Prediction of prognosis in acute ischemic stroke after mechanical thrombectomy based on multimodal MRI radiomics and deep learning

Lei PeiXiaowei HanChenfeng NiJunli Ke^*

• Quzhou City People's Hospital, Quzhou, China

Background: Acute ischemic stroke (AIS) poses a global health threat due to high disability and mortality rates, necessitating early prognostic tools, particularly post-mechanical thrombectomy. This study evaluated multimodal MRI-based radiomics and deep learning for predicting poor AIS prognosis after thrombectomy, aiming to provide an accurate prognostic tool.Methods: Clinical data and baseline MRI of stroke patients were retrospectively analyzed. Logistic regression identified risk factors and built a clinical model. Radiomics features from MRI were extracted, with optimal features selected via LASSO regression and five-fold cross-validation. A radiomics score was calculated, and logistic regression developed the radiomics model. A deep learning model used ResNet101 with transfer learning. Clinical, radiomics, and deep learning models were integrated into the CRD (Clinic-Radiomics-Deep Learning) model. Performance was assessed via ROC curves, calibration curves, and decision curve analysis.Results: Among 222 AIS patients (155 training, 67 validation), admission NIH Stroke Scale score and intracerebral hemorrhage were independent risk factors. From 1,197 radiomics features, 16 were selected. In training, clinical, radiomics, deep learning, and CRD models achieved AUCs of 0.762, 0.755, 0.689, and 0.834, respectively. In validation, AUCs were 0.874 (clinical), 0.805 (radiomics), 0.757 (deep learning), and 0.908 (CRD). Calibration curves confirmed CRD’s accuracy, while decision curves showed its highest net benefit.Conclusions: The CRD model integrating multimodal MRI demonstrated superior efficacy in predicting poor prognosis post-thrombectomy, offering reliable support for clinical decision-making in AIS management.