基于多模态MRI的影像组学和深度学习在评价缺血性脑卒中预后中的应用价值OA

Objective To investigate the application value of radiomics and deep learning approaches based on multimodal magnetic resonance imaging(MRI)in predicting poor prognosis in patients with acute ischemic stroke(AIS).Methods A retrospective analysis was conducted on the clinical data and multimodal MRI imaging data of AIS patients who underwent brain MRI examinations at the Department of Radiology,Quzhou People's Hospital from January 2024 to October 2025.A clinical model was developed using Logistic regression analysis to systematically identify independent clinical risk factors significantly associated with poor stroke outcomes.A radiomics model was constructed using baseline multimodal MRI data,with key features selected via the least absolute shrinkage and selection operator(LASSO)algorithm.Additionally,a deep learning model was built based on ResNet101 architecture combined with transfer learning.Finally,a comprehensive clinic-radiomics-deep learning(CRD)model was established by integrating the above three models.Results A total of 210 stroke patients were included.Clinical data analysis revealed that the National Institute of Health stroke scale score at admission and intracranial hemorrhage were independent predictors of poor prognosis.In the training cohort,the areas under the curve(AUC)for the clinical,radiomics,deep learning,and CRD models were 0.744,0.682,0.531,and 0.795,respectively.In the validation cohort,their corresponding AUC were 0.786,0.716,0.584,and 0.867.The CRD model demonstrated the best predictive performance in both cohorts.Conclusion The CRD comprehensive prediction model,constructed based on multimodal MRI,demonstrates high predictive efficacy and stability in forecasting poor prognosis in patients with AIS,providing a reference for clinical risk assessment and personalized decision-making.