【摘要】 目的 基于T2WI开发并验证一种预测脑胶质瘤患者无进展生存期（progression-free survival, PFS）的深度学习（deep learning, DL）模型。材料与方法 收集了来自三个中心共345名被诊断为胶质瘤患者的MRI图像和临床资料。建立了9个DL模型来预测PFS，使用外部测试集对模型性能进行验证。采用C指数确定最佳模型，并比较了这些模型的性能。根据训练集中计算的风险评分截止值，将患者分层为高风险组和低风险组，评估他们的PFS之间的差异。结果 训练集和测试集分别由249名和96名患者组成。Wide ResNet50-2深度学习模型表现优异，在训练集和测试集的C指数分别为0.694和0.714。将临床特征与DL特征结合后建立的综合模型得到最高的模型表现，在训练集和外部测试集的C指数分别为0.724和0.795。根据深度学习特征计算的风险评分可以用于区分不同风险的患者。结论 基于术前MRI的DL模型能够预测胶质瘤患者的PFS，并可作为术前风险分层工具。更多还原

【Abstract】 Objective: To develop and validate a deep learning(DL) model for predicting progression-free survival(PFS) in patients with glioma based on T2WI. Materials and Methods: MRI and clinical data from 345 patients diagnosed with glioma across three centers were collected. Nine DL models were established to predict PFS, and their performance was validated using an external test set.The best model was determined by the C-index, and the performances of these models were compared. Patients were stratified into high-risk and low-risk groups based on risk score cutoff values calculated from the training set, and differences in PFS between these groups were assessed. Results: The training and test sets consisted of 249 and 96 patients, respectively. Compared with other DL models,the Wide ResNet50-2 DL model performed best, achieving C-indexes of 0.694 and 0.714 in the training and test sets, respectively. A combined model incorporating both clinical and DL features showed the highest performance, with C-indexes of 0.724 and 0.795 in the training and external test sets, respectively. Conclusions: A DL model based on preoperative MRI can predict PFS in patients with glioma and may serve as a preoperative risk stratification tool.更多还原