Development of a multimodal model combining radiomics and deep learning to predict malignant cerebral edema after endovascular thrombectomy

Background Malignant cerebral edema (MCE) represents a severe complication after endovascular thrombectomy (EVT) in treating acute ischemic stroke. This study aimed to develop and validate a multimodal predictive model integrating clinical data, radiomics features, and deep learning (DL)-derived features to improve the accuracy of MCE risk prediction following EVT. Methods A total of 290 patients were included, comprising 189 in the training, 47 in the validation, and 54 in the internal test cohorts. A fusion model was developed by integrating clinical variables, radiomics, and DL features. Separate models based on clinical data, radiomics, and DL features were also constructed for comparison. Model training and evaluation were performed on training, validation, and test cohorts. The predictive performance of the combined model was compared with the ACORNS grading scale using an area under the curve (AUC) analysis to assess clinical effectiveness. Results The combined model exhibited the best predictive performance. Analysis of the receiver operating characteristic curve revealed an AUC of 0.927 95% confidence interval (CI): 0.849–1.000 for predicting MCE in the validation group and an AUC of 0.924 (95% CI: 0.846–1.000) in the test group. Additionally, the fusion model consistently demonstrated higher net benefits across all threshold probabilities than the ACORNS grading scale. Conclusions This study integrated clinical data, radiomics, and DL features to develop a multimodal predictive model with a strong discriminative ability to predict MCE after EVT.