DWI-based deep learning radiomics nomogram for predicting the impaired quality of life in patients with unruptured intracranial aneurysm developing new iatrogenic cerebral infarcts following stent placement: a multicenter cohort study.

This study developed a DWI-based radiomics nomogram to predict impaired health-related quality of life (HRQOL) in patients with unruptured intracranial aneurysms after stent placement, focusing on those who developed new iatrogenic cerebral infarct (NICI). Data from 522 patients across multiple hospitals were divided into a training cohort and two external validation cohorts. Radiomic and deep learning features from DWI-based infarct images were selected through super-resolution reconstruction. Impaired HRQOL was defined as a reduction in any of the five EQ-5D-3L domains. Three signatures (clinical, radiomic, and deep learning) were constructed, with a nomogram developed using multivariable logistic regression. Model performance was assessed using receiver operating characteristic analysis, calibration curves, and decision curve analysis. The clinical signature identified key predictors: NICI lesion count/volume, procedure time, diabetes, hypertension, ischemic stroke history, and multiple stents. The radiomic signature achieved optimal performance through super-resolution reconstruction, while GoogleNet showed the best classification performance among deep learning models. The integrated DLRN model achieved high predictive accuracy across all cohorts (AUCs: 0.960, 0.917, 0.936), outperforming individual signatures and traditional models. Calibration curves and decision curve analysis confirmed the DLRN model's reliability and clinical utility. The DLRN model integrating clinical, radiomic, and DTL features accurately predicted 1-year post-procedural HRQOL impairment, surpassing single-modality models and demonstrating clinical applicability for personalized treatment planning.