This book constitutes the refereed proceedings of the Second MICCAI Workshop on Data Augmentation, Labelling, and Imperfections, DALI 2022, held in conjunction with MICCAI 2022, in Singapore in September 2022.DALI 2022 accepted 12 papers from the 22 submissions that were reviewed. The papers focus on rigorous study of medical data related to machine learning systems.
Author(s): Hien V. Nguyen (editor), Sharon X. Huang (editor), Yuan Xue (editor)
Publisher: Springer
Year: 2022
Language: English
Pages: 136
Preface
Organization
Contents
Image Synthesis-Based Late Stage Cancer Augmentation and Semi-supervised Segmentation for MRI Rectal Cancer Staging
1 Introduction
2 Methodology
2.1 Semi Supervised Learning with T-staging Loss
2.2 Generating Advanced Cancer MRI Image from Labels
3 Experiments and Results
3.1 Dataset
3.2 Implementation Details and Evaluation Metrics
3.3 Results
4 Discussion
5 Conclusions
References
DeepEdit: Deep Editable Learning for Interactive Segmentation of 3D Medical Images
1 Introduction
2 Proposed Method
2.1 User Interaction and Simulated Clicks
2.2 Training DeepEdit
3 Experimental Results
3.1 Prostate Segmentation Tasks
3.2 Abdominal Organ Segmentation
4 Conclusion
References
Long-Tailed Classification of Thorax Diseases on Chest X-Ray: A New Benchmark Study
1 Introduction
2 Long-Tailed Classification of Thorax Diseases
2.1 Task Definition
2.2 Dataset Construction
2.3 Methods for Benchmarking
2.4 Experiments and Evaluation
3 Results and Analysis
4 Discussion and Conclusion
References
Lesser of Two Evils Improves Learning in the Context of Cortical Thickness Estimation Models - Choose Wisely
1 Introduction
2 Methods
3 Experiments and Results
4 Conclusion
References
TAAL: Test-Time Augmentation for Active Learning in Medical Image Segmentation
1 Introduction
2 Method
3 Experiments and Results
3.1 Implementation Details
3.2 Active Learning Setup
3.3 Comparison of Active Learning Strategies
4 Conclusion
References
Disentangling a Single MR Modality
1 Introduction
2 Method
2.1 The Single-Modal Disentangling Network
2.2 A New Metric to Evaluate Disentanglement
3 Experiments and Results
4 Discussion and Conclusion
References
CTooth+: A Large-Scale Dental Cone Beam Computed Tomography Dataset and Benchmark for Tooth Volume Segmentation
1 Introduction
2 CTooth+ Dataset
2.1 Dataset Summary
2.2 Expert Annotation and Quality Assessment
2.3 Potential Research Topics
3 Experiments and Results
3.1 Evaluation Metrics and Implementations on the CTooth+
3.2 Benchmark for Fully-Supervised Tooth Volume Segmentation
3.3 Benchmark for Semi-supervised Tooth Volume Segmentation
3.4 Benchmark for Active Learning Based Tooth Volume Segmentation
4 Conclusion
References
Noisy Label Classification Using Label Noise Selection with Test-Time Augmentation Cross-Entropy and NoiseMix Learning
1 Introduction
2 Methods
2.1 Label Noise Selection with Test-Time Augmentation Cross-Entropy
2.2 Classifier Training with NoiseMix
3 Experiments and Results
3.1 Datasets and Experimental Details
3.2 Results
4 Conclusions
References
CSGAN: Synthesis-Aided Brain MRI Segmentation on 6-Month Infants
1 Introduction
2 Method
2.1 6-Month-Like MR Data Synthesis
2.2 6-month-like MR Data Segmentation
3 Experiments and Results
3.1 Datasets
3.2 Implement Details
3.3 Results
4 Conclusion
References
A Stratified Cascaded Approach for Brain Tumor Segmentation with the Aid of Multi-modal Synthetic Data
1 Introduction
2 Methods
2.1 Stratified Conditional GANs for Brain Tumor Image Synthesis
2.2 Whole Tumor Detection
2.3 Brain Tumor Grade Classification
2.4 Fine-Grained Brain Tumor Region Segmentation
3 Results and Discussion
3.1 Material
3.2 Experimental Setup
3.3 Evaluation
4 Conclusion and Future Work
References
Efficient Medical Image Assessment via Self-supervised Learning
1 Introduction
2 Preliminaries
2.1 Supervised-Learning-Based Data Assessment
2.2 Formulation of Unsupervised-Learning-Based Data Assessment
3 Our Method
3.1 Theoretical Implication
3.2 Data Assessment on Singular Value
3.3 Forming Embedding Space Using Masked Auto-encoding
4 Experiment
4.1 Experiment Setup and Dataset
4.2 Proof of Concept with `Ground-Truth'
4.3 Comparison with Alternative Embedding Methods
4.4 Comparison with Baseline Data Valuation Methods
5 Discussion and Conclusion
References
Few-Shot Learning Geometric Ensemble for Multi-label Classification of Chest X-Rays
1 Introduction
2 Methods
3 Experiments
3.1 Dataset
3.2 Baselines
3.3 Implementation Details
3.4 Results and Evaluation
4 Conclusion
References
Author Index
