A hybrid model based on vessel wall magnetic resonance imaging predicts recurrence in posterior circulation ischemic...

This study aimed to develop and validate a hybrid model integrating clinical features, vessel wall magnetic resonance imaging (VWMRI) characteristics, and radiomic features to predict the recurrence risk of posterior circulation ischemic stroke (PCIS).

This multicenter, retrospective study included 266 PCIS patients with basilar artery atherosclerosis (Institution I: 227; Institution II: 39). Data included clinical information, VWMRI imaging features, and radiomic features. Six predictive models were constructed using the logistic regression algorithm: (1) clinical-imaging model, (2-4) single-sequence radiomics models (2D PDWI, 3D T1WI, 2D T2WI), (5) multi-sequence radiomics model, and (6) hybrid model. Model performance was evaluated using receiver operating characteristic curves, calibration metrics, decision curve analysis across training/internal validation (8:2 split), and external validation cohorts.

The hybrid model exhibited superior predictive performance across all cohorts, with an area under the curve of 0.872 in the training cohort, 0.883 in the internal validation cohort, and 0.869 in the external validation cohort, outperforming all other models. Calibration curves demonstrated excellent fit, with mean absolute errors of 0.03/0.06. Decision curve analysis indicated significant net benefit within clinically relevant threshold ranges. Key risk factors for PCIS recurrence included involvement of perforating arteries, non-dorsal plaque location, elevated low-density lipoprotein cholesterol, high total cholesterol, a history of alcoholism, and 16 radiomic signatures.

The hybrid model enhances the accuracy of predicting PCIS recurrence risk. With robust performance in both internal and external validation cohorts, this model demonstrates strong clinical utility and potential to facilitate early intervention and personalized treatment strategies.