Shape in Medical Imaging: International Workshop, ShapeMI 2023, Held in Conjunction with MICCAI 2023, Vancouver, BC, Canada, October 8, 2023, Proceedings (Lecture Notes in Computer Science)

This volume comprises the proceedings of the International Workshop, ShapeMI 2023, which took place alongside MICCAI 2023 on October 8, 2023, in Vancouver, British Columbia, Canada. The 23 selected full papers deal with all aspects of leading methods and applications for advanced shape analysis and geometric learning in medical imaging. Preface Organization Contents Anatomy Completor: A Multi-class Completion Framework for 3D Anatomy Reconstruction 1 Introduction 2 Methods 2.1 Problem Formulation 2.2 Denoising Auto-Encoder with Residual Connections 2.3 Loss Aggregation for Random Anatomy Completion 2.4 Multi-class Anatomy Completion 3 Experiments and Results 3.1 Dataset and Pre-processing 3.2 Implementation Details 3.3 Experimental Setup 3.4 Results 4 Discussion and Conclusion References C3Fusion: Consistent Contrastive Colon Fusion, Towards Deep SLAM in Colonoscopy 1 Introduction 2 Method Overview 3 Deep-Depth and Deep-Descriptors 3.1 Deep-Depth Self-supervised Training 3.2 Deep-Descriptors 4 Pose Alignment 4.1 Feature Matching 4.2 Hierarchical Pose Optimization 4.3 Registration as Pose-Graph Optimization 5 Scene Reconstruction 6 Results 6.1 Quantitative Results 6.2 Qualitative Results 7 Conclusion A Appendix A.1 Synthetic Data Generation A.2 Depth Training and Implementation Details A.3 Correspondence Matching Qualitative Results A.4 Comparison of the Estimated Trajectories and Ground Truth Trajectories A.5 Extra Qualitative Depth-Map Predictions Results A.6 Extra Qualitative 3D Reconstruction Results on Colon10K A.7 SuperPoint Training A.8 Supplementary Video Results References Anatomy-Aware Masking for Inpainting in Medical Imaging 1 Introduction 2 Methodology 3 Experiments 3.1 Experimental Setup 3.2 Results 4 Discussion and Conclusion References Particle-Based Shape Modeling for Arbitrary Regions-of-Interest 1 Introduction 2 Method 2.1 Quadratic Penalty for Efficient Constrained PDM Construction 2.2 Free-Form Constraints 2.3 Graphical Interface Tool 3 Results 4 Conclusion References Optimal Coronary Artery Segmentation Based on Transfer Learning and UNet Architecture 1 Introduction 2 Methods 2.1 Dataset 2.2 Network Implementation 2.3 Separation of the Coronary Tree into Three Regions: Proximal, Middle and Distal 2.4 Evaluation Metrics 3 Results 4 Discussion and Conclusion References Unsupervised Learning of Cortical Surface Registration Using Spherical Harmonics 1 Introduction 2 Method 2.1 Problem Statement 2.2 Velocity Encoding 2.3 Velocity Field 2.4 Warp Trajectory 2.5 Architecture 3 Experimental Setup 3.1 Imaging Data 3.2 Baseline Methods 3.3 Evaluation Metrics 4 Results 5 Discussion 6 Conclusion References Unsupervised Correspondence with Combined Geometric Learning and Imaging for Radiotherapy Applications 1 Introduction 2 Materials and Method 2.1 Dataset 2.2 Pre-processing 2.3 Model 2.4 Incorporating Imaging Information 2.5 Comparison with Non-rigid Image Registration 2.6 Evaluation Metrics 2.7 Implementation Details 2.8 Experiments 3 Results 4 Discussion 5 Conclusion References ADASSM: Adversarial Data Augmentation in Statistical Shape Models from Images 1 Introduction 2 Methodology 2.1 Adversarial Data Augmentation Block 2.2 Adversary to Image-To-SSM Network 2.3 Image-To-SSM Network: 2.4 Shape Regularization Loss: 3 Results 3.1 Metrics 3.2 Femur 3.3 Left Atrium 3.4 Training Time 4 Conclusion and Future Work References Body Fat Estimation from Surface Meshes Using Graph Neural Networks 1 Introduction 2 Background and Related Work 2.1 Body Fat Estimation from Medical Imaging 2.2 Triangulated Meshes 2.3 Graph Neural Networks 3 Methods 4 Experiments and Results 4.1 Data Processing 4.2 Results 5 Discussion and Conclusion 6 Limitations and Future Work References Geometric Learning-Based Transformer Network for Estimation of Segmentation Errors 1 Introduction 2 Methods 2.1 Architecture 2.2 Pre-training Tasks 3 Dataset Description 3.1 Generation of Training Data 4 Experiments and Results 4.1 Implementation Details 4.2 Results and Discussion 4.3 Ablation Study 5 Conclusion References On the Localization of Ultrasound Image Slices Within Point Distribution Models 1 Introduction 2 Related Work 2.1 SSMs and Image Atlases in Medical Imaging 2.2 Multi-modal Registration 3 Method 3.1 Encoder Training 3.2 Slice Localization 4 Experiments 4.1 Multi-modal Registration 5 Results and Discussion 5.1 Limitations and Future Work 6 Conclusion References FSJP-Net: Foreground and Shape Joint Perception Network for Glomerulus Detection 1 Introduction 2 Method 2.1 Overview of the Framework 2.2 Foreground Perception Branch 2.3 Shape Perception Branch 2.4 Multiscale Fusion Module 2.5 Ellipse Detection Module 3 Experiments 3.1 Dataset and Implementation Details 3.2 Comparison Experiments 3.3 Ablation Experiments 3.4 Morphological Analysis 4 Conclusion References Progressive DeepSSM: Training Methodology for Image-To-Shape Deep Models 1 Introduction 2 Related Works 3 Methodology 3.1 Datasets 3.2 Training Data 3.3 Model Architecture 3.4 Loss Function 3.5 Evaluation Metric 3.6 Training Procedure 4 Results 4.1 Femur 4.2 Left Atrium 5 Ablation Studies 6 Conclusion References Muscle Volume Quantification: Guiding Transformers with Anatomical Priors 1 Introduction 2 Related Work 3 Methodology 3.1 Model 3.2 Prior Anatomical Knowledge 3.3 Loss Function 4 Experimental Validation 4.1 Experimental Settings 4.2 Quantitative Results 4.3 Qualitative Results 5 Conclusion References Geodesic Logistic Analysis of Lumbar Spine Intervertebral Disc Shapes in Supine and Standing Positions 1 Introduction 2 Materials 2.1 Demographics 2.2 MRI Acquisition and Segmentation 2.3 Correspondence Estimation 3 Methods 3.1 Shape Space and Geodesics 3.2 Logistic Regression in Shape Space 3.3 Parameter Estimation 4 Results and Discussions 4.1 Test on Low Dimensional Synthetic Data 4.2 Analysis of IVD Data 4.3 Limitations and Future Work 5 Conclusions References SlicerSALT: From Medical Images to Quantitative Insights of Anatomy 1 Introduction 2 Shape Representations and Correspondence 2.1 Skeletal Representations 2.2 Registration Based Correspondence 2.3 Shape Population Viewer 3 Shape Analysis Methods 3.1 Shape Variation Analyzer 3.2 Distance Weighted Discrimination 3.3 Deep Learning for Geometry: FlyBy CNN 4 Infrastructure 4.1 Pipelines 4.2 Sample Data and Tutorials 5 Discussion References Predicting Shape Development: A Riemannian Method 1 Introduction 2 Method 3 Experiments 3.1 Data and Methodology 3.2 Results 4 Conclusion References AReg IOS: Automatic Registration on IntraOralScans 1 Introduction 2 Related Work 3 Method 3.1 Data 3.2 Automated Standardized Orientation (ASO) 3.3 Pre-processing: Training Reference Patch 3.4 Training Butterfly Patch 3.5 Butterfly Patch Prediction 3.6 Automatic Registration (AReg) 3.7 Evaluation Metrics 4 Result 4.1 Orientation 4.2 Butterfly Patch 4.3 Registration 5 Discussion 6 Conclusion References Modeling Longitudinal Optical Coherence Tomography Images for Monitoring and Analysis of Glaucoma Progression 1 Introduction 2 Methodology 2.1 Subject-Wise Level 2.2 Population Level 3 Experimental Results 3.1 Data Description 3.2 Subject-Wise Geodesic Regression 3.3 Longitudinal Modeling at the Population Level 4 Conclusion References IcoConv: Explainable Brain Cortical Surface Analysis for ASD Classification 1 Introduction 2 Related Work 2.1 ASD Classification 2.2 3D Shape Analysis 2.3 Explainable Artificial Intelligence 3 Materials 4 Method Description 4.1 Rendering the 2D Views 4.2 Architecture 4.3 Training the Models 4.4 Explainability Maps 5 Results 6 Conclusion References DeCA: A Dense Correspondence Analysis Toolkit for Shape Analysis 1 Introduction 2 Methods 3 Experimental Results References 3D Shape Analysis of Scoliosis 1 Introduction 1.1 Related Work 2 The 3D Geometry of the Spine and Vertebral Canal 2.1 Measuring the Deviation of the Curves 2.2 Curvature of the Spine Curve 2.3 Angle of Spinal Axial Rotation 3 Results and Discussion 3.1 Dataset 3.2 Geometry of the Spine: Deviation of the Spine and Vertebral Canal 3.3 Curvature Measurement in MRI and Relation to Axial Plane 4 Conclusion A Segmentation A.1 Segmentation Network B Spline Fitting B.1 2D Spline Fitting B.2 3D Spline Fitting References SADIR: Shape-Aware Diffusion Models for 3D Image Reconstruction 1 Introduction 2 Background: Fréchet Mean via Atlas Building 3 Our Method: SADIR 3.1 Shape-Aware Diffusion Models Based on Atlas Building Network 3.2 Network Loss and Optimization 4 Experimental Evaluation 4.1 Experimental Settings 4.2 Experimental Results 5 Conclusion References Author Index