Uncertainty Quantification and Explainable AI for Aneurysm Rupture Risk Prediction: A Bayesian Deep Learning Approach to Hemodynamic Analysis

Abstract

Aneurysm rupture is a life-threatening medical emergency with high mortality rates, making accurate rupture risk prediction essential for clinical decision-making. While computational hemodynamic modeling and deep learning approaches have shown promise in aneurysm assessment, existing methods often lack robust uncertainty quantification and interpretable predictions that clinicians can trust. This paper proposes a Bayesian deep learning framework for aneurysm rupture risk prediction that combines hemodynamic feature analysis with uncertainty-aware neural networks and explainable AI techniques. Our approach leverages Monte Carlo dropout for epistemic uncertainty estimation and integrates SHAP-based explanations to provide clinicians with transparent, calibrated risk predictions. Through extensive experiments on clinical hemodynamic datasets, we demonstrate that the proposed framework produces well-calibrated probability estimates and provides interpretable explanations that highlight the hemodynamic features most critical for rupture risk. Our work contributes to the growing field of trustworthy AI in healthcare, providing a principled approach to uncertainty-aware, explainable medical decision support for aneurysm management.