The study examines feature selection techniques for identifying key thrombus-related radiomic features from non-contrast CT (NCCT) and CT angiography (CTA) images to differentiate between patients with successful reperfusion and those without after endovascular treatment in acute stroke cases. Using data from 212 patients, 1208 radiomic features were analyzed through a pipeline combining unsupervised and supervised methods, ensuring robustness with 5-fold cross-validation. The study found modality-specific importance in features, noting the relevance of GLCM correlation in CTA, but not NCCT, and identified wavelet-based features as significant across both modalities. The research suggests tailored feature selection to optimize prediction accuracy for different imaging modalities and plans to explore these features’ clinical applicability to outcomes like the first-pass effect and their correlation with thrombus histology. The findings aim to enhance the interpretability and reliability of radiomic analyses, contributing to personalized treatment strategies for acute stroke patients.