Bridging Probabilistic Models and Interpretability in Medical Video and Image Analysis
Speaker:
Magdalena Trędowicz
Data dell'evento:
Thursday, 25 September, 2025 - 14:30
Luogo:
Aula B101, DIAG
Contatto:
bucarelli@diag.uniroma1.it
Abstract
Recent advances in deep learning have brought remarkable progress in medical image and video analysis, but challenges remain in terms of adaptability, real-time performance, and interpretability. In this talk, I will present my research on probabilistic and prototype-based approaches to computer vision in the medical domain. First, I will introduce PrAViC (Probabilistic Adaptation Framework for Real-Time Video Classification), a method designed to improve robustness and adaptability in video-based diagnostic tasks. Next, I will discuss EPIC (Explanation of Pretrained Image Classification Networks via Prototypes), an interpretability framework that leverages prototypes to provide human-understandable explanations of model predictions. Finally, I will briefly highlight my current work with Prometheus MedTech AI, focusing on translational applications of explainable machine learning in clinical settings, particularly in the diagnosis and management of congenital heart defects.
Recent advances in deep learning have brought remarkable progress in medical image and video analysis, but challenges remain in terms of adaptability, real-time performance, and interpretability. In this talk, I will present my research on probabilistic and prototype-based approaches to computer vision in the medical domain. First, I will introduce PrAViC (Probabilistic Adaptation Framework for Real-Time Video Classification), a method designed to improve robustness and adaptability in video-based diagnostic tasks. Next, I will discuss EPIC (Explanation of Pretrained Image Classification Networks via Prototypes), an interpretability framework that leverages prototypes to provide human-understandable explanations of model predictions. Finally, I will briefly highlight my current work with Prometheus MedTech AI, focusing on translational applications of explainable machine learning in clinical settings, particularly in the diagnosis and management of congenital heart defects.
Short Bio
Magdalena Trędowicz is a 2nd year PhD student at the Jagiellonian University (GMUM Team), supervised by prof. Łukasz Struski and prof. Jacek Tabor. Her research focuses on computer vision, medical imaging, and interpretable AI. She is currently involved in the Prometheus MedTech AI project, where she works on developing trustworthy and clinically applicable machine learning solutions.
Magdalena Trędowicz is a 2nd year PhD student at the Jagiellonian University (GMUM Team), supervised by prof. Łukasz Struski and prof. Jacek Tabor. Her research focuses on computer vision, medical imaging, and interpretable AI. She is currently involved in the Prometheus MedTech AI project, where she works on developing trustworthy and clinically applicable machine learning solutions.