Morphological risk factors of parent arteries in intracranial aneurysm formation: an interpretable machine learning analysis.

purposeIntracranial aneurysms (IAs) pose a severe threat to human health. While the morphological features of the parent artery exert a significant influence on their development, the specific mechanisms of influence and contribution remain unclear. To elucidate the underlying mechanisms, this study constructed a risk assessment model for IAs.

methodsSeventy-nine patients with middle cerebral artery aneurysms (experimental group) and 79 healthy individuals (control group) were enrolled. First, five types of indicators—including diameter, angle, tortuosity index, taper, and stenosis degree—were measured. Next, an automated machine learning (AutoML)-based classification model was built to predict the risk of IA occurrence, while SHapley Additive exPlanations (SHAP) was employed to evaluate the contribution of each feature. Finally, model performance was assessed using metrics such as accuracy, F1 score, precision, recall, and area under the receiver operating characteristic (ROC) curve (AUC).

resultsBifurcation angle is key to IA formation. The optimal risk prediction model, an ensemble of ridge regression, linear discriminant analysis, and linear support vector machine, shows strong performance (accuracy: 0.856, AUC: 0.967). Additionally, feature contribution ranking highlights greater contributions from bifurcation angle, bifurcation diameter ratio, smaller daughter vessel taper, and smaller daughter vessel stenosis.

conclusionParent artery morphological characteristics are key to IA formation, with five types—diameter, angle, tortuosity index, taper, and stenosis degree—playing significant roles. The AutoML-built ensemble machine learning model performed well in predicting aneurysm occurrence and contribute to assisting clinical diagnosis and treatment.