Machine learning based models for prediction of rupture status assessment of intracranial aneurysms

Title: Machine learning based models for prediction of rupture status assessment of intracranial aneurysms

Authors: N. Mu, M. Rezaeitaleshmahalleh, Z. Lyu, M. Wang, J. Tang, C. M. Strother, J. J. Gemmete, A. S. Pandey, J. Jiang

Abstract: Unruptured intracranial aneurysms (IAs) are fragile expansions found at critical points in the brain's arterial bifurcations, occurring in approximately 3% of the middle-aged population, indicating that around 168 million individuals globally have an IA. Recent advancements in computational technology have led to the widespread use of numerous machine-learning (ML) models for characterizing IAs, such as rupture risk and stability. However, many machine learning algorithms that offer high prediction accuracy come with the trade-off of model complexity and interpretability challenges. In this presentation, we will first revisit six popular ML algorithms, namely multivariate logistic regression, support vector machine, random forest, extreme gradient boosting, multi-layer perceptron neural network, and Bayesian additive regression trees. Then, we aim to enhance the interpretability of these ML algorithms using techniques such as permutation feature importance, local interpretable model-agnostic explanations (LIME), and the SHapley Additive exPlanations (SHAP) algorithm. Our findings aim to provide better insights into ML algorithms, thereby accelerating their clinical applications.