310 likes | 324 Views
Explore content-based image retrieval, feature extraction, and segmentation in medical imaging. Learn about texture analysis and classification using numerical descriptors for organ and tissue recognition.
E N D
Visual Computing Research @ CTI, DePaul University Daniela Raicu Assistant Professor draicu@cs.depaul.edu http://facweb.cs.depaul.edu/research/vc
Visual Computing Group • CTI Faculty: • Gian Mario Besana • Lucia Dettori • Jacob Furst • Gerald Gordon • Steve Jost • Yakov Keselman • Daniela Raicu • Collaborators: Department of Radiology, Northwestern University & Northwestern Memorial Hospital, Chicago, IL • Dr. David Channin, Chief of Informatics, Department of Radiology Medical Image Processing
Visual Computing Group • Graduate Students: • John Campion, Ramzy Darwish • William Horsthemke, Gabriel Sanchez, Winnie Tsang • Undergraduate Students: • Stelian Aioanei, Andrew Corboy • Jong Lee, Mikhail Kalinin • Lindsay Semler, Dong-Hui Xu • Visual Computing (VC) area: • CSC381/CSC481: Introduction to Image Processing • CSC382/CSC482: Image Analysis and its Applications • CSC384/CSC484: Introduction to Computer Vision • VC research seminar: FallQuarter, Friday, 5:00 - 6:00pm • VC workshop:Spring Quarter, Friday, April 15th , 2005 • Intelligent Multimedia Processing (IMP) lab: http://facweb.cs.depaul.edu/research/vc Medical Image Processing
Research problems Content-based Image Retrieval: Image retrieval systems that permit image searching based on features automatically extracted from the images’ own visual content are called content-based image retrieval (CBIR) systems. Domain-specific features: - fingerprints, human faces • visual features • (primitive or low-level image features) General features: - color, texture, shape Drawback:-lack of expressive power Medical Image Processing
Image Database Content-based Image Retrieval Feature Extraction Semantic Gap ? Mountains and water-falls It is a nice sunset. Meaning: Sunset Text Database Medical Image Processing
Content-based Image Retrieval Feature Representation:Two examples of original images and their representations. Medical Image Processing
Content-based Image Retrieval Two examples of original images and their representations: Medical Image Processing
S(q1,t1) Content-based Image Retrieval Similarity Measure: Image T: Image Q: , bi = masking bit Medical Image Processing
Content-based Image Retrieval Retrieval Results Query Medical Image Processing
Image Search Content-based Image Retrieval Medical Image Processing
Content-based Image Retrieval Medical Image Processing
labels for the organs present in the image heart backbone Medical Imaging Problem statement: Human body organs’ classifications using raw data (pixels) from abdominal and chest CT images. Medical Image Processing
Segmentation Organ/Tissue segmentation in CT images Medical Imaging • - Data: 340 DICOM images • Segmented organs: liver (56), kidneys (55), spleen (39), backbone (140), • & heart (50) • Segmentation algorithm:Active Contour Mappings (Snakes) • A boundary-based segmentation algorithm • Input for the algorithm: a number of initial points & five main • parameters that influence the way the boundary is formed. Medical Image Processing
Segmentation: Matlab Demo • Advantage: it detects complex shapes • Disadvantage: it needs manual selection of the initial points and of the parameters • Our Solution: perform clustering of similar regions using a neural network Medical Image Processing
Segmentation: Examples Medical Image Processing
Segmentation: Examples Medical Image Processing
Texture Analysis & Classification Classification rules for tissue/organs in CT images Calculate numerical texture descriptors for each region [D1, D2,…D21] Organ/Tissue segmentation in CT images IF HGRE <= 0.38 AND ENTROPY > 0.43 AND SRHGE <= 0.20 AND CONTRAST > 0.029 THEN Prediction = Heart Probability = 0.99 Medical Image Processing
Medical Imaging Texture Analysis Entropy Energy Contrast Homogeneity SumMean Variance Correlation Maximum Probability Inverse Difference Moment Cluster Tendency 3.892828 .034692 2.764427 .6345745 11.662886 7.308909 .110921 .112929 .44697 26.471211 Medical Image Processing
Medical Imaging Texture Analysis Entropy Energy Contrast Homogeneity SumMean Variance Correlation Maximum Probability Inverse Difference Moment Cluster Tendency 3.4151415 .108713 6.224426 .631435 13.628323 9.340897 .0723125 .3081855 .280289 31.139159 Medical Image Processing
Medical Imaging Texture Analysis Entropy Energy Contrast Homogeneity SumMean Variance Correlation Maximum Probability Inverse Difference Moment Cluster Tendency 3.38482 .055998 3.49784 .5577785 14.278469 3.737737 .1436305 .1250245 .437988 11.453111 Medical Image Processing
Medical Imaging Texture Analysis Entropy Energy Contrast Homogeneity SumMean Variance Correlation Maximum Probability Inverse Difference Moment Cluster Tendency 3.3099875 .049172 3.066407 .5369255 12.309719 1.634463 .0377875 .0897425 .460422 3.471442 Medical Image Processing
Medical Imaging Texture Analysis Entropy Energy Contrast Homogeneity SumMean Variance Correlation Maximum Probability Inverse Difference Moment Cluster Tendency 2.72509 .091388 1.618982 .6208175 11.755226 0.912752 .123976 .1742075 .506894 2.032082 Medical Image Processing
Texture Descriptors: Matlab Demo Medical Image Processing
Organ/Tissue Classification Classification rules for tissue/organs in CT images Calculate numerical texture descriptors for each region [D1, D2,…D21] IF HGRE <= 0.38 AND ENTROPY > 0.43 AND SRHGE <= 0.20 AND CONTRAST > 0.029 THEN Prediction = Heart Probability = 0.99 Algorithm: - decision trees Output: Decision Rules Performance estimated using: - sensitivity - specificity Advantage: Set of decision rules that can be used for annotation Medical Image Processing
Organ/Tissue Classification • IF (HGRE <= 0.3788) & (CLUSTER <= 0.0383095) & (INVERSE <= 0.768085) & (SUMMEAN <= 0.556015) & (SRLGE <= 0.101655) & (ENEGRY > 0.106715) THEN Prediction = Spleen, Probability = 0.928571 • IF (HGRE <= 0.3788) & (CLUSTER <= 0.0383095) & (INVERSE <= 0.768085) & (SUMMEAN <= 0.556015) & (SRLGE > 0.101655) THEN Prediction = Liver , Probability = 1.000000 • IF (HGRE <= 0.3788) & (CLUSTER <= 0.0383095) & (INVERSE <= 0.768085) & (SUMMEAN > 0.556015) & (GLNU <= 0.087365) THENPrediction = Kidney, Probability = 0.924658 Examples of Decision Tree Rules for Combined Data: Medical Image Processing
Organ/Tissue Classification • IF (HGRE <= 0.3788) & (CLUSTER > 0.0383095) & (GLNU > 0.03184) & (ENTROPY > 0.433185) & (SRHGE <= 0.19935) & (CONTRAST > 0.0295805) THEN Prediction = Heart, Probability = 0.988372 • IF (HGRE <= 0.3788) & (CLUSTER > 0.0383095) & (GLNU <= 0.03184) & (LRE <= 0.123405) THEN Prediction = Backbone, Probability = 1.000000 Examples of Decision Tree Rules for Combined Data: Medical Image Processing
Organ/Tissue Classification Decision Tree Accuracy on Testing Data (Co-occurrence, Run-length, and Combined): Medical Image Processing
Tissue Classification: Matlab Demo Medical Image Processing
Publications (CBIR) [1] Daniela Stan and Ishwar K. Sethi, “Image Retrieval using a Hierarchy of Clusters” in Lecture Notes in Computer Science: Advances in Pattern Recognition – ICAPR 2001, Springer-Verlag Ltd. (Ed), pp. 377-388, 2001. [2] Daniela Stan and Ishwar K. Sethi, “Mapping Low-level Image Features to Semantic Concepts” in Proceedings of SPIE: Storage and Retrieval for Media databases, pp. 172-179, 2001. [3] Ishwar K. Sethi, Ioana Coman, Daniela Stan, “Mining Association Rules between Low-level Image Features and High-level Concepts” in Proceedings of SPIE: Data Mining and Knowledge Discovery III, pp.279-290, 2001. [4] Daniela Stan and Ishwar K. Sethi, “Color Patterns for Pictorial Content Description”, ACM Symposium on Applied Computing, 2002. [5] Daniela Stan and Ishwar K. Sethi, “eID: A System for Exploration of Image Databases”, Information Processing and Management Journal,2002. [6] Daniela Stan and Ishwar K. Sethi, “Synobins: An intermediate level towards Annotation and Semantic Retrieval”, IEEE Trans. Multimedia Journal. Medical Image Processing
Publications (MI) [1] D. Xu, J. Lee, D.S. Raicu, J.D. Furst, D. Channin. "Texture Classification of Normal Tissues in Computed Tomography", The 2005 Annual Meeting of the Society for Computer Applications in Radiology, June 2-5, 2005. (Submitted) [2] D.S. Raicu, W. Tsang, M. Kalinin, D. Xu, J.D. Furst, D. Channin. "Automatic Tissue Context Determination in Computed Tomography", SPIE Medical Imaging, February 12–17, 2005. (Submitted) [3] D. H. Xu, A. Kurani, J. D. Furst, & D. S. Raicu, "Run-length encoding for volumetric texture", The 4th IASTED International Conference on Visualization, Imaging, and Image Processing - VIIP 2004, Marbella, Spain, September 6-8, 2004. [4] D. Channin, D. S. Raicu, J. D. Furst, D. H. Xu, L. Lilly, C. Limpsangsri, "Classification of Tissues in Computed Tomography using Decision Trees", Poster and Demo, The 90th Scientific Assembly and Annual Meeting of Radiology Society of North America (RSNA04), November 28, 2004. [5] A. Kurani, D. H. Xu, J. D. Furst, & D. S. Raicu, "Co-occurrence matrices for volumetric data", The 7th IASTED International Conference on Computer Graphics and Imaging – CGIM, August 16-18, 2004 . [6] D. S. Raicu, J. D. Furst, D. Channin, D. H. Xu, & A. Kurani, "A Texture Dictionary for Human Organs Tissues' Classification", Proceedings of the 8th World Multiconference on Systemics, Cybernetics and Informatics (SCI 2004), July 18-21, 2004. Medical Image Processing
Daniela Raicu Intelligent Multimedia Processing Laboratory School of CTI DePaul University THE END! Medical Image Processing