Left Atrial and Scar Quantification and Segmentation : First Challenge, LAScarQS 2022, Held in Conjunction with MICCAI 2022, Singapore, September 18, 2022, Proceedings

This book constitutes the First Left Atrial and Scar Quantification and Segmentation Challenge, LAScarQS 2022, which was held in conjunction with the 25th International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2022, in Singapore, in September 2022. The 15 papers presented in this volume were carefully reviewed and selected form numerous submissions. The aim of the challenge is not only benchmarking various LA scar segmentation algorithms, but also covering the topic of general cardiac image segmentation, quantification, joint optimization, and model generalization, and raising discussions for further technical development and clinical deployment. Preface Organization Contents LASSNet: A Four Steps Deep Neural Network for Left Atrial Segmentation and Scar Quantification 1 Introduction 2 Methods 2.1 Training and Validation Data 2.2 Data Inspection and Pre-processing 2.3 The Neural Network Architecture 2.4 Implementation 3 Experimental Results 3.1 Segmentation Performances and Discussion 4 Conclusion References Multi-depth Boundary-Aware Left Atrial Scar Segmentation Network 1 Introduction 2 Method 2.1 Network Architecture 2.2 Sobel Fusion Module 2.3 Loss Function 3 Experiment 3.1 Dataset 3.2 Implementations 3.3 Result 4 Conclusion References Self Pre-training with Single-Scale Adapter for Left Atrial Segmentation 1 Introduction 2 Dataset 3 Method 3.1 Self Pre-training with Masked Autoencoders 3.2 Adapter for Downstream Segmentation Task 4 Experiments and Results 4.1 Implementation Details 4.2 Quantitative Results on Validation Set 4.3 Qualitative Results from MAE Reconstruction on Validation Set 4.4 Challenge Results 5 Conclusion References UGformer for Robust Left Atrium and Scar Segmentation Across Scanners 1 Introduction 2 Methodology 2.1 Encoder Block 2.2 Bridge 3 Implementation 3.1 Dataset and Pre-processing 3.2 Training Details 4 Experiment 4.1 Comparison to the State-of-the-art Methods (SOTA) 4.2 Ablation Studies 5 Conclusions References Automatically Segment the Left Atrium and Scars from LGE-MRIs Using a Boundary-Focused nnU-Net 1 Introduction 2 Methods 3 Experiments 4 Results 5 Conclusion References Two Stage of Histogram Matching Augmentation for Domain Generalization: Application to Left Atrial Segmentation 1 Introduction 2 Proposed Method 2.1 ROI-Level Histogram Matching 2.2 nnU-Net 3 Experiment and Results 3.1 Dataset and Training Protocols 3.2 Result 4 Conclusion References Sequential Segmentation of the Left Atrium and Atrial Scars Using a Multi-scale Weight Sharing Network and Boundary-Based Processing 1 Introduction 2 Proposed Approach 2.1 Multi-Scale Weight Sharing Network (MSWS-Net) 2.2 Boundary Processing with the Boundary2Patches Method 3 Dataset Description 4 Experimental Details 4.1 Loss Function 5 Results and Discussion 5.1 LA Cavity Segmentation: Task 2 5.2 LA Cavity and Scars Segmentation: Task 1 6 Conclusion References LA-HRNet: High-Resolution Network for Automatic Left Atrial Segmentation in Multi-center LEG MRI 1 Introduction 2 Methods 2.1 Data Pre-processing 2.2 Network Architectures 3 Results and Discussion 4 Conclusion References Edge-Enhanced Feature Guided Joint Segmentation of Left Atrial and Scars in LGE MRI Images 1 Introduction 2 Methods 2.1 Coarse Segmentation of ROIs 2.2 Fine and Joint Segmentation of LA and Scars 2.3 Loss Function 3 Experiments 3.1 Dataset and Data Preprocessing 3.2 Implementation Details 4 Results and Discussions 4.1 Ablation Experiments 4.2 Comparison Experiments 5 Conclusion References TESSLA: Two-Stage Ensemble Scar Segmentation for the Left Atrium 1 Introduction 2 Methods 2.1 Dataset 2.2 Proposed Model and Implementation 2.3 LA Wall Estimation and Normalisation 3 Results 3.1 Model Implementation and Training 3.2 Validation and Test Set Results 4 Discussion and Conclusion References Deep U-Net Architecture with Curriculum Learning for Left Atrial Segmentation 1 Introduction 2 Methodologies 2.1 Dataset 2.2 Data Pre-processing and Augmentation 2.3 Model 2.4 Loss Function and Post-processing 3 Results 3.1 Implementation and Evaluation Metrics 3.2 Results on 4-Folds Cross Validation Sets 3.3 Results on LAScarQS 2022 Challenge Validation Set 4 Conclusion References Cross-Domain Segmentation of Left Atrium Based on Multi-scale Decision Level Fusion 1 Introduction 2 Methodology 2.1 Data Preprocessing 2.2 Proposed Method 3 Experimental Results and Analysis 4 Conclusion References Using Polynomial Loss and Uncertainty Information for Robust Left Atrial and Scar Quantification and Segmentation 1 Introduction 2 Dataset 3 Methods 3.1 Network Architecture 3.2 Loss Functions 3.3 Data Augmentations 3.4 Training 4 Results and Discussion 5 Conclusion References Automated Segmentation of the Left Atrium and Scar Using Deep Convolutional Neural Networks 1 Introduction 2 Methodology 2.1 Data 2.2 Neural Network Training 3 Results 4 Conclusion References Automatic Semi-supervised Left Atrial Segmentation Using Deep-Supervision 3DResUnet with Pseudo Labeling Approach for LAScarQS 2022 Challenge 1 Introduction 2 Material and Methods 2.1 LAScarQS 2022 Dataset Descriptions 2.2 Proposed Method 3 Results and Discussion 3.1 Quantitative Results 3.2 Qualitative Results 4 Conclusion and Future Work References Author Index