Binning Strategies for Tissue Texture Extraction in DICOM Images

1

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

2
Why Binning ?
? Binning Definition Putting
gray-levels into bins for image compression
e.g. 1,2,3,4 gray levels in Bin 1
5,6,7,8 gray levels in Bin 2?DICOM Images
12 Bit - 4096 intensities?Texture Feature
Calculation All intensities - SLOW?Binning
allows for additional flexibility to trade off
large intensity ranges against computational
speed
COMPUTATION PERFORMANCE
3
Linear Binning
? Linear Binning - Bins of equal size
256 bins for DICOM images produces bin
ranges 0..15 , 16..31 ,4081..4096 ? Quick
and Efficient approach ? Pre -processing step
for Harlick texture feature calculation?
Promising results for classification of tissues
based on Haralick texture
features
4
Disadvantages 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?

5
Analysis of Linear Binning (contd.)
EXAMPLES
Spleen
Liver
6
PROCESS SUMMARY
7
Two Approaches of Non-Linear Binning

• Clipped Binning based on visual inspection of
  gray levels
• Range 0, 856 mapped to Bin 1
• Range 1368 , 4096 mapped to Bin 258
• Range 856, 1368 mapped to 256 linear bins
• e.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
8
Process Flow
Non-linear Binning using K-Means
9
K-Means

• K 256
• 141 Dimensions/Images
• 4096 points/Gray Levels
• Initial Points /Random Centroids
• Similarity Metric Euclidean Distance

• Issues
• 263 Unique Gray Levels Identified
• Multiple Gray Levels Identified in one Cluster
  e.g. Cluster 14 has gray levels from 462 to
  884 Cluster 14 also has gray
  levels from 1540 to 1542

10
Classification Results TRAINING SET
11
Classification Results TESTING SET
12
Graphical User Interface
13
Conclusion

• Non-Linear Binning with K-Means gave
  us the best overall results (
  86.35)
• Results for Liver and Spleen improved from
  73.80 to 91.03 for liver and 70.50 to 74.58
  for spleen
• Clipped Binning performed poorly on testing set
  with overall
  sensitivity of only 68.85
• Results with K-Means improved over Linear
  Binning

14
Future Work

• Experimenting with bins other than 256 such as
  64, 128 etc.
• Exploring other similarity measures such as
  Jeffrey Divergence, Mahalanobis Distance etc.
• Testing other classification algorithms besides
  decision trees, such as Neural Networks,
  Bayesian Networks, Logistic Regression etc.

15
References

• 1 M. Kalinin, D. S. Raicu, J. D. Furst, D. S.
  Channin,, " A Classification Approach for
  Anatomical Regions Segmentation", The IEEE
  International Conference on Image Processing
  (ICIP), September 11-14, 2005. (submitted)
• 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", RSNA, DECEMBER, 2004.
• 3 R.M. Haralick, K. Shanmugam, and I. Dinstein,
  Textural Features for Image Classification,
  IEEE Trans. on Systems,Man, and Cybernetics, vol.
  Smc-3, no.6, pp. 610-621, 1973.
• 4 N. M. Nasrabadi and R. A. King., Image
  Coding Using Vector Quantization A review ,
  IEEE Transaction on Communications,
  36(8)957-971, August 1988. 5 Wei-Ying Ma and
  B. S. Manjunath, A Texture Thesaurus for
  Browsing Large Aerial Photographs, Journal Of
  The American Society For Information Science,
  49(7)633648, 1998.
• 6 Dongqing Chen, Lihong Li, and Zhengrong
  Liang, A Self-adaptive Vector Quantization
  Algorithm for MR Image Segmentation ,
  ISMRM,1999.
• 7 Qixiang Ye2 Wen Gao Wei Zeng1, Color Image
  Segmentation Using Density-Based Clustering,
  ICASSP, 2003, Presentation.

16
References

• 8 Martin Ester, Hans-Peter Kriegel, Jorg
  Sander, XiaoWei Xu, A Density-Based Algorithm
  for Discovering Spatial Databases With Noise,
  Proceedings of 2nd International Conference on
  Knowledge Discovery and Data Mining, 1996.
• 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 Society for Computer Applications in
  Radiology, Orlando, Florida, June 2-5, 2005.
• 10 N. B. Karayiannis, "Soft learning vector
  quantization and clustering algorithms based on
  ordered weighted aggregation operators," IEEE
  Transactions on Neural Networks, vol. 11, no. 5,
  pp. 1093-1105, 2000.
• 11 N. Papamarkos and B. Gatos, "A new approach
  for multithreshold selection", Computer Vision,
  Graphics, and Image Processing-Graphical Models
  and Image Processing, Vol. 56, No. 5, pp.
  357-370, Sept. 1994
• 12 A. Atsalakis, N. Kroupis , D. Soudris, and
  N. Papamarkos, "A window-based color quantization
  technique and its architecture implementation",
  ICIP2002, Rochester, USA.
• 13  N. Papamarkos, A. Atsalakis and C.
  Strouthopoulos, "Adaptive Color Reduction", IEEE
  Trans. on Systems, Man, and Cybernetics-Part B,
  Vol. 32, No. 1, Feb. 2002.