binning strategies for tissue texture extraction in dicom images cti students: bikash bhattacharyya,...

16
Binning Strategies for Tissue Texture Binning Strategies for Tissue Texture Extraction in DICOM Images Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob Furst Advisors: Dr. Daniela Raicu, Dr. Jacob Furst Submitted To: RSNA Conference ‘05, Chicago Submitted To: RSNA Conference ‘05, Chicago , , IL IL

Post on 21-Dec-2015

214 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

Binning Strategies for Tissue Texture Binning Strategies for Tissue Texture Extraction in DICOM ImagesExtraction in DICOM Images

CTI Students: Bikash Bhattacharyya, Kriti JauharCTI Students: Bikash Bhattacharyya, Kriti Jauhar

Advisors: Dr. Daniela Raicu, Dr. Jacob FurstAdvisors: Dr. Daniela Raicu, Dr. Jacob Furst

Submitted To: RSNA Conference ‘05, ChicagoSubmitted To: RSNA Conference ‘05, Chicago,, IL IL

Page 2: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

Why Binning ?Why Binning ?

BACK BONE

0

2

4

6

8

10

32 64 128 256 512 1024

BIN

SEC

ON

DS CPU

ELASPED

HEART

0

0.5

1

1.5

2

32 64 128 256 512 1024

BIN

SE

CO

ND

S CPU

ELAPSED

Binning Definition:Binning Definition: Putting gray-levels into bins for image compression Putting gray-levels into bins for image compression e.g. 1,2,3,4 gray levels in Bin 1 e.g. 1,2,3,4 gray levels in Bin 1 5,6,7,8 gray levels in Bin 2 5,6,7,8 gray levels in Bin 2DICOM Images – 12 Bit - 4096 intensitiesDICOM Images – 12 Bit - 4096 intensitiesTexture Feature Calculation – All intensities - SLOWTexture Feature Calculation – All intensities - SLOWBinning allows for additional flexibility to trade off largeBinning allows for additional flexibility to trade off large intensity ranges against computational speed intensity ranges against computational speed

COMPUTATION PERFORMANCECOMPUTATION PERFORMANCE

Page 3: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

Linear BinningLinear Binning

Linear Binning - Linear Binning - Bins of equal size Bins of equal size 256 bins for DICOM images produces bin 256 bins for DICOM images produces bin ranges [ ranges [0..15] , [16..31] ,…[4081..4096]0..15] , [16..31] ,…[4081..4096]

Quick and Efficient approachQuick and Efficient approach Pre -processing step for Harlick texture feature Pre -processing step for Harlick texture feature calculation calculation Promising results for classification of tissues Promising results for classification of tissues based on Haralick texture featuresbased on Haralick texture features

Page 4: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

Disadvantages of Linear BinningDisadvantages of Linear Binning

Soft tissues with similar intensities may end up in the same bin with linear binning Soft tissues misclassification Accuracy of Liver and Spleen not very high Computed Tomography images contain low number of pixel in the range [1500 – 4096]

Non-Linear Binning – Is it possible to improve accuracy of soft tissues?

Page 5: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

Analysis of Linear Binning (contd.)Analysis of Linear Binning (contd.)

Liver SpleenSpleenEXAMPLESEXAMPLES

Page 6: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

PROCESS SUMMARYPROCESS SUMMARY

IMAGE DB(1374 Images from

2 Patients)BINNING

LINEAR BINNINGBINS= 256

NON LINEAR BINNING

CLIPPED BINNING

BINS= 258K-MEANS CLUSTERING

K=256EUCLIDEAN DISTANCE

CO-OCCURRENCE MATRICES & TEXTURE

DESCRIPTORS

CLASSIFICATION MODEL

(DECISION TREES)

EVALUATE RESULTS

Page 7: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

Two Approaches of Non-LinearTwo Approaches of Non-Linear BinningBinning

Clipped Binning based on visual inspection of gray levelsClipped Binning based on visual inspection of gray levels Range [0, 856] mapped to Bin 1 Range [0, 856] mapped to Bin 1 Range [1368 , 4096] mapped to Bin 258Range [1368 , 4096] mapped to Bin 258 Range [856, 1368] mapped to 256 linear binsRange [856, 1368] mapped to 256 linear bins e.g. 856 to 858 gray levels in Bin 123e.g. 856 to 858 gray levels in Bin 123

Non Linear Binning based on K-Means Clustering

256 Clusters – Compare results of 256 linear-bins Distance Measure – Euclidean Clusters of Gray Level Ranges Gray Level ranges form Non-linear Bins

Page 8: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

Process FlowProcess Flow

Non-linear Binning using K-Means

PATIENT IMAGES141 FILES

COMPUTE FREQUENCY OF GRAY LEVELS FOR EACH

IMAGE

4096 BY 141 VECTORS REPRESENTING ALL IMAGES

K-MEANS CLUSTERS IDENTIFICATION

IS ANY GRAY LEVEL CLUSTER LESS THAN MIN

BIN SIZE?

COMBINE WITH NEXT GRAY LEVEL RANGE

BIN SELECTED IMAGE

DISTANCE MEASURENO OF CLUSTERS

NO

Page 9: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

K-Means K-Means

K =256K =256 141 Dimensions/Images141 Dimensions/Images 4096 points/Gray Levels4096 points/Gray Levels Initial Points /Random CentroidsInitial Points /Random Centroids Similarity Metric = Euclidean DistanceSimilarity Metric = Euclidean Distance

IssuesIssues 263 Unique Gray Levels Identified263 Unique Gray Levels Identified Multiple Gray Levels – Identified in one Cluster Multiple Gray Levels – Identified in one Cluster e.g. Cluster 14 e.g. Cluster 14

has gray levels from 462 to 884 Cluster 14 also has gray levels has gray levels from 462 to 884 Cluster 14 also has gray levels from 1540 to 1542from 1540 to 1542

CLUSTERCLUSTER Start GLStart GL End GLEnd GL

1414 462462 884884

1414 13631363 13651365

1414 13671367 15371537

1414 15391539 15391539

1414 15431543 15431543

139139 15381538 15381538

139139 15401540 15421542

139139 15441544 40954095

Page 10: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

Classification Results Classification Results

TRAINING SETTRAINING SETTRAINING-K-Means Spleen BackboneKidney Heart Liver TotalSensitivity 92.23% 98.10% 92.36% 96.38% 94.08% 94.63%Specificity 100.00% 96.46% 99.48% 99.09% 99.87% 98.98%Precision 91.35% 95.01% 97.08% 95.00% 99.31% 95.55%Accuracy 98.12% 97.13% 98.01% 98.68% 98.90% 98.17%

 TRAINING-Clipped Spleen BackboneKidney Heart Liver TotalSensitivity 88.05% 99.19% 92.59% 97.89% 89.42% 93.43%Specificity 99.07% 99.81% 98.19% 99.35% 100.00% 99.28%Precision 95.24% 99.73% 89.93% 96.53% 83.04% 92.89%Accuracy 97.14% 99.56% 98.02% 99.12% 96.70% 98.11%

 TRAINING Linear Binning Spleen Backbone Kidney Heart Liver TotalSensitivity 79.50% 99.70% 92.70% 84.60% 80.00% 87.30%Specificity 99.50% 99.50% 97.90% 98.50% 96.90% 98.46%Precision 73.60% 99.20% 89.70% 90.60% 83.80% 87.38%Accuracy 94.10% 99.60% 97.10% 96.50% 94.10% 96.28%

Page 11: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

Classification Results Classification Results

TESTING SETTESTING SET

TESTING-KMEANS Liver Backbone Kidney Heart Spleen TotalSensitivity 91.03% 93.97% 88.00% 84.15% 74.58% 86.35%Specificity 96.99% 96.72% 97.61% 98.06% 100.00% 97.88%Precision 86.59% 95.90% 68.75% 90.79% 75.86% 83.58%Accuracy 95.94% 95.49% 97.52% 95.49% 93.45% 95.58%

 TESTING-Clipped Liver Backbone Kidney Heart Spleen TotalSensitivity 52.17% 98.84% 76.81% 85.25% 31.21% 68.85%Specificity 100.00% 92.52% 95.47% 94.81% 97.85% 96.17%Precision 15.58% 88.54% 73.61% 70.27% 84.62% 66.52%Accuracy 83.91% 94.85% 93.13% 93.56% 77.68% 88.63%

TESTING- Linear Binning Liver Backbone Kidney Heart Spleen TotalSensitivity 73.80% 100.00% 86.20% 73.60% 70.50% 80.82%Specificity 95.90% 97.60% 97.80% 97.20% 95.10% 96.72%Precision 76.20% 96.80% 87.50% 84.10% 62.00% 81.32%Accuracy 92.50% 98.60% 96.00% 93.20% 92.50% 94.56%

Page 12: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

Graphical User InterfaceGraphical User Interface

Page 13: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

ConclusionConclusion

Non-Linear Binning with K-Means gave Non-Linear Binning with K-Means gave us the best overall results ( 86.35%) us the best overall results ( 86.35%)

Results for Liver and Spleen improved from Results for Liver and Spleen improved from 73.80% to 91.03% for liver and 70.50% to 74.58% 73.80% to 91.03% for liver and 70.50% to 74.58% for spleenfor spleen

Clipped Binning performed poorly on testing set Clipped Binning performed poorly on testing set with overall sensitivity of only with overall sensitivity of only 68.85%68.85%

Results with K-Means improved over Linear Results with K-Means improved over Linear Binning Binning

Page 14: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

Future WorkFuture Work

Experimenting with bins other than 256 such as: Experimenting with bins other than 256 such as: 64, 128 etc.64, 128 etc.

Exploring other similarity measures such as: Exploring other similarity measures such as: Jeffrey Divergence, Mahalanobis Distance etc.Jeffrey Divergence, Mahalanobis Distance etc.

Testing other classification algorithms besides Testing other classification algorithms besides decision trees, such as: Neural Networks, decision trees, such as: Neural Networks, Bayesian Networks, Logistic Regression etc.Bayesian Networks, Logistic Regression etc.

Page 15: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

ReferencesReferences [1][1] M. Kalinin, D. S. Raicu, J. D. Furst, D. S. ChanninM. Kalinin, D. S. Raicu, J. D. Furst, D. S. Channin,,, , " A Classification " A Classification

Approach for Anatomical Regions Segmentation"Approach for Anatomical Regions Segmentation", The IEEE International , The IEEE International Conference on Image Processing (ICIPConference on Image Processing (ICIP)), September 11-14, 2005. (submitted), September 11-14, 2005. (submitted)

[2] D. Channin, D. S. Raicu, J. D. Furst, D. H. Xu, L. Lilly, C. Limpsangsri, [2] D. Channin, D. S. Raicu, J. D. Furst, D. H. Xu, L. Lilly, C. Limpsangsri, "Classification of Tissues in Computed Tomography using Decision Trees""Classification of Tissues in Computed Tomography using Decision Trees",, RSNA, RSNA, DECEMBER, 2004.DECEMBER, 2004.

[3] R.M. Haralick, K. Shanmugam, and I. Dinstein, “[3] R.M. Haralick, K. Shanmugam, and I. Dinstein, “Textural Features for Image Textural Features for Image ClassificationClassification”, IEEE Trans. on Systems,Man, and Cybernetics, vol. Smc-3, ”, IEEE Trans. on Systems,Man, and Cybernetics, vol. Smc-3, no.6, pp. 610-621, 1973.no.6, pp. 610-621, 1973.

[4] N. M. Nasrabadi and R. A. King., “[4] N. M. Nasrabadi and R. A. King., “Image Coding Using Vector Quantization: Image Coding Using Vector Quantization: A reviewA review” , IEEE Transaction on Communications, 36(8):957-971, August 1988. ” , IEEE Transaction on Communications, 36(8):957-971, August 1988. [5] Wei-Ying Ma and B. S. Manjunath, “[5] Wei-Ying Ma and B. S. Manjunath, “A Texture Thesaurus for Browsing Large A Texture Thesaurus for Browsing Large Aerial PhotographsAerial Photographs”, Journal Of The American Society For Information Science, ”, Journal Of The American Society For Information Science, 49(7):633–648, 1998.49(7):633–648, 1998.

[6] Dongqing Chen, Lihong Li, and Zhengrong Liang, “[6] Dongqing Chen, Lihong Li, and Zhengrong Liang, “A Self-adaptive Vector A Self-adaptive Vector Quantization Algorithm for MR Image SegmentationQuantization Algorithm for MR Image Segmentation” , ISMRM,1999.” , ISMRM,1999.

[7] Qixiang Ye2 Wen Gao Wei Zeng1, “[7] Qixiang Ye2 Wen Gao Wei Zeng1, “Color Image Segmentation Using Color Image Segmentation Using Density-Based ClusteringDensity-Based Clustering”, ICASSP, 2003, Presentation.”, ICASSP, 2003, Presentation.

Page 16: Binning Strategies for Tissue Texture Extraction in DICOM Images CTI Students: Bikash Bhattacharyya, Kriti Jauhar Advisors: Dr. Daniela Raicu, Dr. Jacob

ReferencesReferences [8] Martin Ester, Hans-Peter Kriegel, Jorg Sander, XiaoWei Xu, “[8] Martin Ester, Hans-Peter Kriegel, Jorg Sander, XiaoWei Xu, “A Density-A Density-

Based Algorithm for Discovering Spatial Databases With NoiseBased Algorithm for Discovering Spatial Databases With Noise,” Proceedings ,” Proceedings of 2nd International Conference on Knowledge Discovery and Data Mining, of 2nd International Conference on Knowledge Discovery and Data Mining, 1996.1996.

[9] Texture Classification of Normal Tissues in Computed Tomography D. Xu, [9] Texture Classification of Normal Tissues in Computed Tomography D. Xu, J. Lee, D.S. Raicu, J. D. Furst, D. Channin, The 2005 Annual Meeting of the J. Lee, D.S. Raicu, J. D. Furst, D. Channin, The 2005 Annual Meeting of the Society for Computer Applications in Radiology, Orlando, Florida, June 2-5, Society for Computer Applications in Radiology, Orlando, Florida, June 2-5, 2005.2005.

[10] N. B. Karayiannis, "Soft learning vector quantization and clustering [10] N. B. Karayiannis, "Soft learning vector quantization and clustering algorithms based on ordered weighted aggregation operators," IEEE algorithms based on ordered weighted aggregation operators," IEEE Transactions on Neural Networks, vol. 11, no. 5, pp. 1093-1105, 2000.Transactions on Neural Networks, vol. 11, no. 5, pp. 1093-1105, 2000.

[11] N. Papamarkos and B. Gatos, "[11] N. Papamarkos and B. Gatos, "A new approach for multithreshold A new approach for multithreshold selectionselection", Computer Vision, Graphics, and Image Processing-Graphical ", Computer Vision, Graphics, and Image Processing-Graphical Models and Image Processing, Vol. 56, No. 5, pp. 357-370, Sept. 1994Models and Image Processing, Vol. 56, No. 5, pp. 357-370, Sept. 1994

[12] A. Atsalakis, N. Kroupis , D. Soudris, and N. Papamarkos, "[12] A. Atsalakis, N. Kroupis , D. Soudris, and N. Papamarkos, "A window-A window-based color quantization technique and its architecture implementationbased color quantization technique and its architecture implementation", ", ICIP2002, Rochester, USA.ICIP2002, Rochester, USA.

[13]  N. Papamarkos, A. Atsalakis and C. Strouthopoulos, "[13]  N. Papamarkos, A. Atsalakis and C. Strouthopoulos, "Adaptive Color Adaptive Color ReductionReduction", IEEE Trans. on Systems, Man, and Cybernetics-Part B, Vol. 32, ", IEEE Trans. on Systems, Man, and Cybernetics-Part B, Vol. 32, No. 1, Feb. 2002.No. 1, Feb. 2002.