C-UQ: Conflict-based uncertainty quantification—A case study in lung cancer classification

Computers in Biology and Medicine, Journal Year: 2025, Volume and Issue: 188, P. 109825 - 109825

Published: Feb. 20, 2025

Uncertainty quantification is crucial in deep learning, especially medical diagnostics, to measure model prediction confidence and ensure reliable clinical decisions. This study introduces a novel conflict-based uncertainty approach, applied as case lung cancer classification, leveraging Dempster-Shafer Theory conjunction with Deep Ensemble methods. The proposed method aggregates predictions from multiple neural network models using conflict an measure. By converting softmax outputs into Basic Belief Assignments applying the rule of combination, this effectively quantifies uncertainty: high values indicate requiring expert review, low are considered reliable. Evaluations on LIDC-IDRI dataset additional 3D biomedical datasets show that achieved accuracy (0.957) URecall (0.819) for classification. sensitivity analysis further revealed increasing ensemble size enhanced performance even though computational demands may challenge real-time applications. In contrast, entropy-based smoothing effect limits improvement traditional addition, Out-of-Distribution detection AUC scores up 0.864 across various datasets. Future work will focus optimising efficiency exploring alternative combination rules hybrid models.

Language: Английский

Dempster-shafer deep capsule attention model (DDCAM)

Multimedia Tools and Applications, Journal Year: 2025, Volume and Issue: unknown

Published: April 30, 2025

Machine learning prediction of heating and cooling loads based on Athenian residential buildings’ simulation dataset

Energy and Buildings, Journal Year: 2025, Volume and Issue: unknown, P. 115808 - 115808

Published: May 1, 2025

Language: Английский