基于高分辨T2WI及弥散加权成像的放射组学模型预测Ⅱ~Ⅲ期直肠癌患者微卫星稳定性状态  

Radiomics-based prediction of microsatellite instability in stageⅡandⅢrectal cancer patients based on T2WI MRI and diffusion-weighted imaging

作　　者：相帅 郑龙波 朱玲 高源[1] 王东升[1] 刘尚龙[1] 张帅[3] 王童语 卢云[1] Xiang Shuai;Zheng Longbo;Zhu Ling;Gao Yuan;Wang Dongsheng;Liu Shanglong;Zhang Shuai;Wang Tongyu;Lu Yun(Department of Gastrointestinal Surgery,Affiliated Hospital of Qingdao University,Qingdao 266000,China;Shandong Provincial Key Laboratory of Digital Medicine and Computer-Assisted Surgery,Qingdao 266000,China;Department of Radiology,Affiliated Hospital of Qingdao University,Qingdao 266000,China)

机构地区：[1]青岛大学附属医院胃肠外科,青岛266000 [2]山东省数字医学与计算机辅助手术重点实验室,青岛266000 [3]青岛大学附属医院放射科,青岛266000

出　　处：《中华外科杂志》2023年第9期782-787,共6页Chinese Journal of Surgery

摘　　要：目的探讨基于高分辨T2WI及弥散加权成像(DWI)的放射组学模型在预测Ⅱ~Ⅲ期直肠癌患者微卫星稳定性状态中的价值。方法回顾性收集青岛大学附属医院2016年2月至2020年10月连续收治的175例Ⅱ~Ⅲ期直肠癌患者资料,男性119例,女性56例,年龄(63.9±9.4)岁(范围:37~85岁)。微卫星稳定性患者152例,微卫星不稳定性患者23例。以7∶3的比例将患者随机分为训练组123例和验证组52例。利用ITK-SNAP软件在每位患者的T2WI及DWI图像中标记感兴趣区域,用PyRadiomics提取7类放射组学特征,在去除冗余特征及特征归一化后,使用最小绝对收缩和选择算子进行特征选择。基于支持向量机在训练组中构建1个临床模型、3个放射组学模型和1个临床-放射组学模型,并使用受试者工作特征曲线下面积、灵敏度、特异度、准确性在验证组中评估模型性能。结果筛选出与直肠癌患者微卫星状态最相关的3个临床特征(年龄、肿瘤分化程度、肿瘤下缘距肛缘距离)及6个放射组学特征(2个DWI相关特征和4个T2WI相关特征)。临床-放射组学模型在训练组中的曲线下面积为0.95,验证组中为0.81,优于临床模型(0.68,Z=0.71,P=0.04),与T2WI+DWI模型诊断效能相当(0.82,Z=-0.21,P=0.83)。结论术前T2WI与DWI的放射组学特征与Ⅱ~Ⅲ期直肠癌患者微卫星稳定性状态相关,基于此构建的模型表现出了较高的分类效能。Objective To examine the radiomics model based on high-resolution T2WI and diffusion weighted imaging(DWI)in predicting microsatellite stability in patients with stageⅡandⅢrectal cancer.Methods From February 2016 to October 2020,175 patients with stageⅡandⅢrectal cancer who met the inclusion criteria were retrospectively collected.There were 119 males and 56 females,aged(63.9±9.4)years(range:37 to 85 years),including 152 patients with microsatellite stability and 23 patients with microsatellite instability.All patients were randomly divided into the training group(n=123)and the validation group(n=52)with a ratio of 7∶3.The region of interest was labeled on the T2WI and DWI images of each patient using the ITK-SNAP software,and PyRadiomics was used to extract seven kinds of radiomics features.After removing redundant features and normalizing features,the least absolute shrinkage and selection operation were used for feature selection.One clinical model,three radiomics models and one clinical-radiomics model were constructed in the training group based on a support vector machine.The area under receiver operating characteristic curve(AUC),sensitivity,specificity,and accuracy were used to evaluate the performance of the models in the verification group.Results Three clinical features(age,degree of tumor differentiation,and distance from the lower edge of the tumor to the anal edge)and six radiomics features(two DWI-related features and four T2WI-related features)most related to microsatellite status of rectal cancer patients were selected.The AUC of the clinical-radiomics model in the training group was 0.95.In the validation group,the AUC was 0.81,better than the clinical model(0.68,Z=0.71,P=0.04),and equivalent to the T2WI+DWI model(0.82,Z=0.21,P=0.83).Conclusions Radiomic features based on preoperative T2WI and DWI were related to microsatellite stability in patients with stageⅡandⅢrectal cancer and showed a high classification efficiency.The model based on the features provided a noninvasive and c

关 键 词：直肠肿瘤 微卫星不稳定性 诊断 计算机辅助 机器学习 放射组学 

分 类 号：R735.37[医药卫生—肿瘤]