Development and validation of a radiomics model based on the ASPECTS framework using CT imaging for predicting...

Accurate delineation of the infarct region on acute-phase Computed Tomography (CT) remains challenging, and radiomics applications in stroke are limited. We aimed to develop and validate a multimodal prediction model for malignant cerebral edema (MCE) using clinical and radiomic features extracted based on the Alberta Stroke Program Early Computed Tomography Score (ASPECTS) framework, eliminating the need for manual infarct segmentation.

This multicenter retrospective study included patients with acute ischemic stroke (AIS) from five stroke centers who underwent Non-Contrast Computed Tomography (NCCT) and computed tomography angiography (CTA). Radiomic features were extracted from ASPECTS regions. Clinical, imaging-alone, and fused models were developed using machine learning. Model performance was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC) analysis, accuracy, calibration curves, and decision curve analysis (DCA). Shapley Additive exPlanations (SHAP) was used to interpret feature contributions.

A total of 708 patients were included (median age: 67 years; interquartile range (IQR): 60-76; 448 men, 63.3 %). In the training cohort, the fused model (AUC = 0.91, 95 %CI: 0.88-0.95) outperformed the clinical (AUC = 0.72, 95 %CI: 0.66-0.78, p < 0.001) and CTA_Features (AUC = 0.83, 95 %CI: 0.79-0.88, p < 0.001) models. In the internal validation cohort, the fused model (AUC = 0.78, 95 %CI: 0.70-0.87) outperformed the clinical model (AUC = 0.68, 95 %CI: 0.59-0.77, p = 0.030). In the external validation, the fused model (AUC = 0.88, 95 %CI: 0.82-0.94) outperformed the clinical (AUC = 0.71, 95 %CI: 0.59-0.83, p = 0.010) and CT_Features (AUC = 0.70, 95 %CI: 0.56-0.85, p = 0.012) models. SHAP analysis identified ASPECTS, National Institutes of Health Stroke Scale (NIHSS) score, and collateral score (CS) as top predictors. The fused model demonstrated the highest specificity (82.5 %) and accuracy (78.3 %).

The fused model integrating clinical, radiological, and radiomic features extracted using the ASPECTS framework, demonstrated superior and generalizable predictive performance for early MCE prediction. As it uses routinely acquired baseline NCCT and CTA with readily available admission variables and avoids manual segmentation through an ASPECTS-based framework, it can be integrated into clinical workflows to enable rapid and consistent MCE risk estimation.