基于自动乳腺全容积成像冠状面超声图像的影像组学模型预测乳腺肿瘤良恶性的价值

Value of radiomics models based on coronal plane of automated breast volume scanner in predicting benign and malignant breast tumors

目的分析基于自动乳腺全容积成像(ABVS)冠状面超声图像的影像组学模型预测乳腺肿瘤良恶性的临床价值。方法选取我院经病理证实的乳腺肿瘤患者152例,基于ABVS冠状面超声图像的感兴趣区(ROI)提取影像组学特征,进行筛选后构建随机森林、支持向量机、多层感知器、决策树、逻辑回归、K-近邻6种影像组学模型;绘制受试者工作特征(ROC)曲线分析并比较各影像组学模型与不同年资超声医师(医师A具有2年工作经验、医师B具有5年工作经验)对乳腺肿瘤良恶性的诊断效能。结果最终纳入28个最优特征构建影像组学模型,其中随机森林模型对乳腺肿瘤良恶性的诊断效能最高,其曲线下面积为0.87,高于超声医师A和超声医师B的曲线下面积(0.76和0.83)。结论应用基于ABVS冠状面超声图像的影像组学模型预测乳腺肿瘤良恶性具有良好的价值。

Objective To investigate the clinical value of radiomics models based on the coronal plane of automated breast volume scanner(ABVS)in predicting benign and malignant breast tumors.Methods A total of 152 patients with pathologically confirmed breast tumors in our hospital were collected.The radiomics features were extracted and screening from region of interest(ROI)of ABVS coronal ultrasound images,and six prediction models including random forest,support vector machine,multilayer perceptron,decision tree,Logistic regression and K-nearest Neighbor were constructed.Receiver operating characteristic(ROC)curve was drawn to analyze and compare the diagnostic efficacy of radiomics models and different seniority sonographers(doctor A with 2 years of working experience,and doctor B with 5 years of working experience)for benign and malignant breast tumors.Results Totally 28 features were retained for the model building.Random forest has the highest diagnostic efficacy for benign and malignant breast tumors,with an AUC of 0.87,which was higher than that of sonographers A and sonographers B(0.76,0.83).Conclusion Radiomics models based on coronal plane of ABVS shows a good value in predicting benign and malignant breast tumors.