摘要
目的:探索MRI影像组学特征在手术前预测胃肠道间质瘤病理危险性分级的应用价值。方法:筛选了132例通过手术病理确认的胃肠道间质瘤病例,并按7:3的比例随机分配为训练集和测试集。从DWI、T2WIfs和T1WI序列抽取关联的影像学特征,联合CE-T1WIfs及临床数据特征,并通过方差阈值方法、相关性筛选及单变量逻辑回归法筛选出具有重要意义的影像学特征和临床数据,利用这些数据特征构筑预测模型,应用受试者工作特征(ROC)曲线下面积(AUC)在训练集和测试集上对模型效能开展评估。结果:单因素预测模型中,各序列预测能力不一,其中由DWI构筑的模型评估效能最好,其AUC、灵敏度及特异度分别为0.960、80.0%和100.0%;在多序列联合模型中,由T2WIfs联合DWI构筑的模型评估效能最好,其AUC、灵敏度及特异度分别为0.859、87.5%和75.0%;在单序列联合增强模型及影像组学联合临床数据模型中,两者的预测效能较单因素构建模型均有不同程度提高,以T1WI联合CE-T1WIfs及DWI联合病灶最大直径这2个模型预测效能最高,两者测试集AUC、灵敏度及特异度分别为0.834、70.4%及85.2%和0.969、96.0%及88.0%。结论:MRI不同序列组学特征所构筑模型均能不同程度地预测胃肠道间质瘤的病理危险性分级,其中DWI序列展示出优秀的预测效能。
Purpose:To explore the value of MRI radiomics in predicting risk classication of gastrointestinal stromal tumor before operation.Methods:A total of 132 patients with pathologically confirmed gastrointestinal stromal tumors were retrospectively analyzed.Patients were randomly divided into a training group and a test group at a ratio of 7∶3.The radiomics features of DWI,T2WIfs and T1WI sequence were extracted,and were combined with CE-T1WIfs and clinical data characteristics.Variance threshold method,correlation screening and single factor logistic regression method were used to analyzed significant radiomics and clinical features.And the data features selected were used to construct a predictive model.The area under ROC curve(AUC)was used to evaluate the model performance on the training and test sets.Results:In the single-factor prediction model,the prediction ability of each sequence was different,and the evaluation efficiency of DWI model was the best,with AUC,sensitivity and specificity of 0.960,80.0%and 100.0%,respectively.Among the multi-sequence combination models,the model constructed by T2WIfs combined with DWI had the best evaluation efficiency,with AUC,sensitivity and specificity of 0.859,87.5%and 75.0%,respectively.Among the single sequence combined with enhancement model and the radiomics combined with clinical data model,the prediction efficiency of the two models was improved to some extent compared with that of the single factor model,and the prediction efficiency of T1WI combining with CE-T1 WIfs and DWI combining with maximum lesion diameter was the highest.The AUC,sensitivity and specificity of the two test sets were 0.834,70.4%and 85.2%,0.969,96.0%and 88.0%,respectively.Conclusion:The models constructed by different MRI sequences could be used to predict the pathological risk grade of gastrointestinal stromal tumors to varying degrees,and DWI sequences showed excellent prediction efficiency.
作者
潘桂海
陈克敏
周飞
胡佳慧
邸小青
陈东
PAN Guihai;CHEN Kemin;ZHOU Fei;HU Jiahui;DI Xiaoqing;CHEN Dong(Department of Radiology,Affiliated Hospital of Guangdong Medical University,Zhanjang 524001,China;Department of Radiology,Rujin Hospital,Shanghai Jiao Tong University School of Medicine;Department of Endocrinology,Afiliated Hospital of Guangdong Medical University;Department of Pathology,Affliated Hospital of Guangdong Medical University)
出处
《中国医学计算机成像杂志》
北大核心
2025年第1期65-71,共7页
Chinese Computed Medical Imaging
基金
湛江市科技发展专项资金竞争性分配项目(2022A702-1)。
关键词
胃肠道间质瘤
危险性分级
磁共振成像
影像组学
Gastrointestinal stromal tumor
Risk classification
Magnetic resonance imaging
Radiomics
