An Interpretable Machine Learning-Based Prognostic Prediction Model for Acute Ischemic Stroke Patients with Large Vessel Occlusion

Abstract Objective This study aimed to develop and validate a nomogram combining radiomics scores (Rad-score), clinical characteristics, and machine learning (ML) models to predict 3-month functional outcomes in acute ischemic stroke (AIS) patients with large vessel occlusion (LVO). Methods We analyzed 240 AIS patients with anterior circulation LVO who underwent one-stop stroke imaging (non-contrast computed tomography (NCCT), computed tomography perfusion (CTP), and multiphase CT angiography (mCTA)) between October 2019 and December 2024. Fifty-three variables, including pretreatment clinical features, conventional and advanced imaging, and angiographic characteristics, were assessed. The modified Rankin Scale (mRS) at 90 days post-admission (mRS-90) defined the prognostic endpoint (good outcome: mRS-90 ≤ 2; poor outcome: mRS-90 > 2). Patients were randomly allocated to training (80%, n = 192) and testing (20%, n = 48) cohorts. Least absolute shrinkage and selection operator (LASSO) regression derived the Rad-score, which, along with key clinical predictors, was incorporated into a nomogram. Separate clinical, imaging, and hybrid models were developed and evaluated using area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA). SHapley Additive Explanations (SHAP) identified influential features, with SHAP plots visualizing feature importance and interactions. Results The hybrid model exhibited robust discriminative performance for poor outcomes, with an AUC of 0.956 (95% CI: 0.931–0.981). SHAP analysis highlighted Rad-score, admission NIHSS score, symptom onset-to-CT time, and age as the strongest predictors. Conclusions Integrating clinical, laboratory, and imaging data with machine learning enabled precise functional outcome prediction in AIS-LVO patients.