瘤周超声影像组学对乳腺结节良恶性的鉴别诊断价值

The value of ultrasound-based peri-tumoral radiomics in differential diagnosis of benign and malignant breast nodules

目的:探索瘤周超声影像组学对乳腺结节良恶性的鉴别诊断价值。方法:回顾并收集于上海交通大学医学院附属第六人民医院进行常规超声检查且有明确病理学诊断结果的300例乳腺结节患者。选取二维超声图像上病灶最大层面勾画感兴趣区,同时自动适形向外扩展2 mm,提取基于二维超声的瘤内及瘤周影像组学特征。将纳入患者按7∶3随机分为训练组(210例)和验证组(90例),而后采取最小绝对收缩与选择算法(least absolute shrinkage and selection operator,LASSO)对其进行特征筛选,得到最优特征组合。影像组学特征经降维后,保留纳入模型的最优特征,利用支持向量机(support vector machine,SVM)模型进行乳腺结节良恶性分类,分别建立瘤内、瘤周、临床变量、瘤内联合瘤周、瘤内瘤周联合临床变量模型,通过受试者工作特征(receiver operating characteristic,ROC)曲线评估模型对超声乳腺结节良恶性的诊断效能。结果:在纳入研究的300例乳腺结节患者中,术后病理学检查结果为良性199例,恶性101例。瘤内超声影像组学模型在训练组曲线下面积(area under curve,AUC)为0.927(95%CI 0.889~0.965),验证组的AUC为0.808(95%CI 0.710~0.905),验证组的准确度、灵敏度、特异度、F1值、精确度分别为0.753、0.731、0.763、0.644、0.576。瘤周超声影像组学模型在训练组AUC为0.930(95%CI 0.891~0.969),验证组的AUC为0.857(95%CI 0.763~0.949),验证组的准确度、灵敏度、特异度、F1值、精确度分别为0.812、0.846、0.797、0.733、0.647。瘤内联合瘤周超声影像组学特征在训练组SVM模型的AUC为0.941(95%CI 0.843~0.967),验证组AUC为0.865(95%CI 0.781~0.949),验证组的准确度、灵敏度、特异度、F1值、精确度分别为0.824、0.692、0.881、0.706、0.720。瘤内瘤周超声影像组学特征结合临床变量的模型在训练集AUC为0.952(95%CI 0.924~0.979),验证组AUC为0.873(95%CI 0.788~0.958),验证组的准确度、灵敏度、特异度、F1值、精确度分别为0.859、0.692、0.932、0.750、0.818。瘤内瘤周联合临床变量模型的诊断效能均优于临床变量组、瘤内影像组学模型,差异有统计学意义(P<0.05);高于瘤周、瘤内结合瘤周模型,但差异无统计学意义(P>0.05)。结论:瘤内、瘤周超声影像组学对乳腺结节的良恶性均有较高的诊断价值,瘤内瘤周超声影像组学特征联合临床变量特征可以降低乳腺癌的漏诊率,避免不必要的穿刺活检。

Objective:To investigate the value of ultrasound-based peri-tumoral radiomics in discriminating benign and malignant breast nodules.Methods:A total of 300 cases of breast masses patients who were screened by regular ultrasound examination in The Sixth People’s Hospital,Shanghai Jiao Tong University School of Medicine were retrospectively collected.For the lesion on the regular ultrasound image,the largest dimension was selected to outline the region of interest.Subsequently,this area was automatically expanded by 2 mm in all directions,conformally and outwardly,to extract intra-and peritumor radiomics features.The included cases were randomly divided into a training group(210 cases)and a validation group(90 cases)in a ratio of 7∶3.Apply the least absolute shrinkage and selection operator(LASSO)to perform feature selection and obtain the optimal feature combination.The optimal features of the included models were retained by dimensionality reduction of the imaging omics features.The support vector machine(SVM)model was used to classify benign and malignant breast nodules,establish the intra-tumoral,peritumoral,clinical variables,intra-tumoral+peritumoral,intra-tumoral+peritumoral+clinical variables respectively,and evaluate the diagnostic efficacy of ultrasonic breast nodules by the receiver operating characteristics(ROC)curve.Results:Among 300 breast nodules,101 were malignant nodules and 199 were benign nodules.The ultrasound-based intra-tumoral radiomics model had an area under curve(AUC)value of 0.927(95%CI 0.889-0.965)in the training group and 0.808(95%CI 0.710-0.905)in the validation group.The accuracy,sensitivity,specificity,F1 value,and precision were 0.753,0.731,0.763,0.644,and 0.576 in the validation group,respectively in the ultrasound-based intra-tumoral radiomics model.The ultrasound-based peri-tumoral radiomics model had an AUC value of 0.930(95%CI 0.891-0.969)in the training group and 0.857(95%CI 0.763-0.949)in the validation group,and the accuracy,sensitivity,specificity,F1 value,and precision of this model were 0.812、0.846、0.797、0.733、0.647 for the validation group,respectively in the ultrasound-based peri-tumoral radiomics model.The intratumorally combined with peritumoral ultrasound imaging histological features had an AUC value of 0.941(95%CI 0.843-0.967)in the training group and 0.865(95%CI 0.781-0.949)in the validation group,the accuracy,sensitivity,specificity,F1 value,and precision of the model were 0.824,0.692,0.881,0.706,0.720 in the validation group,respectively.The model with intra-perineural radiomics features combined with clinical variables had an AUC value of 0.952(95%CI 0.924-0.979)in the training set and an AUC value of 0.873(95%CI:0.788 to 0.958)in the validation group,and the accuracy,sensitivity,specificity,F1 value,and precision of the validation group were 0.859,0.692,0.932,0.750,and 0.818,respectively.The diagnostic efficacy of the intra-peri-tumoral combined with clinical variables model was better than that of the clinical variables group and intratumoral imaging histology,with statistically significant differences(P<0.05);it was higher than that of the peri-tumoral and intratumoral combined with peri-tumoral models,but the differences were not statistically significant(P>0.05).Conclusion:Both intra-tumoral and peri-tumoral ultrasound radiomics have high value in the diagnosis of benign and malignant breast nodules.The application of intra-tumoral and peri-tumoral radiomics can reduce the missed rate of breast cancer and unnecessary biopsies.