Core Concepts
Bayesian Uncertainty Estimation using Hamiltonian Monte Carlo improves segmentation accuracy and uncertainty estimation in cardiac MRI.
Abstract
Deep learning methods in medical image segmentation can be miscalibrated and overconfident, leading to "silent failures" in clinical applications.
Bayesian statistics offer a solution for failure detection based on posterior probability estimation.
Proposed HMC-CP framework enhances uncertainty estimation and segmentation accuracy.
Cyclical annealing strategy in HMC captures local and global geometries for efficient Bayesian DNN training.
Functional space diversity analysis reveals superior uncertainty estimation with proposed method.
Evaluation on ACDC, M&M, and QMRI datasets shows improved performance compared to baseline methods.
Stats
Bayesian statistics provide an intuitive approach to DL failure detection.
HMC-CP framework improves segmentation accuracy and uncertainty estimation.
Quotes
"Our results show that the proposed method improves both segmentation accuracy and uncertainty estimation for in- and out-of-domain data."