Uncertainty for Safe Utilization of Machine Learning in Medical Imaging: 5th International Workshop, UNSURE 2023, Held in Conjunction with MICCAI 2023, Vancouver, BC, Canada, October 12, 2023, Proceedings

Preface
Organization
Contents
Propagation and Attribution of Uncertainty in Medical Imaging Pipelines
	1 Introduction
	2 Related Work
	3 Methods
		3.1 Upstream Model
		3.2 Downstream Model and Joint Pipeline
		3.3 Loss and Parametrization
	4 Experiments and Results
	5 Conclusions
	References
RR-CP: Reliable-Region-Based Conformal Prediction for Trustworthy Medical Image Classification
	1 Introduction
	2 Reliable-Region-Based Conformal Prediction
		2.1 Principles and Overall Ideas
		2.2 Algorithms and Details
		2.3 Theoretical Analyses
		2.4 Limitations
	3 Experiments
		3.1 Datasets and Setups
		3.2 Results
	4 Conclusions
	References
Bayesian Uncertainty Estimation in Landmark Localization Using Convolutional Gaussian Processes
	1 Introduction
	2 Methods
		2.1 Stage 1: Coarse Prediction Using U-Net
		2.2 Stage 2: Fine Prediction Using CGP
	3 Experiments and Analysis
		3.1 Dataset
		3.2 Metrics
		3.3 Implementation Details
		3.4 Results and Discussion
	4 Conclusion
	References
TriadNet: Sampling-Free Predictive Intervals for Lesional Volume in 3D Brain MR Images
	1 Introduction
	2 Problem Definition
	3 Our Solution: TriadNet
	4 Material and Methods
		4.1 Datasets
		4.2 Comparison with Known Approaches
		4.3 Post-hoc PI Calibration
		4.4 Evaluation
		4.5 Implementation Details
	5 Results and Discussion
	6 Conclusion
	References
Examining the Effects of Slice Thickness on the Reproducibility of CT Radiomics for Patients with Colorectal Liver Metastases
	1 Introduction
	2 Methods
		2.1 CT Imaging and Segmentation
		2.2 Radiomic Feature Extraction
		2.3 Reproducibility Analysis
		2.4 Survival Analysis on Independent Data Set
	3 Results
	4 Discussion
	References
Benchmarking Scalable Epistemic Uncertainty Quantification in Organ Segmentation
	1 Introduction
	2 Epistemic Uncertainty Quantification Techniques
	3 Experimental Design
		3.1 Datasets
		3.2 Metrics
	4 Results
		4.1 Spleen
		4.2 Pancreas
		4.3 Scalability Comparison
	5 Discussion and Conclusion
	References
Numerical Uncertainty of Convolutional Neural Networks Inference for Structural Brain MRI Analysis
	1 Introduction
	2 Materials and Methods
		2.1 Random Rounding
		2.2 Numerical Uncertainty Metrics
		2.3 Non-linear Registration
		2.4 Whole-Brain Segmentation
		2.5 Dataset and Processing
	3 Results
	4 Conclusion
	References
How Inter-rater Variability Relates to Aleatoric and Epistemic Uncertainty: A Case Study with Deep Learning-Based Paraspinal Muscle Segmentation
	1 Introduction
	2 Materials and Methodology
		2.1 Inter-rater Variability
		2.2 Aleatoric and Epistemic Uncertainty Assessment
		2.3 Network Architectures and Label Fusion
		2.4 Dataset
	3 Experiments and Results
		3.1 Experimental Set-Up and Implementation Details
		3.2 Results
	4 Discussion
	5 Conclusion
	References
Uncertainty Estimation and Propagation in Accelerated MRI Reconstruction
	1 Introduction
	2 Methods
	3 Experiments and Results
	4 Discussion
	References
Uncertainty-Based Quality Assurance of Carotid Artery Wall Segmentation in Black-Blood MRI
	1 Introduction
	2 Methods
		2.1 Data Set
		2.2 Segmentation Method
		2.3 Simulation of Inputs
		2.4 Training and Evaluation of the Networks
		2.5 Uncertainty Quantification and Quality Assessment
	3 Results
	4 Discussion and Conclusion
	References
Multi-layer Aggregation as a Key to Feature-Based OOD Detection
	1 Introduction
	2 Compared Methods
		2.1 Feature-Based Methods
		2.2 Adapting Single-Layer Methods to Multi-layer Methods
	3 Material and Method
		3.1 In-Distribution Datasets
		3.2 Out-of-distribution Datasets
		3.3 Influence of the Segmentation Model Architecture
		3.4 Evaluation Setting
	4 Results and Discussion
	References
Feature-Based Pipeline for Improving Unsupervised Anomaly Segmentation on Medical Images
	1 Introduction
	2 Pipeline
		2.1 Basic PatchCore Method
		2.2 Data
		2.3 The Proposed PatchCore3D Method
		2.4 The Proposed Pipeline
	3 Results and Discussion
		3.1 Experimental Setup
		3.2 Results
	4 Conclusion
	References
Redesigning Out-of-Distribution Detection on 3D Medical Images
	1 Introduction
	2 Background
		2.1 Problem Setting
		2.2 OOD Detection Metrics
	3 Expected Performance Drop
	4 Experiments and Results
		4.1 Datasets and Methods
		4.2 Results and Discussion
	5 Conclusion
	References
On the Use of Mahalanobis Distance for Out-of-distribution Detection with Neural Networks for Medical Imaging
	1 Introduction
	2 Methods
	3 Investigation of the Use of D M with Synthetic Artefacts
	4 Investigation of the Use of D M with Real Artefacts
	5 Conclusion
	References
Dimensionality Reduction for Improving Out-of-Distribution Detection in Medical Image Segmentation
	1 Introduction
	2 Methods
		2.1 Data
		2.2 Segmentation Model
		2.3 Out-of-Distribution Detection
	3 Results
	4 Conclusion
	References
Breaking Down Covariate Shift on Pneumothorax Chest X-Ray Classification
	1 Introduction
	2 Related Work
	3 Methods
		3.1 Data
		3.2 Experimental Setup
		3.3 Isolating Domain Shift Factors
	4 Results
		4.1 Isolating Domain Shift Factors
		4.2 LISA-topK
	5 Outlook and Broader Impact
	References
Robustness Stress Testing in Medical Image Classification
	1 Introduction
	2 Stress Testing via Image Perturbations
	3 Experimental Setup
	4 Results and Findings
	5 Discussion and Conclusion
	References
Confidence-Aware and Self-supervised Image Anomaly Localisation
	1 Introduction
	2 Method
	3 Evaluation and Results
	4 Conclusion
	References
Uncertainty Estimation in Liver Tumor Segmentation Using the Posterior Bootstrap
	1 Introduction
	2 Theory
	3 Methods
		3.1 Experiment Setup
		3.2 Evaluation Metrics
	4 Results
		4.1 Dataset Level
		4.2 Subject Level
		4.3 Visual Evaluation
	5 Discussion
	6 Conclusion
	References
Pitfalls of Conformal Predictions for Medical Image Classification
	1 Introduction
	2 Conformal Predictions
	3 Conformal Prediction for Histopathology and Dermatology
	4 Guarantees of Conditional and Marginal Coverage
	5 Conformal Predictions Under Domain Shift
	6 Conformal Predictions and Selective Classification
	7 Classification with Few Classes
	8 Conclusion
	References
Proper Scoring Loss Functions Are Simple and Effective for Uncertainty Quantification of White Matter Hyperintensities
	1 Introduction
	2 Background
	3 Materials and Methods
		3.1 Dataset and Preprocessing
		3.2 Evaluation Metrics
		3.3 Implementation Test Bench
	4 Results
	5 Discussion
	References
Author Index