Research

Uncertainty-Aware Deep Learning for Angiographic Image Analysis

Uncertainty-aware deep learning aims to develop reliable AI systems for medical image analysis by explicitly modeling the inherent ambiguity in expert annotations. In clinical practice, image interpretation often involves subjective decisions, leading to variability between experts. For example, the delineation of stenotic regions in angiographic images can differ depending on the individual assessment of the observer.

This research focuses on capturing and learning such inter-observer variability using probabilistic and uncertainty-aware deep learning approaches. By incorporating uncertainty estimation into segmentation and detection models, the goal is to develop AI-based support systems that better reflect the complexity of clinical decision-making and provide more reliable assistance in medical image analysis.

Knowledge-Informed Deep Learning for Robust Ultrasound Image Analysis

Ultrasound imaging provides valuable information for medical diagnosis and intervention, but its analysis remains challenging due to limited field of view, strong appearance variability, and the dependency on operator experience. This research explores how anatomical knowledge can be integrated into deep learning models to enable more robust and intelligent ultrasound image analysis.

This research investigates how anatomical priors and learned representations can be incorporated into deep learning models to overcome limitations of conventional ultrasound image analysis. The long-term vision is to enable intelligent and autonomous ultrasound systems capable of navigating the human body and acquiring clinically relevant information with reduced operator dependency.