Deep Generative Models, and Data Augmentation, Labelling, and Imperfections

DGM4MICCAI 2021 Preface
DGM4MICCAI 2021 Organization
DALI 2021 Preface
DALI 2021 Organization
Contents
Image-to-Image Translation, Synthesis
Frequency-Supervised MR-to-CT Image Synthesis
	1 Introduction
	2 Method
		2.1 Frequency-Supervised Synthesis Network
		2.2 High-Frequency Adversarial Learning
	3 Experiments and Results
		3.1 Experimental Setup
		3.2 Results
	4 Conclusion
	References
Ultrasound Variational Style Transfer to Generate Images Beyond the Observed Domain
	1 Introduction
	2 Methods
		2.1 Style Encoder
		2.2 Content Encoder
		2.3 Decoder
		2.4 Loss Functions
		2.5 Implementation Details
	3 Experiments
		3.1 Qualitative Results
		3.2 Quantitative Results
	4 Conclusion
	References
3D-StyleGAN: A Style-Based Generative Adversarial Network for Generative Modeling of Three-Dimensional Medical Images
	1 Introduction
	2 Methods
		2.1 3D-StyleGAN
	3 Results
	4 Discussion
	References
Bridging the Gap Between Paired and Unpaired Medical Image Translation
	1 Introduction
	2 Methods
	3 Experiments
		3.1 Comparison with Baselines
		3.2 Ablation Studies
	4 Conclusion
	References
Conditional Generation of Medical Images via Disentangled Adversarial Inference
	1 Introduction
	2 Method
		2.1 Overview
		2.2 Dual Adversarial Inference (DAI)
		2.3 Disentanglement Constrains
	3 Experiments
		3.1 Generation Evaluation
		3.2 Style-Content Disentanglement
		3.3 Ablation Studies
	4 Conclusion
	A  Disentanglement Constrains
		A.1  Content-Style Information Minimization
		A.2  Self-supervised Regularization
	B  Implementation Details
		B.1  Implementation Details
		B.2  Generating Hybrid Images
	C  Datasets
		C.1  HAM10000
		C.2  LIDC
	D  Baselines
		D.1  Conditional InfoGAN
		D.2  cAVAE
		D.3  Evaluation Metrics
	E  Related Work
		E.1  Connection to Other Conditional GANs in Medical Imaging
		E.2  Disentangled Representation Learning
	References
CT-SGAN: Computed Tomography Synthesis GAN
	1 Introduction
	2 Methods
	3 Datasets and Experimental Design
		3.1 Dataset Preparation
	4 Results and Discussion
		4.1 Qualitative Evaluation
		4.2 Quantitative Evaluation
	5 Conclusions
	A Sample Synthetic CT-scans from CT-SGAN
	B Nodule Injector and Eraser
	References
Applications and Evaluation
Hierarchical Probabilistic Ultrasound Image Inpainting via Variational Inference
	1 Introduction
	2 Methods
		2.1 Learning
		2.2 Inference
		2.3 Objectives
		2.4 Implementation
	3 Experiments
		3.1 Inpainting on Live-Pig Images
		3.2 Filling in Artifact Regions After Segmentation
		3.3 Needle Tracking
	4 Conclusion
	References
CaCL: Class-Aware Codebook Learning for Weakly Supervised Segmentation on Diffuse Image Patterns
	1 Introduction
	2 Methods
		2.1 Class-Aware Codebook Based Feature Encoding
		2.2 Loss Definition
		2.3 Training Strategy
		2.4 Weakly Supervised Learning Segmentation
	3 Data and Experiments
	4 Results
	5 Discussion
	References
BrainNetGAN: Data Augmentation of Brain Connectivity Using Generative Adversarial Network for Dementia Classification
	1 Introduction
		1.1 Related Work
		1.2 BrainNetGAN
	2 Methods
		2.1 Structural Brain Networks
		2.2 Data Augmentation Using BrainNetGAN
		2.3 Data Acquisition and Experimental Setup
	3 Numerical Results
	4 Discussion and Conclusion
	References
Evaluating GANs in Medical Imaging
	1 Introduction
	2 Methods
		2.1 Competing GANs
	3 Materials
	4 Experimental Results
	5 Conclusions
	References
AdaptOR Challenge
Improved Heatmap-Based Landmark Detection
	1 Introduction
	2 Materials and Methods
		2.1 Data Set
		2.2 Outline of the Proposed Method
		2.3 Pre-processing
		2.4 Point Detection
		2.5 Post-processing
		2.6 Evaluation
	3 Results
	4 Conclusions
	References
Cross-Domain Landmarks Detection in Mitral Regurgitation
	1 Introduction
	2 Method
		2.1 Generating Heatmap of Key Points for Training
		2.2 Inference Procedure
	3 Experiments
		3.1 Datasets
		3.2 Implementation Details
		3.3 Results
	4 Conclusion
	References
DALI 2021
Scalable Semi-supervised Landmark Localization for X-ray Images Using Few-Shot Deep Adaptive Graph
	1 Introduction
	2 Method
		2.1 Deep Adaptive Graph
		2.2 Few-Shot DAG
	3 Results
	4 Conclusion
	References
Semi-supervised Surgical Tool Detection Based on Highly Confident Pseudo Labeling and Strong Augmentation Driven Consistency
	1 Introduction
	2 Methodology
		2.1 Dataset
		2.2 Methods
	3 Experiments
		3.1 Comparative Results
		3.2 Ablation Study
	4 Conclusion
	References
One-Shot Learning for Landmarks Detection
	1 Introduction
	2 Method
		2.1 Overview
		2.2 Offline One-Shot CNN Training
		2.3 Online Structure Detection
		2.4 Online Image Patch Registration
	3 Experiment
		3.1 Dataset
		3.2 Network Architecture and Training Details
	4 Results
	5 Conclusion
	References
Compound Figure Separation of Biomedical Images with Side Loss
	1 Introduction
	2 Related Work
	3 Methods
		3.1 Anchor Based Detection
		3.2 Compound Figure Simulation
		3.3 Side Loss for Compound Figure Separation
	4 Data and Implementation Details
	5 Results
		5.1 Ablation Study
		5.2 Comparison with State-of-the-Art
	6 Conclusion
	References
Data Augmentation with Variational Autoencoders and Manifold Sampling
	1 Introduction
	2 Variational Autoencoder
	3 Some Elements on Riemannian Geometry
	4 The Proposed Method
		4.1 The Wrapped Normal Distribution
		4.2 Riemannian Random Walk
		4.3 Discussion
	5 Data Augmentation Experiments for Classification
		5.1 Augmentation Setting
		5.2 Results
	6 Conclusion
	References
Medical Image Segmentation with Imperfect 3D Bounding Boxes
	1 Introduction
	2 Related Work
	3 Methodology
		3.1 Bounding Box Correction
		3.2 Bounding Boxes for Weakly Supervised Segmentation
		3.3 Implementation Details
	4 Experiments and Discussion
		4.1 Weakly-Supervised Segmentation of 3D CT Volume Using Bounding Box Correction
	5 Conclusions and Discussions
	References
Automated Iterative Label Transfer Improves Segmentation of Noisy Cells in Adaptive Optics Retinal Images
	1 Introduction
	2 Methodology
		2.1 Cell Segmentation Initialization
		2.2 Cell-to-Cell Correspondence Using Graph Matching
		2.3 Data Augmentation Through Iterative Label Transfer
		2.4 Data Collection and Validation Methods
	3 Experimental Results
		3.1 Iterative Cell Segmentation in Noisy Images
		3.2 Purposeful Data Augmentation Improves Training Results
	4 Conclusion and Future Work
	References
How Few Annotations are Needed for Segmentation Using a Multi-planar U-Net?
	1 Introduction
	2 Methods
	3 Datasets
	4 Experiments and Results
	5 Discussion
	References
FS-Net: A New Paradigm of Data Expansion for Medical Image Segmentation
	1 Introduction
	2 Proposed FS-Net
		2.1 Images Channel Coding and Re-Encoding the Ground Truth
		2.2 FS Module
		2.3 Weighted Loss
	3 Experiments
		3.1 Datasets
		3.2 Baselines and Implementation
		3.3 Results
		3.4 Ablation Study
	4 Conclusions
	References
An Efficient Data Strategy for the Detection of Brain Aneurysms from MRA with Deep Learning
	1 Introduction
	2 Materials and Methods
		2.1 Dataset and Data Annotation
		2.2 Model Implementation
		2.3 Patch Generation and Data Augmentation
		2.4 Metrics and Performance Evaluation
	3 Experiments and Results
		3.1 Ablation Study
		3.2 5-Fold Validation
	4 Discussion
	5 Conclusion
	References
Evaluation of Active Learning Techniques on Medical Image Classification with Unbalanced Data Distributions
	1 Introduction
		1.1 Active Learning in Medical Imaging
		1.2 Active Learning Methodology
	2 Methods
		2.1 Datasets
		2.2 Scoring Functions
		2.3 Sampling Strategies
	3 Experiments
		3.1 Experimental Setup
		3.2 Results
	4 Discussion
	5 Conclusion
	References
Zero-Shot Domain Adaptation in CT Segmentation by Filtered Back Projection Augmentation
	1 Introduction
	2 Materials and Methods
		2.1 Filtered Back-Projection Augmentation
		2.2 Comparison Augmentation Approaches
		2.3 Datasets
		2.4 Quality Metrics
	3 Experiments
		3.1 Experimental Pipeline
		3.2 Network Architecture and Training Setup
	4 Results
	5 Conclusion
	References
Label Noise in Segmentation Networks: Mitigation Must Deal with Bias
	1 Introduction
	2 Segmentation Models
	3 Model Performance on Corrupted Labels
		3.1 Random Warp
		3.2 Constant Shift
		3.3 Random Crop
		3.4 Permutation
	4 Limitations and Future Work
	5 Conclusion
	References
DeepMCAT: Large-Scale Deep Clustering for Medical Image Categorization
	1 Introduction
	2 Related Work
	3 Method
	4 Results and Discussion
	5 Conclusion
	References
MetaHistoSeg: A Python Framework for Meta Learning in Histopathology Image Segmentation
	1 Introduction
	2 MetaHistoSeg Framework
		2.1 Histopathology Task Dataset Preprocessing
		2.2 Task and Instance Level Batch Sampling
		2.3 Task-Specific Heads and Multi-GPU Support
	3 Experiments
		3.1 Implementation Details
		3.2 Results
	4 Conclusions
	References
Author Index