Domain Adaptation and Representation Transfer, and Affordable Healthcare and AI for Resource Diverse Global Health. Third MICCAI Workshop, DART 2021 and First MICCAI Workshop, FAIR 2021 Held in Conjunction with MICCAI 2021 Strasbourg, France, September 27 and October 1, 2021 Proceedings

Preface DART 2021
Preface FAIR 2021
Organization
Contents
Domain Adaptation and Representation Transfer
A Systematic Benchmarking Analysis of Transfer Learning for Medical Image Analysis
	1 Introduction
	2 Transfer Learning Setup
	3 Transfer Learning Benchmarking and Analysis
	4 Conclusion and Future Work
	References
Self-supervised Multi-scale Consistency for Weakly Supervised Segmentation Learning
	1 Introduction
	2 Related Work
	3 Proposed Approach
	4 Experiments
		4.1 Data
		4.2 Evaluation Protocol
		4.3 Results
	5 Conclusion
	References
FDA: Feature Decomposition and Aggregation for Robust Airway Segmentation
	1 Introduction
	2 Method
		2.1 Learning Transferable Features
		2.2 Refinement of Transferable Features
		2.3 Training Loss Functions
	3 Experiments and Results
	4 Conclusion
	References
Adversarial Continual Learning for Multi-domain Hippocampal Segmentation
	1 Introduction
	2 Related Work
	3 Methods
	4 Datasets and Experiments
	5 Result and Discussion
	6 Conclusion
	References
Self-supervised Multimodal Generalized Zero Shot Learning for Gleason Grading
	1 Introduction
	2 Method
		2.1 Feature Extraction and Transformation
		2.2 CVAE Based Feature Generator Using Self Supervision
	3 Experimental Results
		3.1 Generalized Zero Shot Learning Results
	4 Conclusion
	References
Self-supervised Learning of Inter-label Geometric Relationships for Gleason Grade Segmentation
	1 Introduction
	2 Method
		2.1 Geometry Aware Shape Generation
		2.2 Sample Diversity from Uncertainty Sampling
	3 Experimental Results
		3.1 Dataset Description
		3.2 Experimental Setup, Baselines and Metrics
		3.3 Segmentation Results on Gleason Training Data
	4 Conclusion
	References
Stop Throwing Away Discriminators! Re-using Adversaries for Test-Time Training
	1 Introduction
	2 Related Work
	3 Method
		3.1 Re-usable Discriminators: Challenges and Proposed Solutions
		3.2 Architectures and Training Objectives for s and d
		3.3 Adversarial Test-Time Training: Adapting w
	4 Experiments
		4.1 Results and Discussion
	5 Conclusion
	References
Transductive Image Segmentation: Self-training and Effect of Uncertainty Estimation
	1 Introduction
	2 Methodology
		2.1 Transductive Learning via Self-training
		2.2 Analysing Information Gain and Improving Self-training
	3 Experimental Evaluation
		3.1 Data and Model Configuration
		3.2 Comparing Supervised and Transductive Learning
		3.3 Comparing Inductive and Transductive Semi-supervised Learning
		3.4 Blinded Comparison via Manual Refinement of Segmentations
	4 Conclusion
	References
Unsupervised Domain Adaptation with Semantic Consistency Across Heterogeneous Modalities for MRI Prostate Lesion Segmentation
	1 Introduction
	2 Method
		2.1 Problem Formulation
		2.2 Implementation Details
	3 Datasets
	4 Results
	5 Conclusion
	References
Cohort Bias Adaptation in Aggregated Datasets for Lesion Segmentation
	1 Introduction
	2 Related Work
	3 Methods
	4 Implementation Details
		4.1 Network Architecture and Training Parameters
		4.2 Data Set
		4.3 Evaluation Metrics
	5 Experiments and Results
		5.1 Trial Conditioning
		5.2 Fine-Tuning to New Cohort Bias
		5.3 Accounting for Complex Cohort Biases - Missing Small Lesions
	6 Conclusions
	References
Exploring Deep Registration Latent Spaces
	1 Introduction
	2 Related Work
	3 Methodology
		3.1 Deep Learning-Based Registration Scheme
		3.2 Decomposition of Latent Space
		3.3 Implementation and Training Details
	4 Experiments and Results
		4.1 Qualitative Evaluation
	5 Discussion and Conclusion
	References
Learning from Partially Overlapping Labels: Image Segmentation Under Annotation Shift
	1 Introduction
	2 Learning from Heterogeneously Labeled Data
		2.1 Problem Definition and Label-Contradiction Issue
		2.2 Adaptive Cross Entropy for Learning from Data with Heterogeneous Annotations
		2.3 Learning from Non-annotated Regions via Mean Teacher
	3 Experiments
		3.1 Data and Model Configuration
		3.2 Results
	4 Conclusion
	References
Unsupervised Domain Adaption via Similarity-Based Prototypes for Cross-Modality Segmentation
	1 Introduction
	2 Methodology
		2.1 Motivation
		2.2 Proposed Framework
		2.3 Feature Prototypes and Class-Wise Similarity Loss
		2.4 Contrastive Loss via Feature Dictionaries
	3 Experiments
		3.1 Datasets and Details
		3.2 Results and Analysis
	4 Conclusion
	References
Affordable AI and Healthcare
Classification and Generation of Microscopy Images with Plasmodium Falciparum via Artificial Neural Networks Using Low Cost Settings
	1 Introduction
	2 Methods and Materials
	3 Results
	4 Discussions
	5 Conclusions
	References
Contrast and Resolution Improvement of POCUS Using Self-consistent CycleGAN
	1 Introduction
	2 Method
	3 Experimental Results
		3.1 Qualitative Evaluation
		3.2 Quantitative Evaluation
		3.3 Comparative Evaluation
	4 Conclusion
	References
Low-Dose Dynamic CT Perfusion Denoising Without Training Data
	1 Introduction and Related Work
	2 Methods
		2.1 Problem Formulation and Strategy
		2.2 Self-supervised Low-Dose Sinogram-Space Denoising
		2.3 Unsupervised CBF Map Denoising Using CTP Information
		2.4 Low-Dose Simulation
		2.5 DNN Architectures
	3 Data and Experiments
	4 Evaluation, Results, and Discussion
	5 Conclusion
	References
Recurrent Brain Graph Mapper for Predicting Time-Dependent Brain Graph Evaluation Trajectory
	1 Introduction
	2 Proposed Method
	3 Results and Discussion
	4 Conclusion
	References
COVID-Net US: A Tailored, Highly Efficient, Self-attention Deep Convolutional Neural Network Design for Detection of COVID-19 Patient Cases from Point-of-Care Ultrasound Imaging
	1 Introduction
	2 Related Work
	3 Methods
		3.1 COVIDx-US Dataset
		3.2 Network Design
		3.3 Explanation-Driven Performance Validation
	4 Results and Discussion
		4.1 Quantitative Analysis
		4.2 Qualitative Analysis
	5 Conclusions
	References
Inter-domain Alignment for Predicting High-Resolution Brain Networks Using Teacher-Student Learning
	1 Introduction
	2 Methodology
	3 Results and Discussion
	4 Conclusion
	References
Sickle Cell Disease Severity Prediction from Percoll Gradient Images Using Graph Convolutional Networks
	1 Introduction
	2 Methodology
		2.1 Model
		2.2 Feature Extraction
		2.3 Graph Convolution Network
		2.4 Hemoglobin Density Estimation
		2.5 Similarity Metric
	3 Experiments
		3.1 Dataset
		3.2 Implementation Details
		3.3 Results
		3.4 Ablation Study
		3.5 Discussion
	4 Conclusion
	References
Continual Domain Incremental Learning for Chest X-Ray Classification in Low-Resource Clinical Settings
	1 Introduction
	2 Related Work
	3 Method
		3.1 Proposed Approach
	4 Experiments and Results
		4.1 Results
	5 Conclusion
	References
Deep Learning Based Automatic Detection of Adequately Positioned Mammograms
	1 Introduction
	2 Data
		2.1 Data Labeling
	3 Predicting the PEC and PNL on the MLO View
	4 Detecting the BB (Nipple) and the PNL on CC View
	5 Results
		5.1 Predicting PEC and PNL Lines
		5.2 Predicting the Adequacy of MLO
		5.3 Predicting the Positioning of CC View of the Mammogram
		5.4 Predicting the Adequacy of the MLO/CC Pair
		5.5 Generating an Automated Report on the Positioning of the Breast: Real-World Application
	6 Discussion and Future Work
	References
Can Non-specialists Provide High Quality Gold Standard Labels in Challenging Modalities?
	1 Introduction
	2 Method
	3 Experiments and Results
	4 Discussion
	5 Conclusion
	References
Author Index