Statistical Atlases and Computational Models of the Heart. Regular and CMRxMotion Challenge Papers: 13th International Workshop, STACOM 2022 Held in Conjunction with MICCAI 2022 Singapore, September 18, 2022 Revised Selected Papers

Description

This book constitutes the proceedings of the 13th International Workshop on Statistical Atlases and Computational Models of the Heart, STACOM 2022, held in conjunction with the 25th MICCAI conference. The 34 regular workshop papers included in this volume were carefully reviewed and selected after being revised and deal with topics such as: common cardiac segmentation and modelling problems to more advanced generative modelling for ageing hearts, learning cardiac motion using biomechanical networks, physics-informed neural networks for left atrial appendage occlusion, biventricular mechanics for Tetralogy of Fallot, ventricular arrhythmia prediction by using graph convolutional network, and deeper analysis of racial and sex biases from machine learning-based cardiac segmentation. In addition, 14 papers from the CMRxMotion challenge are included in the proceedings which aim to assess the effects of respiratory motion on cardiac MRI (CMR) imaging quality and examine the robustness of segmentation models in face of respiratory motion artefacts. A total of 48 submissions to the workshop was received. Preface Organization Contents Regular Papers Generative Modelling of the Ageing Heart with Cross-Sectional Imaging and Clinical Data 1 Introduction 1.1 Related Work 1.2 Contributions 2 Methods 2.1 Problem Formulation 2.2 Conditional Generative Modelling 2.3 Training Scheme 3 Experiments 3.1 Datasets 3.2 Experimental Setup 3.3 Experiments and Results 4 Conclusion References Learning Correspondences of Cardiac Motion from Images Using Biomechanics-Informed Modeling 1 Introduction 2 Method 2.1 Biomechanics-Informed Modeling 2.2 Loss Function 3 Experiments 3.1 Dataset Details 3.2 Experimental Setup 3.3 Evaluation Metrics 4 Results 4.1 Visual and Quantitative Assessments 4.2 Effect of Hyperparameters 5 Discussion 6 Conclusion References Multi-modal Latent-Space Self-alignment for Super-Resolution Cardiac MR Segmentation 1 Introduction 2 Methodology 2.1 Interpolation via ABME for 2D MR Images 2.2 Segmentation Network 2.3 Multi-modal Latent-Space Self-alignment Network 2.4 Training Scheme and Implementation Details 3 Experiment 3.1 Datasets 3.2 Segmentation Results 4 Conclusion References Towards Real-Time Optimization of Left Atrial Appendage Occlusion Device Placement Through Physics-Informed Neural Networks*-4pt 1 Introduction 2 Methods 2.1 Computational Fluid Dynamics Simulations 2.2 Domain Setup 2.3 Physics-Informed Neural Network 3 Results 4 Discussion 5 Conclusion References Haemodynamic Changes in the Fetal Circulation Post-connection to an Artificial Placenta: A Computational Modelling Study 1 Introduction 2 Methodology 2.1 In vivo experiments 2.2 Lumped Model 2.3 Wave Reflections 2.4 Signal Analysis of Pressure and Flow Traces 3 Results 3.1 In vivo Pressure and Flow Signal Analysis 3.2 Lumped Model of the Artificial Placenta 3.3 Wave Reflections 4 Discussion 5 Conclusions References Personalized Fast Electrophysiology Simulations to Evaluate Arrhythmogenicity of Ventricular Slow Conduction Channels 1 Introduction 2 Material and Methods 2.1 Clinical Data 2.2 Digital Twin Construction 2.3 Cardiac Electrophysiology Modeling 2.4 Batch Simulations 3 Results 4 Discussion and Conclusions References Self-supervised Motion Descriptor for Cardiac Phase Detection in 4D CMR Based on Discrete Vector Field Estimations 1 Introduction 2 Related Work 3 Methods 3.1 Model Definition 3.2 Loss Function and Key Frame Extraction 3.3 Deep Learning Framework 3.4 Datasets 4 Evaluation and Experiments 5 Results 6 Discussion and Conclusion A Dataset Properties B Visual Examples References Going Off-Grid: Continuous Implicit Neural Representations for 3D Vascular Modeling 1 Introduction 2 Method 2.1 Signed Distance Functions 2.2 Implicit Neural Representations 2.3 Optimising an INR 2.4 Fitting Multiple Functions in an INR 2.5 Constructive Solid Geometry 2.6 Data 3 Experiments and Results 3.1 Robustness 3.2 Reconstructing Nested Shapes 3.3 Constructive Geometry 4 Discussion and Conclusion References Comparison of Semi- and Un-Supervised Domain Adaptation Methods for Whole-Heart Segmentation 1 Introduction 1.1 Problem 1.2 Related Work 1.3 Contributions and Aim 2 Method 2.1 Dataset 2.2 Experiments 3 Results 4 Conclusion 4.1 Future Work References Automated Quality Controlled Analysis of 2D Phase Contrast Cardiovascular Magnetic Resonance Imaging 1 Introduction 2 Methods 3 Materials 4 Experiments and Results 5 Discussion References An Atlas-Based Analysis of Biventricular Mechanics in Tetralogy of Fallot 1 Introduction 2 Methods 2.1 Study Population and Geometry Fitting 2.2 Atlas-Based Analysis of Systolic Wall Motion 2.3 Sensitivity of Biventricular Function to Systolic Wall Motion 2.4 Association of End-Diastolic Shape with Systolic Wall Motion 2.5 Finite Element Analysis of Biventricular Biomechanics 2.6 Statistical Analysis 3 Results 3.1 Atlas of Systolic Wall Motion and Associations with End-Diastolic Shape 3.2 Finite Element Analysis of Biventricular Biomechanics 4 Discussion 4.1 Shape Determinants of Biventricular Function 4.2 Limitations 5 Conclusions 6 Competing Interests References Review of Data Types and Model Dimensionality for Cardiac DTI SMS-Related Artefact Removal 1 Introduction 2 Background 2.1 Cardiac Diffusion Tensor Imaging 2.2 Simultaneous Multi-slice Acquisition (SMS) 2.3 Deep Learning in DTI 3 Methods 3.1 Complex Neural Networks 3.2 Experimental Setting 3.3 Results Evaluation 4 Results 5 Discussion 6 Conclusion References Improving Echocardiography Segmentation by Polar Transformation 1 Introduction 2 Methodology 2.1 Polar Transformation of Imaging Region 2.2 Joint Training and Testing 3 Experiments 3.1 Data 3.2 Experimental Details 3.3 Experimental Results 4 Conclusion References Spatiotemporal Cardiac Statistical Shape Modeling: A Data-Driven Approach 1 Introduction 2 Methods 2.1 Notation 2.2 Proposed Spatiotemporal Optimization Scheme 2.3 Image-Based Comparison Method 2.4 LDS Evaluation Metrics 3 Results 3.1 4D Left Atrium Data 3.2 Evaluation 4 Conclusion Appendix A Linear Dynamical System A.1 Time-Variant Model A.2 EM Algorithm References Interpretable Prediction of Post-Infarct Ventricular Arrhythmia Using Graph Convolutional Network 1 Introduction 2 Method 2.1 Image Processing 2.2 Graph Convolutional Network Model 2.3 Interpretability Study 3 Experimental Setup 3.1 Baseline Models 4 Results 4.1 Model Interpretability 5 Discussion 6 Conclusion References Unsupervised Echocardiography Registration Through Patch-Based MLPs and Transformers 1 Introduction 2 Methodology 2.1 Diffeomorphic Registration 2.2 Proposed Frameworks 2.3 Multi-scale Features 3 Experiments and Results 3.1 Dataset 3.2 Implementation 3.3 Experiments 3.4 Results 4 Conclusion References Sensitivity Analysis of Left Atrial Wall Modeling Approaches and Inlet/Outlet Boundary Conditions in Fluid Simulations to Predict Thrombus Formation 1 Introduction 2 Material and Methods 2.1 Clinical Data and 3D Model Generation 2.2 Computational Fluid Dynamic Simulations 2.3 In-silico Haemodynamic Indices 3 Results 4 Discussion and Conclusions A Appendix References APHYN-EP: Physics-Based Deep Learning Framework to Learn and Forecast Cardiac Electrophysiology Dynamics 1 Introduction 2 Learning Framework 3 Experimental Settings 3.1 In Silico Data 3.2 Ex Vivo Data 4 Results 4.1 In Silico Data 4.2 Ex Vivo Data 5 Discussion and Conclusion References Unsupervised Machine Learning Exploration of Morphological and Haemodynamic Indices to Predict Thrombus Formation in the Left Atrial Appendage 1 Introduction 2 Materials and Methods 2.1 Data 2.2 Methodological Pipeline 2.3 Data Pre-processing 2.4 Morphological Parameter Extraction 2.5 Unsupervised MKL Analysis 3 Results 3.1 Latent Space Exploration 3.2 Cluster Analysis 3.3 Validation of Cluster Analysis Results 4 Discussion and Conclusions References Geometrical Deep Learning for the Estimation of Residence Time in the Left Atria 1 Introduction 2 Methods 2.1 Dataset 2.2 Generation of Residence Time Fields for Training 2.3 Mesh Pre-processing 2.4 Geometric PointNet 2.5 Mesh U-Net 2.6 Hyperparameter Tuning 2.7 Model Evaluation 3 Results 4 Discussion 5 Conclusion References Explainable Electrocardiogram Analysis with Wave Decomposition: Application to Myocardial Infarction Detection 1 Introduction 2 Methods 2.1 Data Preprocessing 2.2 Cascaded FMMnet 3 Experiments and Results 3.1 Datasets 3.2 Reconstruction 3.3 Classification 4 Discussion and Conclusion References A Systematic Study of Race and Sex Bias in CNN-Based Cardiac MR Segmentation 1 Introduction 2 Materials 2.1 Experimental Setup 3 Methods 3.1 Model Evaluation 4 Results 4.1 Experiment 1: Male vs. Female 4.2 Experiment 2: White vs. Black 4.3 Experiment 3: White vs. Asian 5 Discussion References Mesh U-Nets for 3D Cardiac Deformation Modeling 1 Introduction 2 Methods 2.1 Dataset and Preprocessing 2.2 Network Design 2.3 Implementation and Training 3 Experiments and Results 3.1 Prediction Quality 3.2 Clinical Evaluation 3.3 Subpopulation-Specific Deformations 4 Discussion and Conclusion A 3D U-Net B Subpopulation-Specific Deformations References Skeletal Model-Based Analysis of the Tricuspid Valve in Hypoplastic Left Heart Syndrome 1 Introduction 2 Materials 2.1 Subjects 3 Methods 3.1 Image Segmentation and Model Creation 3.2 Skeletal Representations 3.3 Analysis of s-reps 3.4 Normalization 4 Evaluation 4.1 Principal Component Analysis 4.2 Distance Weighted Discrimination 5 Discussion References Simplifying Disease Staging Models into a Single Anatomical Axis - A Case Study of Aortic Coarctation In-utero 1 Introduction 2 Methods 2.1 Clinical Background and Dataset 2.2 Shape Encoding 2.3 Dimensionality Reduction 2.4 Phenotype Exploration 2.5 The Uniqueness of a Disease Staging Axis 3 Results 4 Discussion and Conclusions References Point2Mesh-Net: Combining Point Cloud and Mesh-Based Deep Learning for Cardiac Shape Reconstruction 1 Introduction 2 Methods 2.1 Overview 2.2 Datasets and Preprocessing 2.3 Network Architecture 2.4 Training and Implementation 3 Experiments and Results 3.1 Surface Reconstruction on Synthetic Dataset 3.2 Surface Reconstruction Pipeline on Real Dataset 4 Discussion and Conclusion References Post-Infarction Risk Prediction with Mesh Classification Networks 1 Introduction 2 Methods 2.1 Overview 2.2 Dataset and Preprocessing 2.3 Network Architecture and Training 3 Experiments and Results 3.1 End-Systolic Shape-Based Prediction 3.2 Contraction-Based Prediction 3.3 Effect of Class Imbalance 4 Discussion and Conclusion References Statistical Shape Modeling of Biventricular Anatomy with Shared Boundaries 1 Introduction 2 Methods 2.1 Background: Particle-Based Shape Modeling 2.2 Shared Boundary Extraction 2.3 Particle-Based Shape Modeling with Shared Boundaries 3 Experiments and Results 4 Conclusion A Appendix A.1 Modes of Variation References Computerized Analysis of the Human Heart to Guide Targeted Treatment of Atrial Fibrillation 1 Introduction 2 Methods 2.1 LGE-MRI Data and Automatic Segmentation 2.2 Standardized Atrial Geometry Fattening 2.3 Semi-automatic Rule-Based Fiber Generation 2.4 Efficient Electrophysiology Computer Modeling 3 Results 3.1 CNN-Based LGE-MRI Segmentation 3.2 Atrial Flattening 3.3 Three-Dimensional Fiber Orientation Generation 3.4 Modeling 4 Conclusion References 3D Mitral Valve Surface Reconstruction from 3D TEE via Graph Neural Networks 1 Introduction 2 Materials and Method 2.1 Dataset 2.2 Method 3 Results 4 Discussion and Conclusion References Efficient MRI Reconstruction with Reinforcement Learning for Automatic Acquisition Stopping 1 Introduction 2 Related Works 3 Methods 3.1 Reconstruction Network 3.2 Policy Network 4 Experimental Results 4.1 Results Analysis 5 Discussion and Conclusions References Unsupervised Cardiac Segmentation Utilizing Synthesized Images from Anatomical Labels 1 Introduction 2 Method 2.1 Unsupervised Segmentation Network with Intensity Constraint 2.2 Strong Shape Constraints 2.3 Overall Architecture 3 Experiment 3.1 Materials and Experimental Setups 3.2 Performance of Segmentation Network 3.3 Performance of Generator 4 Conclusion References PAT-CNN: Automatic Segmentation and Quantification of Pericardial Adipose Tissue from T2-Weighted Cardiac Magnetic Resonance Images 1 Introduction 2 Methods 3 Results 4 Discussion 5 Conclusions References Deep Computational Model for the Inference of Ventricular Activation Properties 1 Introduction 2 Methodology 2.1 Geometrical Triangular Mesh Generation 2.2 Simulated Data Generation via Eikonal Models 2.3 Patient Specific Deep Computational Model 3 Experiments and Results 3.1 Materials 3.2 Results 4 Conclusion References CMRxMotion Challenge Papers Semi-supervised Domain Generalization for Cardiac Magnetic Resonance Image Segmentation with High Quality Pseudo Labels 1 Introduction 2 Methodology 2.1 Data Augmentation Based on Fourier Transformation 2.2 Ensemble of Two Confidence-Aware Cross Pseudo Supervision 3 Experiments and Results 3.1 Dataset and Preprocessing 3.2 Experiment Setting 3.3 Experiment Results 4 Conclusion References Cardiac Segmentation Using Transfer Learning Under Respiratory Motion Artifacts 1 Introduction 2 Method 2.1 Data Resampling, Preprocessing and Normalisation 2.2 Architecture Study 2.3 Data Augmentation 2.4 Cardiac MRI Dataset 3 Experiment Settings 4 Results 4.1 Validation Results 4.2 Evaluation Results 4.3 Test Results 5 Conclusions References Deep Learning Based Classification and Segmentation for Cardiac Magnetic Resonance Imaging with Respiratory Motion Artifacts 1 Introduction 2 Materials and Methods 2.1 CMRxMotion Data 2.2 Deep Learning Algorithms 3 Results 3.1 Experimental Setup 3.2 Results 4 Discussion 5 Conclusion References Multi-task Swin Transformer for Motion Artifacts Classification and Cardiac Magnetic Resonance Image Segmentation 1 Introduction 2 Related Works 3 Method 3.1 Multitask Swin UNETR 4 Experiments and Results 4.1 Dataset 4.2 Loss Function 4.3 Implementation Details 4.4 Evaluation Metrics 4.5 Motion Artifacts Classification 4.6 CMR Segmentation 5 Conclusions References Automatic Quality Assessment of Cardiac MR Images with Motion Artefacts Using Multi-task Learning and K-Space Motion Artefact Augmentation 1 Introduction 2 Dataset 3 Methods 3.1 Cardiac Image Quality Classification 3.2 Cardiac Segmentation 3.3 Training 4 Results and Discussion 5 Conclusion References Motion-Related Artefact Classification Using Patch-Based Ensemble and Transfer Learning in Cardiac MRI 1 Introduction 2 Methods 2.1 Image Pre-processing Using Gradient Magnitude 2.2 Image Patch Sampling 2.3 Transfer-Learning-Based Model Ensemble 3 Experiments and Results 3.1 Dataset 3.2 Batch Balanced Training 3.3 Bias Voting in Testing 3.4 Other Parameter Setting 3.5 Experiments and Results 4 Conclusions and Discussion References Automatic Image Quality Assessment and Cardiac Segmentation Based on CMR Images 1 Introduction 2 Methods 2.1 Task I: CMR Image Quality Classification 2.2 Task II: CMR Image Segmentation 3 Results 3.1 Task I: CMR Image Quality Classification 3.2 Task II: CMR Image Segmentation 4 Discussions and Conclusions 4.1 Task I: Classification 4.2 Task II: Segmentation References Detecting Respiratory Motion Artefacts for Cardiovascular MRIs to Ensure High-Quality Segmentation 1 Introduction 2 Method 2.1 Image Quality Assessment of Respiratory Motion Artefacts 2.2 CMR Image Segmentation with Realistic Respiratory Motion 3 Results 3.1 Image Quality Assessment of Respiratory Motion Artefacts 3.2 CMR Image Segmentation with Realistic Respiratory Motion 4 Discussion 5 Conclusion References 3D MRI Cardiac Segmentation Under Respiratory Motion Artifacts 1 Introduction 1.1 Related Works 2 Methods 2.1 DenseBiasNet Part 2.2 VAE Part 2.3 Loss 2.4 Optimization 2.5 Data Pre-processing 3 Experiments and Results 4 Discussions and Conclusion References Cardiac MR Image Segmentation and Quality Control in the Presence of Respiratory Motion Artifacts Using Simulated Data 1 Introduction 2 Methods 2.1 Simulation of Respiratory Motion Artifact 2.2 Image Quality Assessment 2.3 Cardiac Image Segmentation 2.4 Data 3 Results 3.1 Image Quality Assessment 3.2 Cardiac Image Segmentation 4 Discussion and Conclusion References Combination Special Data Augmentation and Sampling Inspection Network for Cardiac Magnetic Resonance Imaging Quality Classification 1 Introduction 2 Materials and Methods 2.1 Dataset 2.2 Data Augmentation 2.3 Sampling Inspection Network Architecture 3 Experiments and Results 4 Conclusion References Automatic Cardiac Magnetic Resonance Respiratory Motions Assessment and Segmentation 1 Introduction 2 Respiratory Motions Assessment and Segmentation 3 Dataset and Evaluation Measures 4 Implementation Details 4.1 Pre-processing 4.2 Post-processing 4.3 Environments and Requirements 4.4 Training Protocols 4.5 Testing Protocols 5 Results 6 Conclusion References Robust Cardiac MRI Segmentation with Data-Centric Models to Improve Performance via Intensive Pre-training and Augmentation 1 Introduction 2 Methods 2.1 Dataset 2.2 Network Architecture 2.3 Pre-training 2.4 Data Augmentation 2.5 Training Protocol 3 Experiments 3.1 Architectural Variants 3.2 Data-Driven Methods with Pre-training and Augmentation 3.3 Ensemble 4 Discussion and Conclusion References A Deep Learning-Based Fully Automatic Framework for Motion-Existing Cine Image Quality Control and Quantitative Analysis 1 Introduction 2 Methods 2.1 Datasets 2.2 Quality Control 2.3 Segmentation 3 Results 4 Conclusions References Author Index