基于DBT建立的早期预测乳腺癌HER-2状态的模型研究

A Nomogram Prediction Model Study Based on Breast DBT Images to Establish Early Prediction of HER-2 Status of Breast Cancer

目的 使用基于数字乳腺断层摄影(DBT)图像的影像组学特征早期预测乳腺癌患者的HER-2状态。方法 搜集符合纳入标准的160例女性乳腺癌患者,其中HER-2阳性50例,HER-2阴性110例。手动在DBT图像上勾画感兴趣区,提取影像组学特征,经过数据归一化和降维,共选取1~10个特征建立逻辑回归模型,进行10折交叉验证,选取交叉验证集曲线下面积最高的模型为最佳组学模型。分析临床特征得到独立预测因子并建立临床预测模型,将最佳组学模型预测值与临床独立预测因子结合建立综合模型并绘制诺莫图,诺莫图的拟合度用Hosmer-Lemeshow拟合优度检验评价,绘制决策曲线评价诺莫图的净收益。结果 基于4个组学特征的LR模型为最佳组学模型,此模型在测试集中的曲线下面积为0.810(95%CI 0.729~0.891)。综合模型在测试集中的曲线下面积为0.829(95%CI 0.751~0.907),诺莫图的校正曲线有较好的一致性,综合模型的决策曲线也有良好的净收益。结论 结合影像组学特征和临床特征绘制的诺莫图具有早期预测乳腺癌患者HER-2状态的能力。

Objective Early prediction of the HER-2 status of breast cancer patients using radiomic features based on DBT images. Methods A total of 160 female breast cancer patients meeting the inclusion criteria were collected, of which 50 were positive for HER-2 and 110 were negative for HER-2.Manually delineate the region of interest on the DBT image to extract the image radiomics features.After data normalization and dimensionality reduction, a total of 1-10 features were selected to establish a logistic regression model.Ten fold cross validation was carried out.The model with the highest AUC value in cross validation set was selected as the best radiomic model.The clinical characteristics were analyzed to obtain independent predictors andbuild a clinical prediction model, and the predictive values of the best radiomic model were combined with the clinical independent predictors to build a comprehensive model and plot a Nomogram.The fitting degree of Nomogram was evaluated by Hosmer-Lemeshow test, and the decision curve was drawn to evaluate the Nomogram. Results The LR model based on four radiomic features was the best radiomic model, and the AUC value of this model in the test set was 0.810(95%CI 0.729-0.891).The AUC of the comprehensive model in the test set was 0.829(95% CI 0.751-0.907),The calibration curve of the Nomogram has good consistency, and the decision curve of the comprehensive model also has a good net income. Conclusion TheNomogram drawn by combining the radiomic and clinical features has a better ability to predict the HER-2 status early in breast cancer patients.