Joo-Hwee Lim is the Head of the Visual Computing Department at the Institute for Infocomm Research (I2R), A*STAR, Singapore, and an Adjunct Associate Professor at the School of Computer Engineering, Nanyang Technological University, Singapore. He is the co-Director of IPAL (Image & Pervasive Access Laboratory), a French-Singapore Joint Lab. He established the medical image analysis group at I2R in 2006, collaborating with clinicians closely, resulting in strong competency in ocular imaging, brain image analysis, cell image understanding etc at the institute. He has published over 200 journal and conference papers and owns 17 patents in the areas of computer vision, cognitive vision, pattern recognition, and medical image analysis.Sim-Heng Ong is an Associate Professor in the Departments of Electrical Engineering and Bioengineering at the National University of Singapore. He received his PhD from the University of Sydney, Australia. His major research areas are computer vision and medical image analysis and visualization. He has worked extensively with clinicians in developing algorithms for a variety of medical applications, and has publications in many highly respected journals and conferences.Wei Xiong is a Research Scientist at the Institute for Infocomm Research (I2R), A*STAR, Singapore. He obtained his PhD degree from the National University of Singapore.  His research interest is in computer vision, image processing, pattern classification and acoustic imaging. Dr. Xiong has published over 60 technical papers.

• List of Contributors xvPreface xixAcronyms xxiiiPART I INTRODUCTION 11 Overview of Biomedical Image Understanding Methods 3Wei Xiong, Jierong Cheng, Ying Gu, Shimiao Li and Joo Hwee Lim1.1 Segmentation and Object Detection 51.1.1 Methods Based on Image Processing Techniques 61.1.2 Methods Using Pattern Recognition and Machine Learning Algorithms 71.1.3 Model and Atlas-Based Segmentation 81.1.4 Multispectral Segmentation 91.1.5 User Interactions in Interactive Segmentation Methods 101.1.6 Frontiers of Biomedical Image Segmentation 111.2 Registration 111.2.1 Taxonomy of Registration Methods 121.2.2 Frontiers of Registration for Biomedical Image Understanding 151.3 Object Tracking 161.3.1 Object Representation 171.3.2 Feature Selection for Tracking 181.3.3 Object Tracking Technique 191.3.4 Frontiers of Object Tracking 191.4 Classification 201.4.1 Feature Extraction and Feature Selection 211.4.2 Classifiers 221.4.3 Unsupervised Classification 231.4.4 Classifier Combination 241.4.5 Frontiers of Pattern Classification for Biomedical Image Understanding 251.5 Knowledge-Based Systems 261.5.1 Semantic Interpretation and Knowledge-Based Systems 261.5.2 Knowledge-Based Vision Systems 271.5.3 Knowledge-Based Vision Systems in Biomedical Image Analysis 281.5.4 Frontiers of Knowledge-Based Systems 29References 29PARTII SEGMENTATION AND OBJECT DETECTION 472 Medical Image Segmentation and its Application in Cardiac MRI 49Dong Wei, Chao Li, and Ying Sun2.1 Introduction 502.2 Background 512.2.1 Active Contour Models 512.2.2 Parametric and Nonparametric Contour Representation 522.2.3 Graph-Based Image Segmentation 532.2.4 Summary 542.3 Parametric Active Contours – The Snakes 542.3.1 The Internal Spline Energy Eint 542.3.2 The Image-Derived Energy Eimg 552.3.3 The External Control Energy Econ 552.3.4 Extension of Snakes and Summary of Parametric Active Contours 572.4 Geometric Active Contours – The Level Sets 582.4.1 Variational Level Set Methods 582.4.2 Region-Based Variational Level Set Methods 602.4.3 Summary of Level Set Methods 642.5 Graph-Based Methods – The Graph Cuts 652.5.1 Basic Graph Cuts Formulation 652.5.2 Patch-Based Graph Cuts 662.5.3 An Example of Graph Cuts 682.5.4 Summary of Graph Cut Methods 722.6 Case Study: Cardiac Image Segmentation Using A Dual Level Sets Model 732.6.1 Introduction 732.6.2 Method 742.6.3 Experimental Results 792.6.4 Conclusion of the Case Study 812.7 Conclusion and Near-Future Trends 81References 833 Morphometric Measurements of the Retinal Vasculature in Fundus Images With Vampire 91Emanuele Trucco, Andrea Giachetti, Lucia Ballerini, Devanjali Relan, Alessandro Cavinato, and Tom Macgillivray3.1 Introduction 923.2 Assessing Vessel Width 933.2.1 Previous Work 933.2.2 Our Method 943.2.3 Results 953.2.4 Discussion 963.3 Artery or Vein? 983.3.1 Previous Work 983.3.2 Our Solution 993.3.3 Results 1013.3.4 Discussion 1033.4 Are My Program’s Measurements Accurate? 1043.4.1 Discussion 106References 1074 Analyzing Cell and Tissue Morphologies Using Pattern Recognition Algorithms 113Hwee Kuan Lee, Yan Nei Law, Chao-Hui Huang, and Choon Kong Yap 4.1 Introduction, 1134.2 Texture Segmentation of Endometrial Images Using the Subspace Mumford–Shah Model 1154.2.1 Subspace Mumford–Shah Segmentation Model 1164.2.2 Feature Weights 1184.2.3 Once-and-For-All Approach 1194.2.4 Results 1194.3 Spot Clustering for Detection of Mutants in Keratinocytes 1204.3.1 Image Analysis Framework 1234.3.2 Results 1244.4 Cells and Nuclei Detection 1244.4.1 Model 1254.4.2 Neural Cells and Breast Cancer Cells Data 1274.4.3 Performance Evaluation 1274.4.4 Robustness Study 1274.4.5 Results 1284.5 Geometric Regional Graph Spectral Feature 1344.5.1 Conversion of Image Patches into Region Signatures 1344.5.2 Comparing Region Signatures 1354.5.3 Classification of Region Signatures 1364.5.4 Random Masking and Object Detection 1364.5.5 Results 1374.6 Mitotic Cells in the H&E Histopathological Images of Breast Cancer Carcinoma 1384.6.1 Mitotic Index Estimation 1394.6.2 Mitotic Candidate Selection 1404.6.3 Exclusive Independent Component Analysis (XICA) 1404.6.4 Classification Using Sparse Representation 1434.6.5 Training and Testing Over Channels 1444.6.6 Results 1464.7 Conclusions 147References 147PARTIII REGISTRATION AND MATCHING 1535 3D Nonrigid Image Registration by Parzen-Window-Based Normalized Mutual Information and its Application on Mr-Guided Microwave Thermocoagulation of Liver Tumors 155Rui Xu, Yen-Wei Chen, Shigehiro Morikawa, and Yoshimasa Kurumi5.1 Introduction 1555.2 Parzen-Window-Based Normalized Mutual Information 1575.2.1 Definition of Parzen-Window Method 1575.2.2 Parzen-Window-Based Estimation of Joint Histogram 1585.2.3 Normalized Mutual Information and its Derivative 1605.3 Analysis of Kernel Selection 1635.3.1 The Designed Kernel 1635.3.2 Comparison in Theory 1675.3.3 Comparison by Experiments 1705.4 Application on MR-Guided Microwave Thermocoagulation of Liver Tumors 1745.4.1 Introduction of MR-Guided Microwave Thermocoagulation of Liver Tumors 1745.4.2 Nonrigid Registration by Parzen-Window-Based Mutual Information 1755.4.3 Evaluation on Phantom Data 1775.4.4 Evaluation on Clinical Cases 1805.5 Conclusion 185Acknowledgements 186References 1876 2D/3D Image Registration For Endovascular Abdominal Aortic Aneurysm (AAA) Repair 189Shun Miao and Rui Liao6.1 Introduction 1896.2 Background 1906.2.1 Image Modalities 1906.2.2 2D/3D Registration Framework 1926.2.3 Feature-Based Registration 1946.2.4 Intensity-Based Registration 1966.2.5 Number of Imaging Planes 1976.2.6 2D/3D Registration for Endovascular AAA Repair 1986.3 Smart Utilization of Two X-Ray Images for Rigid-Body 2D/3D Registration 1996.3.1 2D/3D Registration: Challenges in EVAR 1996.3.2 3D Image Processing and DRR Generation 2026.3.3 2D Image Processing 2036.3.4 Similarity Measure 2056.3.5 Optimization 2076.3.6 Validation 2106.4 Deformable 2D/3D Registration 2116.4.1 Problem Formulation 2126.4.2 Graph-Based Difference Measure 2136.4.3 Length Preserving Term 2156.4.4 Smoothness Term 2156.4.5 Optimization 2166.4.6 Validation 2176.5 Visual Check of Patient Movement Using Pelvis Boundary Detection 2206.6 Discussion and Conclusion 222References 223PARTIV OBJECT TRACKING 2297 Motion Tracking in Medical Images 231Chuqing Cao, Chao Li, and Ying Sun7.1 Introduction 2327.1.1 Point-Based Tracking 2337.1.2 Silhouette-Based Tracking 2337.1.3 Kernel-Based Tracking 2337.2 Background 2347.2.1 Point-Based Tracking 2347.2.2 Silhouette-Based Tracking 2367.2.3 Kernel-Based Tracking 2377.2.4 Summary 2387.3 Bayesian Tracking Methods 2387.3.1 Kalman Filters 2397.3.2 Particle Filters 2407.3.3 Summary of Bayesian Tracking Methods 2417.4 Deformable Models 2417.4.1 Mathematical Foundations of Deformable Models 2417.4.2 Energy-Minimizing Deformable Models 2427.4.3 Probabilistic Deformable Models 2447.4.4 Summary of Deformable Models 2457.5 Motion Tracking Based on the Harmonic Phase Algorithm 2467.5.1 HARP Imaging 2467.5.2 HARP Tracking 2487.5.3 Summary 2497.6 Case Study: Pseudo Ground Truth-Based Nonrigid Registration of MRI for Tracking the Cardiac Motion 2507.6.1 Data Fidelity Term 2517.6.2 Spatial Smoothness Constraint 2527.6.3 Temporal Smoothness Constraint 2537.6.4 Energy Minimization 2547.6.5 Preliminary Results 2557.6.6 Nonrigid Registration of Myocardial Perfusion MRI 2557.6.7 Experimental Results 2597.7 Discussion 2647.8 Conclusion and Near-Future Trends 265References 267PARTV CLASSIFICATION 2758 Blood Smear Analysis, Malaria Infection Detection, and Grading from Blood Cell Images 277Wei Xiong, Sim-Heng Ong, Joo-Hwee Lim, Jierong Cheng, and Ying Gu8.1 Introduction 2788.2 Pattern Classification Techniques 2828.2.1 Supervised and Nonsupervised Learning 2828.2.2 Bayesian Decision Theory 2838.2.3 Clustering 2848.2.4 Support Vector Machines 2868.3 GWA Detection 2878.3.1 Image Analysis 2888.3.2 Association between the Object Area and the Number of Cells Per Object 2898.3.3 Clump Splitting 2918.3.4 Clump Characterization 2938.3.5 Classification 2958.4 Dual-Model-Guided Image Segmentation and Recognition 2958.4.1 Related Work 2968.4.2 Strategies and Object Functions 2978.4.3 Endpoint Adjacency Map Construction and Edge Linking 2998.4.4 Parsing Contours and Their Convex Hulls 3008.4.5 A Recursive and Greedy Splitting Approach 3018.4.6 Incremental Model Updating and Bayesian Decision 3018.5 Infection Detection and Staging 3028.5.1 Related Work 3028.5.2 Methodology 3038.6 Experimental Results 3058.6.1 GWA Classification 3058.6.2 RBC Segmentation 3108.6.3 RBC Classification 3158.7 Summary 320References 3219 Liver Tumor Segmentation Using SVM Framework and Pathology Characterization Using Content-Based Image Retrieval 325Jiayin Zhou, Yanling Chi, Weimin Huang, Wei Xiong, Wenyu Chen, Jimin Liu, and Sudhakar K. Venkatesh9.1 Introduction 3259.2 Liver Tumor Segmentation Under a Hybrid SVM Framework 3279.2.1 Fundamentals of SVM for Classification 3279.2.2 SVM Framework for Liver Tumor Segmentation and the Problems 3309.2.3 A Three-Stage Hybrid SVM Scheme for Liver Tumor Segmentation 3319.2.4 Experiment 3349.2.5 Evaluation Metrics 3359.2.6 Results 3369.3 Liver Tumor Characterization by Content-Based Image Retrieval 3389.3.1 Existing Work and the Rationale of Using CBIR 3399.3.2 Methodology Overview and Preprocessing 3409.3.3 Tumor Feature Representation 3419.3.4 Similarity Query and Tumor Pathological Type Prediction 3439.3.5 Experiment 3459.3.6 Results 3469.4 Discussion 3519.4.1 About Liver Tumor Segmentation Using Machine Learning 3519.4.2 About Liver Tumor Characterization Using CBIR 3539.5 Conclusion 356References 35710 Benchmarking Lymph Node Metastasis Classification for Gastric Cancer Staging 361Su Zhang, Chao Li, Shuheng Zhang, Lifang Pang, and Huan Zhang10.1 Introduction 36210.1.1 Introduction of GSI-CT 36310.1.2 Imaging Findings of Gastric Cancer 36610.2 Related Feature Selection, Metric Learning, and Classification Methods 36710.2.1 Feature Extraction 36710.2.2 KNN 36710.2.3 Feature Selection 36910.2.4 AdaBoost and EAdaBoost Algorithms 37410.3 Preprocessing Method for GSI-CT Data 37710.3.1 Data Acquisition for GSI-CT Data 37710.3.2 Univariate Analysis 37810.4 Classification Results For GSI-CT Data of Gastric Cancer 37910.4.1 Experimental Results of mRMR-KNN 37910.4.2 Experimental Results of SFS-KNN 38310.4.3 Experimental Results of Metric Learning 38510.4.4 Experiments Results of AdaBoost and EAdaBoost 38510.4.5 Experiment Analysis 38810.5 Conclusion and Future Work 388Acknowledgment 388References 388PARTVI KNOWLEDGE-BASED SYSTEMS 39311 The Use of Knowledge in Biomedical Image Analysis 395Florence Cloppet11.1 Introduction 39511.2 Data, Information, and Knowledge? 39711.2.1 Data Versus Information 39711.2.2 Knowledge Versus Information 39811.3 What Kind of Information/Knowledge Can be Introduced? 39911.4 How to Introduce Information in Computer Vision Systems? 40011.4.1 Nature of Prior Information/Knowledge 40211.4.2 Frameworks Allowing Prior Information Introduction 40811.5 Conclusion 418References 41812 Active Shape Model for Contour Detection of Anatomical Structure 429Huiqi Li and Qing Nie12.1 Introduction 42912.2 Background 43012.2.1 Free-Form Deformable Models 43012.2.2 Parametrically Deformable Models 43212.3 Methodology 43412.3.1 Point Distribution Model 43412.3.2 Active Shape Model (ASM) 43612.3.3 A Modified ASM 43812.4 Applications 44012.4.1 Boundary Detection of Optic Disk 44012.4.2 Lens Structure Detection 45012.5 Summary 456Acknowledgment 457References 457Index 463

Joo-Hwee Lim is the Head of the Visual Computing Department at the Institute for Infocomm Research (I2R), A*STAR, Singapore, and an Adjunct Associate Professor at the School of Computer Engineering, Nanyang Technological University, Singapore. He is the co-Director of IPAL (Image & Pervasive Access Laboratory), a French-Singapore Joint Lab. He established the medical image analysis group at I2R in 2006, collaborating with clinicians closely, resulting in strong competency in ocular imaging, brain image analysis, cell image understanding etc at the institute. He has published over 200 journal and conference papers and owns 17 patents in the areas of computer vision, cognitive vision, pattern recognition, and medical image analysis.Sim-Heng Ong is an Associate Professor in the Departments of Electrical Engineering and Bioengineering at the National University of Singapore. He received his PhD from the University of Sydney, Australia. His major research areas are computer vision and medical image analysis and visualization. He has worked extensively with clinicians in developing algorithms for a variety of medical applications, and has publications in many highly respected journals and conferences.Wei Xiong is a Research Scientist at the Institute for Infocomm Research (I2R), A*STAR, Singapore. He obtained his PhD degree from the National University of Singapore.  His research interest is in computer vision, image processing, pattern classification and acoustic imaging. Dr. Xiong has published over 60 technical papers.

• List of Contributors xvPreface xixAcronyms xxiiiPART I INTRODUCTION 11 Overview of Biomedical Image Understanding Methods 3Wei Xiong, Jierong Cheng, Ying Gu, Shimiao Li and Joo Hwee Lim1.1 Segmentation and Object Detection 51.1.1 Methods Based on Image Processing Techniques 61.1.2 Methods Using Pattern Recognition and Machine Learning Algorithms 71.1.3 Model and Atlas-Based Segmentation 81.1.4 Multispectral Segmentation 91.1.5 User Interactions in Interactive Segmentation Methods 101.1.6 Frontiers of Biomedical Image Segmentation 111.2 Registration 111.2.1 Taxonomy of Registration Methods 121.2.2 Frontiers of Registration for Biomedical Image Understanding 151.3 Object Tracking 161.3.1 Object Representation 171.3.2 Feature Selection for Tracking 181.3.3 Object Tracking Technique 191.3.4 Frontiers of Object Tracking 191.4 Classification 201.4.1 Feature Extraction and Feature Selection 211.4.2 Classifiers 221.4.3 Unsupervised Classification 231.4.4 Classifier Combination 241.4.5 Frontiers of Pattern Classification for Biomedical Image Understanding 251.5 Knowledge-Based Systems 261.5.1 Semantic Interpretation and Knowledge-Based Systems 261.5.2 Knowledge-Based Vision Systems 271.5.3 Knowledge-Based Vision Systems in Biomedical Image Analysis 281.5.4 Frontiers of Knowledge-Based Systems 29References 29PARTII SEGMENTATION AND OBJECT DETECTION 472 Medical Image Segmentation and its Application in Cardiac MRI 49Dong Wei, Chao Li, and Ying Sun2.1 Introduction 502.2 Background 512.2.1 Active Contour Models 512.2.2 Parametric and Nonparametric Contour Representation 522.2.3 Graph-Based Image Segmentation 532.2.4 Summary 542.3 Parametric Active Contours – The Snakes 542.3.1 The Internal Spline Energy Eint 542.3.2 The Image-Derived Energy Eimg 552.3.3 The External Control Energy Econ 552.3.4 Extension of Snakes and Summary of Parametric Active Contours 572.4 Geometric Active Contours – The Level Sets 582.4.1 Variational Level Set Methods 582.4.2 Region-Based Variational Level Set Methods 602.4.3 Summary of Level Set Methods 642.5 Graph-Based Methods – The Graph Cuts 652.5.1 Basic Graph Cuts Formulation 652.5.2 Patch-Based Graph Cuts 662.5.3 An Example of Graph Cuts 682.5.4 Summary of Graph Cut Methods 722.6 Case Study: Cardiac Image Segmentation Using A Dual Level Sets Model 732.6.1 Introduction 732.6.2 Method 742.6.3 Experimental Results 792.6.4 Conclusion of the Case Study 812.7 Conclusion and Near-Future Trends 81References 833 Morphometric Measurements of the Retinal Vasculature in Fundus Images With Vampire 91Emanuele Trucco, Andrea Giachetti, Lucia Ballerini, Devanjali Relan, Alessandro Cavinato, and Tom Macgillivray3.1 Introduction 923.2 Assessing Vessel Width 933.2.1 Previous Work 933.2.2 Our Method 943.2.3 Results 953.2.4 Discussion 963.3 Artery or Vein? 983.3.1 Previous Work 983.3.2 Our Solution 993.3.3 Results 1013.3.4 Discussion 1033.4 Are My Program’s Measurements Accurate? 1043.4.1 Discussion 106References 1074 Analyzing Cell and Tissue Morphologies Using Pattern Recognition Algorithms 113Hwee Kuan Lee, Yan Nei Law, Chao-Hui Huang, and Choon Kong Yap 4.1 Introduction, 1134.2 Texture Segmentation of Endometrial Images Using the Subspace Mumford–Shah Model 1154.2.1 Subspace Mumford–Shah Segmentation Model 1164.2.2 Feature Weights 1184.2.3 Once-and-For-All Approach 1194.2.4 Results 1194.3 Spot Clustering for Detection of Mutants in Keratinocytes 1204.3.1 Image Analysis Framework 1234.3.2 Results 1244.4 Cells and Nuclei Detection 1244.4.1 Model 1254.4.2 Neural Cells and Breast Cancer Cells Data 1274.4.3 Performance Evaluation 1274.4.4 Robustness Study 1274.4.5 Results 1284.5 Geometric Regional Graph Spectral Feature 1344.5.1 Conversion of Image Patches into Region Signatures 1344.5.2 Comparing Region Signatures 1354.5.3 Classification of Region Signatures 1364.5.4 Random Masking and Object Detection 1364.5.5 Results 1374.6 Mitotic Cells in the H&E Histopathological Images of Breast Cancer Carcinoma 1384.6.1 Mitotic Index Estimation 1394.6.2 Mitotic Candidate Selection 1404.6.3 Exclusive Independent Component Analysis (XICA) 1404.6.4 Classification Using Sparse Representation 1434.6.5 Training and Testing Over Channels 1444.6.6 Results 1464.7 Conclusions 147References 147PARTIII REGISTRATION AND MATCHING 1535 3D Nonrigid Image Registration by Parzen-Window-Based Normalized Mutual Information and its Application on Mr-Guided Microwave Thermocoagulation of Liver Tumors 155Rui Xu, Yen-Wei Chen, Shigehiro Morikawa, and Yoshimasa Kurumi5.1 Introduction 1555.2 Parzen-Window-Based Normalized Mutual Information 1575.2.1 Definition of Parzen-Window Method 1575.2.2 Parzen-Window-Based Estimation of Joint Histogram 1585.2.3 Normalized Mutual Information and its Derivative 1605.3 Analysis of Kernel Selection 1635.3.1 The Designed Kernel 1635.3.2 Comparison in Theory 1675.3.3 Comparison by Experiments 1705.4 Application on MR-Guided Microwave Thermocoagulation of Liver Tumors 1745.4.1 Introduction of MR-Guided Microwave Thermocoagulation of Liver Tumors 1745.4.2 Nonrigid Registration by Parzen-Window-Based Mutual Information 1755.4.3 Evaluation on Phantom Data 1775.4.4 Evaluation on Clinical Cases 1805.5 Conclusion 185Acknowledgements 186References 1876 2D/3D Image Registration For Endovascular Abdominal Aortic Aneurysm (AAA) Repair 189Shun Miao and Rui Liao6.1 Introduction 1896.2 Background 1906.2.1 Image Modalities 1906.2.2 2D/3D Registration Framework 1926.2.3 Feature-Based Registration 1946.2.4 Intensity-Based Registration 1966.2.5 Number of Imaging Planes 1976.2.6 2D/3D Registration for Endovascular AAA Repair 1986.3 Smart Utilization of Two X-Ray Images for Rigid-Body 2D/3D Registration 1996.3.1 2D/3D Registration: Challenges in EVAR 1996.3.2 3D Image Processing and DRR Generation 2026.3.3 2D Image Processing 2036.3.4 Similarity Measure 2056.3.5 Optimization 2076.3.6 Validation 2106.4 Deformable 2D/3D Registration 2116.4.1 Problem Formulation 2126.4.2 Graph-Based Difference Measure 2136.4.3 Length Preserving Term 2156.4.4 Smoothness Term 2156.4.5 Optimization 2166.4.6 Validation 2176.5 Visual Check of Patient Movement Using Pelvis Boundary Detection 2206.6 Discussion and Conclusion 222References 223PARTIV OBJECT TRACKING 2297 Motion Tracking in Medical Images 231Chuqing Cao, Chao Li, and Ying Sun7.1 Introduction 2327.1.1 Point-Based Tracking 2337.1.2 Silhouette-Based Tracking 2337.1.3 Kernel-Based Tracking 2337.2 Background 2347.2.1 Point-Based Tracking 2347.2.2 Silhouette-Based Tracking 2367.2.3 Kernel-Based Tracking 2377.2.4 Summary 2387.3 Bayesian Tracking Methods 2387.3.1 Kalman Filters 2397.3.2 Particle Filters 2407.3.3 Summary of Bayesian Tracking Methods 2417.4 Deformable Models 2417.4.1 Mathematical Foundations of Deformable Models 2417.4.2 Energy-Minimizing Deformable Models 2427.4.3 Probabilistic Deformable Models 2447.4.4 Summary of Deformable Models 2457.5 Motion Tracking Based on the Harmonic Phase Algorithm 2467.5.1 HARP Imaging 2467.5.2 HARP Tracking 2487.5.3 Summary 2497.6 Case Study: Pseudo Ground Truth-Based Nonrigid Registration of MRI for Tracking the Cardiac Motion 2507.6.1 Data Fidelity Term 2517.6.2 Spatial Smoothness Constraint 2527.6.3 Temporal Smoothness Constraint 2537.6.4 Energy Minimization 2547.6.5 Preliminary Results 2557.6.6 Nonrigid Registration of Myocardial Perfusion MRI 2557.6.7 Experimental Results 2597.7 Discussion 2647.8 Conclusion and Near-Future Trends 265References 267PARTV CLASSIFICATION 2758 Blood Smear Analysis, Malaria Infection Detection, and Grading from Blood Cell Images 277Wei Xiong, Sim-Heng Ong, Joo-Hwee Lim, Jierong Cheng, and Ying Gu8.1 Introduction 2788.2 Pattern Classification Techniques 2828.2.1 Supervised and Nonsupervised Learning 2828.2.2 Bayesian Decision Theory 2838.2.3 Clustering 2848.2.4 Support Vector Machines 2868.3 GWA Detection 2878.3.1 Image Analysis 2888.3.2 Association between the Object Area and the Number of Cells Per Object 2898.3.3 Clump Splitting 2918.3.4 Clump Characterization 2938.3.5 Classification 2958.4 Dual-Model-Guided Image Segmentation and Recognition 2958.4.1 Related Work 2968.4.2 Strategies and Object Functions 2978.4.3 Endpoint Adjacency Map Construction and Edge Linking 2998.4.4 Parsing Contours and Their Convex Hulls 3008.4.5 A Recursive and Greedy Splitting Approach 3018.4.6 Incremental Model Updating and Bayesian Decision 3018.5 Infection Detection and Staging 3028.5.1 Related Work 3028.5.2 Methodology 3038.6 Experimental Results 3058.6.1 GWA Classification 3058.6.2 RBC Segmentation 3108.6.3 RBC Classification 3158.7 Summary 320References 3219 Liver Tumor Segmentation Using SVM Framework and Pathology Characterization Using Content-Based Image Retrieval 325Jiayin Zhou, Yanling Chi, Weimin Huang, Wei Xiong, Wenyu Chen, Jimin Liu, and Sudhakar K. Venkatesh9.1 Introduction 3259.2 Liver Tumor Segmentation Under a Hybrid SVM Framework 3279.2.1 Fundamentals of SVM for Classification 3279.2.2 SVM Framework for Liver Tumor Segmentation and the Problems 3309.2.3 A Three-Stage Hybrid SVM Scheme for Liver Tumor Segmentation 3319.2.4 Experiment 3349.2.5 Evaluation Metrics 3359.2.6 Results 3369.3 Liver Tumor Characterization by Content-Based Image Retrieval 3389.3.1 Existing Work and the Rationale of Using CBIR 3399.3.2 Methodology Overview and Preprocessing 3409.3.3 Tumor Feature Representation 3419.3.4 Similarity Query and Tumor Pathological Type Prediction 3439.3.5 Experiment 3459.3.6 Results 3469.4 Discussion 3519.4.1 About Liver Tumor Segmentation Using Machine Learning 3519.4.2 About Liver Tumor Characterization Using CBIR 3539.5 Conclusion 356References 35710 Benchmarking Lymph Node Metastasis Classification for Gastric Cancer Staging 361Su Zhang, Chao Li, Shuheng Zhang, Lifang Pang, and Huan Zhang10.1 Introduction 36210.1.1 Introduction of GSI-CT 36310.1.2 Imaging Findings of Gastric Cancer 36610.2 Related Feature Selection, Metric Learning, and Classification Methods 36710.2.1 Feature Extraction 36710.2.2 KNN 36710.2.3 Feature Selection 36910.2.4 AdaBoost and EAdaBoost Algorithms 37410.3 Preprocessing Method for GSI-CT Data 37710.3.1 Data Acquisition for GSI-CT Data 37710.3.2 Univariate Analysis 37810.4 Classification Results For GSI-CT Data of Gastric Cancer 37910.4.1 Experimental Results of mRMR-KNN 37910.4.2 Experimental Results of SFS-KNN 38310.4.3 Experimental Results of Metric Learning 38510.4.4 Experiments Results of AdaBoost and EAdaBoost 38510.4.5 Experiment Analysis 38810.5 Conclusion and Future Work 388Acknowledgment 388References 388PARTVI KNOWLEDGE-BASED SYSTEMS 39311 The Use of Knowledge in Biomedical Image Analysis 395Florence Cloppet11.1 Introduction 39511.2 Data, Information, and Knowledge? 39711.2.1 Data Versus Information 39711.2.2 Knowledge Versus Information 39811.3 What Kind of Information/Knowledge Can be Introduced? 39911.4 How to Introduce Information in Computer Vision Systems? 40011.4.1 Nature of Prior Information/Knowledge 40211.4.2 Frameworks Allowing Prior Information Introduction 40811.5 Conclusion 418References 41812 Active Shape Model for Contour Detection of Anatomical Structure 429Huiqi Li and Qing Nie12.1 Introduction 42912.2 Background 43012.2.1 Free-Form Deformable Models 43012.2.2 Parametrically Deformable Models 43212.3 Methodology 43412.3.1 Point Distribution Model 43412.3.2 Active Shape Model (ASM) 43612.3.3 A Modified ASM 43812.4 Applications 44012.4.1 Boundary Detection of Optic Disk 44012.4.2 Lens Structure Detection 45012.5 Summary 456Acknowledgment 457References 457Index 463

Joo-Hwee Lim is the Head of the Visual Computing Department at the Institute for Infocomm Research (I2R), A*STAR, Singapore, and an Adjunct Associate Professor at the School of Computer Engineering, Nanyang Technological University, Singapore. He is the co-Director of IPAL (Image & Pervasive Access Laboratory), a French-Singapore Joint Lab. He established the medical image analysis group at I2R in 2006, collaborating with clinicians closely, resulting in strong competency in ocular imaging, brain image analysis, cell image understanding etc at the institute. He has published over 200 journal and conference papers and owns 17 patents in the areas of computer vision, cognitive vision, pattern recognition, and medical image analysis.Sim-Heng Ong is an Associate Professor in the Departments of Electrical Engineering and Bioengineering at the National University of Singapore. He received his PhD from the University of Sydney, Australia. His major research areas are computer vision and medical image analysis and visualization. He has worked extensively with clinicians in developing algorithms for a variety of medical applications, and has publications in many highly respected journals and conferences.Wei Xiong is a Research Scientist at the Institute for Infocomm Research (I2R), A*STAR, Singapore. He obtained his PhD degree from the National University of Singapore.  His research interest is in computer vision, image processing, pattern classification and acoustic imaging. Dr. Xiong has published over 60 technical papers.

• List of Contributors xvPreface xixAcronyms xxiiiPART I INTRODUCTION 11 Overview of Biomedical Image Understanding Methods 3Wei Xiong, Jierong Cheng, Ying Gu, Shimiao Li and Joo Hwee Lim1.1 Segmentation and Object Detection 51.1.1 Methods Based on Image Processing Techniques 61.1.2 Methods Using Pattern Recognition and Machine Learning Algorithms 71.1.3 Model and Atlas-Based Segmentation 81.1.4 Multispectral Segmentation 91.1.5 User Interactions in Interactive Segmentation Methods 101.1.6 Frontiers of Biomedical Image Segmentation 111.2 Registration 111.2.1 Taxonomy of Registration Methods 121.2.2 Frontiers of Registration for Biomedical Image Understanding 151.3 Object Tracking 161.3.1 Object Representation 171.3.2 Feature Selection for Tracking 181.3.3 Object Tracking Technique 191.3.4 Frontiers of Object Tracking 191.4 Classification 201.4.1 Feature Extraction and Feature Selection 211.4.2 Classifiers 221.4.3 Unsupervised Classification 231.4.4 Classifier Combination 241.4.5 Frontiers of Pattern Classification for Biomedical Image Understanding 251.5 Knowledge-Based Systems 261.5.1 Semantic Interpretation and Knowledge-Based Systems 261.5.2 Knowledge-Based Vision Systems 271.5.3 Knowledge-Based Vision Systems in Biomedical Image Analysis 281.5.4 Frontiers of Knowledge-Based Systems 29References 29PARTII SEGMENTATION AND OBJECT DETECTION 472 Medical Image Segmentation and its Application in Cardiac MRI 49Dong Wei, Chao Li, and Ying Sun2.1 Introduction 502.2 Background 512.2.1 Active Contour Models 512.2.2 Parametric and Nonparametric Contour Representation 522.2.3 Graph-Based Image Segmentation 532.2.4 Summary 542.3 Parametric Active Contours – The Snakes 542.3.1 The Internal Spline Energy Eint 542.3.2 The Image-Derived Energy Eimg 552.3.3 The External Control Energy Econ 552.3.4 Extension of Snakes and Summary of Parametric Active Contours 572.4 Geometric Active Contours – The Level Sets 582.4.1 Variational Level Set Methods 582.4.2 Region-Based Variational Level Set Methods 602.4.3 Summary of Level Set Methods 642.5 Graph-Based Methods – The Graph Cuts 652.5.1 Basic Graph Cuts Formulation 652.5.2 Patch-Based Graph Cuts 662.5.3 An Example of Graph Cuts 682.5.4 Summary of Graph Cut Methods 722.6 Case Study: Cardiac Image Segmentation Using A Dual Level Sets Model 732.6.1 Introduction 732.6.2 Method 742.6.3 Experimental Results 792.6.4 Conclusion of the Case Study 812.7 Conclusion and Near-Future Trends 81References 833 Morphometric Measurements of the Retinal Vasculature in Fundus Images With Vampire 91Emanuele Trucco, Andrea Giachetti, Lucia Ballerini, Devanjali Relan, Alessandro Cavinato, and Tom Macgillivray3.1 Introduction 923.2 Assessing Vessel Width 933.2.1 Previous Work 933.2.2 Our Method 943.2.3 Results 953.2.4 Discussion 963.3 Artery or Vein? 983.3.1 Previous Work 983.3.2 Our Solution 993.3.3 Results 1013.3.4 Discussion 1033.4 Are My Program’s Measurements Accurate? 1043.4.1 Discussion 106References 1074 Analyzing Cell and Tissue Morphologies Using Pattern Recognition Algorithms 113Hwee Kuan Lee, Yan Nei Law, Chao-Hui Huang, and Choon Kong Yap 4.1 Introduction, 1134.2 Texture Segmentation of Endometrial Images Using the Subspace Mumford–Shah Model 1154.2.1 Subspace Mumford–Shah Segmentation Model 1164.2.2 Feature Weights 1184.2.3 Once-and-For-All Approach 1194.2.4 Results 1194.3 Spot Clustering for Detection of Mutants in Keratinocytes 1204.3.1 Image Analysis Framework 1234.3.2 Results 1244.4 Cells and Nuclei Detection 1244.4.1 Model 1254.4.2 Neural Cells and Breast Cancer Cells Data 1274.4.3 Performance Evaluation 1274.4.4 Robustness Study 1274.4.5 Results 1284.5 Geometric Regional Graph Spectral Feature 1344.5.1 Conversion of Image Patches into Region Signatures 1344.5.2 Comparing Region Signatures 1354.5.3 Classification of Region Signatures 1364.5.4 Random Masking and Object Detection 1364.5.5 Results 1374.6 Mitotic Cells in the H&E Histopathological Images of Breast Cancer Carcinoma 1384.6.1 Mitotic Index Estimation 1394.6.2 Mitotic Candidate Selection 1404.6.3 Exclusive Independent Component Analysis (XICA) 1404.6.4 Classification Using Sparse Representation 1434.6.5 Training and Testing Over Channels 1444.6.6 Results 1464.7 Conclusions 147References 147PARTIII REGISTRATION AND MATCHING 1535 3D Nonrigid Image Registration by Parzen-Window-Based Normalized Mutual Information and its Application on Mr-Guided Microwave Thermocoagulation of Liver Tumors 155Rui Xu, Yen-Wei Chen, Shigehiro Morikawa, and Yoshimasa Kurumi5.1 Introduction 1555.2 Parzen-Window-Based Normalized Mutual Information 1575.2.1 Definition of Parzen-Window Method 1575.2.2 Parzen-Window-Based Estimation of Joint Histogram 1585.2.3 Normalized Mutual Information and its Derivative 1605.3 Analysis of Kernel Selection 1635.3.1 The Designed Kernel 1635.3.2 Comparison in Theory 1675.3.3 Comparison by Experiments 1705.4 Application on MR-Guided Microwave Thermocoagulation of Liver Tumors 1745.4.1 Introduction of MR-Guided Microwave Thermocoagulation of Liver Tumors 1745.4.2 Nonrigid Registration by Parzen-Window-Based Mutual Information 1755.4.3 Evaluation on Phantom Data 1775.4.4 Evaluation on Clinical Cases 1805.5 Conclusion 185Acknowledgements 186References 1876 2D/3D Image Registration For Endovascular Abdominal Aortic Aneurysm (AAA) Repair 189Shun Miao and Rui Liao6.1 Introduction 1896.2 Background 1906.2.1 Image Modalities 1906.2.2 2D/3D Registration Framework 1926.2.3 Feature-Based Registration 1946.2.4 Intensity-Based Registration 1966.2.5 Number of Imaging Planes 1976.2.6 2D/3D Registration for Endovascular AAA Repair 1986.3 Smart Utilization of Two X-Ray Images for Rigid-Body 2D/3D Registration 1996.3.1 2D/3D Registration: Challenges in EVAR 1996.3.2 3D Image Processing and DRR Generation 2026.3.3 2D Image Processing 2036.3.4 Similarity Measure 2056.3.5 Optimization 2076.3.6 Validation 2106.4 Deformable 2D/3D Registration 2116.4.1 Problem Formulation 2126.4.2 Graph-Based Difference Measure 2136.4.3 Length Preserving Term 2156.4.4 Smoothness Term 2156.4.5 Optimization 2166.4.6 Validation 2176.5 Visual Check of Patient Movement Using Pelvis Boundary Detection 2206.6 Discussion and Conclusion 222References 223PARTIV OBJECT TRACKING 2297 Motion Tracking in Medical Images 231Chuqing Cao, Chao Li, and Ying Sun7.1 Introduction 2327.1.1 Point-Based Tracking 2337.1.2 Silhouette-Based Tracking 2337.1.3 Kernel-Based Tracking 2337.2 Background 2347.2.1 Point-Based Tracking 2347.2.2 Silhouette-Based Tracking 2367.2.3 Kernel-Based Tracking 2377.2.4 Summary 2387.3 Bayesian Tracking Methods 2387.3.1 Kalman Filters 2397.3.2 Particle Filters 2407.3.3 Summary of Bayesian Tracking Methods 2417.4 Deformable Models 2417.4.1 Mathematical Foundations of Deformable Models 2417.4.2 Energy-Minimizing Deformable Models 2427.4.3 Probabilistic Deformable Models 2447.4.4 Summary of Deformable Models 2457.5 Motion Tracking Based on the Harmonic Phase Algorithm 2467.5.1 HARP Imaging 2467.5.2 HARP Tracking 2487.5.3 Summary 2497.6 Case Study: Pseudo Ground Truth-Based Nonrigid Registration of MRI for Tracking the Cardiac Motion 2507.6.1 Data Fidelity Term 2517.6.2 Spatial Smoothness Constraint 2527.6.3 Temporal Smoothness Constraint 2537.6.4 Energy Minimization 2547.6.5 Preliminary Results 2557.6.6 Nonrigid Registration of Myocardial Perfusion MRI 2557.6.7 Experimental Results 2597.7 Discussion 2647.8 Conclusion and Near-Future Trends 265References 267PARTV CLASSIFICATION 2758 Blood Smear Analysis, Malaria Infection Detection, and Grading from Blood Cell Images 277Wei Xiong, Sim-Heng Ong, Joo-Hwee Lim, Jierong Cheng, and Ying Gu8.1 Introduction 2788.2 Pattern Classification Techniques 2828.2.1 Supervised and Nonsupervised Learning 2828.2.2 Bayesian Decision Theory 2838.2.3 Clustering 2848.2.4 Support Vector Machines 2868.3 GWA Detection 2878.3.1 Image Analysis 2888.3.2 Association between the Object Area and the Number of Cells Per Object 2898.3.3 Clump Splitting 2918.3.4 Clump Characterization 2938.3.5 Classification 2958.4 Dual-Model-Guided Image Segmentation and Recognition 2958.4.1 Related Work 2968.4.2 Strategies and Object Functions 2978.4.3 Endpoint Adjacency Map Construction and Edge Linking 2998.4.4 Parsing Contours and Their Convex Hulls 3008.4.5 A Recursive and Greedy Splitting Approach 3018.4.6 Incremental Model Updating and Bayesian Decision 3018.5 Infection Detection and Staging 3028.5.1 Related Work 3028.5.2 Methodology 3038.6 Experimental Results 3058.6.1 GWA Classification 3058.6.2 RBC Segmentation 3108.6.3 RBC Classification 3158.7 Summary 320References 3219 Liver Tumor Segmentation Using SVM Framework and Pathology Characterization Using Content-Based Image Retrieval 325Jiayin Zhou, Yanling Chi, Weimin Huang, Wei Xiong, Wenyu Chen, Jimin Liu, and Sudhakar K. Venkatesh9.1 Introduction 3259.2 Liver Tumor Segmentation Under a Hybrid SVM Framework 3279.2.1 Fundamentals of SVM for Classification 3279.2.2 SVM Framework for Liver Tumor Segmentation and the Problems 3309.2.3 A Three-Stage Hybrid SVM Scheme for Liver Tumor Segmentation 3319.2.4 Experiment 3349.2.5 Evaluation Metrics 3359.2.6 Results 3369.3 Liver Tumor Characterization by Content-Based Image Retrieval 3389.3.1 Existing Work and the Rationale of Using CBIR 3399.3.2 Methodology Overview and Preprocessing 3409.3.3 Tumor Feature Representation 3419.3.4 Similarity Query and Tumor Pathological Type Prediction 3439.3.5 Experiment 3459.3.6 Results 3469.4 Discussion 3519.4.1 About Liver Tumor Segmentation Using Machine Learning 3519.4.2 About Liver Tumor Characterization Using CBIR 3539.5 Conclusion 356References 35710 Benchmarking Lymph Node Metastasis Classification for Gastric Cancer Staging 361Su Zhang, Chao Li, Shuheng Zhang, Lifang Pang, and Huan Zhang10.1 Introduction 36210.1.1 Introduction of GSI-CT 36310.1.2 Imaging Findings of Gastric Cancer 36610.2 Related Feature Selection, Metric Learning, and Classification Methods 36710.2.1 Feature Extraction 36710.2.2 KNN 36710.2.3 Feature Selection 36910.2.4 AdaBoost and EAdaBoost Algorithms 37410.3 Preprocessing Method for GSI-CT Data 37710.3.1 Data Acquisition for GSI-CT Data 37710.3.2 Univariate Analysis 37810.4 Classification Results For GSI-CT Data of Gastric Cancer 37910.4.1 Experimental Results of mRMR-KNN 37910.4.2 Experimental Results of SFS-KNN 38310.4.3 Experimental Results of Metric Learning 38510.4.4 Experiments Results of AdaBoost and EAdaBoost 38510.4.5 Experiment Analysis 38810.5 Conclusion and Future Work 388Acknowledgment 388References 388PARTVI KNOWLEDGE-BASED SYSTEMS 39311 The Use of Knowledge in Biomedical Image Analysis 395Florence Cloppet11.1 Introduction 39511.2 Data, Information, and Knowledge? 39711.2.1 Data Versus Information 39711.2.2 Knowledge Versus Information 39811.3 What Kind of Information/Knowledge Can be Introduced? 39911.4 How to Introduce Information in Computer Vision Systems? 40011.4.1 Nature of Prior Information/Knowledge 40211.4.2 Frameworks Allowing Prior Information Introduction 40811.5 Conclusion 418References 41812 Active Shape Model for Contour Detection of Anatomical Structure 429Huiqi Li and Qing Nie12.1 Introduction 42912.2 Background 43012.2.1 Free-Form Deformable Models 43012.2.2 Parametrically Deformable Models 43212.3 Methodology 43412.3.1 Point Distribution Model 43412.3.2 Active Shape Model (ASM) 43612.3.3 A Modified ASM 43812.4 Applications 44012.4.1 Boundary Detection of Optic Disk 44012.4.2 Lens Structure Detection 45012.5 Summary 456Acknowledgment 457References 457Index 463