Analysis of a clot-based combined radiomics model for predicting embolic etiology in acute ischemic stroke patients.

Background and objectives: Acute ischemic stroke (AIS) remains the primary cause of mortality and disability among adults in China. Different etiologies of acute ischemic stroke (AIS) are considered important factors affecting neurological function. The accurate etiological classification of AIS prior to surgery is crucial. To investigate and explore the predictive value of a clot-based combined radiomics model for identifying the etiological subtypes of acute ischemic stroke. Materials and methods: A total of 263 patients with acute ischemic stroke caused by anterior circulation large artery occlusion were retrospectively enrolled. These were grouped into training (180), testing (45), and external validation cohorts (38). NCCT and CTA scans were adopted to segment region of interest (ROI) of clots. Feature selection was conducted to establish Clinical model, radiomics models (NCCT, CTA, and NCCT&CTA), and combined model. Results: The AUCs of the clinical model and radiomics models (NCCT, CTA and NCCT&CTA) in the testing cohort were 0.8288(95 % CI: 0.7174-0.9403), 0.8133(95 % CI:0.6853-0.9414), 0.8075(95 % CI:0.6844-0.9307) and 0.8535(95 % CI:0.6774-1), respectively. The combined model achieved a greater AUC than the other four models in the testing cohort (0.9077 [95 % CI: 0.821-0.9944]). Clinical decision curve analysis (DCA) demonstrated that the radiomics (NCCT&CTA) model and combined model show better net benefits within a relatively wide range of threshold probabilities. Conclusion: The combined radiomics model achieved good predictive efficacy for distinguishing the etiological subtypes of acute ischemic stroke and can provide valuable information for the precise selection of recanalization strategies in clinical practice.