Medical Image Learning with Limited and Noisy Data: First International Workshop, MILLanD 2022, Held in Conjunction with MICCAI 2022, Singapore, ... (Lecture Notes in Computer Science, 13559)

This book constitutes the proceedings of the First Workshop on Medical Image Learning with Limited and Noisy Data, MILLanD 2022, held in conjunction with MICCAI 2022. The conference was held in Singapore. For this workshop, 22 papers from 54 submissions were accepted for publication. They selected papers focus on the challenges and limitations of current deep learning methods applied to limited and noisy medical data and present new methods for training models using such imperfect data. Preface Organization Contents Efficient and Robust Annotation Strategies Heatmap Regression for Lesion Detection Using Pointwise Annotations 1 Introduction 2 Related Work 3 Method 3.1 Training via Heatmap Regression 3.2 Detection During Inference 3.3 Segmentation Transfer Learning 4 Experiments and Results 4.1 Experimental Setup 4.2 Lesion Detection Results 4.3 Lesion Segmentation via Transfer Learning 5 Discussion and Conclusion References Partial Annotations for the Segmentation of Large Structures with Low Annotation Cost 1 Introduction 2 Method 2.1 Selective Dice Loss 2.2 Optimization 3 Experimental Results 4 Conclusion References Abstraction in Pixel-wise Noisy Annotations Can Guide Attention to Improve Prostate Cancer Grade Assessment 1 Introduction 2 Materials and Method 2.1 Data 2.2 Architecture 2.3 Multiple Instance Learning for Cancer Grade Assessment 2.4 Noisy Labels and Weak Supervision 3 Experiments 3.1 Implementation and Evaluation 3.2 Results 4 Conclusion References Meta Pixel Loss Correction for Medical Image Segmentation with Noisy Labels 1 Introduction 2 Methodology 2.1 Meta Pixel Loss Correction 2.2 Optimization Algorithm 3 Experiment Results 3.1 Dataset 3.2 Experiment Setting 3.3 Experimental Results 3.4 Limitation 4 Conclusion References Re-thinking and Re-labeling LIDC-IDRI for Robust Pulmonary Cancer Prediction 1 Introduction 2 Materials 3 Study Design 4 Methods 4.1 Label Induction Using Machine Annotator 4.2 Similar Nodule Retrieval Using Metric Learning 5 Experiments and Results 5.1 Implementation 5.2 Quantitative Evaluation 5.3 Discussion 6 Conclusion and Future Work References Weakly-Supervised, Self-supervised, and Contrastive Learning Universal Lesion Detection and Classification Using Limited Data and Weakly-Supervised Self-training 1 Introduction 2 Methods 3 Experiments and Results 4 Discussion and Conclusion References BoxShrink: From Bounding Boxes to Segmentation Masks 1 Introduction 2 Related Work 3 Boxshrink Framework 3.1 Main Components 3.2 rapid-BoxShrink 3.3 robust-BoxShrink 4 Experiments 4.1 Qualitative and Quantitative Experiments 4.2 Reproducibility Details 5 Discussion 6 Conclusion References Multi-Feature Vision Transformer via Self-Supervised Representation Learning for Improvement of COVID-19 Diagnosis 1 Introduction 2 Methods 3 Experiments and Results 4 Conclusion References SB-SSL: Slice-Based Self-supervised Transformers for Knee Abnormality Classification from MRI 1 Introduction 2 Related Works 3 Methodology 3.1 Vision Transformer 3.2 Self-supervised Pretraining 4 Experimental Results 4.1 Implementation Details 4.2 Results 4.3 Ablation Studies 5 Conclusion References Optimizing Transformations for Contrastive Learning in a Differentiable Framework 1 Introduction 2 Transformation Network 2.1 Optimizing Transformations 2.2 Differentiable Formulation of the Transformations 2.3 Experimental Settings 2.4 Linear Evaluation 3 Results and Discussion 4 Conclusions and Perspectives References Stain Based Contrastive Co-training for Histopathological Image Analysis 1 Introduction 2 Stain Based Contrastive Co-training 2.1 Stain Separation 2.2 Contrastive Co-training 3 Experiments 3.1 Datasets 3.2 Model Selection, Training and Hyperparameters 3.3 Results 3.4 Co-training View Analysis 3.5 Ablation Studies 4 Conclusion References Active and Continual Learning .26em plus .1em minus .1emCLINICAL: Targeted Active Learning for Imbalanced Medical Image Classification 1 Introduction 1.1 Related Work 1.2 Our Contributions 2 Preliminaries 2.1 Examples of Smi Functions 3 Clinical: Our Targeted Active Learning Framework for Binary and Long-Tail Imbalance 4 Experiments 4.1 Binary Imbalance 4.2 Long-Tail Imbalance 5 Conclusion References Real Time Data Augmentation Using Fractional Linear Transformations in Continual Learning 1 Introduction 2 Methodology 3 Experiments, Results and Discussion 4 Conclusion References DIAGNOSE: Avoiding Out-of-Distribution Data Using Submodular Information Measures 1 Introduction 1.1 Problem Statement: OOD Scenarios in Medical Data 1.2 Related Work 1.3 Our Contributions 2 Preliminaries 3 Leveraging Submodular Information Measures for Multiple Out-of-Distribution Scenarios 4 Experimental Results 4.1 Scenario A - Unrelated Images 4.2 Scenario B - Incorrectly Acquired Images 4.3 Scenario C - Mixed View Images 5 Conclusion References Transfer Representation Learning Auto-segmentation of Hip Joints Using MultiPlanar UNet with Transfer Learning 1 Introduction 2 Method 3 Data and Experiments 4 Results 4.1 Numerical Validation and Ablation Study 5 Conclusion References Asymmetry and Architectural Distortion Detection with Limited Mammography Data 1 Introduction 2 Related Work 3 Method 4 Experiment Design 5 Experimental Results 5.1 Comparison with Other Methods 5.2 Ablation Study 6 Conclusions References Imbalanced Data and Out-of-Distribution Generalization Class Imbalance Correction for Improved Universal Lesion Detection and Tagging in CT 1 Introduction 2 Methods 3 Experiments 4 Results and Discussion 5 Conclusion References CVAD: An Anomaly Detector for Medical Images Based on Cascade VAE 1 Introduction 2 Method 2.1 CVAD Architecture 2.2 Combined Loss Function 2.3 Network Details 3 Experiments 3.1 Datasets and Implementation Details 3.2 Results 4 Conclusion References Approaches for Noisy, Missing, and Low Quality Data Visual Field Prediction with Missing and Noisy Data Based on Distance-Based Loss 1 Introduction 2 Method 2.1 Distance-Based Loss 3 Experiments 3.1 Dataset and Implementation 3.2 Results 4 Conclusion References Image Quality Classification for Automated Visual Evaluation of Cervical Precancer 1 Introduction 2 Image Quality Labeling Criteria and Data 2.1 The Labeling Criteria 2.2 Datasets 3 Methods 3.1 Cervix Detection 3.2 Quality Classification 3.3 Mislabel Identification 4 Experimental Results and Discussion 5 Conclusions Appendix References A Monotonicity Constrained Attention Module for Emotion Classification with Limited EEG Data 1 Introduction 2 Related Work 3 Proposed Method 4 Experiments 4.1 Models' Scalp Attention Pattern 4.2 Models' Sensitivity of Prediction on Inputs' Frequency 4.3 Model Sensitivity on Morphisms Between Samples 5 Conclusion References Automated Skin Biopsy Analysis with Limited Data 1 Introduction 2 Methods 2.1 Dataset 2.2 Nerve Labeling 2.3 Dermis-Epidermis Boundary Detection 2.4 Nerve Crossing Identification 3 Experimental Setup 3.1 Evaluating the Nerve Tracing Model 3.2 Evaluating Dermis Model 4 Results 4.1 Nerve Labeling Results 4.2 Dermis Labeling Results 4.3 Crossing Count Results 5 Discussion References Author Index