Deformable Meshes for Medical Image Segmentation (eBook)

Dagmar Kainmueller works as a research scientist at the Max Planck Institute of Molecular Cell Biology and Genetics in Dresden, Germany, with a focus on bio image analysis.

Preface by the Series Editor 7
Foreword 8
Acknowledgments 9
Abstract 10
Contents 11
Chapter 1 Introduction 15
1.1 Motivation 15
1.1.1 Segmentation of Medical Image Data 16
1.1.2 Automation of the Segmentation Task 18
1.1.3 Segmentation Accuracy 19
1.2 Scope of this Thesis 19
1.2.1 Automated Segmentation with Deformable Models 19
1.2.2 Selection of Anatomical Structures and Imaging Modalities 23
1.2.3 Contribution 23
1.2.4 Topics Not Discussed 25
1.3 Structure of this Thesis 26
Part I The Segmentation Framework 28
Chapter 2 Basic Terms and Notation 29
2.1 Images, Segmentations, and Surface Meshes 30
2.1.1 Three-dimensional Medical Images 30
2.1.2 Segmentations of Three-dimensional Medical Images 30
2.1.3 Triangle Surface Meshes 31
2.1.4 From Segmentations to Surface Meshes and Back 33
2.2 Deformable Surface Meshes 33
2.2.1 Displacement Fields and Sets of Candidate Displacements 34
2.2.2 Appearance Cost 34
Chapter 3 Deformable Meshes for Automatic Segmentation 36
3.1 Statistical Shape Models (SSMs) for Segmentation 38
3.1.1 Generation of SSMs 39
3.1.2 Prerequisites: Shape Correspondences and Alignment 40
3.1.3 Image Segmentation via SSM Deformation 41
3.1.4 Initial Shape Detection 43
3.1.5 Lack of Image Features: SSMs for Extrapolation 44
3.2 A Simple Heuristic Appearance Model 45
3.2.1 Appearance Cost Function 45
3.2.2 Intensity Parameter Estimation 46
3.3 Local Search for Appearance Match 46
3.3.1 Unidirectional Displacements 46
3.3.2 Optimal Displacement Fields 47
3.3.3 Intensity Profiles 48
3.4 Shape-constrained Free Mesh Deformations 48
3.4.1 Free Deformation within a Narrow Band 50
3.4.2 Free Deformation with Bounded Displacement Differences 51
3.5 Simultaneous Free Deformations of Multiple Meshes 54
3.5.1 Multi-object Graph-based Deformation of Coupled Meshes 55
3.5.2 Coupling Adjacent Surface Meshes 56
3.6 Conclusion 60
Chapter 4 Omnidirectional Displacements for Deformable Surfaces (ODDS) 61
4.1 The Visibility Problem 62
4.2 ODDS: Free Mesh Deformations with All-around Visibility 63
4.2.1 Omnidirectional Displacements 64
4.2.2 The Mesh Deformation Problem 65
4.2.3 Optimal Mesh Deformation via MRF Energy Minimization 66
4.2.4 Refined Regularization 66
4.2.5 Proof of Concept Synthetic Experiments 67
4.3 FastODDS 70
4.3.1 Where to use Omnidirectional Displacements 71
4.3.2 The Hybrid Mesh Deformation Problem 72
4.3.3 Optimal Hybrid Mesh Deformation 72
4.3.4 Multi-object FastODDS 74
4.3.5 Appendix: Automatic Ridge Detection 74
4.4 Conclusion 75
Chapter 5 From Surface Mesh Deformations to Volume Deformations 77
5.1 Mesh-based Extrapolation 78
5.1.1 Introduction 78
5.1.2 Affine Transformations 80
5.1.3 Polyaffine Transformations 80
5.1.4 Mean Value Coordinates 82
5.2 Atlas-based Segmentation 83
5.2.1 Image-to-image Registration 84
5.2.2 Application of Volume Deformations to Atlases 86
5.3 Conclusion 86
Part IIApplications to Medical Image Data 88
Chapter 6 Fundamentals of Quantitative Evaluation 89
6.1 Measures of Segmentation Accuracy 90
6.2 Presentation of Results 92
6.3 Comparison of Methods 93
6.4 Generalization to New Image Data 94
6.5 Parameter Settings 94
Chapter 7 Single-object Segmentation of Anatomical Structures 96
7.1 Segmentation of the Liver in Contrast-enhanced CT 98
7.1.1 Statistical Shape Model of the Liver 99
7.1.2 Application-specific Initialization 99
7.1.3 Heuristic Appearance Model for Displacement Computation 99
7.1.4 Segmentation Pipeline 102
7.1.5 Results and Discussion 103
7.2 Segmentation of the Pelvic Bones in CT 107
7.2.1 Image Data and SSM of the Pelvic Bones 108
7.2.2 Segmentation Pipeline 109
7.2.3 Results and Discussion 109
7.3 Segmentation of the Mandibular Bone and Nerve in CBCT 115
7.3.1 Image Data and Compound SSM of Mandible and Nerves 116
7.3.2 SSM-Based Reconstruction of Bone and Nerve 116
7.3.3 Image-based Refinement of Nerve Delineation 117
7.3.4 Results and Discussion 119
7.4 Conclusion 121
Chapter 8 Multi-object Segmentation of Joints 123
8.1 Segmentation of the Hip Joint in CT Data 124
8.2 Segmentation of Knee Bones and Cartilage in MR Data 127
8.2.1 SSMs of Femur and Tibia and Cartilage Thickness Model 128
8.2.2 Appearance Cost Functions and Parameter Estimation 129
8.2.3 Multi-object Segmentation Pipeline 131
8.2.4 Results and Discussion 132
8.3 Conclusion 132
Chapter 9 ODDS for Segmentation of Highly Curved Structures 136
9.1 Experimental Setup 137
9.1.1 Identification of the Mandibular Coronoid Process 138
9.1.2 Acetabular Rim Delineation on Surface Meshes 139
9.2 Results 140
9.2.1 Mandibular Coronoid Process 140
9.2.2 Acetabular Rim and Hip Bones 142
9.2.3 Run-time and Memory Requirements 143
9.3 Discussion 144
9.3.1 Segmentation Accuracy 144
9.3.2 Comparability of Regularization 146
9.3.3 Influence of Mesh Resolution 147
9.3.4 Consistency of Deformed Meshes 148
9.3.5 Run-time and Memory Requirements 149
9.4 Conclusion 150
Chapter 10 Extrapolation and Atlas-based Segmentation of Leg Muscles 151
10.1 Extraction of Anatomical Landmarks of the Pelvic Bones 152
10.1.1 Methods 153
10.1.2 Results and Discussion 155
10.2 Segmentation of Leg Muscles 156
10.2.1 Segmentation Pipeline 158
10.2.2 Results and Discussion 160
10.3 Conclusion 163
Conclusions 165
Publications 167
Bibliography 169