Abstract
OBJECTIVE: This study aimed to develop a multi-omics nomogram that combines clinical parameters, radiomics, and deep transfer learning (DTL) features of hyperattenuated imaging markers (HIM) from computed tomography scans immediately following mechanical thrombectomy (MT) to predict functional outcomes at discharge. METHODS: This study enrolled 246 patients with HIM who underwent MT. Patients were randomly assigned to a training cohort (n = 197, 80%) and a validation cohort (n = 49, 20%), with an additional internal prospective test cohort (n = 57). A total of 1,834 radiomics features and 25,088 DTL features were extracted from HIM images. Feature selection was conducted using analysis of variance (ANOVA), Pearson's correlation, principal component analysis (PCA), and least absolute shrinkage and selection operator (LASSO) regression. A support vector machine (SVM)-based nomogram integrating clinical, radiomics, and DTL features was developed to predict functional outcomes at discharge. Its performance was evaluated based on accuracy, sensitivity, specificity, receiver operating characteristic (ROC) curve and area under the curve (AUC) analysis, decision curve analysis (DCA), and the DeLong test. RESULTS: The nomogram achieved AUCs of 0.995 (95% CI: 0.989-1.000) in training, 0.959 (95% CI: 0.910-1.000) in validation, and 0.894 (95% CI: 0.807-0.981) in test cohorts. Our nomogram significantly outperformed clinical, radiomics, and DTL models, as well as physician assessments (senior physicians: 0.693, p = 0.001; junior physicians: 0.600, p < 0.001). CONCLUSION: This multi-omics nomogram, integrating HIM-derived, clinical, radiomic, and DTL features, accurately predicts post-MT discharge outcomes, enabling early identification of high-risk patients and optimizing management to improve prognosis.