Multimodal CT radiomics-clinical ensemble machine learning model effectively predicts futile recanalization after endovascular treatment of acute ischemic stroke.

Backgrounds: Futile recanalization (FR) poses a significant challenge in endovascular treatment and there is a lack of reliable predictive models for assessing treatment outcomes in stroke. The aim of this study is to develop a robust CT radiomics-clinical ensemble model that predicts FR in patients with acute ischemic stroke (AIS) following endovascular treatment (EVT) utilizing machine learning techniques. Methods: This study enrolled 101 patients diagnosed with AIS who underwent successful EVT. A total of 946 radiomics features were, respectively, extracted from non-contrast CT (NCCT), contrast-enhanced CT (CECT), and various CT perfusion maps (CBF, CBV, MTT, and TTP) using PyRadiomics prior to the endovascular intervention. Demographic characteristics, along with baseline clinical, laboratory, and angiographic variables, were incorporated as clinical features in the model analysis. Feature engineering was performed using SelectKBest. Five traditional machine learning algorithms were employed for modeling. The dataset was randomly split into a training cohort ( Results: Among the 101 patients, FR occurred in 66 individuals (65%), as determined by the modified Rankin Scale (mRS) at 90 days. The ensemble model integrating clinical data, NCCT, and CBV achieved the highest performance, with an area under the curve (AUC) of 0.918 using the CatBoost algorithm. Conclusion: The multimodal CT radiomics-clinical ensemble machine learning model demonstrated excellent predictive capability for identifying FR in AIS patients with large vessel occlusion prior to EVT.