Machine Learning for Medical Image Reconstruction: 4th International Workshop, MLMIR 2021, Held in Conjunction with MICCAI 2021, Strasbourg, France, ... Vision, Pattern Recognition, and Graphics)

This book constitutes the refereed proceedings of the 4th International Workshop on Machine Learning for Medical Reconstruction, MLMIR 2021, held in conjunction with MICCAI 2021, in October 2021. The workshop was planned to take place in Strasbourg, France, but was held virtually due to the COVID-19 pandemic. The 13 papers presented were carefully reviewed and selected from 20 submissions. The papers are organized in the following topical sections: deep learning for magnetic resonance imaging and deep learning for general image reconstruction. Preface 6 Organization 7 Contents 8 Deep Learning for Magnetic Resonance Imaging 10 HyperRecon: Regularization-Agnostic CS-MRI Reconstruction with Hypernetworks 11 1 Introduction 11 2 Background 12 2.1 Amortized Optimization of CS-MRI 12 2.2 Hypernetworks 13 3 Proposed Method 13 3.1 Regularization-Agnostic Reconstruction Network 13 3.2 Training 14 4 Experiments 15 4.1 Hypernetwork Capacity and Hyperparameter Sampling 17 4.2 Range of Reconstructions 18 5 Conclusion 19 References 19 Efficient Image Registration Network for Non-Rigid Cardiac Motion Estimation 22 1 Introduction 22 2 Method 24 2.1 Network Architecture 24 2.2 Self-supervised Loss Function 24 2.3 Enhancement Mask (EM) 26 3 Experiments 27 4 Results 27 5 Discussion 29 6 Conclusion 30 References 30 .26em plus .1em minus .1emEvaluation of the Robustness of Learned MR Image Reconstruction to Systematic Deviations Between Training and Test Data for the Models from the fastMRI Challenge*-6pt 33 1 Introduction 34 2 Methods 35 2.1 Image Perturbations 35 2.2 Description of 2019 fastMRI Approaches 36 3 Results 37 4 Discussion and Conclusion 40 References 41 Self-supervised Dynamic MRI Reconstruction 43 1 Introduction 43 2 Theory 45 2.1 Dynamic MRI Reconstruction 45 2.2 Self-supervised Learning 46 3 Methods 47 4 Experimental Results 48 5 Conclusion 50 References 51 A Simulation Pipeline to Generate Realistic Breast Images for Learning DCE-MRI Reconstruction 53 1 Introduction 53 2 Method 54 2.1 DCE-MRI Data Acquisition 54 2.2 Pharmacokinetics Model Analysis and Simulation 55 2.3 MR Acquisition Simulation 55 2.4 Testing with ML Reconstruction 56 3 Result 57 4 Discussion 59 5 Conclusion 60 References 61 Deep MRI Reconstruction with Generative Vision Transformers 62 1 Introduction 62 2 Theory 64 2.1 Deep Unsupervised MRI Reconstruction 64 2.2 Generative Vision Transformers 64 3 Methods 67 4 Results 69 5 Discussion 70 6 Conclusion 70 References 70 Distortion Removal and Deblurring of Single-Shot DWI MRI Scans 73 1 Introduction 73 2 Background 76 2.1 Distortion Removal Framework 76 2.2 EDSR Architecture 76 3 Distortion Removal and Deblurring of EPI-DWI 76 3.1 Data 76 3.2 Distortion Removal Using Structural Images 77 3.3 Pre-processing for Super-Resolution 77 3.4 Data Augmentation 77 3.5 Architectures Explored for EPI-DWI Deblurring 78 4 Experiments and Results 78 4.1 Computer Hardware Details 78 4.2 Training Details 79 4.3 Baselines 79 4.4 Evaluation Metrics 79 4.5 Results 79 5 Conclusion 81 References 81 One Network to Solve Them All: A Sequential Multi-task Joint Learning Network Framework for MR Imaging Pipeline 84 1 Introduction 85 2 Method 86 2.1 SampNet: The Sampling Pattern Learning Network 86 2.2 ReconNet: The Reconstruction Network 86 2.3 SegNet: The Segmentation Network 87 2.4 SemuNet: The Sequential Multi-task Joint Learning Network Framework 87 3 Experiments and Discussion 89 3.1 Experimental Details 89 3.2 Experiments Results 90 4 Limitation, Discussion and Conclusion 92 References 92 Physics-Informed Self-supervised Deep Learning Reconstruction for Accelerated First-Pass Perfusion Cardiac MRI 94 1 Introduction 94 2 Methods 95 2.1 Conventional FPP-CMR Reconstruction 95 2.2 Supervised Learning Reconstruction: MoDL 96 2.3 SECRET Reconstruction 96 2.4 Dataset 97 2.5 Implementation Details 98 3 Results and Discussion 98 4 Conclusion 102 References 102 Deep Learning for General Image Reconstruction 104 Noise2Stack: Improving Image Restoration by Learning from Volumetric Data 105 1 Introduction and Related Work 105 2 Methods 107 3 Experiments 108 3.1 MRI 108 3.2 Microscopy 109 4 Discussion 111 5 Conclusion 112 References 113 Real-Time Video Denoising to Reduce Ionizing Radiation Exposure in Fluoroscopic Imaging 115 1 Introduction 115 1.1 Background 116 1.2 Our Contributions 117 2 Methods 117 2.1 Data 117 2.2 Training Pair Simulation 117 2.3 Denoising Model 118 2.4 Model Training 119 3 Experiments 120 3.1 Reader Study 120 3.2 Video Quality 120 3.3 Runtime 121 4 Conclusion 122 References 124 A Frequency Domain Constraint for Synthetic and Real X-ray Image Super Resolution 126 1 Introduction 126 2 Related Work 127 3 Methods 128 3.1 Frequency Domain Analysis 129 3.2 Frequency Domain Loss 129 4 Experiments 130 4.1 Dataset 130 4.2 Training Details 131 4.3 Results 131 4.4 Ablation Study 133 5 Conclusion 134 References 134 Semi- and Self-supervised Multi-view Fusion of 3D Microscopy Images Using Generative Adversarial Networks 136 1 Introduction 136 2 Related Work 137 3 Methods 137 4 Experiments and Results 140 4.1 Datasets 140 4.2 Existing Methods for Comparison 140 4.3 CNN-Based Multi-View Deconvolution and Fusion 142 5 Conclusions 143 References 144 Author Index 146