自动乳腺全容积扫描影像组学联合临床和超声特征列线图鉴别良、恶性乳腺导管内病变

Nomogram based on automatic breast volume scanner radiomics combined with clinical and ultrasonic features for differentiating benign or malignant breast intraductal lesions

目的 观察自动乳腺全容积扫描(ABVS)影像组学联合临床及超声特征列线图鉴别良、恶性乳腺导管内病变的价值。方法 回顾性分析144例经病理证实乳腺导管内病变女性患者的临床及超声资料;按照2∶1比例将其随机分为训练集(n=96)及验证集(n=48)。基于ABVS图像提取并筛选最优影像组学特征,构建影像组学模型,计算影像组学评分(Radscore);将临床、超声特征及Radscore纳入单因素和多因素logistic回归分析,筛选鉴别良、恶性乳腺导管内病变的独立影响因素,构建临床-超声模型,并联合影像组学模型构建列线图模型;以受试者工作特征(ROC)曲线评估各模型鉴别良、恶性乳腺导管内病变的效能。结果 患者年龄[OR(95%CI)=1.104(1.045,1.180),P=0.001]、病变边缘[OR(95%CI)=0.273(0.075,0.917),P=0.039]、微小钙化灶[OR(95%CI)=9.759(2.240,60.730),P=0.006]及Radscore[OR(95%CI)=3.818(1.435,11.994),P=0.012]均为良、恶性乳腺导管内病变的独立影响因素。影像组学模型、临床-超声模型及列线图模型鉴别良、恶性乳腺导管内病变的曲线下面积(AUC)在训练集分别为0.766、0.866及0.901,在验证集分别为0.770、0.765及0.854。结论 ABVS影像组学联合临床及超声特征列线图鉴别良、恶性乳腺导管内病变效能良好。

Objective To observe the value of nomogram based on automated breast volume scanner(ABVS)radiomics combined with clinical and ultrasonic features for differentiating benign or malignant breast intraductal lesions.Methods Clinical and ultrasonic data of 144 female patients with pathologically confirmed breast intraductal lesions were retrospectively analyzed.The patients were randomly divided into training set(n=96)or validation set(n=48)at the ratio of 2∶1.The optimal radiomics features were extracted and screened based on ABVS images,then radiomics model was constructed,and Radscore was calculated.Univariate and multivariate logistic regression analysis of clinical,ultrasonic features and Radscores were performed to screen the independent impact factors of benign or malignant breast intraductal lesions,and clinic-ultrasound model was established.Nomogram model was constructed by combining the clinic-ultrasound model with radiomics.Receiver operating characteristic(ROC)curve was used to evaluate the efficacy of each model for differentiating benign or malignant breast intraductal lesions.Results Patients'age(OR[95%CI]=1.104[1.045,1.180],P=0.001),lesion's margin(OR[95%CI]=0.273[0.075,0.917],P=0.039),microcalcification(OR[95%CI]=9.759[2.240,60.730],P=0.006)and Radscore(OR[95%CI]=3.818[1.435,11.994],P=0.012)were all independent impact factors for benign or malignant intraductal lesions.The area under the curve(AUC)of radiomics model,clinic-ultrasound model and nomogram model for differentiating benign or malignant breast intraductal lesions was 0.766,0.866 and 0.901 in training set,while 0.770,0.765 and 0.854 in validation set,respectively.Conclusion Nomogram based on ABVS radiomics combined with clinical and ultrasonic features had good efficacy for differentiating benign or malignant breast intraductal lesions.