Medical Image Computing and Computer Assisted Intervention - MICCAI 2021 24th International Conference, Strasbourg, France, September 27-October 1, 2021, Proceedings, Part III. Image Processing, Computer Vision, Pattern Recognition, and Graphics

The eight-volume set LNCS 12901, 12902, 12903, 12904, 12905, 12906, 12907, and 12908 constitutes the refereed proceedings of the 24th International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2021, held in Strasbourg, France, in September/October 2021.* The 531 revised full papers presented were carefully reviewed and selected from 1630 submissions in a double-blind review process. The papers are organized in the following topical sections: Part I: image segmentation Part II: machine learning - self-supervised learning; machine learning - semi-supervised learning; and machine learning - weakly supervised learning Part III: machine learning - advances in machine learning theory; machine learning - attention models; machine learning - domain adaptation; machine learning - federated learning; machine learning - interpretability / explainability; and machine learning - uncertainty Part IV: image registration; image-guided interventions and surgery; surgical data science; surgical planning and simulation; surgical skill and work flow analysis; and surgical visualization and mixed, augmented and virtual reality Part V: computer aided diagnosis; integration of imaging with non-imaging biomarkers; and outcome/disease prediction Part VI: image reconstruction; clinical applications - cardiac; and clinical applications - vascular Part VII: clinical applications - abdomen; clinical applications - breast; clinical applications - dermatology; clinical applications - fetal imaging; clinical applications - lung; clinical applications - neuroimaging - brain development; clinical applications - neuroimaging - DWI and tractography; clinical applications - neuroimaging - functional brain networks; clinical applications - neuroimaging – others; and clinical applications - oncology Part VIII: clinical applications - ophthalmology; computational (integrative) pathology; modalities - microscopy; modalities - histopathology; and modalities - ultrasound *The conference was held virtually. Preface Organization Contents – Part III Machine Learning - Advances in Machine Learning Theory Towards Robust General Medical Image Segmentation 1 Introduction 2 Methodology 2.1 Adversarial Robustness 2.2 Generic Medical Segmentation 3 Experiments 3.1 Adversarial Robustness Assessment 3.2 General Medical Segmentation 4 Results 4.1 Adversarial Robustness 4.2 Results on the Medical Segmentation Decathlon 5 Conclusion References Joint Motion Correction and Super Resolution for Cardiac Segmentation via Latent Optimisation 1 Introduction 2 Methods 2.1 Problem Formulation and Preliminaries 2.2 Proposed Latent Optimisation Framework 2.3 Implementation 3 Experiments 3.1 Datasets 3.2 Experimental Design 4 Results and Discussion 4.1 Benchmark on Simulated Degradation 4.2 Real-World Performance 5 Conclusion References Targeted Gradient Descent: A Novel Method for Convolutional Neural Networks Fine-Tuning and Online-Learning 1 Introduction 2 Methodology 2.1 Targeted Gradient Descent Layer 3 TGD in PET Denoising 3.1 Fine-Tuning 3.2 Online-Learning 4 Experiments 4.1 Evaluation of TGD on Whole-Body PET Denoising 5 Conclusion References A Hierarchical Feature Constraint to Camouflage Medical Adversarial Attacks 1 Introduction 2 Related Work 3 Why Are Medical AEs Easy to Detect? 4 Adversarial Attack with a Hierarchical Feature Constraint 5 Experiments 5.1 Setup 5.2 Experimental Results 6 Conclusion References Group Shift Pointwise Convolution for Volumetric Medical Image Segmentation 1 Introduction 2 Method 2.1 Grouping 2.2 Shift 2.3 Formulation of Group Shift 3 Experiments 3.1 Results on PROMISE12 3.2 Results on BraTS18 4 Conclusion References Machine Learning - Attention Models UTNet: A Hybrid Transformer Architecture for Medical Image Segmentation 1 Introduction 2 Method 2.1 Revisiting Self-attention Mechanism 2.2 Efficient Self-attention Mechanism 2.3 Relative Positional Encoding 2.4 Network Architecture 3 Experiment 3.1 Experiment Setup 3.2 Implementation Detail 3.3 Segmentation Results 4 Conclusion References AlignTransformer: Hierarchical Alignment of Visual Regions and Disease Tags for Medical Report Generation 1 Introduction 2 Related Works 3 Approach 3.1 Problem Formulation 3.2 Align Hierarchical Attention 3.3 Multi-grained Transformer 4 Experiments 5 Analysis 6 Conclusions References Continuous-Time Deep Glioma Growth Models 1 Introduction 2 Methods 2.1 Model 2.2 Data and Evaluation 3 Results 4 Discussion References Spine-Transformers: Vertebra Detection and Localization in Arbitrary Field-of-View Spine CT with Transformers 1 Introduction 2 Method 2.1 Vertebrae Detection as an One-to-one Set Prediction 2.2 The Spine-Transformers 2.3 Implementation Details 3 Experiments and Results 4 Conclusion References Multi-view Analysis of Unregistered Medical Images Using Cross-View Transformers 1 Introduction 2 Related Work 3 Methods 3.1 Baseline Models 3.2 Cross-View Transformer Models 4 Data 5 Experiments 6 Results 7 Discussion and Conclusion References Machine Learning - Domain Adaptation Stain Mix-Up: Unsupervised Domain Generalization for Histopathology Images 1 Introduction 2 Method 2.1 Stain Separation via SNMF 2.2 Stain Mix-Up Augmentation 3 Experiments 3.1 Datasets 3.2 Implementation Details 3.3 Results on the CAMELYON17 Dataset 3.4 Results on the Hema Dataset 4 Conclusions References A Unified Hyper-GAN Model for Unpaired Multi-contrast MR Image Translation 1 Introduction 2 Method 2.1 Network Architecture 2.2 Training Loss 3 Experiments 3.1 Data Sets 3.2 Experimental Results 4 Conclusion References Generative Self-training for Cross-Domain Unsupervised Tagged-to-Cine MRI Synthesis 1 Introduction 2 Methodology 2.1 Generative Self-training UDA 2.2 Bayesian Uncertainty Mask for Target Samples 2.3 Training Protocol 3 Experiments and Results 3.1 Cross-Scanner Tagged-to-Cine MR Image Synthesis 3.2 Cross-Center Tagged-to-Cine MR Image Synthesis 4 Discussion and Conclusion References Cooperative Training and Latent Space Data Augmentation for Robust Medical Image Segmentation 1 Introduction 2 Methodology 2.1 Overview of the Framework 2.2 Standard Training 2.3 Latent Space Data Augmentation for Hard Example Generation 2.4 Cooperative Training 3 Experiments and Results 4 Conclusion References Controllable Cardiac Synthesis via Disentangled Anatomy Arithmetic 1 Introduction 2 Generative Model Architecture 3 Experiments 4 Conclusions References CoTr: Efficiently Bridging CNN and Transformer for 3D Medical Image Segmentation 1 Introduction 2 Materials 3 Methods 3.1 CNN-encoder 3.2 DeTrans-Encoder 3.3 Decoder 3.4 Implementation Details 4 Results 5 Discussion on Hyper-parameter Settings 6 Conclusion References Harmonization with Flow-Based Causal Inference 1 Introduction 2 Related Work 3 Method 3.1 Building Blocks 3.2 Harmonization Using Counterfactual Inference in a Flow-Based SCM 4 Experimental Results 4.1 Setup 4.2 Evaluation of the Learned Flow-Based SCM 4.3 Age Prediction 4.4 Classification of Alzheimer's Disease 5 Conclusion References Uncertainty-Aware Label Rectification for Domain Adaptive Mitochondria Segmentation 1 Introduction 2 Method 3 Experiments and Results 4 Analyses and Discussions 5 Conclusion References Semantic Consistent Unsupervised Domain Adaptation for Cross-Modality Medical Image Segmentation 1 Introduction 2 Method 2.1 Overview of Our Method 2.2 Learning Objective 3 Experiments and Results 3.1 Datasets and Evaluation Metrics 3.2 Experimental Results 4 Discussion and Conclusion References Anatomy of Domain Shift Impact on U-Net Layers in MRI Segmentation 1 Introduction 2 Related Work 3 Method 4 Experiments 4.1 Technical Details 4.2 Experimental Setup 4.3 Results and Discussion 5 Conclusion References FoldIt: Haustral Folds Detection and Segmentation in Colonoscopy Videos 1 Introduction 2 Related Works 3 Data 4 Methods 5 Results 6 Limitations and Future Work References Reference-Relation Guided Autoencoder with Deep CCA Restriction for Awake-to-Sleep Brain Functional Connectome Prediction 1 Introduction 2 Method 2.1 Model Description 2.2 Validation of Functional Connectome Prediction 3 Experiments 3.1 Data Description 3.2 Validation of R2AE-dCCA 4 Conclusion References Domain Composition and Attention for Unseen-Domain Generalizable Medical Image Segmentation 1 Introduction 2 Methods 2.1 Domain Composition and Attention Block 2.2 DCA-Net 3 Experiments and Results 4 Conclusion References Fully Test-Time Adaptation for Image Segmentation 1 Introduction 2 Fully Test-Time Adaptation for Image Segmentation 2.1 Objective Function 2.2 Optimization Parameters 3 Experiments 3.1 Datasets 3.2 Implementation Details 3.3 Results 4 Conclusion References OLVA: Optimal Latent Vector Alignment for Unsupervised Domain Adaptation in Medical Image Segmentation 1 Introduction 2 Method 2.1 VAEs for Segmentation 2.2 Optimal Transport for Latent Vector Alignment (OLVA) 2.3 Learning OLVA with a Stochastic Approximation 3 Experiments and Results 4 Conclusion References Prototypical Interaction Graph for Unsupervised Domain Adaptation in Surgical Instrument Segmentation 1 Introduction 2 Approach 2.1 Prototypical Inner-Interaction Graph (PI-Graph) 2.2 Prototypical Cross-Interaction Graph (PC-Graph) 2.3 Training Objective 3 Experiment 4 Conclusions References Unsupervised Domain Adaptation for Small Bowel Segmentation Using Disentangled Representation 1 Introduction 2 Method 2.1 Dataset 2.2 Unsupervised Disentangling of Intensity and Non-intensity Representations 2.3 Unsupervised Domain Adaptation Using Disentangled Representations 2.4 Evaluation Details 3 Results 3.1 Quantitative Evaluation 3.2 Qualitative Evaluation 4 Conclusion References Data-Driven Mapping Between Functional Connectomes Using Optimal Transport 1 Introduction 2 Methods 2.1 Optimal Transport 2.2 Proposed Algorithm for Mapping Atlases Using Optimal Transport 3 Results 3.1 Datasets 3.2 Intrinsic Evaluation 3.3 Extrinsic Evaluation 4 Discussion and Conclusions References EndoUDA: A Modality Independent Segmentation Approach for Endoscopy Imaging 1 Introduction 2 Method 2.1 VAE Training on Source Domain 2.2 Segmentation Module 2.3 VAE Latent Search Optimization via New Joint Loss 3 Experiments and Results 3.1 Benchmark Comparison and Ablation Study 3.2 Comparison with SOTA Method 3.3 Effect of Out-of-Distribution Data in Supervised Training 4 Conclusion References Style Transfer Using Generative Adversarial Networks for Multi-site MRI Harmonization 1 Introduction 2 Method 2.1 The Architecture of Style-Encoding GAN 2.2 Network Training 2.3 Experimental Setup 3 Results 3.1 Validation of the Removal of Cross-Sites Variances 3.2 Validation of the Preservation of Brain Anatomical Information 3.3 Hyperparameter Selection 3.4 Harmonization for MR Images from Unseen Dataset 4 Discussion and Conclusion References Machine Learning - Federated Learning Federated Semi-supervised Medical Image Classification via Inter-client Relation Matching 1 Introduction 2 Method 2.1 Backbone Federated Semi-supervised Learning Framework 2.2 Disease Relation Estimation at Labeled and Unlabeled Clients 2.3 Objective of Inter-client Relation Matching 3 Experiments 3.1 Dataset and Experimental Setup 3.2 Comparison with State-of-the-Arts 3.3 Analytical Studies of Our Method 4 Conclusion References FedPerl: Semi-supervised Peer Learning for Skin Lesion Classification 1 Introduction 2 Methodology 2.1 Semi-Supervised Federated Learning (SSFL) 2.2 Federated Peer Supervised Learning (FedPerl) 3 Experiments and Results 3.1 Peer Learning Results 3.2 Building Communities Results 3.3 The Influence of Peer Learning on Clients 3.4 Class Level Results 4 Discussion and Conclusion References Personalized Retrogress-Resilient Framework for Real-World Medical Federated Learning 1 Introduction 2 Personalized Retrogress-Resilient Framework 2.1 Overview 2.2 Progressive Fourier Aggregation in Server 2.3 Deputy-Enhanced Transfer in Client 3 Experiment 3.1 Dataset and Implementation Details 3.2 Comparison with State-of-the-art Methods 3.3 Ablation Study 3.4 Out-of-distribution Generalization 4 Conclusion References Federated Whole Prostate Segmentation in MRI with Personalized Neural Architectures 1 Introduction 2 Method 3 Experiments and Results 4 Discussion and Conclusions References Federated Contrastive Learning for Volumetric Medical Image Segmentation 1 Introduction 2 Background and Related Work 3 Method 3.1 Contrastive Learning with Feature Exchange 3.2 Global Structural Matching 4 Experiments 4.1 Results of Local Fine-Tuning 4.2 Results of Federated Fine-Tuning 5 Conclusion and Future Work References Federated Contrastive Learning for Decentralized Unlabeled Medical Images 1 Introduction 2 Method 2.1 Intra-Node Contrastive Learning 2.2 Metadata Transfer 2.3 Self-adaptive Aggregation 3 Experiments 3.1 Experimental Setup 3.2 Evaluation 4 Conclusion References Machine Learning - Interpretability/Explainability Explaining COVID-19 and Thoracic Pathology Model Predictions by Identifying Informative Input Features 1 Introduction 1.1 Contribution Statement 2 Methodology 2.1 Inverse IBA 2.2 Regression IBA 2.3 Multi-layer IBA 2.4 Chest X-Ray Models 3 Experiments and Results 3.1 Feature Importance (Human-Agnostic) Evaluations 3.2 Ground Truth Based (Human-centric) Evaluations 4 Discussion 5 Conclusion References Demystifying T1-MRI to FDG18-PET Image Translation via Representational Similarity 1 Introduction 2 Related Works 3 Image Translation Model, Dataset, and Analysis Tool 4 Demystification from a Medical Perspective 4.1 Brain Tissues Are Segmented in the Early Encoding Stage 4.2 Brain Tissue Information Is the Key to PET Image Synthesis 4.3 Brain Regions Are Recognized Later in the Translation 5 Explainable and Simplified Image Translation Model 5.1 Extraction of Regional Gray Matter Volume Information from Brain Tissue Maps 5.2 Regional Attention on Metabolic Variation in Aging and AD 6 Conclusion References Fairness in Cardiac MR Image Analysis: An Investigation of Bias Due to Data Imbalance in Deep Learning Based Segmentation 1 Introduction 2 Background 3 Methods 4 Materials and Experiments 5 Discussion References An Interpretable Approach to Automated Severity Scoring in Pelvic Trauma 1 Introduction 2 Related Work 3 Methods 4 Experiments and Results 4.1 Dataset and Metrics 4.2 Implementation Details 4.3 Experimental Results and Discussion 5 Conclusion References Scalable, Axiomatic Explanations of Deep Alzheimer's Diagnosis from Heterogeneous Data 1 Introduction 2 Methods 2.1 Wide and Deep PointNet 2.2 Shapley Value 2.3 Efficient Estimation of the Approximate Shapley Value 3 Experiments 3.1 Synthetic Shape Data Set 3.2 Alzheimer's Disease Diagnosis 4 Conclusion References SPARTA: An Integrated Stability, Discriminability, and Sparsity Based Radiomic Feature Selection Approach 1 Introduction 2 Previous Work and Novel Contributions 3 Methodology 4 Experimental Results and Discussion 4.1 Data and Implementation 4.2 Experiment 1: Assessing Parameter Sensitivity and Radiomic Feature Contributions 4.3 Experiment 2: Comparative Strategies 5 Concluding Remarks References The Power of Proxy Data and Proxy Networks for Hyper-parameter Optimization in Medical Image Segmentation 1 Introduction 2 Related Work 2.1 Contributions 3 Method 4 Experiments and Data 4.1 Datasets 4.2 Experimental Design 5 Results 6 Discussion and Conclusions References Fighting Class Imbalance with Contrastive Learning 1 Introduction 2 Related Work 3 Supervised Representation Disentanglement via Contrastive Learning 4 Experiments 4.1 Experimental Setup 4.2 Baselines 4.3 Implementation Details 4.4 Results and Discussion 4.5 Ablation Study 5 Conclusion References Interpretable Gender Classification from Retinal Fundus Images Using BagNets 1 Introduction 2 Related Work 3 Methods 3.1 Data and Preprocessing 3.2 Network Architecture and Training 3.3 Generation of Saliency Maps 3.4 Embedding of Image Patches 4 Results 5 Discussion References Explainable Classification of Weakly Annotated Wireless Capsule Endoscopy Images Based on a Fuzzy Bag-of-Colour Features Model and Brain Storm Optimization 1 Introduction 2 XFBoW Feature Extraction Model 2.1 Explainable Fuzzy Bag-Of-Words Model 2.2 Model Optimization 2.3 Explanation Extraction 3 Experiment and Results 3.1 Performance Evaluation 3.2 Statistical Analysis 3.3 Explainability Analysis 4 Discussion and Conclusion References Towards Semantic Interpretation of Thoracic Disease and COVID-19 Diagnosis Models 1 Introduction 2 Methodology 2.1 Setup: Chest X-ray Models: 2.2 Background: Network Dissection ch47bau2017network 2.3 Semantic Attribution 3 Results and Discussion 3.1 Semantics of Thoracic Classification Models 3.2 Semantics of COVID-19 Regression Models 3.3 Semantics of Models Trained on Weakly Labeled Datasets 3.4 Evolution of Semantics 3.5 Semantic Attribution 4 Conclusion References A Principled Approach to Failure Analysis and Model Repairment: Demonstration in Medical Imaging 1 Introduction 2 Methods 2.1 Phase I: Identification of Failure Types 2.2 Phase II: Repairment of Failure Types 3 Datasets and Implementation Details 4 Experiments and Results 4.1 Comparison of Failure Subtyping Methods 4.2 Analysis of Automatically Discovered Failure Subtypes 4.3 Model Repairment 5 Conclusion References Using Causal Analysis for Conceptual Deep Learning Explanation 1 Introduction 2 Method 2.1 Concept Associations 2.2 Causal Concept Ranking 2.3 Surrogate Explanation Function 3 Experiments 3.1 Evaluation of Concept Classifiers 3.2 Evaluating Causal Concepts Using Explanation Function 4 Conclusion References A Spherical Convolutional Neural Network for White Matter Structure Imaging via dMRI 1 Introduction 2 Theory 3 Methods 3.1 Denoising Layers 3.2 Convolutional Layers 3.3 Non-linearity and Pooling 3.4 Rotation Invariant Feature Vector 4 Experiments 5 Results and Conclusions References Sharpening Local Interpretable Model-Agnostic Explanations for Histopathology: Improved Understandability and Reliability 1 Introduction 2 Methods 2.1 Datasets 2.2 Network Architectures and Training 2.3 LIME and Sharp-LIME 2.4 Evaluation 3 Results 3.1 Improved Understandability 3.2 Improved Reliability 4 Discussion 5 Conclusions References Improving the Explainability of Skin Cancer Diagnosis Using CBIR 1 Introduction 2 Proposed Approach 2.1 Classification-CBIR Model Overview 2.2 Feature-Space Regularization 3 Experimental Results 3.1 Experimental Setup 3.2 Results 4 Conclusions References PAC Bayesian Performance Guarantees for Deep (Stochastic) Networks in Medical Imaging 1 Introduction 2 PAC-Bayesian Theory and Generalization 2.1 Formal Setup 2.2 The PAC-Bayesian Theorem 3 Experiments 3.1 Setup 3.2 Results 4 Conclusion References Machine Learning – Uncertainty Medical Matting: A New Perspective on Medical Segmentation with Uncertainty 1 Introduction 2 Datasets 3 Methodology 3.1 Mask Generator 3.2 Matting Network 3.3 Multi-task Loss 4 Experiments 5 Conclusions References Confidence-Aware Cascaded Network for Fetal Brain Segmentation on MR Images 1 Introduction 2 Methods 2.1 Coarse Segmentation with Slice Confidence 2.2 Confidence-Aware Fine Fetal Brain Segmentation 3 Experiments 3.1 Dataset and Evaluation Metrics 3.2 Implementation Details 3.3 Comparison with State-of-the-art Methods 3.4 Ablation Study 4 Conclusion References Orthogonal Ensemble Networks for Biomedical Image Segmentation 1 Introduction 2 Related Works 3 Orthogonal Ensemble Networks for Image Segmentation 4 Experimental Framework 5 Results and Discussion 6 Conclusions References Learning to Predict Error for MRI Reconstruction 1 Introduction 2 Notation 3 An Anatomy of Prediction Error 4 What Does the Predictive Uncertainty Quantify? 5 A Two-Step Estimation Method 6 Experiments 6.1 Datasets 6.2 Single Image Super-Resolution 6.3 MRI Reconstruction 7 Related Work 8 Conclusion References Uncertainty-Guided Progressive GANs for Medical Image Translation 1 Introduction 2 Related Works 3 Uncertainty-Guided Progressive GAN (UP-GAN) 4 Experiments 4.1 Experimental Setup 4.2 Results and Analysis 5 Conclusion References Variational Topic Inference for Chest X-Ray Report Generation 1 Introduction 2 Methodology 2.1 Problem Formulation 2.2 Variational Topic Inference 2.3 Implementation Using Neural Networks 3 Experiments 3.1 Datasets and Implementation Details 3.2 Results and Discussion 4 Conclusion References Uncertainty Aware Deep Reinforcement Learning for Anatomical Landmark Detection in Medical Images 1 Introduction and Related Work 2 Methods 2.1 Landmark Annotation 2.2 Agent Architecture 2.3 Inference and Uncertainty Estimation 3 Experiment and Results 3.1 Experiment 1: Landmark Detection Accuracy 3.2 Experiment 2: Uncertainty Estimation 3.3 Experiment 3: Landmark Prediction Error Detection 3.4 Failure Modes 4 Discussion References Author Index The eight-volume set LNCS 12901, 12902, 12903, 12904, 12905, 12906, 12907, and 12908 constitutes the refereed proceedings of the 24th International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2021, held in Strasbourg, France, in September/October 2021.* The 542 revised full papers presented were carefully reviewed and selected from 1809 submissions in a double-blind review process. The papers are organized in the following topical sections: Part I: image segmentation Part II: machine learning - self-supervised learning; machine learning - semi-supervised learning; and machine learning - weakly supervised learning Part III: machine learning - advances in machine learning theory; machine learning - domain adaptation; machine learning - federated learning; machine learning - interpretability / explainability; and machine learning - uncertainty Part IV: image registration; image-guided interventions and surgery; surgical data science; surgical planning and simulation; surgical skill and work flow analysis; and surgical visualization and mixed, augmented and virtual reality Part V: computer aided diagnosis; integration of imaging with non-imaging biomarkers; and outcome/disease prediction Part VI: image reconstruction; clinical applications - cardiac; and clinical applications - vascular Part VII: clinical applications - abdomen; clinical applications - breast; clinical applications - dermatology; clinical applications - fetal imaging; clinical applications - lung; clinical applications - neuroimaging - brain development; clinical applications - neuroimaging - DWI and tractography; clinical applications - neuroimaging - functional brain networks; clinical applications - neuroimaging - others; and clinical applications - oncology Part VIII: clinical applications - ophthalmology; computational (integrative) pathology; modalities - microscopy; modalities - histopathology; and modalities - ultrasound *The conference was held virtually