双参数MRI影像组学模型对卵巢浆液性交界性肿瘤与良性囊性病变的鉴别价值  

Differential Value of the Biparametric MRI Radiomics Model for Serous Borderline Ovarian Tumor and Ovarian Benign Cystic Lesion

作　　者：聂赫 林欣 贺晨 刘伟 闫锐 

机构地区：[1]西安医学院研究生院,陕西 西安 [2]西安邮电大学计算机学院,陕西 西安 [3]西北妇女儿童医院医学影像中心,陕西 西安

出　　处：《临床医学进展》2023年第10期15475-15483,共9页Advances in Clinical Medicine

摘　　要：目的:探讨基于联合T2WI + DWI图像的影像组学特征鉴别卵巢浆液性交界性肿瘤(SBOT)和良性囊性病变(OBCL)的应用价值。方法:回顾性分析经病理证实的98例患者(SBOT 42例、OBCL 56例)的临床及MRI资料。通过3D Slicer手动勾画兴趣区(ROI),经Python进行特征提取和Lasso算法特征降维。建立逻辑回归模型(LR)对筛选出的特征参数进行分类训练,采用留一法交叉验证评估模型性能,绘制ROC曲线评价模型的效能。结果:共提取2446个影像组学特征,通过降维后得到39个特征。T2WI + DWI联合影像组学模型诊断效能的敏感度、特异度、准确度分别为90.4%、96.4%、93.8%,其AUC值(0.98)高于影像医师诊断的AUC值(0.79)。结论:基于联合T2WI + DWI图像的影像组学模型在鉴别SBOT和OBCL中具有重要的临床价值。Objective: To explore the application value of radiomics features based on combined T2WI and DWI images in distinguishing serous borderline ovarian tumor (SBOT) and ovarian benign cystic lesion (OBCL). Method: Clinical and MRI data of 98 patients (42 SBOT and 56 OBCL) confirmed by patholo-gy were retrospectively analyzed. Manually sketch the region of interest (ROI) using the 3D Slicer, extract features using Python, and reduce dimensionality using the Lasso algorithm. Establish a lo-gistic regression model (LR) for classification training of the selected feature parameters, use leave one out cross validation for cross validation to evaluate model performance, and draw ROC curves to evaluate model performance. A total of 2446 radiomics features were extracted, and 39 features were obtained after dimensionality reduction. The sensitivity, specificity, and accuracy of the T2WI and DWI combined radiomics model for diagnostic efficacy were 90.4%, 96.4%, and 93.8%, respec-tively. Its AUC value (0.98) was higher than the AUC value diagnosed by imaging physicians (0.79). Conclusion: The imaging omics model based on combined T2WI and DWI images has important clinical value in distinguishing SBOT and OBCL.

关 键 词：卵巢交界性肿瘤 影像组学 诊断效能 磁共振成像 

分 类 号：R73[医药卫生—肿瘤]