基于超声的影像组学结合临床参数术前预测子宫内膜癌淋巴结转移的价值

Value of ultrasound-based radiomics combined with clinical parameters in predicting lymph node metastasis of endometrial cancer

目的探讨基于超声的影像组学结合临床参数预测子宫内膜癌淋巴结转移的价值。方法回顾性分析2017年1月至2021年2月在广西医科大学第一附属医院255例术后确诊为子宫内膜癌患者的临床资料及术前超声图像。采用ITK-SNAP软件手动勾勒肿瘤的边界即感兴趣区(region of interest,ROI),并从中提取特征。通过随机抽样将数据以7∶3的比例分为用于构建模型的训练组和用于评估模型可靠性的验证组(例数比177∶78)。训练组采用Spearman相关系数以0.95为阈值去除相关性高的特征,决策树模型对特征进行重要性排序,选择重要性前6%的特征建立逻辑回归模型,计算每例患者的影像组学评分。在逻辑回归分析的基础上,建立综合预测模型,绘制列线图,并通过ROC曲线下面积(AUC)、校准曲线及决策曲线分析对其预测效果进行评价。结果经降维及逻辑回归分析后最终获得20项特征和2项临床参数(CA125、CA153)用于构建模型。影像组学模型和综合预测模型中训练组和验证组的AUC分别为0.81、0.78和0.89、0.85。临床决策曲线显示两者均有较好的临床实用性。结论基于超声图像构建的影像组学模型及结合相关临床参数构建的综合模型对子宫内膜癌淋巴结转移均有较好的预测效能。

Objective To explore the value of ultrasound-based radiomics combined with clinical parameters in predicting lymph node metastasis of endometrial cancer.Methods The clinical data and preoperative ultrasound images of 255 patients with endometrial cancer diagnosed after operation in the First Affiliated Hospital of Guangxi Medical University from January 2017 to February 2021 were analyzed retrospectively.The boundary of the tumor,that is,the region of interest(ROI),was determined manually by ITK-SNAP software,and the features were extracted from it.Through random sampling,the data were divided into training group used to build the model and verification group used to evaluate the reliability of the model at the proportion of 7∶3.In the training group,the spearman correlation coefficient was used as the threshold of 0.95 to remove the features with high correlation.The importance of the features were ranked by the decision tree model,the features of the top 6%importance were selected to establish a logical regression model,and then calculated the Radiomics score of each patient.On the basis of logical regression analysis,a comprehensive prediction model was established,the line diagram was drawn,and the prediction effect was evaluated by the analysis of the area under the subject working curve(AUC),calibration curve and the decision curve.Results After dimensionality reduction and logical regression analysis,20 features and 2 clinical parameters(CA125,CA153)were obtained to build the model.The AUC of training group and verification group in radiomics model and comprehensive prediction model were 0.81,0.78 and 0.89,0.85,respectively.The decision curve analysis showed that both of them had good clinical utility.Conclusions The ultrasound-based radiomics model and the comprehensive model combined with relevant clinical parameters are effective in predicting lymph node metastasis of endometrial cancer.