jaya algorithm guided procedure to segment tumor from...

Research ArticleJaya Algorithm Guided Procedure to Segment Tumor fromBrain MRI

Suresh Chandra Satapathy 1 and Venkatesan Rajinikanth 2

1School of Computer Engineering Kalinga Institute of Industrial Technology (Deemed to be University)Bhubaneswar 751024 Odisha India2Department of Electronics and Instrumentation Engineering St Josephrsquos College of Engineering Chennai 600119 India

Correspondence should be addressed to Venkatesan Rajinikanth rajinikanthvstjosephsacin

Received 31 May 2018 Accepted 31 October 2018 Published 14 November 2018

Academic Editor Wlodzimierz Ogryczak

Copyright copy 2018 Suresh Chandra Satapathy and Venkatesan Rajinikanth This is an open access article distributed under theCreative Commons Attribution License which permits unrestricted use distribution and reproduction in any medium providedthe original work is properly cited

Brain abnormality is a cause for the chief risk factors in human society with larger morbidity rate Identification of tumor in itsearly stage is essential to provide necessary treatment procedure to save the patient In this work Jaya Algorithm (JA) and OtsursquosFunction (OF) guided method is presented to mine the irregular section of brain MRI recorded with Flair and T2 modality Thiswork implements a two-step process to examine the brain tumor from the axial sagittal and coronal views of the two-dimensional(2D)MRI slicesThis paper presents a detailed evaluation of thresholding procedure with varied threshold levels (Th=2345) skullstripping process beforeafter the thresholding practice and the tumor extraction based on the Chan-Vese approach Superiority ofJA is confirmed among other prominent heuristic approaches found in literatureThe outcome of implemented study confirms thatJaya Algorithm guided method is capable of presenting superior values of Jaccard-Index Dice-Coefficient sensitivity specificityaccuracy and precision on the BRATS 2015 dataset

1 Introduction

Image processing procedure plays a vital role in therapeuticdata examination [1 2] Brain abnormalities such as tumorand stroke are the chief risk factors in human societywhich affects the people irrespective of their gender andgeographical location Usually the severity and the causefor the brain abnormalities are diagnosed in medical clinicsbased on a predefined procedures constructed by experienceddoctorsThis procedures involves in the recording the signalsand the images of brain under a controlled environmentAfter recording the essential information from the patient aprescreening procedure is to be executed using the manual orautomated procedures to assess the cause and the harshnessof the severity After the prescreening process a prescribedclinical evaluation procedure is then executed with thepatient in order to arrange for further handling procedure tocontrol and cure the abnormalities [3]

Normally the condition of the brain can be evaluatedbased on a single-channel and multichannel EEG signals

recorded using the external electrodes or the brain imagesrecorded with MRI and CT imaging approaches [4ndash7]Previous study confirms that information existing in brainimage is essential to categorize and localize the abnormalitycompared to the brain signals [8] If needed a mapping ofbrain signal and image can also be carried to locate the causeand region of the brain abnormality

In literature a considerable amount of conventional andsoft-computing guided measures are widely employed toexamine the brain pictures recorded with MRI and CT [910] The information available in MRI is better compared toCT because of its multimodality since the MRI supports arange of modalities like Flair T1 T1C T2 and DW [11ndash14]The visibility of the abnormality in Flair and T2 modalityis superior to other approaches Hence in this paper thebrain images recorded with the Flair and T2 are used for theexamination

This work implements a two-stage scheme to mine andappraise the tumor using the Jaya Algorithm (JA) andOtsursquos Function (OF) based thresholding and the Chan-Vese

HindawiJournal of OptimizationVolume 2018 Article ID 3738049 12 pageshttpsdoiorg10115520183738049

2 Journal of Optimization

2D slicesof brain

MRI

Pre-processing(Initial imageenhancement)

Post-processing(Extraction of

abnormal section)

Extractedtumor

Ground-truth

Validation Decision

making

Figure 1 Stages in the proposed brain MRI examination scheme

(CV) based segmentation procedure to extract the abnormalsection from the two-dimension (2D) slices of the brainMRI The performance of JA is confirmed against otherheuristic algorithms existing in the literature Correctnessof the proposed approach is also confirmed by computingthe image likeliness measures based on a comparative studylinking the mined tumor section and the ground-truth (GT)image The experimental results of this work confirm thatthe JA assisted procedure tenders improved result on the 2Dbrain slices recorded with Flair and T2 modalities

2 Related Previous Works

In brain imaging literature a significant number of con-ventional and soft-computing (SC) based techniques arediscussed and executed to extract and evaluate the brainabnormality The earlier works also suggests that traditionalapproaches may implement a single-step or two-step proce-dure to extract the abnormal section from the chosen MRIslice [11 15 16]Most of these procedures aremodality specificand works well only for few modality cases such as Flairand T2 To overcome these limitations heuristic algorithmassisted two-step events are broadly adopted to examine thebrain abnormality recorded using MRI of various modalitiesPalani et al (2016) implemented a SC approach to segmentthe soft brain tissues using a two-step procedure basedon Otsursquos thresholding [8] Chaddad (2015) implementedGaussian mixture model technique to support automatedfeature extraction in brain tumor extracted from MRI [17]Dey et al (2015) applied genetic algorithm tuned intervalfilter to remove noise in brain MRI [6] Rajinikanth andSatapathy (2018) implemented a detailed evaluation of thebrain MRI with various segmentation approaches [9] Theworks of Rajinikanth et al (2017) also implemented variousheuristic evaluation procedures to separate the abnormal areafrom brain MRI recorded with a range of modalities [18]The work of Rajinikanth et al (2018) also confirmed thatthe need for image fusion methods to enhance the accuracyin brain MRI evaluation [13] Recently Amin et al (20172018) proposed a distinctive approach as well as the deep-learning procedure to extract the abnormal sections fromthe brain MRI [4 5] Every machine-learning and deep-learning method in the literature discuss about their ownmerit and demerits The brain MRI examination procedurealways needs efficient as well as reliable procedures to mineand evaluate the strange segment from the brain picture [19]

Researchers have already implemented a considerablenumber of swarm-basedevolutionary approaches to exam-ine the brain MRI [9 13]This paper attempts a methodology

to implement the recently developed approach called theJaya Algorithm (JA) The literature also confirms that imple-mentation of JA is quite simple contrast to former heuristicschemes [20ndash23] Further this paper implements the Chan-Vese method to mine the tumor section in preprocessedpicture Finally the superiority of executed approach is con-firmed using a qualified investigation among the extractedtumor section and the GT picture [9]

3 Materials and Methods

This division of the manuscript provides the details of thetwo-step procedure employed to inspect the abnormality ofbrain MRI This section presents the summary of the pre-and postprocessing methods implemented to examine thetest pictures of 2D brain MRI

Figure 1 shows the stages involved in the proposed workThis technique initially considers a 2D MRI slice as the testpictures to be processed The preprocessing stage enhancesthe abnormal region of the test picture by implementing amultithresholding scheme with or without the skull sectionRecent work of Rajinikanth and Satapathy (2018) discussedthat skull-elimination is essential to implement fully auto-mated disease examination schemes [9] The outcome ofthe initial stage is then processed by the postprocessingstage which implements a chosen image separation methodto extort the abnormal section The performance of MRIexamination method is then established by a relative analysisamong extracted tumor section and GT picture Final resultof this scheme is then considered to support the decisionmaking and treatment planning procedure

31 Preprocessing Stage It is the initial step in the two-stage process This implements a multithresholding processto enhance the tumor section of 2D test picture This stageimplements the Jaya Algorithm (JA) assisted Otsursquos between-class variance procedure for different threshold (Th) levels(Th = 2345) on the chosen test image Later the imagequality metrics are computed for these images to discover thebest threshold for the image enhancement

311 JAYA Algorithm JA is a newly invented evolutionaryalgorithm by Rao (2016) [20] and its theory and the workingprinciple can be found in [21ndash23] The chief merit of theJA compared with other considered heuristicevolutionaryapproaches such as the Firefly Algorithm (FA) Teach-ing-Learning Based Optimization (TLBO) Particle-Swarm-Optimization (PSO) Bacterial-Optimization (BFO)

Journal of Optimization 3

and Bat-Algorithm (BA) involves minimal early parametersto be assigned [2 12ndash14 19 24ndash27]

Let G(x)119898119886119909 is the preferred objective function R1=R2are subjective quantities of option [01] N is populationamount (ien=12 N) and i andD indicates the amount ofiterations and dimensions (ie j=12 D) correspondingly

At some occurrence in the algorithm G(x)119887119890119904119905 andG(x)119908119900119903119904119905 signify the finest and worst results reached throughthe contestant in population

The prime equation of JA is presented below

1198831119895119899119894 = 119883119895119899119894 +R1119895119894 (119883119895119887119890119904119905119894 minus10038161003816100381610038161003816119883119895119899119894

10038161003816100381610038161003816)

minusR2119895119894 (119883119895119908119900119903119904119905119894 minus10038161003816100381610038161003816119883119895119899119894

10038161003816100381610038161003816)(1)

where 119883119895119899119894 signify the j119905ℎ variable of n119905ℎ contestant at i119905ℎ

iteration and 1198831119895119899119894 indicate the modernized value In thiswork JA is utilized to discover the maximized value of theOtsursquos between-class variance value for a chosen lsquoThrsquo

Additional details concerning the conventional JA and itsrecent advancements can be found in recent literature [28ndash35]

312 Otsursquos Function From the year 1979 Otsursquos thresh-olding is broadly considered to enhance the traditional andmedical images [36] The theory of Otsu is as followsfor multithresholding problem (Th=2345) input image isseparated into various groups fromH0 andHTh-1 based on thechosen Th The class H0 encloses the gray levels in the range0 to t-1 and class H119879ℎminus1 encircles the gray levels from t to L ndash1 The probability distributions for gray levels H0 and HThminus1can be articulated as [37 38]

1198670 =1199010

1205960 (119905)

119901119905minus11205960 (119905)

119867119879ℎ-1

=119901119905

120596119879ℎ-1 (119905)

119901119871minus1120596119879ℎ-1 (119905)

(2)

where 1205960(119905) = sum119905minus1119894=0 119901119894 120596119879ℎminus1(119905) = sum119871minus1119894=119905 119901119894 and L=256The mean levels m0 m119879ℎminus1 for 1198670 119867119879ℎ-1 can be

articulated as

1198980 =119905minus1

sum119894=0

1198941199011198941205960 (119905)

119898119879ℎminus1 =119871minus1

sum119894=119905

119894119901119894120596119879ℎminus1 (119905)

(3)

The mean intensity (m119879) of total picture can be representedas

119898119879 = 12059601198980 + + 120596119879ℎminus1119898119879ℎminus1

1205960 + + 120596119879ℎminus1 = 1(4)

The objective function for the bilevel thresholding problemcan be expressed as

Maximize 119869 (119879ℎ) = 1205900 + + 120590119879ℎ-1 (5)

where 1205900=1205960( 1198980- 119898119879)2 120590119879ℎ-1 = 120596119879ℎ-1(120583119879ℎ-1- 120583119879)

2Additional details regarding Otsu can be found from [24

25 39ndash41]

313 Realization During the realization of the preprocessingstage the initial JA factors are allotted as follows agents size(N) = 20 number of iterations=2000 dimension of explore(D) =Th and the stopping criterion if the maximized Otsursquosvalue

Initially the 2D slices of Cerebrix and Brainix images ofsize 256x256 pixels are considered for the examination [42]These images are associated with the skull section Hence theproposed paper implements the skull-elimination procedureunder the following conditions (i) skull stripping before thethresholding and (ii) skull stripping after the thresholdingSimilarly in order to confirm the best threshold duringthe initial picture processing various thresholds are chosen(Th=2345) and the superiority of the outcome is assessedusing the image qualitymeasure values such asRMSE PSNRSSIM NCC AD and SC (Satapathy et al 2018) furtherthe CPU runtime in seconds also recorded to choose thethreshold level in order to get the better throughput duringthe image analysis The thresholding procedure is initiallyimplementedwith JA+Otsu Later other heuristic approachesare considered to validate the superiority of JA comparedto the chosen heuristic procedures Finally the abnormalsection of the brain MRI is extracted using the Chan-Vesesegmentation procedure

32 Postprocessing This section implements the Chan-Vese(CV) active-contour practice to mine the irregular sectionfrom preprocessed MRI [9] CV is a bounding box basedsemiautomated technique implemented by Chan and Vese toextract the information from the test pictures [9] RecentlyCV is widely adopted to extract the chosen regions ofmedicinal image under assessment [43ndash45] The working ofCV is similar to the level-set procedure in which the edgesof the box are corrected incessantly based on the allocatediteration quantity If grouping of pixel inside the box is overthe area within the contour will be extracted and displayed Inthis work CV is utilized tomine the irregular section (tumor)from preprocessed brain MRI of Flair and T2 modality [1546]

33 Evaluation The major aim of this paper is to extracttumor in the 2D brain MRI and to compute the features oftumor region for further study In this work two kinds ofdata such as brain MRI with and without the ground truth(GT) are considered Image likeness values such as Jaccard-Index (JI) Dice-Coefficient (DC) False-Positive-Rate (FPR)and False-Negative-Rate (FNR) are computed for MRI withGT [2 9 47]

The arithmetic expression is shown below

119869119868 (119868119866119879 119868119874) = 119868119866119879 cap 119868119874119868119866119879 cup 119868119874

(6)

119863119862 (119868119866119879 119868119874) = 2 (119868119866119879 cap 119868119874)|119868119866119879| cup |119868119874|

(7)

119865119875119877 (119868119866119879 119868119874) = (119868119866119879119868119874)(119868119866119879 cup 119868119874)

(8)


119865119873119877 (119868119866119879 119868119874) = (119868119874119868119866119879)(119868119866119879 cup 119868119874)

(9)

where IGT symbolize theGTand IO symbolizemined regionFurther values like sensitivity (SE) specificity (SP)

accuracy (AC) precision (PR) Balanced-Classification-Rate(BCR) and Balanced-Error-Rate (BER) aremeasuredMath-ematical expression for these parameters is given below

119878119864 = 119879119875(119879119875 + 119865119873)

(10)

119878119875 = 119879119873(119879119873 + 119865119875)

(11)

119860119862 =(119879119875 + 119879119873)

(119879119875 + 119879119873 + 119865119875 + 119865119873)(12)

119875119877 = 119879119875(119879119875 + 119865119875)

(13)

119861119862119877 = 12 (119879119875 (119879119875 + 119865119873) + 119879119873 (119879119873 + 119865119875))

(14)

119861119864119877 = 1 minus 119861119862119877 (15)

where T119873 T119875 F119873 and F119875 indicates true-negative true-positive false-negative and false-positive correspondingly

4 Results and Discussion

This division of the paper presents the investigational out-come of the proposed methodology using the brain MRIof Brainix (256x256 pixels) Cerebrix (256x256 pixels) andBRATS 2015 (216x160 pixels) database [48] The work isexecuted in Matlab7 using a system AMD C70 Dual Core1GHz CPU with 4GB of RAM The work implemented inthis paper is as follows (i) selecting the 2D brain MRI slices(ii) implementation of skull stripping procedure (iii) thresh-olding based enhancement with JA+Otsu (iv) extraction ofthe tumor using CV and (v) computation of picture similarityvalues to validate the implemented procedure

Initially the chosen test images of Brainix dataset isconsidered for the evaluation and the various modality based2D slices of this dataset is depicted in Figure 2 Later askull-elimination procedure discussed by Rajinikanth et al(2017) is implemented for this test pictures and its outcome ispresented in Figure 3 [18]

Figures 3(a) and 3(c) show the skull stripped soft brainregions and Figures 3(b) and 3(d) show the stripped skullsection The gray histogram of Flair modality image ofFigure 2(a) is presented in Figure 4 Figure 4(a) shows thehistogram with the skull section and Figure 4(b) shows thehistogram without the skull From these images it can beobserved that the peak histogram distribution is approx-imately similar in both the cases This confirms that themultithresholding result will be approximately similar forboth the image cases From this it is clear that the skull regionwill not affect the quality of the multithresholding output ofthe brain MRIs Similar results are obtained for other cases

Flai

rT1

T2

(a) Slice1

Flai

rT1

T2

(b) Slice2

Figure 2 Axial view brain MRI of size 256x256 pixels

such as T1 and T2 modalities The visibility of tumor in T1modality is poor hence in this paper only the Flair and T2modality MRIs are considered for the evaluation

Initially in order to know the best threshold level for thepreprocessing of the image the JA+Otsu based approach isimplemented with various threshold values like Th=2345and the equivalent outcomes are described in Figure 5 Itconfirms that the skull stripping task can also be implementedafter the thresholding task To decide the best thresholdlevel for the MRI the image similarity values are calculatedbased on a relative analysis among the original test imageand the thresholded image and its outcomes are describedin Table 1 When Th=2 the visibility of tumor section inthe preprocessed image is very poor When threshold levelincreases the image quality measure as well as the CPU runtime increases From this study it can be noted that Th=3tender enhanced thresholding outcomewithminimised CPUtime compared withTh=4 and 5 Hence in this paper trilevelthreshold is chosen to preprocess the test pictures consideredin this study

The thresholding performance of the JA algorithm is thentested against other heuristicevolutionary approaches suchas FA TLBO PSO BFO and BA and the convergence ofthis optimization search for Th=2 (axial view Flair MRI) ispresented in Figure 6 This result confirms that the averageCPU time taken by the JA is lesser compared to the BFO andPSO algorithms for the chosen test image The FA TLBOand BA offer better CPU time for the chosen images butthe number of initial parameters to be assigned in JA issmaller than these approaches This result confirms that JAcan be chosen to preprocess the brain MRI compared to thealternative approaches chosen in this paper

After implementing the preprocessing with JA+Otsu theCV segmentation procedure is then implemented to extractthe tumor section from the axial view brain MRI Figure 7shows the outcome of the postprocessing section obtained


Flai

rT1

T2

(a) Slice1

Flai

rT1

T2

(b) Skull of Slice1

Flai

rT1

T2

(c) Slice2

Flai

rT1

T2

(d) Skull of Slice2

Figure 3 Skull eliminated brain MRI

0 50 100 150 200 2500

100200300400500600700

(a) Histogram with skull0 50 100 150 200 250

0100200300400500600700

(b) Histogram without skull

Figure 4 Gray histogram of a chosen test image (Flair) with and without skull section

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(a) Th=2

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(b) Th=3

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(c) Th=4

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(d) Th=5

Figure 5 Skull section removal after the multithresholding


Otsu

rsquos fu

nctio

nIteration number

2160

2140

2120

2100

2080

2060

0 50 100 150 200 250 300 350 400 450 500

JAYAFATLBOPSOBFOBA

Figure 6 Convergence of heuristic search (Th=2)

Imag

e1Im

age2

(a) Preprocessed image

Imag

e1Im

age2

(b) Initial bounding box

Imag

e1Im

age2

(c) Final CVIm

age1

Imag

e2(d) Extracted tumor

Figure 7 Results of CV based segmentation

with a sample test picture As shown in Figure 7(b) initiallya bounding box over the tumor region is assigned manuallyWhen the Chan-Vese approach is executed the bounding boxwill shrink toward the inner region in order to identify allthe possible pixel values of the tumor section This shrinkingprocess will be executed based on the assigned iterationvalue after completing the iteration the pixel region boundedby the CV contour is extracted and is presented as thetumor section Figure 7(c) depicts the final contour of the CVand Figure 7(d) shows the extracted tumor section Similarprocedure is implemented for other test images consideredin this paper Further the segmentation performance of theCV is also confirmed against active-contour (AC) and RegionGrowing (RG) techniques existing in the literature

Figure 8 shows the sample images of coronal and sagittalviews of Cerebrix dataset When compared with the axialview skull stripping and the tumor extraction are quitedifficult in these views due to its image complexity Recentworks confirm that an efficient skull-elimination procedurewill help to overcome this difficulty [9] Initially the skull-elimination is implemented to separate the soft brain sectionfrom the high intensity skull using the procedure discussedin [15] Later the preprocessing and the postprocessing

approaches are implemented for these images in order toextract the tumor section

Figure 9 presents the outcome of the proposed techniquewith the sample Cerebrix MRIs Figure 9(a) presents thepreprocessed test picture and Figures 9(b) 9(c) and 9(d)show the extracted tumor sections with CV AC and RGrespectively The Cerebrix dataset does not have the ground-truth (GT) image Hence in order to validate the perfor-mance of proposed approach grand challenge benchmarkimage dataset called the BRATS2015 dataset is considered

The main advantage of the BRATS2015 dataset comparedto the Brainix and Cerebrix is as follows (i) it has a skullstripped 3D brain MRI from which the required amount of2D slices can be extracted (ii) it supports themodalities suchas Flair T1 T1C and T2 and (iii) it has a commonGT pictureoffered by an expert member Due to this reason BRATSimages are widely adopted by most of the researchers to testtheir disease examination tool [11]

Figure 10 shows the sample test images of axial view MRIrecorded with various modalities and its corresponding GTIn this paper the Flair and T2modality images are consideredfor the study Figure 11 shows the JA+Otsu based trilevelthresholding result for the chosen BRATS dataset and also


Table 1 Image quality measures for the chosen test image (Flair) for variousTh values

Test image RMSE PSNR SSIM NCC AD SC CPU timeThreshold2 326324 178578 07674 04546 160188 42512 246264Threshold3 259206 198579 08097 05777 141803 27689 313827Threshold4 173467 233465 08798 07241 87682 18508 588926Threshold5 149659 246288 08912 07688 79169 16496 733028

Cor

onal

vie

wSa

gitta

l vie

w

(a) Slice1

Cor

onal

vie

wSa

gitta

l vie

w

(b) Slice2

Cor

onal

vie

wSa

gitta

l vie

w

(c) Slice3

Figure 8 Chosen test images of coronal and sagittal viewMRIs

the segmentation results Figure 11(a) shows the slice numberFigures 11(b) and 11(d) depict the preprocessed picture andFigures 11(c) and 11(e) show the extracted tumor region withCV technique

Later the superiority of implemented procedure is con-firmed using a comparative study between the extractedtumor section and the GT and the essential image similarityvalues and statistical values are computed for both the Flairand T2 modality cases Tables 2 and 3 show the resultsobtained with the Flair modality test pictures Similar resultsare attained for the T2 modality imagesThese tables confirmthat proposed technique is efficient in attaining better valuesof JI DC sensitivity specificity accuracy and precisioncompared with the segmentation procedures like AC and RGconsidered in this study

The overall performance (average of the image qual-ity measures obtained for the trilevel thresholding) of thealgorithms considered in this paper is depicted in Figure 12and it confirms that the optimization performance of theJA is superior to other methods considered in this paperIn Figure 12 the x-axis numbers 1 to 6 denote the soft-computing methods like JA FA TLBO PSO BFO and BAcorrespondingly

The main limitation of the proposed procedure is asfollows it is a semiautomated approach and requires theassistance of the operator to fix the threshold value during thepreprocessing task and also requires the operators assistance

during the bounding box initiation while implementing theCV segmentation In future the tumor segmentation andevaluation procedure can be automated by implementing therecent procedure known deep-learning approach But thecomplexity in training the required task is more compared tothe approach discussed in this paperThe results of this paperconfirm that proposed approach is very efficient in extractingthe tumor section from various datasets considered in thispaperThis paper also confirms that this approach works wellon Flair and T2 modality MRIs

In future a feature extraction and classification procedurefor the extracted tumor section can be implemented toclassify the brain tumors such as benign and malignantFurther the adopted brain MRI images can be examinedusing the deep-learning procedures based on the NeuralNetwork (NN) [49ndash51] and its outcome can be compared andvalidated with the machine-learning procedure discussed inthis work

5 Conclusions

This paper implements the combination of a preprocessingand a postprocessing to mine the irregular section of thebrain MRI The preprocessing is implemented to enhancethe tumor section using a multithresholding based on theJaya Algorithm and Otsursquos function and the postprocessingimplements the Chan-Vese (CV) segmentation procedure


Axi

alC

oron

alSa

gitta

l

(a) Test picture

Axi

alC

oron

alSa

gitta

l

(b) CV

Axi

alC

oron

alSa

gitta

l

(c) AC

Axi

alC

oron

alSa

gitta

l

(d) RG

Figure 9 Segmented tumor section with CV AC and RG

Table 2 Picture likeness measures obtained for the flair modality images

Image TPR TNR FPR FNR JI DCF100 08196 09946 00054 01804 07628 08655F110 09132 09933 00067 00868 08376 09116F120 09072 09921 00079 00928 08187 09003F130 08446 09926 00074 01554 07357 08477T2100 07533 09953 00047 02467 07071 08284T2110 09006 09936 00064 00994 08293 09067T2120 08433 09921 00079 01567 07610 08643T2130 08114 09930 00070 01886 07127 08322Average 08492 09933 00067 01508 07706 08696


100

110

120

130

(a) Slice (b) Flair (c) T2 (d) T1C (e) T1 (f) G2

Figure 10 Sample test images of BRATS2015

100

110

120

130

(a) Slice (b) Flair (c) Flair-tumor (d)T2 (e)T2-tumor

Figure 11 Segmentation results obtained for BRATS2015 dataset

85

855

86

865

87

875

88

1 2 3 4 5 6

Ove

rall

perf

orm

ance

()

So-computing methods

Figure 12 Performance evaluation of considered heuristic algorithms


Table 3 Picture statistical measures obtained for the flair modality images

Image Sensitivity Specificity Accuracy PrecisionF100 08196 09946 09829 09167F110 09132 09933 09877 09101F120 09072 09921 09863 08935F130 08446 09926 09855 08508T2100 07533 09953 09790 09201T2110 09006 09936 09872 09128T2120 08433 09921 09820 08864T2130 08114 09930 09843 08542Average 08492 09933 09844 08931

to extract the tumor During this work the brain MRIimages are obtained from the benchmark datasets such asBrainix Cerebrix and BRATS2015 Initially an excrementis implemented to recognize the best threshold value forthe chosen problem and the results of the preprocessingconfirm that the trilevel thresholding offers better resultfor the brain tumor evaluation for both the Flair and T2modalities Later the performance of CV is compared againsttheAC andRGproceduresThe segmentation result confirmsthat CV offers better result compared to AC and RG Finallythe overall performance of the proposed approach is testedand confirmed with the BRATS2015 brain MRI databaseThe investigational outcome of this study substantiates thatJA+Otsu and CV based procedure offer improved pictureexcellence measures image likeness measures and imagestatistical measures for the considered dataset In future CVsegmentation can be validated against other segmentationprocedures existing in the medical imaging literature

Data Availability

Previously reported Brain MRI data were used to supportthis study and are available at httpwwwosirix-viewercomdatasets and httphalinriafrhal-00935640 Theseprior studies and datasets are cited at relevant places withinthe text as [11 42 48]

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

References

[1] A Pillai R Soundrapandiyan S Satapathy S C Satapathy K-H Jung and R Krishnan ldquoLocal diagonal extrema numberpattern A new feature descriptor for face recognitionrdquo FutureGeneration Computer Systems vol 81 pp 297ndash306 2018

[2] N S M Raja V Rajinikanth S L Fernandes and S C Sat-apathy ldquoSegmentation of breast thermal images using kapurrsquosentropy and hidden markov random fieldrdquo Journal of MedicalImaging andHealth Informatics vol 7 no 8 pp 1825ndash1829 2017

[3] V Bhateja M Misra and S Urooj ldquoUnsharp maskingapproaches for HVS based enhancement of mammographic

masses A comparative evaluationrdquo Future Generation Com-puter Systems vol 82 pp 176ndash189 2018

[4] J Amin M Sharif M Yasmin and S L Fernandes ldquoAdistinctive approach in brain tumor detection and classificationusing MRIrdquo Pattern Recognition Letters 2017

[5] J Amin M Sharif M Yasmin and S L Fernandes ldquoBig dataanalysis for brain tumor detection Deep convolutional neuralnetworksrdquo Future Generation Computer Systems vol 87 pp290ndash297 2018

[6] NDey A Ashour S Beagumet al ldquoParameter optimization forlocal polynomial approximation based intersection confidenceinterval filter using genetic algorithm an application for brainMRI image de-noisingrdquo Journal of Imaging vol 1 no 1 pp 60ndash84 2015

[7] S L Fernandes V P Gurupur H Lin and R J Martis ldquoA novelfusion approach for early lung cancer detection using computeraided diagnosis techniquesrdquo Journal of Medical Imaging andHealth Informatics vol 7 no 8 pp 1841ndash1850 2017

[8] P T Krishnan P Balasubramanian and C Krishnan ldquoSegmen-tation of brain regions by integrating meta heuristic multilevelthreshold with markov random fieldrdquo Current Medical ImagingReviews vol 12 no 1 pp 4ndash12 2016

[9] V Rajinikanth and S C Satapathy ldquoSegmentation of Ischemicstroke lesion in brain mri based on social group optimizationand fuzzy-Tsallis entropyrdquo Arabian Journal for Science andEngineering vol 43 no 8 pp 4365ndash4378 2018

[10] V Rajinikanth S C Satapathy N Dey and R VijayarajanldquoDWT-PCA image fusion technique to improve segmentationaccuracy in brain tumor analysisrdquo Lecture Notes in ElectricalEngineering vol 471 pp 453ndash462 2018

[11] Menze ldquoThe multimodal brain tumor image segmentationbenchmark (BRATS)rdquo IEEE T Med Imaging vol 34 no 10 pp1993ndash2024 2015

[12] N S Raja S L Fernandes N Dey S C Satapathy and VRajinikanth ldquoContrast enhancedmedicalMRI evaluationusingTsallis entropy and region growing segmentationrdquo Journal ofAmbient Intelligence and Humanized Computing pp 1ndash12 2018

[13] V Rajinikanth N Dey S C Satapathy and A S AshourldquoAn approach to examine Magnetic Resonance Angiographybased on Tsallis entropy and deformable snake modelrdquo FutureGeneration Computer Systems vol 85 pp 160ndash172 2018

[14] V Rajinikanth S L Fernandes B Bhushan Harisha and N RSunder ldquoSegmentation and analysis of brain tumor using tsallisentropy and regularised level setrdquo Lecture Notes in ElectricalEngineering vol 434 pp 313ndash321 2018

[15] I T Roopini M Vasanthi V Rajinikanth M Rekha and MSangeetha ldquoSegmentation of tumor frombrainMRI using fuzzy


entropy and distance regularised level setrdquo in ComputationalSignal Processing and Analysis vol 490 of Lecture Notes inElectrical Engineering pp 297ndash304 Springer Singapore 2018

[16] A Srivastava V Bhateja H Tiwari and S C SatapathyldquoRestoration algorithm for gaussian corrupted MRI using non-local averagingrdquo in Information Systems Design and IntelligentApplications vol 340 of Advances in Intelligent Systems andComputing pp 831ndash840 Springer India New Delhi 2015

[17] A Chaddad ldquoAutomated feature extraction in brain tumor bymagnetic resonance imaging using gaussian mixture modelsrdquoInternational Journal of Biomedical Imaging vol 2015 ArticleID 868031 11 pages 2015

[18] V Rajinikanth N S M Raja and K Kamalanand ldquoFireflyalgorithm assisted segmentation of tumor from brain MRIusing Tsallis function and Markov Random Fieldrdquo ControlEngineering and Applied Informatics vol 19 no 3 pp 97ndash1062017

[19] Y Xin-She Nature-Inspired Metaheuristic Algorithms LuniverPress Frome 2nd edition 2008

[20] R V Rao ldquoJaya a simple and new optimization algorithm forsolving constrained andunconstrained optimization problemsrdquoInternational Journal of Industrial Engineering Computationsvol 7 no 1 pp 19ndash34 2016

[21] R Rao and K More ldquoDesign optimization and analysis ofselected thermal devices using self-adaptive Jaya algorithmrdquoEnergy Conversion and Management vol 140 pp 24ndash35 2017

[22] R V Rao and A Saroj ldquoA self-adaptive multi-populationbased Jaya algorithm for engineering optimizationrdquo Swarm andEvolutionary Computation vol 37 pp 1ndash26 2017

[23] R V Rao and A Saroj ldquoConstrained economic optimizationof shell-and-tube heat exchangers using elitist-Jaya algorithmrdquoEnergy vol 128 pp 785ndash800 2017

[24] N S M Raja V Rajinikanth and K Latha ldquoOtsu basedoptimal multilevel image thresholding using firefly algorithmrdquoModelling and Simulation in Engineering vol 2014 Article ID794574 17 pages 2014

[25] V Rajinikanth and M S Couceiro ldquoOptimal multilevel imagethreshold selection using a novel objective functionrdquo in Infor-mation Systems Design and Intelligent Applications vol 340 ofAdvances in Intelligent Systems and Computing pp 177ndash186Springer India New Delhi 2015

[26] Y Xin-She and H Xingshi ldquoBat algorithm literature reviewand applicationsrdquo International Journal of Bio-Inspired Compu-tation vol 5 no 3 pp 141ndash149 2013

[27] X-S Yang and A H Gandomi ldquoBat algorithm A novelapproach for global engineering optimizationrdquo EngineeringComputations vol 29 no 5 pp 464ndash483 2012

[28] R V Rao ldquoJaya optimization algorithm and its variantsrdquo inJaya An Advanced Optimization Algorithm and its EngineeringApplications Springer Cham 2019

[29] RV RaoD P Rai and J Balic ldquoOptimization of abrasivewater-jet machining process using multi-objective jaya algorithmrdquoMaterials Today Proceedings vol 5 no 2 pp 4930ndash4938 2018

[30] R V Rao and H S Keesari ldquoMulti-team perturbation guidingJaya algorithm for optimization of wind farm layoutrdquo AppliedSoft Computing vol 71 pp 800ndash815 2018

[31] R V Rao A Saroj P Oclon J Taler and D Taler ldquoSingle-and Multi-Objective Design Optimization of Plate-Fin HeatExchangers Using Jaya Algorithmrdquo Heat Transfer Engineeringvol 39 no 13-14 pp 1201ndash1216 2017

[32] R V Rao and A Saroj ldquoAn elitism-based self-adaptive multi-population Jaya algorithm and its applicationsrdquo Soft Computingpp 1ndash24 2018

[33] R V Rao and D P Rai ldquoOptimisation of welding processesusing quasi-oppositional-based Jaya algorithmrdquo Journal ofExperimental amp Theoretical Artificial Intelligence vol 29 no 5Article ID 0952813 pp 1099ndash1117 2017

[34] R V Rao K C More L S Coelho and V C Mariani ldquoMulti-objective optimization of the Stirling heat engine through self-adaptive Jaya algorithmrdquo Journal of Renewable and SustainableEnergy vol 9 no 3 p 033703 2017

[35] R V Rao and G G Waghmare ldquoA new optimization algorithmfor solving complex constraineddesign optimizationproblemsrdquoEngineering Optimization vol 49 no 1 pp 60ndash83 2017

[36] N Otsu ldquoA threshold selection method from gray-level his-togramsrdquo IEEE Transactions on Systems Man and Cyberneticsvol 9 no 1 pp 62ndash66 1979

[37] K Kamalanand and P M Jawahar ldquoComparison of swarmintelligence techniques for estimation of HIV-1 viral loadrdquo IETETechnical Review vol 32 no 3 pp 188ndash195 2015

[38] K Kamalanand and P Mannar Jawahar ldquoComparison of par-ticle swarm and bacterial foraging optimization algorithms fortherapy planning in HIVAIDS patientsrdquo International Journalof Biomathematics vol 9 no 2 1650024 10 pages 2016

[39] V Rajinikanth N S MadhavaRaja S C Satapathy and S LFernandes ldquoOtsursquosmulti-thresholding and active contour snakemodel to segment dermoscopy imagesrdquo Journal of MedicalImaging andHealth Informatics vol 7 no 8 pp 1837ndash1840 2017

[40] S C Satapathy N S M Raja and V Rajinikanth ldquoMulti-level image thresholding using Otsu and chaotic bat algorithmrdquoNeural Computing and Applications vol 29 no 12 pp 1285ndash1307 2018

[41] G H Sudhan R G Aravind K Gowri and V RajinikanthldquoOptic disc segmentation based on Otsursquos thresholding andlevel setrdquo in Proceedings of the 2017 International Conferenceon Computer Communication and Informatics (ICCCI) pp 1ndash5Coimbatore January 2017

[42] ldquoBrain Tumor Database (CEREBRIX and BRAINIX)rdquo httpwwwosirix-viewercomdatasets

[43] N Dey V Rajinikanth A S Ashour and J M R S TavaresldquoSocial group optimization supported segmentation and evalu-ation of skin melanoma imagesrdquo Symmetry vol 10 no 2 2018

[44] Z Li N Dey A S Ashour et al ldquoConvolutional neural networkbased clustering and manifold learning method for diabeticplantar pressure imaging datasetrdquo Journal of Medical Imagingand Health Informatics vol 7 no 3 pp 639ndash651 2017

[45] K Manickavasagam S Sutha and K Kamalanand ldquoDevel-opment of systems for classification of different plasmodiumspecies in thin blood smear microscopic imagesrdquo Journal ofAdvanced Microscopy Research vol 9 no 2 pp 86ndash92 2014

[46] ANamburu S K Samay and S R Edara ldquoSoft fuzzy rough set-based MR brain image segmentationrdquo Applied Soft Computingvol 54 pp 456ndash466 2017

[47] R FMoghaddam andMCheriet ldquoAmulti-scale framework foradaptive binarization of degraded document imagesrdquo PatternRecognition vol 43 no 6 pp 2186ndash2198 2010

[48] ldquoBrats (Brain Tumor Database (BraTS-MICCAI))rdquo httphalinriafrhal-00935640

[49] K S Manic I S Al Naimi F N Hasoon and V RajinikanthldquoJaya algorithm-assisted evaluation of tooth elements using dig-ital bitewing radiography imagesrdquo inComputational Techniques


for Dental Image Analysis Advances in Medical Technologiesand Clinical Practice pp 107ndash128 IGI Global 2019

[50] Y Wang Y Chen N Yang et al ldquoClassification of mice hepaticgranuloma microscopic images based on a deep convolutionalneural networkrdquo Applied Soft Computing vol 74 pp 40ndash502019

[51] A Bakiya K Kamalanand V Rajinikanth R S Nayak andS Kadry ldquoDeep neural network assisted diagnosis of time-frequency transformed electromyogramsrdquo Multimedia Toolsand Applications pp 1ndash17 2018

Hindawiwwwhindawicom Volume 2018

MathematicsJournal of


Mathematical Problems in Engineering

Applied MathematicsJournal of


Probability and StatisticsHindawiwwwhindawicom Volume 2018

Journal of


Mathematical PhysicsAdvances in

Complex AnalysisJournal of


OptimizationJournal of



Engineering Mathematics

International Journal of


Operations ResearchAdvances in

Journal of


Function SpacesAbstract and Applied AnalysisHindawiwwwhindawicom Volume 2018

International Journal of Mathematics and Mathematical Sciences


Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Hindawiwwwhindawicom Volume 2018Volume 2018

Numerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisAdvances inAdvances in Discrete Dynamics in

Nature and SocietyHindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom

Dierential EquationsInternational Journal of

Volume 2018


Decision SciencesAdvances in


AnalysisInternational Journal of


Stochastic AnalysisInternational Journal of

Submit your manuscripts atwwwhindawicom


2D slicesof brain

MRI

Pre-processing(Initial imageenhancement)

Post-processing(Extraction of

abnormal section)

Extractedtumor

Ground-truth

Validation Decision

making

Figure 1 Stages in the proposed brain MRI examination scheme

(CV) based segmentation procedure to extract the abnormalsection from the two-dimension (2D) slices of the brainMRI The performance of JA is confirmed against otherheuristic algorithms existing in the literature Correctnessof the proposed approach is also confirmed by computingthe image likeliness measures based on a comparative studylinking the mined tumor section and the ground-truth (GT)image The experimental results of this work confirm thatthe JA assisted procedure tenders improved result on the 2Dbrain slices recorded with Flair and T2 modalities

2 Related Previous Works

In brain imaging literature a significant number of con-ventional and soft-computing (SC) based techniques arediscussed and executed to extract and evaluate the brainabnormality The earlier works also suggests that traditionalapproaches may implement a single-step or two-step proce-dure to extract the abnormal section from the chosen MRIslice [11 15 16]Most of these procedures aremodality specificand works well only for few modality cases such as Flairand T2 To overcome these limitations heuristic algorithmassisted two-step events are broadly adopted to examine thebrain abnormality recorded using MRI of various modalitiesPalani et al (2016) implemented a SC approach to segmentthe soft brain tissues using a two-step procedure basedon Otsursquos thresholding [8] Chaddad (2015) implementedGaussian mixture model technique to support automatedfeature extraction in brain tumor extracted from MRI [17]Dey et al (2015) applied genetic algorithm tuned intervalfilter to remove noise in brain MRI [6] Rajinikanth andSatapathy (2018) implemented a detailed evaluation of thebrain MRI with various segmentation approaches [9] Theworks of Rajinikanth et al (2017) also implemented variousheuristic evaluation procedures to separate the abnormal areafrom brain MRI recorded with a range of modalities [18]The work of Rajinikanth et al (2018) also confirmed thatthe need for image fusion methods to enhance the accuracyin brain MRI evaluation [13] Recently Amin et al (20172018) proposed a distinctive approach as well as the deep-learning procedure to extract the abnormal sections fromthe brain MRI [4 5] Every machine-learning and deep-learning method in the literature discuss about their ownmerit and demerits The brain MRI examination procedurealways needs efficient as well as reliable procedures to mineand evaluate the strange segment from the brain picture [19]

Researchers have already implemented a considerablenumber of swarm-basedevolutionary approaches to exam-ine the brain MRI [9 13]This paper attempts a methodology

to implement the recently developed approach called theJaya Algorithm (JA) The literature also confirms that imple-mentation of JA is quite simple contrast to former heuristicschemes [20ndash23] Further this paper implements the Chan-Vese method to mine the tumor section in preprocessedpicture Finally the superiority of executed approach is con-firmed using a qualified investigation among the extractedtumor section and the GT picture [9]

3 Materials and Methods

This division of the manuscript provides the details of thetwo-step procedure employed to inspect the abnormality ofbrain MRI This section presents the summary of the pre-and postprocessing methods implemented to examine thetest pictures of 2D brain MRI

Figure 1 shows the stages involved in the proposed workThis technique initially considers a 2D MRI slice as the testpictures to be processed The preprocessing stage enhancesthe abnormal region of the test picture by implementing amultithresholding scheme with or without the skull sectionRecent work of Rajinikanth and Satapathy (2018) discussedthat skull-elimination is essential to implement fully auto-mated disease examination schemes [9] The outcome ofthe initial stage is then processed by the postprocessingstage which implements a chosen image separation methodto extort the abnormal section The performance of MRIexamination method is then established by a relative analysisamong extracted tumor section and GT picture Final resultof this scheme is then considered to support the decisionmaking and treatment planning procedure

31 Preprocessing Stage It is the initial step in the two-stage process This implements a multithresholding processto enhance the tumor section of 2D test picture This stageimplements the Jaya Algorithm (JA) assisted Otsursquos between-class variance procedure for different threshold (Th) levels(Th = 2345) on the chosen test image Later the imagequality metrics are computed for these images to discover thebest threshold for the image enhancement

311 JAYA Algorithm JA is a newly invented evolutionaryalgorithm by Rao (2016) [20] and its theory and the workingprinciple can be found in [21ndash23] The chief merit of theJA compared with other considered heuristicevolutionaryapproaches such as the Firefly Algorithm (FA) Teach-ing-Learning Based Optimization (TLBO) Particle-Swarm-Optimization (PSO) Bacterial-Optimization (BFO)






1198831119895119899119894 = 119883119895119899119894 +R1119895119894 (119883119895119887119890119904119905119894 minus10038161003816100381610038161003816119883119895119899119894

10038161003816100381610038161003816)

minusR2119895119894 (119883119895119908119900119903119904119905119894 minus10038161003816100381610038161003816119883119895119899119894

10038161003816100381610038161003816)(1)





1198670 =1199010

1205960 (119905)

119901119905minus11205960 (119905)

119867119879ℎ-1

=119901119905

120596119879ℎ-1 (119905)

119901119871minus1120596119879ℎ-1 (119905)

(2)


articulated as

1198980 =119905minus1

sum119894=0

1198941199011198941205960 (119905)

119898119879ℎminus1 =119871minus1

sum119894=119905

119894119901119894120596119879ℎminus1 (119905)

(3)


119898119879 = 12059601198980 + + 120596119879ℎminus1119898119879ℎminus1

1205960 + + 120596119879ℎminus1 = 1(4)


Maximize 119869 (119879ℎ) = 1205900 + + 120590119879ℎ-1 (5)

where 1205900=1205960( 1198980- 119898119879)2 120590119879ℎ-1 = 120596119879ℎ-1(120583119879ℎ-1- 120583119879)


25 39ndash41]






119869119868 (119868119866119879 119868119874) = 119868119866119879 cap 119868119874119868119866119879 cup 119868119874

(6)

119863119862 (119868119866119879 119868119874) = 2 (119868119866119879 cap 119868119874)|119868119866119879| cup |119868119874|

(7)

119865119875119877 (119868119866119879 119868119874) = (119868119866119879119868119874)(119868119866119879 cup 119868119874)

(8)


119865119873119877 (119868119866119879 119868119874) = (119868119874119868119866119879)(119868119866119879 cup 119868119874)

(9)



119878119864 = 119879119875(119879119875 + 119865119873)

(10)

119878119875 = 119879119873(119879119873 + 119865119875)

(11)

119860119862 =(119879119875 + 119879119873)

(119879119875 + 119879119873 + 119865119875 + 119865119873)(12)

119875119877 = 119879119875(119879119875 + 119865119875)

(13)

119861119862119877 = 12 (119879119875 (119879119875 + 119865119873) + 119879119873 (119879119873 + 119865119875))

(14)

119861119864119877 = 1 minus 119861119862119877 (15)






Flai

rT1

T2

(a) Slice1

Flai

rT1

T2

(b) Slice2







Flai

rT1

T2

(a) Slice1

Flai

rT1

T2

(b) Skull of Slice1

Flai

rT1

T2

(c) Slice2

Flai

rT1

T2

(d) Skull of Slice2


0 50 100 150 200 2500

100200300400500600700


0100200300400500600700



Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(a) Th=2

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(b) Th=3

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(c) Th=4

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(d) Th=5



Otsu

rsquos fu

nctio

nIteration number

2160

2140

2120

2100

2080

2060

0 50 100 150 200 250 300 350 400 450 500

JAYAFATLBOPSOBFOBA


Imag

e1Im

age2


Imag

e1Im

age2


Imag

e1Im

age2

(c) Final CVIm

age1

Imag












Cor

onal

vie

wSa

gitta

l vie

w

(a) Slice1

Cor

onal

vie

wSa

gitta

l vie

w

(b) Slice2

Cor

onal

vie

wSa

gitta

l vie

w

(c) Slice3








5 Conclusions



Axi

alC

oron

alSa

gitta

l

(a) Test picture

Axi

alC

oron

alSa

gitta

l

(b) CV

Axi

alC

oron

alSa

gitta

l

(c) AC

Axi

alC

oron

alSa

gitta

l

(d) RG





100

110

120

130



100

110

120

130



85

855

86

865

87

875

88

1 2 3 4 5 6

Ove

rall

perf

orm

ance

()







Data Availability




References
































































Journal of












Journal of







Volume 2018






Volume 2018













1198831119895119899119894 = 119883119895119899119894 +R1119895119894 (119883119895119887119890119904119905119894 minus10038161003816100381610038161003816119883119895119899119894

10038161003816100381610038161003816)

minusR2119895119894 (119883119895119908119900119903119904119905119894 minus10038161003816100381610038161003816119883119895119899119894

10038161003816100381610038161003816)(1)





1198670 =1199010

1205960 (119905)

119901119905minus11205960 (119905)

119867119879ℎ-1

=119901119905

120596119879ℎ-1 (119905)

119901119871minus1120596119879ℎ-1 (119905)

(2)


articulated as

1198980 =119905minus1

sum119894=0

1198941199011198941205960 (119905)

119898119879ℎminus1 =119871minus1

sum119894=119905

119894119901119894120596119879ℎminus1 (119905)

(3)


119898119879 = 12059601198980 + + 120596119879ℎminus1119898119879ℎminus1

1205960 + + 120596119879ℎminus1 = 1(4)


Maximize 119869 (119879ℎ) = 1205900 + + 120590119879ℎ-1 (5)

where 1205900=1205960( 1198980- 119898119879)2 120590119879ℎ-1 = 120596119879ℎ-1(120583119879ℎ-1- 120583119879)


25 39ndash41]






119869119868 (119868119866119879 119868119874) = 119868119866119879 cap 119868119874119868119866119879 cup 119868119874

(6)

119863119862 (119868119866119879 119868119874) = 2 (119868119866119879 cap 119868119874)|119868119866119879| cup |119868119874|

(7)

119865119875119877 (119868119866119879 119868119874) = (119868119866119879119868119874)(119868119866119879 cup 119868119874)

(8)


119865119873119877 (119868119866119879 119868119874) = (119868119874119868119866119879)(119868119866119879 cup 119868119874)

(9)



119878119864 = 119879119875(119879119875 + 119865119873)

(10)

119878119875 = 119879119873(119879119873 + 119865119875)

(11)

119860119862 =(119879119875 + 119879119873)

(119879119875 + 119879119873 + 119865119875 + 119865119873)(12)

119875119877 = 119879119875(119879119875 + 119865119875)

(13)

119861119862119877 = 12 (119879119875 (119879119875 + 119865119873) + 119879119873 (119879119873 + 119865119875))

(14)

119861119864119877 = 1 minus 119861119862119877 (15)






Flai

rT1

T2

(a) Slice1

Flai

rT1

T2

(b) Slice2







Flai

rT1

T2

(a) Slice1

Flai

rT1

T2

(b) Skull of Slice1

Flai

rT1

T2

(c) Slice2

Flai

rT1

T2

(d) Skull of Slice2


0 50 100 150 200 2500

100200300400500600700


0100200300400500600700



Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(a) Th=2

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(b) Th=3

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(c) Th=4

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(d) Th=5



Otsu

rsquos fu

nctio

nIteration number

2160

2140

2120

2100

2080

2060

0 50 100 150 200 250 300 350 400 450 500

JAYAFATLBOPSOBFOBA


Imag

e1Im

age2


Imag

e1Im

age2


Imag

e1Im

age2

(c) Final CVIm

age1

Imag












Cor

onal

vie

wSa

gitta

l vie

w

(a) Slice1

Cor

onal

vie

wSa

gitta

l vie

w

(b) Slice2

Cor

onal

vie

wSa

gitta

l vie

w

(c) Slice3








5 Conclusions



Axi

alC

oron

alSa

gitta

l

(a) Test picture

Axi

alC

oron

alSa

gitta

l

(b) CV

Axi

alC

oron

alSa

gitta

l

(c) AC

Axi

alC

oron

alSa

gitta

l

(d) RG





100

110

120

130



100

110

120

130



85

855

86

865

87

875

88

1 2 3 4 5 6

Ove

rall

perf

orm

ance

()







Data Availability




References
































































Journal of












Journal of







Volume 2018






Volume 2018









119865119873119877 (119868119866119879 119868119874) = (119868119874119868119866119879)(119868119866119879 cup 119868119874)

(9)



119878119864 = 119879119875(119879119875 + 119865119873)

(10)

119878119875 = 119879119873(119879119873 + 119865119875)

(11)

119860119862 =(119879119875 + 119879119873)

(119879119875 + 119879119873 + 119865119875 + 119865119873)(12)

119875119877 = 119879119875(119879119875 + 119865119875)

(13)

119861119862119877 = 12 (119879119875 (119879119875 + 119865119873) + 119879119873 (119879119873 + 119865119875))

(14)

119861119864119877 = 1 minus 119861119862119877 (15)






Flai

rT1

T2

(a) Slice1

Flai

rT1

T2

(b) Slice2







Flai

rT1

T2

(a) Slice1

Flai

rT1

T2

(b) Skull of Slice1

Flai

rT1

T2

(c) Slice2

Flai

rT1

T2

(d) Skull of Slice2


0 50 100 150 200 2500

100200300400500600700


0100200300400500600700



Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(a) Th=2

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(b) Th=3

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(c) Th=4

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(d) Th=5



Otsu

rsquos fu

nctio

nIteration number

2160

2140

2120

2100

2080

2060

0 50 100 150 200 250 300 350 400 450 500

JAYAFATLBOPSOBFOBA


Imag

e1Im

age2


Imag

e1Im

age2


Imag

e1Im

age2

(c) Final CVIm

age1

Imag












Cor

onal

vie

wSa

gitta

l vie

w

(a) Slice1

Cor

onal

vie

wSa

gitta

l vie

w

(b) Slice2

Cor

onal

vie

wSa

gitta

l vie

w

(c) Slice3








5 Conclusions



Axi

alC

oron

alSa

gitta

l

(a) Test picture

Axi

alC

oron

alSa

gitta

l

(b) CV

Axi

alC

oron

alSa

gitta

l

(c) AC

Axi

alC

oron

alSa

gitta

l

(d) RG





100

110

120

130



100

110

120

130



85

855

86

865

87

875

88

1 2 3 4 5 6

Ove

rall

perf

orm

ance

()







Data Availability




References
































































Journal of












Journal of







Volume 2018






Volume 2018









Flai

rT1

T2

(a) Slice1

Flai

rT1

T2

(b) Skull of Slice1

Flai

rT1

T2

(c) Slice2

Flai

rT1

T2

(d) Skull of Slice2


0 50 100 150 200 2500

100200300400500600700


0100200300400500600700



Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(a) Th=2

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(b) Th=3

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(c) Th=4

Pre-

proc

esse

d im

age

Skul

lstr

ippe

d im

age

Skul

l sec

tion

(d) Th=5



Otsu

rsquos fu

nctio

nIteration number

2160

2140

2120

2100

2080

2060

0 50 100 150 200 250 300 350 400 450 500

JAYAFATLBOPSOBFOBA


Imag

e1Im

age2


Imag

e1Im

age2


Imag

e1Im

age2

(c) Final CVIm

age1

Imag












Cor

onal

vie

wSa

gitta

l vie

w

(a) Slice1

Cor

onal

vie

wSa

gitta

l vie

w

(b) Slice2

Cor

onal

vie

wSa

gitta

l vie

w

(c) Slice3








5 Conclusions



Axi

alC

oron

alSa

gitta

l

(a) Test picture

Axi

alC

oron

alSa

gitta

l

(b) CV

Axi

alC

oron

alSa

gitta

l

(c) AC

Axi

alC

oron

alSa

gitta

l

(d) RG





100

110

120

130



100

110

120

130



85

855

86

865

87

875

88

1 2 3 4 5 6

Ove

rall

perf

orm

ance

()







Data Availability




References
































































Journal of












Journal of







Volume 2018






Volume 2018









Otsu

rsquos fu

nctio

nIteration number

2160

2140

2120

2100

2080

2060

0 50 100 150 200 250 300 350 400 450 500

JAYAFATLBOPSOBFOBA


Imag

e1Im

age2


Imag

e1Im

age2


Imag

e1Im

age2

(c) Final CVIm

age1

Imag












Cor

onal

vie

wSa

gitta

l vie

w

(a) Slice1

Cor

onal

vie

wSa

gitta

l vie

w

(b) Slice2

Cor

onal

vie

wSa

gitta

l vie

w

(c) Slice3








5 Conclusions



Axi

alC

oron

alSa

gitta

l

(a) Test picture

Axi

alC

oron

alSa

gitta

l

(b) CV

Axi

alC

oron

alSa

gitta

l

(c) AC

Axi

alC

oron

alSa

gitta

l

(d) RG





100

110

120

130



100

110

120

130



85

855

86

865

87

875

88

1 2 3 4 5 6

Ove

rall

perf

orm

ance

()







Data Availability




References
































































Journal of












Journal of







Volume 2018






Volume 2018











Cor

onal

vie

wSa

gitta

l vie

w

(a) Slice1

Cor

onal

vie

wSa

gitta

l vie

w

(b) Slice2

Cor

onal

vie

wSa

gitta

l vie

w

(c) Slice3








5 Conclusions



Axi

alC

oron

alSa

gitta

l

(a) Test picture

Axi

alC

oron

alSa

gitta

l

(b) CV

Axi

alC

oron

alSa

gitta

l

(c) AC

Axi

alC

oron

alSa

gitta

l

(d) RG





100

110

120

130



100

110

120

130



85

855

86

865

87

875

88

1 2 3 4 5 6

Ove

rall

perf

orm

ance

()







Data Availability




References
































































Journal of












Journal of







Volume 2018






Volume 2018









Axi

alC

oron

alSa

gitta

l

(a) Test picture

Axi

alC

oron

alSa

gitta

l

(b) CV

Axi

alC

oron

alSa

gitta

l

(c) AC

Axi

alC

oron

alSa

gitta

l

(d) RG





100

110

120

130



100

110

120

130



85

855

86

865

87

875

88

1 2 3 4 5 6

Ove

rall

perf

orm

ance

()







Data Availability




References
































































Journal of












Journal of







Volume 2018






Volume 2018









100

110

120

130



100

110

120

130



85

855

86

865

87

875

88

1 2 3 4 5 6

Ove

rall

perf

orm

ance

()







Data Availability




References
































































Journal of












Journal of







Volume 2018






Volume 2018












Data Availability




References
































































Journal of












Journal of







Volume 2018






Volume 2018























































Journal of












Journal of







Volume 2018






Volume 2018








jaya algorithm guided procedure to segment tumor from...

Documents