探讨影像组学应用于子宫内膜息肉样病变的价值

The value of radiomics in endometrial polypoid lesions

目的:探讨影像组学在子宫内膜息肉样病变的应用价值。方法:回顾性分析广西医科大学第一附属医院193例经病理证实的子宫内膜息肉样病变的超声图像资料。从超声图像提取影像组学特征并生成数据集。数据集以病变类型按7比3的比例拆分为训练集和验证集。训练集通过假设检验筛选与病变类型相关的差异特征,并提交到L1正则化逻辑回归方法筛选最终构建模型的特征。最后,以决策树机器学习算法构建预测模型。验证集用于评估模型。使用受试者工作特征(ROC)曲线及曲线下面积(AUC)评估模型的性能。结果:最终选择了39个特征进行模型构建。训练集和验证集的影像组学评分在良性和恶性子宫内膜息肉样病变上显著差异(训练集,P<0.0001,验证集,P<0.0001),训练集和验证集的AUC分别为0.91和0.84,准确度、灵敏度及特异度分别为0.93和0.86,0.84和0.74,0.97和0.92,显示出中度以上的鉴别性能。结论:影像组学在子宫内膜息肉样病变的鉴别诊断中具有一定的价值,可作为鉴别诊断子宫内膜息肉样病变的补充手段。

Objective:To investigate the value of radiomics in endometrial polypoid lesions.Methods:The ultrasonographic data of 193 pathologically confirmed cases of endometrial polyposis in the First Affiliated Hospital of Guangxi Medical University were retrospectively analyzed.Radiomics features were extracted from ultrasonic images and data set was generated.The data set was divided into training set and verification set according to the ratio of 7 to 3 for lesion type.In the training set,the differential features related to lesion type were screened by hypothesis testing,and the L1 regularization logistic regression method was presented to screen the features of the final model.Finally,a decision tree machine learning algorithm was used to construct the prediction model.Validation set was used to evaluate the model.The performance of the model was assessed using the receiver operating characteristic(ROC)curve and area under the curve(AUC).Results:In the end,39 features were selected for model construction.Radiomics scores of the training set and the validation set were significantly different in benign and malignant endometrial polypoid lesions(training set,P<0.0001,validation set,P<0.0001).The AUC of training set and verification set were 0.91 and 0.84,respectively.The accuracy,sensitivity and specificity were 0.93 and 0.86,0.84 and 0.74,0.97 and 0.92 respectively,showing moderate or higher discriminant performance.Conclusions:Radiomics has a certain value in the differential diagnosis of endometrial polypoid lesions and can be used as a supplementary method in the differential diagnosis of endometrial polypoid lesions.