Segmentation, classification, and registration of multi-modality medical imaging data : MICCAI 2020 challenges, ABCs 2020, L2R 2020, TN-SCUI 2020, held in conjunction with MICCAI 2020, Lima, Peru, October 4-8, 2020 : proceedings

This book constitutes three challenges that were held in conjunction with the 23rd International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2020, in Lima, Peru, in October 2020*: the Anatomical Brain Barriers to Cancer Spread: Segmentation from CT and MR Images Challenge, the Learn2Reg Challenge, and the Thyroid Nodule Segmentation and Classification in Ultrasound Images Challenge. The 19 papers presented in this volume were carefully reviewed and selected form numerous submissions. The ABCs challenge aims to identify the best methods of segmenting brain structures that serve as barriers to the spread of brain cancers and structures to be spared from irradiation, for use in computer assisted target definition for glioma and radiotherapy plan optimization. The papers of the L2R challenge cover a wide spectrum of conventional and learning-based registration methods and often describe novel contributions. The main goal of the TN-SCUI challenge is tofind automatic algorithms to accurately segment and classify the thyroid nodules in ultrasound images. *The challenges took place virtually due to the COVID-19 pandemic. ABCs 2020 Preface Organization L2R 2020 Preface Organization TN-SCUI 2020 Preface Organization Contents ABCs – Anatomical Brain Barriers to Cancer Spread: Segmentation from CT and MR Images Cross-Modality Brain Structures Image Segmentation for the Radiotherapy Target Definition and Plan Optimization 1 Introduction 2 Challenge Setup 2.1 Imaging 2.2 Structure Labeling 2.3 Data Pre-processing 2.4 Segmentation Accuracy Evaluation and Scoring 2.5 Organization 3 Results 3.1 Inter-Rater Variability Study 3.2 Performance of the Algorithms 4 Summary and Conclusion References Domain Knowledge Driven Multi-modal Segmentation of Anatomical Brain Barriers to Cancer Spread 1 Introduction 2 Methods 2.1 Data Description 2.2 Network Architecture 2.3 Label Merging for Symmetric Structures 2.4 Multi-modality Ensemble 2.5 Training Protocol 3 Results 4 Discussion 5 Conclusion References Ensembled ResUnet for Anatomical Brain Barriers Segmentation 1 Introduction 2 Method 2.1 Encoder Design 2.2 Decoder Design 2.3 Loss Function 2.4 Pseudo Training with Model Ensemble 2.5 Optimization 3 Experiments 3.1 Implementation Details and Data Processing 3.2 Quantitative and Qualitative Analysis 4 Conclusions References An Enhanced Coarse-to-Fine Framework for the Segmentation of Clinical Target Volume 1 Introduction 2 Method 2.1 F-Loss to Keep the High Recall Rate of Coarse Segmentation 2.2 Iterative Refinement to Iteratively Refine the Results 2.3 Ensemble Refinement to Fuse Multiple Information to Get Finer Results 3 Experiments 3.1 DataSet 3.2 Training Details 4 Results References Automatic Segmentation of Brain Structures for Treatment Planning Optimization and Target Volume Definition 1 Introduction 2 Materials and Methods 2.1 Pre-processing 2.2 Training 2.3 Prediction 2.4 Post-processing 3 Results 4 Discussion and Conclusion References A Bi-directional, Multi-modality Framework for Segmentation of Brain Structures 1 Introduction 2 Methods 2.1 Data 2.2 Training Task 1 2.3 Training Task 2 2.4 Evaluation Metrics 3 Results 3.1 Challenge Results 4 Post-challenge Processing 5 Discussion 6 Conclusion References L2R – Learn2Reg: Multitask and Multimodal 3D Medical Image Registration Large Deformation Image Registration with Anatomy-Aware Laplacian Pyramid Networks 1 Introduction 2 Methods 2.1 Laplacian Pyramid Image Registration Networks 2.2 Anatomical Label Supervision 2.3 Data Preprocessing and Augmentation 3 Results 4 Conclusion References Discrete Unsupervised 3D Registration Methods for the Learn2Reg Challenge 1 Motivation and Background 2 Methods 2.1 Dense Displacement Sampling (Deeds) with Discrete Spanning Tree Optimisation 2.2 Probabilistic Dense Displacement (PDD) Net 3 Experiments and Results 4 Conclusion References Variable Fraunhofer MEVIS RegLib Comprehensively Applied to Learn2Reg Challenge 1 Introduction 2 Method and Results 3 Conclusion References Learning a Deformable Registration Pyramid 1 Introduction 2 Method 2.1 Architecture 2.2 Loss Function 3 Experiment 4 Conclusion and Future Work References Deep Learning Based Registration Using Spatial Gradients and Noisy Segmentation Labels 1 Introduction 2 Methodology 2.1 Training Strategy and Implementation Details 3 Experimental Results 4 Conclusions References Multi-step, Learning-Based, Semi-supervised Image Registration Algorithm 1 Introduction 2 Methods 2.1 Method 2.2 Dataset and Experimental Setup 3 Results 4 Conclusion References Using Elastix to Register Inhale/Exhale Intrasubject Thorax CT: A Unsupervised Baseline to the Task 2 of the Learn2Reg Challenge 1 Introduction 2 Material and Method 2.1 Dataset 2.2 Method 2.3 Evaluation Metrics 3 Results 4 Discussion and Conclusion References TN-SCUI – Thyroid Nodule Segmentation and Classification in Ultrasound Images Automatic Segmentation and Classification of Thyroid Nodules in Ultrasound Images with Convolutional Neural Networks 1 Introduction 2 Method 2.1 Data Preprocessing 2.2 Cascaded Segmentation Framework 2.3 Dual-Attention ResNet Framework 2.4 Data Augmentation and Test Time Augmentation 3 Experiments 4 Results 4.1 Segmentation Results 4.2 Classification Results 5 Conclusions References LRTHR-Net: A Low-Resolution-to-High-Resolution Framework to Iteratively Refine the Segmentation of Thyroid Nodule in Ultrasound Images 1 Introduction 2 Methods 2.1 Refining High-Resolution Segmentations Based on Multi-scale Results 2.2 Refining Results Based on Multiple Models 3 Experiments 3.1 Dataset 3.2 Evaluation Metrics 3.3 Implementation Details 4 Results 5 Conclusion References Coarse to Fine Ensemble Network for Thyroid Nodule Segmentation 1 Introduction 2 Method 2.1 Architecture of Coarse to Fine Network 2.2 Loss Function 2.3 Data Preprocess and Augmentation 2.4 Inference Test Time Augmentation 2.5 Ensemble 3 Experiments and Results 3.1 Dataset 3.2 Result Evaluation 4 Conclusion References Cascade UNet and CH-UNet for Thyroid Nodule Segmentation and Benign and Malignant Classification 1 Introduction 2 Method 2.1 Cascade UNet 2.2 Classification with Auxiliary Task 2.3 Multi-scale Test and Test Time Augmentation 3 Experiment 3.1 Implementation Details 3.2 Ablative Evaluation on Segmentation 3.3 Ablative Evaluation on Classification 3.4 TNSCUI 2020 Challenge Results 4 Summary References Identifying Thyroid Nodules in Ultrasound Images Through Segmentation-Guided Discriminative Localization 1 Introduction 2 Method 2.1 Cascaded U-Net for Segmentation Ensemble 2.2 Online Attention Module for Classification 2.3 Data Augmentation 3 Experiments and Results 3.1 Ablation Study on Segmentation Ensemble 3.2 Performance of Segmentation-Guided Classification 3.3 Ablation Study on Classifier 4 Conclusion References Cascaded Networks for Thyroid Nodule Diagnosis from Ultrasound Images 1 Introduction 2 Method 3 Experiment 3.1 Data Set and Augmentation Techniques 3.2 Ablation Study on Ensembled Segmentation 3.3 Ablation Study on Classifier 3.4 Comparison with Detection-Based Classification 4 Conclusion References Author Index