Machine Learning-Based Classification of Anterior Circulation Cerebral Infarction Using Computational Fluid Dynamics and CT Perfusion Metrics<span style="mso-fareast-language: ZH-CN;">

Background: Intracranial atherosclerotic stenosis (ICAS) is a primary cause of ischemic stroke and understanding its pathogenesis aids in therapeutic decision-making. This study integrates computational fluid dynamics (CFD), CT perfusion (CTP) metrics, anatomical indexes, and machine learning to classify anterior circulation ICAS within the Chinese Ischemic Stroke Subclassification (CISS) framework and identify characteristics of different stroke mechanisms. Methods: A total of 118 ICAS patients were classified based on CISS criteria. Key indicators were identified through one-way ANOVA, correlation, and effect size analysis. A decision tree model established thresholds for stroke mechanisms, and six machine learning models were evaluated for classification performance using confusion matrices, ROC curves, and PR curves. Results: Time to Maximum (Tmax)&gt; 4.0s, Area stenosis rate (AS%), wall shear stress ratio (WSSR), and pressure ratio (PR) were key indicators for classification, while cerebral blood flow (CBF) and cerebral blood volume (CBV) showed no significant differences across infarction types. Thresholds identified included Tmax &gt; 4.0s = 49.85 ml, WSSR = 52.79/86.51, PR = 0.69, and AS% = 0.73. Logistic regression outperformed other models (AUC = 0.923, AP = 0.872), followed by the ensemble model. Conclusion: Combining CFD and CTP metrics with machine learning effectively classifies anterior circulation ICAS-related stroke mechanisms within CISS typing, offering a reliable approach for precise diagnosis and individualized treatment of ischemic stroke.