基于多模态常规MRI的影像组学预测胶质瘤术后早期复发的价值  

Predictive value of multi-modal conventional MRI radiomics for early postoperative glioma recurrence

作　　者：张裕辉 杨颖思 范伟雄[1] 江桂华[2] 熊小丽 杨日辉 ZHANG Yuhui;YANG Yingsi;FAN Weixiong;JIANG Guihua;XIONG Xiaoli;YANG Rihui(Department of Magnetic Resonance Imaging,Meizhou People's Hospital,Meizhou 514031,China;Department of Imaging,Guangdong Second Provincial General Hospital,Guangzhou 510403,China)

机构地区：[1]梅州市人民医院磁共振科,广东梅州514031 [2]广东省第二人民医院影像科,广东广州510403

出　　处：《中国医学物理学杂志》2025年第2期208-212,共5页Chinese Journal of Medical Physics

基　　金：广东省医学科研基金(A2024768)。

摘　　要：目的:探讨基于多模态常规MRI在术前无创预测胶质瘤术后早期复发的价值。方法:回顾性分析83例符合纳入标准的脑胶质瘤患者的临床及MRI资料。使用Kruskal-Wallis检验方法比较复发组与非复发组的临床因素。通过联影公司开发的深度学习VB-Net算法实现胶质瘤患者全瘤病灶的自动分割并采用URP平台在术前T_(1)CE及T_(2)WI图像上提取影像组学特征,利用最大相关和最小冗余和最小绝对收缩选择算子筛选最佳的特征组合。采用Logistic回归及五折交叉验证分析影像组学特征并构建4个预测模型:T_(2)WI模型,T_(1)CE模型,T_(2)WI+T_(1)CE联合模型,影像-临床融合模型。以受试者工作曲线下面积(AUC)评估各模型诊断效能,并统计模型敏感度及特异度。使用Delong检验比较模型诊断效能。结果:脑胶质瘤术后复发40例,非复发43例。临床因素中胶质瘤级别在两组中差异具有统计学意义(P<0.05),而性别、年龄差异无统计学意义(P>0.05)。在单序列影像组学模型中,T_(1)CE模型(AUC:0.804)优于T_(2)WI模型(AUC:0.702);多模态联合模型AUC高于单序列预测模型(AUC、敏感度、特异度分别为0.849、72.5%、79.1%)。此外,影像-临床融合模型在预测胶质瘤术后早期复发中亦具有良好预测效能,AUC、敏感度、特异度分别为0.839、72.5%、79.1%,但与多模态联合模型效能比较差异无统计学意义(P=0.303)。结论:多模态常规MRI组合模型可以更好地预测胶质瘤术后早期复发,纳入胶质瘤级别的影像-临床融合模型诊断效能并未优于影像组学模型。Objective To explore the preoperative non-invasive prediction of early postoperative glioma recurrence using multi-modal conventional MRI radiomics.Methods A retrospective analysis of the clinical and MRI data of 83 glioma patients who met the inclusion criteria was conducted.The Kruskal-Wallis test was used to compare clinical factors between recurrent and non-recurrent groups.The automated segmentation of the entire tumor lesion for glioma patients was accomplished with VB-Net algorithm,a deep learning approach developed by United Imaging Healthcare;and the extraction of radiomics features from preoperative T_(1)CE and T_(2)WI images was carried out on URP platform.The optimal feature combination was determined using the maximum relevance and minimum redundancy and least absolute shrinkage and selection operator methods.Logistic regression and five-fold cross-validation were employed to analyze radiomics features and construct 4 prediction models,namely T_(2)WI model,T_(1)CE model,T_(2)WI+T_(1)CE model,and imaging-clinical fusion model.The diagnostic performances of these models were evaluated and compared using the area under the receiver operating characteristic curve(AUC)and the Delong test.In addition,the model sensitivity and specificity were calculated.Results Postoperatively,there were 40 recurrent cases and 43 non-recurrent cases.The clinical factors such as glioma grade showed statistical significance between two groups(P<0.05),while gender and age did not show significant statistical differences(P>0.05).For the single-sequence radiomics models,T_(1)CE model(AUC:0.804)outperformed T_(2)WI model(AUC:0.702).The multi-modal combined model exhibited a higher AUC than the single-sequence prediction models,with an AUC of 0.849,a sensitivity of 72.5%,and a specificity of 79.1%.The imaging-clinical fusion model whose predictive efficiency was close to that of multi-modal combined model(P=0.303)also performed well in forecasting postoperative glioma recurrence,with an AUC of 0.839,a sensitivity of 72.5%,and a s

关 键 词：胶质瘤 肿瘤复发 磁共振成像 影像组学 多模态 

分 类 号：R816.1[医药卫生—放射医学]