kidney insufficiency pid patientsrdquoChinaModern Medicine vol 15 pp 82ndash83 2010

[22] M Yang C-D Li W-N Liang and H Li ldquoCorrelationalstudy on pathological changes of liver qi depression and kidneydeficiency tongue coating exfoliative cells and sex hormonesin perimenopausal syndromerdquo China Journal of TraditionalChinese Medicine and Pharmacy vol 9 pp 1984ndash1986 2011

[23] P Viola and M J Jones ldquoRobust real-time face detectionrdquoInternational Journal of Computer Vision vol 57 no 2 pp 137ndash154 2004

[24] F Valbusa L Bertolini S Bonapace et al ldquoRelation of elevatedserum uric acid levels to incidence of atrial fibrillation inpatients with type 2 diabetes mellitusrdquoThe American Journal ofCardiology vol 112 no 4 pp 499ndash504 2013

[25] D A Grimes and K F Schulz ldquoAn overview of clinical researchthe lay of the landrdquo The Lancet vol 359 no 9300 pp 57ndash612002

[26] D A Grimes and K F Schulz ldquoCompared to what Findingcontrols for case-control studiesrdquoTheLancet vol 365 no 9468pp 1429ndash1433 2005

[27] M M Gorkani and R W Picard ldquo Texture orientation forsorting photos lsquoat a glancersquordquo inProceedings of the 12th IAPR Inter-national Conference on Pattern Recognition vol 1mdashConferenceA Computer Vision amp Image Processing pp 459ndash464 IEEEOctober 1994

[28] R Core Team R A Language and Environment for StatisticalComputing 2012

[29] R Nuzzo ldquoStatistical errorsrdquoNature vol 506 no 7487 pp 150ndash152 2014

[30] R Kanawong T Obafemi-Ajayi J Yu D Xu S Li and YDuan ldquoZheng classification in traditional chinese medicinebased on modified specular-free tongue imagesrdquo in Proceed-ings of the IEEE International Conference on Bioinformaticsand Biomedicine Workshops (BIBMW rsquo12) pp 288ndash294 IEEEOctober 2012

[31] R Kanawong T Obafemi-Ajayi T Ma D Xu S Li andY Duan ldquoAutomated tongue feature extraction for ZHENGclassification in traditional Chinese medicinerdquo Evidence-BasedComplementary and Alternative Medicine vol 2012 Article ID912852 14 pages 2012

[32] V Vitart I Rudan C Hayward et al ldquoSLC2A9 is a newly iden-tified urate transporter influencing serum urate concentrationurate excretion and goutrdquo Nature Genetics vol 40 no 4 pp437ndash442 2008

[33] M Tomita SMizunoH Yamanaka et al ldquoDoes hyperuricemiaaffect mortality A prospective cohort study of Japanese maleworkersrdquo Journal of epidemiologyJapan Epidemiological Associ-ation vol 10 no 6 pp 403ndash409 2000

[34] J Fang andMH Alderman ldquoSerumuric acid and cardiovascu-lar mortality the nhanes i epidemiologic follow-up study 1971ndash1992rdquo Journal of the American Medical Association vol 283 no18 pp 2404ndash2410 2000

[35] O Toprak M Cirit E Esi N Postaci M Yesil and S BayataldquoHyperuricemia as a risk factor for contrast-induced nephropa-thy in patientswith chronic kidney diseaserdquoCatheterization andCardiovascular Interventions vol 67 no 2 pp 227ndash235 2006

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

8 BioMed Research International

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixelsy

-pix

els

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 11 Cumulative image of 50 Haar-like features in the green plain

P df

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

0

0

20

20

40

40

60

60

80

80

100

100

120

x-pixels

y-p

ixel

s

Figure 12 Cumulative image of 50 Haar-like features in the blue plain

standard deviations of color value These features prohibitdetailed medical interpretation because they do not considerdifferences among parts of the tongue In a previous study theexamination of tongue portions resulted in the identificationof some features that were located outside of the tongue [16]In our study we scanned the entire tongue image and foundthat all meaningful features (those with the smallest 119875 valuesand largest 119889119891 values) were located within the tongue area Incontrast to those of previous studies our results indicate that

tongue image preprocessing does not require perfect tonguesegmentation

Tongue image preprocessing using Haar-like features is anew method that not only resolves the segmentation issuebut also provides a novel means of interpreting tongueimages A face detection study using Haar-like featuresprovided the intuitive explanation that the most decisivefeatures include the eyes and nose [23] In our study wefound that the most decisive features for the diagnosis of

BioMed Research International 9

hyperuricemia are centralized on the tongue root A previousstudy described the results of tongue image analysis for thediagnosis of metastatic cancer [19] but applicable quantita-tive image analysis was not available at the time the studywas performed and the study also lacked a control groupAnother study focused on elderly subjects [20] In our studywe calculated quantitative feature values (119875 and119889119891) to expressdifferences between subjects with andwithout hyperuricemiausing a case-control design and including subjects from allage groups

The features identified in this study can be interpretedwithin the framework of TCM because they are based onpixels All features that performed well in this study werecentralized around the tongue root the area consideredto reflect kidney disease in TCM This study provideddirect evidence of the relationship between changes in thetongue root and kidney disease The kidney filters bloodand excretes metabolic waste products including uric acidIn the human body 70 of urate is disposed of via thekidneys [32] Hyperuricemia is not only related to severaldiseases but is also a risk factor for kidney injury [33ndash35] The diagnosis of hyperuricemia thus provides earlywarning of kidney injury However the determination ofserum urea nitrogen creatinine carbon dioxide and uricacid concentrations requires time-consuming biochemicalexamination An intuitive inexpensive and noninvasivemethod of hyperuricemia would thus be of benefit TCMprovides examination tools fulfilling these requirements Ourstudy provided direct evidence supporting the TCMmethodof diagnosing hyperuricemia based on tongue features

However this study has several limitations First theROCanalysiswas performedusing the same sample In futurestudies a test dataset will be collected to confirm the findingsof this study Second given that the use of tongue imagesis a complementary and alternative diagnostic method themethod described in this study should be combined withother available variables associated with hyperuricemia suchas body mass index and alcohol intake We plan to takethis approach in a further study Third because the samplewas carefully selected and patients with underlying diseasesassociated with hyperuricemia were excluded the use oftongue images for the diagnosis of hyperuricemia should berestricted

5 Conclusions

Haar-like features extracted from tongue images differed sig-nificantly between subjects with and without hyperuricemiaThe locations of these features are consistent with interpre-tations of tongue appearance in TCM andWestern medicineindicating the existence of a relationship between tongue rootcolor and hyperuricemia in our sample

Conflict of Interests

The authors declare that they have no conflict of interestsrelated to this work

Authorsrsquo Contribution

Yan Cui completed the statistical analysis and wrote thepaper Shizhong Liao performed critical revision of the paperand HongwuWang performed the mathematical models andmethods of this studyHongyuLiuWenhuaWang andLiqunYin validated health information for all subjects

Acknowledgment

This work was supported in part by the National BasicResearch Program of China (973 Program Grant no2011CB505406)

References

[1] H K Choi K Atkinson E W Karlson W Willett and GCurhan ldquoPurine-rich foods daily and protein intake and therisk of gout in menrdquoThe New England Journal of Medicine vol350 no 11 pp 1093ndash1103 2004

[2] K Kaneko Y Aoyagi T Fukuuchi K Inazawa andNYamaokaldquoTotal purine and purine base content of common foodstuffsfor facilitating nutritional therapy for gout and hyperuricemiardquoBiological and Pharmaceutical Bulletin vol 37 no 5 pp 709ndash721 2014

[3] J T Scott ldquoDrug-induced goutrdquo Baillierersquos Clinical Rheumatol-ogy vol 5 no 1 pp 39ndash60 1991

[4] A Brandstater S Kiechl B Kollerits et al ldquoSex-specificassociation of the putative fructose transporter SLC2A9 variantswith uric acid levels is modified by BMIrdquo Diabetes Care vol 31no 8 pp 1662ndash1667 2008

[5] L-H Lai S-Y Chou F-Y Wu J J-H Chen and H-W KuoldquoRenal dysfunction and hyperuricemia with low blood leadlevels and ethnicity in community-based studyrdquo Science of theTotal Environment vol 401 no 1-3 pp 39ndash43 2008

[6] Q Liu Z Li H Wang et al ldquoHigh prevalence and associatedrisk factors for impaired renal function and urinary abnormal-ities in a rural adult population from Southern Chinardquo PLoSONE vol 7 no 10 Article ID e47100 2012

[7] W-G Fang X-M Huang Y Wang et al ldquoA cross-sectionalstudy of hyperuricemia in state-employees in Beijing preva-lence and risk factorsrdquo National Medical Journal of China vol86 no 25 pp 1764ndash1768 2006

[8] Y S Kim J P Guevara K M Kim H K Choi D F Heitjanand D A Albert ldquoHyperuricemia and coronary heart diseasea systematic review and meta-analysisrdquo Arthritis Care andResearch vol 62 no 2 pp 170ndash180 2010

[9] P C Grayson S Y Kim M LaValley and H K ChoildquoHyperuricemia and incident hypertension a systematic reviewand meta-analysisrdquo Arthritis Care and Research vol 63 no 1pp 102ndash110 2011

[10] M Li W Hou X Zhang L Hu and Z Tang ldquoHyperuricemiaand risk of stroke a systematic review and meta-analysis ofprospective studiesrdquoAtherosclerosis vol 232 no 2 pp 265ndash2702014

[11] M R Carnethon S P Fortmann L Palaniappan B B Dun-can M I Schmidt and L E Chambless ldquoRisk factors forprogression to incident hyperinsulinemia the atherosclerosisrisk in communities study 1987ndash1998rdquo American Journal ofEpidemiology vol 158 no 11 pp 1058ndash1067 2003

10 BioMed Research International

[12] B Liu T Wang H Zhao et al ldquoThe prevalence of hyper-uricemia in china a meta-analysisrdquo BMC Public Health vol 11no 1 article 832 2011

[13] E Krishnan ldquoInteraction of inflammation hyperuricemia andthe prevalence of hypertension among adults free of metabolicsyndromeNHANES 2009ndash2010rdquo Journal of the AmericanHeartAssociation vol 3 no 2 Article ID e000157 2014

[14] B Zhang XWang J You andD Zhang ldquoTongue color analysisfor medical applicationrdquo Evidence-Based Complementary andAlternativeMedicine vol 2013 Article ID 264742 11 pages 2013

[15] B Pang D Zhang and K Wang ldquoTongue image analysis forappendicitis diagnosisrdquo Information Sciences vol 175 no 3 pp160ndash176 2005

[16] J Kim J Son S Jang et al ldquoAvailability of tongue diagno-sis system for assessing tongue coating thickness in patientswith functional dyspepsiardquoEvidence-BasedComplementary andAlternativeMedicine vol 2013 Article ID 348272 6 pages 2013

[17] XWang B Zhang Z Yang HWang andD Zhang ldquoStatisticalanalysis of tongue images for feature extraction and diagnos-ticsrdquo IEEE Transactions on Image Processing vol 22 no 12 pp5336ndash5347 2013

[18] C J Jung J H Nam Y J Jeon and K H Kim ldquoColor distribu-tion differences in the tongue in sleep disorderrdquo Evidence-BasedComplementary and Alternative Medicine vol 2014 Article ID323645 8 pages 2014

[19] A H Friedlander and R Singer ldquoRenal adenocarcinoma ofthe kidney with metastasis to the tonguerdquo The Journal of theAmerican Dental Association vol 97 no 6 pp 989ndash991 1978

[20] K A Amir R K Bobba B Clarke et al ldquoTongue discolorationin an elderly kidney transplant recipient treatment-relatedadverse eventrdquoThe American Journal Geriatric Pharmacother-apy vol 4 no 3 pp 260ndash263 2006

[21] Q Tang J Chen and J Li ldquoAnalysis of the tongue characteristicsof liver depression and kidney insufficiency pid patientsrdquoChinaModern Medicine vol 15 pp 82ndash83 2010

[22] M Yang C-D Li W-N Liang and H Li ldquoCorrelationalstudy on pathological changes of liver qi depression and kidneydeficiency tongue coating exfoliative cells and sex hormonesin perimenopausal syndromerdquo China Journal of TraditionalChinese Medicine and Pharmacy vol 9 pp 1984ndash1986 2011

[23] P Viola and M J Jones ldquoRobust real-time face detectionrdquoInternational Journal of Computer Vision vol 57 no 2 pp 137ndash154 2004

[24] F Valbusa L Bertolini S Bonapace et al ldquoRelation of elevatedserum uric acid levels to incidence of atrial fibrillation inpatients with type 2 diabetes mellitusrdquoThe American Journal ofCardiology vol 112 no 4 pp 499ndash504 2013

[25] D A Grimes and K F Schulz ldquoAn overview of clinical researchthe lay of the landrdquo The Lancet vol 359 no 9300 pp 57ndash612002

[26] D A Grimes and K F Schulz ldquoCompared to what Findingcontrols for case-control studiesrdquoTheLancet vol 365 no 9468pp 1429ndash1433 2005

[27] M M Gorkani and R W Picard ldquo Texture orientation forsorting photos lsquoat a glancersquordquo inProceedings of the 12th IAPR Inter-national Conference on Pattern Recognition vol 1mdashConferenceA Computer Vision amp Image Processing pp 459ndash464 IEEEOctober 1994

[28] R Core Team R A Language and Environment for StatisticalComputing 2012

[29] R Nuzzo ldquoStatistical errorsrdquoNature vol 506 no 7487 pp 150ndash152 2014

[30] R Kanawong T Obafemi-Ajayi J Yu D Xu S Li and YDuan ldquoZheng classification in traditional chinese medicinebased on modified specular-free tongue imagesrdquo in Proceed-ings of the IEEE International Conference on Bioinformaticsand Biomedicine Workshops (BIBMW rsquo12) pp 288ndash294 IEEEOctober 2012

[31] R Kanawong T Obafemi-Ajayi T Ma D Xu S Li andY Duan ldquoAutomated tongue feature extraction for ZHENGclassification in traditional Chinese medicinerdquo Evidence-BasedComplementary and Alternative Medicine vol 2012 Article ID912852 14 pages 2012

[32] V Vitart I Rudan C Hayward et al ldquoSLC2A9 is a newly iden-tified urate transporter influencing serum urate concentrationurate excretion and goutrdquo Nature Genetics vol 40 no 4 pp437ndash442 2008

[33] M Tomita SMizunoH Yamanaka et al ldquoDoes hyperuricemiaaffect mortality A prospective cohort study of Japanese maleworkersrdquo Journal of epidemiologyJapan Epidemiological Associ-ation vol 10 no 6 pp 403ndash409 2000

[34] J Fang andMH Alderman ldquoSerumuric acid and cardiovascu-lar mortality the nhanes i epidemiologic follow-up study 1971ndash1992rdquo Journal of the American Medical Association vol 283 no18 pp 2404ndash2410 2000

[35] O Toprak M Cirit E Esi N Postaci M Yesil and S BayataldquoHyperuricemia as a risk factor for contrast-induced nephropa-thy in patientswith chronic kidney diseaserdquoCatheterization andCardiovascular Interventions vol 67 no 2 pp 227ndash235 2006

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

BioMed Research International 9

hyperuricemia are centralized on the tongue root A previousstudy described the results of tongue image analysis for thediagnosis of metastatic cancer [19] but applicable quantita-tive image analysis was not available at the time the studywas performed and the study also lacked a control groupAnother study focused on elderly subjects [20] In our studywe calculated quantitative feature values (119875 and119889119891) to expressdifferences between subjects with andwithout hyperuricemiausing a case-control design and including subjects from allage groups

The features identified in this study can be interpretedwithin the framework of TCM because they are based onpixels All features that performed well in this study werecentralized around the tongue root the area consideredto reflect kidney disease in TCM This study provideddirect evidence of the relationship between changes in thetongue root and kidney disease The kidney filters bloodand excretes metabolic waste products including uric acidIn the human body 70 of urate is disposed of via thekidneys [32] Hyperuricemia is not only related to severaldiseases but is also a risk factor for kidney injury [33ndash35] The diagnosis of hyperuricemia thus provides earlywarning of kidney injury However the determination ofserum urea nitrogen creatinine carbon dioxide and uricacid concentrations requires time-consuming biochemicalexamination An intuitive inexpensive and noninvasivemethod of hyperuricemia would thus be of benefit TCMprovides examination tools fulfilling these requirements Ourstudy provided direct evidence supporting the TCMmethodof diagnosing hyperuricemia based on tongue features

However this study has several limitations First theROCanalysiswas performedusing the same sample In futurestudies a test dataset will be collected to confirm the findingsof this study Second given that the use of tongue imagesis a complementary and alternative diagnostic method themethod described in this study should be combined withother available variables associated with hyperuricemia suchas body mass index and alcohol intake We plan to takethis approach in a further study Third because the samplewas carefully selected and patients with underlying diseasesassociated with hyperuricemia were excluded the use oftongue images for the diagnosis of hyperuricemia should berestricted

5 Conclusions

Haar-like features extracted from tongue images differed sig-nificantly between subjects with and without hyperuricemiaThe locations of these features are consistent with interpre-tations of tongue appearance in TCM andWestern medicineindicating the existence of a relationship between tongue rootcolor and hyperuricemia in our sample

Conflict of Interests

The authors declare that they have no conflict of interestsrelated to this work

Authorsrsquo Contribution

Yan Cui completed the statistical analysis and wrote thepaper Shizhong Liao performed critical revision of the paperand HongwuWang performed the mathematical models andmethods of this studyHongyuLiuWenhuaWang andLiqunYin validated health information for all subjects

Acknowledgment

This work was supported in part by the National BasicResearch Program of China (973 Program Grant no2011CB505406)

References

[1] H K Choi K Atkinson E W Karlson W Willett and GCurhan ldquoPurine-rich foods daily and protein intake and therisk of gout in menrdquoThe New England Journal of Medicine vol350 no 11 pp 1093ndash1103 2004

[2] K Kaneko Y Aoyagi T Fukuuchi K Inazawa andNYamaokaldquoTotal purine and purine base content of common foodstuffsfor facilitating nutritional therapy for gout and hyperuricemiardquoBiological and Pharmaceutical Bulletin vol 37 no 5 pp 709ndash721 2014

[3] J T Scott ldquoDrug-induced goutrdquo Baillierersquos Clinical Rheumatol-ogy vol 5 no 1 pp 39ndash60 1991

[4] A Brandstater S Kiechl B Kollerits et al ldquoSex-specificassociation of the putative fructose transporter SLC2A9 variantswith uric acid levels is modified by BMIrdquo Diabetes Care vol 31no 8 pp 1662ndash1667 2008

[5] L-H Lai S-Y Chou F-Y Wu J J-H Chen and H-W KuoldquoRenal dysfunction and hyperuricemia with low blood leadlevels and ethnicity in community-based studyrdquo Science of theTotal Environment vol 401 no 1-3 pp 39ndash43 2008

[6] Q Liu Z Li H Wang et al ldquoHigh prevalence and associatedrisk factors for impaired renal function and urinary abnormal-ities in a rural adult population from Southern Chinardquo PLoSONE vol 7 no 10 Article ID e47100 2012

[7] W-G Fang X-M Huang Y Wang et al ldquoA cross-sectionalstudy of hyperuricemia in state-employees in Beijing preva-lence and risk factorsrdquo National Medical Journal of China vol86 no 25 pp 1764ndash1768 2006

[8] Y S Kim J P Guevara K M Kim H K Choi D F Heitjanand D A Albert ldquoHyperuricemia and coronary heart diseasea systematic review and meta-analysisrdquo Arthritis Care andResearch vol 62 no 2 pp 170ndash180 2010

[9] P C Grayson S Y Kim M LaValley and H K ChoildquoHyperuricemia and incident hypertension a systematic reviewand meta-analysisrdquo Arthritis Care and Research vol 63 no 1pp 102ndash110 2011

[10] M Li W Hou X Zhang L Hu and Z Tang ldquoHyperuricemiaand risk of stroke a systematic review and meta-analysis ofprospective studiesrdquoAtherosclerosis vol 232 no 2 pp 265ndash2702014

[11] M R Carnethon S P Fortmann L Palaniappan B B Dun-can M I Schmidt and L E Chambless ldquoRisk factors forprogression to incident hyperinsulinemia the atherosclerosisrisk in communities study 1987ndash1998rdquo American Journal ofEpidemiology vol 158 no 11 pp 1058ndash1067 2003

10 BioMed Research International

[12] B Liu T Wang H Zhao et al ldquoThe prevalence of hyper-uricemia in china a meta-analysisrdquo BMC Public Health vol 11no 1 article 832 2011

[13] E Krishnan ldquoInteraction of inflammation hyperuricemia andthe prevalence of hypertension among adults free of metabolicsyndromeNHANES 2009ndash2010rdquo Journal of the AmericanHeartAssociation vol 3 no 2 Article ID e000157 2014

[14] B Zhang XWang J You andD Zhang ldquoTongue color analysisfor medical applicationrdquo Evidence-Based Complementary andAlternativeMedicine vol 2013 Article ID 264742 11 pages 2013

[15] B Pang D Zhang and K Wang ldquoTongue image analysis forappendicitis diagnosisrdquo Information Sciences vol 175 no 3 pp160ndash176 2005

[16] J Kim J Son S Jang et al ldquoAvailability of tongue diagno-sis system for assessing tongue coating thickness in patientswith functional dyspepsiardquoEvidence-BasedComplementary andAlternativeMedicine vol 2013 Article ID 348272 6 pages 2013

[17] XWang B Zhang Z Yang HWang andD Zhang ldquoStatisticalanalysis of tongue images for feature extraction and diagnos-ticsrdquo IEEE Transactions on Image Processing vol 22 no 12 pp5336ndash5347 2013

[18] C J Jung J H Nam Y J Jeon and K H Kim ldquoColor distribu-tion differences in the tongue in sleep disorderrdquo Evidence-BasedComplementary and Alternative Medicine vol 2014 Article ID323645 8 pages 2014

[19] A H Friedlander and R Singer ldquoRenal adenocarcinoma ofthe kidney with metastasis to the tonguerdquo The Journal of theAmerican Dental Association vol 97 no 6 pp 989ndash991 1978

[20] K A Amir R K Bobba B Clarke et al ldquoTongue discolorationin an elderly kidney transplant recipient treatment-relatedadverse eventrdquoThe American Journal Geriatric Pharmacother-apy vol 4 no 3 pp 260ndash263 2006

[21] Q Tang J Chen and J Li ldquoAnalysis of the tongue characteristicsof liver depression and kidney insufficiency pid patientsrdquoChinaModern Medicine vol 15 pp 82ndash83 2010

[22] M Yang C-D Li W-N Liang and H Li ldquoCorrelationalstudy on pathological changes of liver qi depression and kidneydeficiency tongue coating exfoliative cells and sex hormonesin perimenopausal syndromerdquo China Journal of TraditionalChinese Medicine and Pharmacy vol 9 pp 1984ndash1986 2011

[23] P Viola and M J Jones ldquoRobust real-time face detectionrdquoInternational Journal of Computer Vision vol 57 no 2 pp 137ndash154 2004

[24] F Valbusa L Bertolini S Bonapace et al ldquoRelation of elevatedserum uric acid levels to incidence of atrial fibrillation inpatients with type 2 diabetes mellitusrdquoThe American Journal ofCardiology vol 112 no 4 pp 499ndash504 2013

[25] D A Grimes and K F Schulz ldquoAn overview of clinical researchthe lay of the landrdquo The Lancet vol 359 no 9300 pp 57ndash612002

[26] D A Grimes and K F Schulz ldquoCompared to what Findingcontrols for case-control studiesrdquoTheLancet vol 365 no 9468pp 1429ndash1433 2005

[27] M M Gorkani and R W Picard ldquo Texture orientation forsorting photos lsquoat a glancersquordquo inProceedings of the 12th IAPR Inter-national Conference on Pattern Recognition vol 1mdashConferenceA Computer Vision amp Image Processing pp 459ndash464 IEEEOctober 1994

[28] R Core Team R A Language and Environment for StatisticalComputing 2012

[29] R Nuzzo ldquoStatistical errorsrdquoNature vol 506 no 7487 pp 150ndash152 2014

[30] R Kanawong T Obafemi-Ajayi J Yu D Xu S Li and YDuan ldquoZheng classification in traditional chinese medicinebased on modified specular-free tongue imagesrdquo in Proceed-ings of the IEEE International Conference on Bioinformaticsand Biomedicine Workshops (BIBMW rsquo12) pp 288ndash294 IEEEOctober 2012

[31] R Kanawong T Obafemi-Ajayi T Ma D Xu S Li andY Duan ldquoAutomated tongue feature extraction for ZHENGclassification in traditional Chinese medicinerdquo Evidence-BasedComplementary and Alternative Medicine vol 2012 Article ID912852 14 pages 2012

[32] V Vitart I Rudan C Hayward et al ldquoSLC2A9 is a newly iden-tified urate transporter influencing serum urate concentrationurate excretion and goutrdquo Nature Genetics vol 40 no 4 pp437ndash442 2008

[33] M Tomita SMizunoH Yamanaka et al ldquoDoes hyperuricemiaaffect mortality A prospective cohort study of Japanese maleworkersrdquo Journal of epidemiologyJapan Epidemiological Associ-ation vol 10 no 6 pp 403ndash409 2000

[34] J Fang andMH Alderman ldquoSerumuric acid and cardiovascu-lar mortality the nhanes i epidemiologic follow-up study 1971ndash1992rdquo Journal of the American Medical Association vol 283 no18 pp 2404ndash2410 2000

[35] O Toprak M Cirit E Esi N Postaci M Yesil and S BayataldquoHyperuricemia as a risk factor for contrast-induced nephropa-thy in patientswith chronic kidney diseaserdquoCatheterization andCardiovascular Interventions vol 67 no 2 pp 227ndash235 2006

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

10 BioMed Research International

[12] B Liu T Wang H Zhao et al ldquoThe prevalence of hyper-uricemia in china a meta-analysisrdquo BMC Public Health vol 11no 1 article 832 2011

[13] E Krishnan ldquoInteraction of inflammation hyperuricemia andthe prevalence of hypertension among adults free of metabolicsyndromeNHANES 2009ndash2010rdquo Journal of the AmericanHeartAssociation vol 3 no 2 Article ID e000157 2014

[14] B Zhang XWang J You andD Zhang ldquoTongue color analysisfor medical applicationrdquo Evidence-Based Complementary andAlternativeMedicine vol 2013 Article ID 264742 11 pages 2013

[15] B Pang D Zhang and K Wang ldquoTongue image analysis forappendicitis diagnosisrdquo Information Sciences vol 175 no 3 pp160ndash176 2005

[16] J Kim J Son S Jang et al ldquoAvailability of tongue diagno-sis system for assessing tongue coating thickness in patientswith functional dyspepsiardquoEvidence-BasedComplementary andAlternativeMedicine vol 2013 Article ID 348272 6 pages 2013

[17] XWang B Zhang Z Yang HWang andD Zhang ldquoStatisticalanalysis of tongue images for feature extraction and diagnos-ticsrdquo IEEE Transactions on Image Processing vol 22 no 12 pp5336ndash5347 2013

[18] C J Jung J H Nam Y J Jeon and K H Kim ldquoColor distribu-tion differences in the tongue in sleep disorderrdquo Evidence-BasedComplementary and Alternative Medicine vol 2014 Article ID323645 8 pages 2014

[19] A H Friedlander and R Singer ldquoRenal adenocarcinoma ofthe kidney with metastasis to the tonguerdquo The Journal of theAmerican Dental Association vol 97 no 6 pp 989ndash991 1978

[20] K A Amir R K Bobba B Clarke et al ldquoTongue discolorationin an elderly kidney transplant recipient treatment-relatedadverse eventrdquoThe American Journal Geriatric Pharmacother-apy vol 4 no 3 pp 260ndash263 2006

[21] Q Tang J Chen and J Li ldquoAnalysis of the tongue characteristicsof liver depression and kidney insufficiency pid patientsrdquoChinaModern Medicine vol 15 pp 82ndash83 2010

[22] M Yang C-D Li W-N Liang and H Li ldquoCorrelationalstudy on pathological changes of liver qi depression and kidneydeficiency tongue coating exfoliative cells and sex hormonesin perimenopausal syndromerdquo China Journal of TraditionalChinese Medicine and Pharmacy vol 9 pp 1984ndash1986 2011

[23] P Viola and M J Jones ldquoRobust real-time face detectionrdquoInternational Journal of Computer Vision vol 57 no 2 pp 137ndash154 2004

[24] F Valbusa L Bertolini S Bonapace et al ldquoRelation of elevatedserum uric acid levels to incidence of atrial fibrillation inpatients with type 2 diabetes mellitusrdquoThe American Journal ofCardiology vol 112 no 4 pp 499ndash504 2013

[25] D A Grimes and K F Schulz ldquoAn overview of clinical researchthe lay of the landrdquo The Lancet vol 359 no 9300 pp 57ndash612002

[26] D A Grimes and K F Schulz ldquoCompared to what Findingcontrols for case-control studiesrdquoTheLancet vol 365 no 9468pp 1429ndash1433 2005

[27] M M Gorkani and R W Picard ldquo Texture orientation forsorting photos lsquoat a glancersquordquo inProceedings of the 12th IAPR Inter-national Conference on Pattern Recognition vol 1mdashConferenceA Computer Vision amp Image Processing pp 459ndash464 IEEEOctober 1994

[28] R Core Team R A Language and Environment for StatisticalComputing 2012

[29] R Nuzzo ldquoStatistical errorsrdquoNature vol 506 no 7487 pp 150ndash152 2014

[30] R Kanawong T Obafemi-Ajayi J Yu D Xu S Li and YDuan ldquoZheng classification in traditional chinese medicinebased on modified specular-free tongue imagesrdquo in Proceed-ings of the IEEE International Conference on Bioinformaticsand Biomedicine Workshops (BIBMW rsquo12) pp 288ndash294 IEEEOctober 2012

[31] R Kanawong T Obafemi-Ajayi T Ma D Xu S Li andY Duan ldquoAutomated tongue feature extraction for ZHENGclassification in traditional Chinese medicinerdquo Evidence-BasedComplementary and Alternative Medicine vol 2012 Article ID912852 14 pages 2012

[32] V Vitart I Rudan C Hayward et al ldquoSLC2A9 is a newly iden-tified urate transporter influencing serum urate concentrationurate excretion and goutrdquo Nature Genetics vol 40 no 4 pp437ndash442 2008

[33] M Tomita SMizunoH Yamanaka et al ldquoDoes hyperuricemiaaffect mortality A prospective cohort study of Japanese maleworkersrdquo Journal of epidemiologyJapan Epidemiological Associ-ation vol 10 no 6 pp 403ndash409 2000

[34] J Fang andMH Alderman ldquoSerumuric acid and cardiovascu-lar mortality the nhanes i epidemiologic follow-up study 1971ndash1992rdquo Journal of the American Medical Association vol 283 no18 pp 2404ndash2410 2000

[35] O Toprak M Cirit E Esi N Postaci M Yesil and S BayataldquoHyperuricemia as a risk factor for contrast-induced nephropa-thy in patientswith chronic kidney diseaserdquoCatheterization andCardiovascular Interventions vol 67 no 2 pp 227ndash235 2006

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Molecular Biology International

GenomicsInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioinformaticsAdvances in

Marine BiologyJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Signal TransductionJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

BioMed Research International

Evolutionary BiologyInternational Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Genetics Research International

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Enzyme Research

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of

Microbiology