Resource-Efficient Medical Image Analysis: First MICCAI Workshop, REMIA 2022, Singapore, September 22, 2022, Proceedings

Description

This book constitutes the refereed proceedings of the first MICCAI Workshop on Resource-Efficient Medical Image Analysis, REMIA 2022, held in conjunction with MICCAI 2022, in September 2022 as a hybrid event. REMIA 2022 accepted 13 papers from the 19 submissions received. The workshop aims at creating a discussion on the issues for practical applications of medical imaging systems with data, label and hardware limitations. Preface Organization Contents Multi-task Semi-supervised Learning for Vascular Network Segmentation and Renal Cell Carcinoma Classification 1 Introduction 2 Related Works 3 Dataset and Methods 3.1 Dataset Building 3.2 Multi-task Learning Pipeline 3.3 Evaluation 4 Experiments and Results 4.1 Backbone and MTL-SSL Method Choice 4.2 Segmentation Benchmarks of Vascular Network 4.3 Test on New Subtype of RCC and Other Cancers Dataset 5 Conclusion References Self-supervised Antigen Detection Artificial Intelligence (SANDI) 1 Introduction 2 Method 2.1 Datasets 2.2 Single-cell Patches Sampling 2.3 Patch Cropping and Pairing 2.4 Network Architecture and Training 2.5 Reference-based Cell Classification 2.6 Automatic Expansion of the Reference Set 3 Experimental Results 3.1 Classification Performance with Different Size of Randomly Selected Reference Set 3.2 Classification Performance with the Automatic Expanding Reference Set 4 Conclusion References RadTex: Learning Efficient Radiograph Representations from Text Reports 1 Introduction 2 Method 2.1 Network Architecture 2.2 Adapting VirTex to the Radiology Domain 3 Experimental Results 3.1 Datasets 3.2 Training Details 3.3 Training with Fewer Labeled Images in Downstream Tasks 3.4 Pretrained Representation Quality 3.5 Proxy Task: Generating Radiology Reports 4 Conclusions References Single Domain Generalization via Spontaneous Amplitude Spectrum Diversification 1 Introduction 2 Method 2.1 Spectrum Diversification Module 2.2 Adversarial Sample Generation 3 Experiments 3.1 Performance Evaluation 4 Conclusion References Triple-View Feature Learning for Medical Image Segmentation 1 Introduction 2 Methodology 2.1 Training Setup 2.2 Label Processing 2.3 Loss Function 3 Experiments and Results 3.1 Datasets and Experimental Setup 3.2 Evaluation and Results 4 Conclusion References Classification of 4D fMRI Images Using ML, Focusing on Computational and Memory Utilization Efficiency 1 Introduction 2 Related Work 3 Dataset 4 Region of Interest (ROI) 5 CNN Models 6 Transformer Model 7 Experiments 8 Training 8.1 Training Performance 9 Classification Accuracy 10 Conclusion References An Efficient Defending Mechanism Against Image Attacking on Medical Image Segmentation Models 1 Introduction 2 Our Methods 2.1 Improve Robustness Against Adversarial Example for Segmentation Models 2.2 Attacking Segmentation Model 3 Experiment and Results 3.1 Evaluation 4 Conclusions References Leverage Supervised and Self-supervised Pretrain Models for Pathological Survival Analysis via a Simple and Low-cost Joint Representation Tuning 1 Introduction 2 Methods 2.1 Supervised and Self-supervised Pretraining 2.2 Joint Representation Tuning 2.3 Evaluate Different Strategies of Using Pretrained Model 3 Experiments 3.1 Data Description 3.2 Experimental Setting 4 Results 4.1 Performance on Classification and Survival Prediction 4.2 Computational Resource 5 Ablation Studies 5.1 Different Strategies of Using Pretrained Models 5.2 Effect of Transformer 5.3 Effect of Data Augmentation 6 Conclusion References Pathological Image Contrastive Self-supervised Learning 1 Introduction 2 Revisiting Contrastive Learning 3 Histopathological Contrastive Learning 3.1 Properties of Histopathological Images 3.2 Stain Perturbation 3.3 Pipeline of Transforms 4 Experiments 4.1 Dataset 4.2 Implementation Details 4.3 Results 5 Conclusion References Investigation of Training Multiple Instance Learning Networks with Instance Sampling 1 Introduction 2 MIL and Attention-Based MIL Networks 2.1 MIL Problem Formulation 2.2 Attention-Based MIL 3 Sampling Strategies for Attention-Based MIL 3.1 Random Sampling 3.2 Adaptive Sampling 3.3 Top-k Sampling 4 Dataset, Network Architecures, and Tuning Hyper-parameters 4.1 Datasets 4.2 Network Architecture 4.3 Tuning Hyper-parameters 5 Results 6 Discussion References Masked Video Modeling with Correlation-Aware Contrastive Learning for Breast Cancer Diagnosis in Ultrasound 1 Introduction 2 Methodology 2.1 Masked Video Modeling 2.2 Correlation-Aware Contrastive Learning 3 Experiments and Results 4 Conclusions References A Self-attentive Meta-learning Approach for Image-Based Few-Shot Disease Detection 1 Introduction 2 Related Work 3 Proposed Approach 4 Experiments and Results 4.1 Experimental Settings 4.2 Few-Shot Diseases Detection Results 5 Conclusion References Facing Annotation Redundancy: OCT Layer Segmentation with only 10 Annotated Pixels per Layer 1 Introduction 2 Investigation on Less Annotated Data 2.1 Experimental Setting 2.2 Results and Observations 3 Annotation-Efficient Learning 4 Experiment 4.1 Experiment Setting 4.2 Performance Comparison 4.3 Ablation Study 5 Discussion and Conclusion References Author Index