Machine learning models integrating intracranial artery calcification to predict outcomes of mechanical thrombectomy

Abstract

Objective: To investigate whether intracranial artery calcification (IAC) serves as a reliable imaging predictor of mechanical thrombectomy (MT) outcomes and to develop robust machine learning (ML) models incorporating preoperative emergency data to predict outcomes in patients with acute ischemic stroke (AIS). Methods: This retrospective study included patients with AIS and anterior circulation occlusion who underwent MT at the Second Affiliated Hospital of Dalian Medical University and the Central Hospital Affiliated to Dalian University of Technology between January 2017 and December 2024. Patients were categorized into favorable [modified Rankin Scale (mRS) 0–2] and poor outcome (mRS 3–6) groups based on their 90-day functional independence. Preoperative clinical and radiological data, including a quantitative assessment of IAC, were systematically collected. Eleven ML algorithms were trained and validated using Python, and external validation and performance evaluations were conducted. The Shapley additive explanation (SHAP) method was used to interpret the optimal model. Results: A total of 823 eligible patients were enrolled and stratified into training (n = 437), internal validation (n = 188), and external testing (n = 198) cohorts. The Extra Trees model demonstrated the highest predictive accuracy. The top three predictors were a history of hypertension, serum albumin level, and total calcified volume. Conclusion: The total volume of IAC is a critical imaging biomarker for predicting MT outcomes in patients with anterior circulation AIS. The ML models developed using preoperative emergency data demonstrated strong predictive performance, providing a valuable tool to help clinicians identify suitable MT candidates with greater precision.