Colloquium: Robust and trustworthy deep learning-based disease detection and risk assessment on MRI and histopathological images that exceeds predictive performance of human experts

8 October 2024

PhD Colloquium by Fabian Westhäußer, Computer Vision Group, University of Hamburg

Title:

Robust and trustworthy deep learning-based disease detection and risk assessment on MRI and histopathological images that exceeds predictive performance of human experts.

Abstract:

Medical deep learning models often struggle to meet requirements for clinical applicability such as human-level performance, robustness to data variability, and interpretability. This research proposes solutions to these challenges in the context of prostate cancer aggressiveness grading through the development of the Prostate Cancer Aggressiveness Index (PCAI). Trained on a diverse dataset, PCAI demonstrates superior predictive capabilities over several medical experts by predicting cancer risk based on objective patient outcomes. Besides generalizing across several clinical environments, the model quantifies its predictive confidence, enhancing trustworthiness by detecting problematic input and making it a potential blueprint for future clinical decision support systems.

Date and time: 08 October 2024, 09:00 a.m., Albert-Einstein-Ring 8-10, Raum 0005/0010 or via Zoom.