Testing the Trust: Verification and Validation of Bayesian Segmentation under Uncertainty

Giuseppina Carannante, Nidhal Bouaynaya, Dimah Dera, Hassan Fathallah-Shaykh, Ghulam Rasool

Keywords: Bayesian deep learning, Robustness, Segmentation, Trustworthiness

TL;DR: A large-scale robustness and uncertainty analysis of medical segmentation models across datasets, noise types, and adversarial attacks.

Abstract:

Deep learning has achieved state-of-the-art performance in medical image segmentation, yet safe clinical deployment requires rigorous verification and validation of model robustness, reliability, and uncertainty behavior. Bayesian segmentation methods are often viewed as more trustworthy because they provide uncertainty estimates that can support human decision-making, flag unreliable predictions, and mitigate risks in downstream clinical workflows. However, most prior studies evaluate these models primarily on clean test data, with limited assessment of robustness to perturbations, and without examining whether the predicted uncertainty meaningfully correlates with segmentation quality.

In this work, we conduct a comprehensive and systematic evaluation of state-of-the-art deterministic and Bayesian segmentation models across multiple datasets, corruption types, and performance metrics. Beyond accuracy-based metrics such as DSC and HD95, we analyze over- and under-segmentation trends, predictive variance, and the relationship between uncertainty and segmentation correctness. Our results show that while all models behave similarly on clean or mildly corrupted data, performance diverges significantly as perturbations increase. Models that learn and propagate uncertainty during training consistently provide both improved robustness and more clinically meaningful uncertainty estimates, making them stronger candidates for safe and reliable deployment in medical imaging

Primary Subject Area: Safe and Trustworthy Learning-assisted Solutions for Medical Imaging

Secondary Subject Area: Uncertainty Estimation

Registration Requirement: Yes

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Replace NNN with your OpenReview submission ID., Includes \documentclass{midl}, \jmlryear{2026}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package., Did not use the times package., Use the correct spelling and format, avoid Unicode characters, and use LaTeX equivalents instead., Any math in the title and abstract must be enclosed within $...$., Did not override the bibliography style defined in midl.cls and did not use \begin{thebibliography} directly to insert references., Avoid using \scalebox; use \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., No separate supplementary PDF uploads., Acknowledgements, references, and appendix must start after the main content.

Submission Number: 15