瘤内与瘤周超声影像组学模型对乳腺癌淋巴血管浸润状态的术前预测效能

Intratumoral and peritumoral ultrasound radiomics models in predicting lymphovascular invasion of breast cancer

目的探讨基于瘤内与瘤周的超声影像组学模型对乳腺癌淋巴血管浸润(LVI)状态的术前预测效能。方法回顾性选择2018年4月至2024年3月在丽水市中心医院乳腺外科收治的女性乳腺癌患者553例,按7∶3随机分为训练集387例和验证集166例。训练集和验证集患者均按照LVI状态分为LVI阴性组和LVI阳性组。收集患者的一般资料和病理资料。采集超声图像,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5 mm,分别提取瘤内、瘤周和瘤内+瘤周区域的超声影像组学特征。采用Mann-Whitney U检验和最小绝对收缩与选择算子回归进行超声影像组学特征筛选。利用支持向量机算法分别构建瘤内、瘤周和瘤内+瘤周超声影像组学模型。采用单因素及多因素logistic回归筛选临床危险因素并构建临床模型,结合超声影像组学特征和临床因素构建列线图模型。通过ROC曲线分析评估模型对乳腺癌LVI状态的术前预测效能,并计算AUC、灵敏度、特异度和准确度。结果在临床特征中,月经状态、后方回声和淋巴结状态是预测LVI状态的危险因素。在训练集和验证集中,瘤内+瘤周超声影像组学模型均表现出最高的预测效能,AUC分别为0.875、0.823。进一步结合临床危险因素构建的列线图模型,在训练集中的AUC、灵敏度、特异度、准确度分别为0.901、0.836、0.858、0.850,在验证集中的AUC、灵敏度、特异度、准确度分别为0.847、0.776、0.843、0.819。结论瘤内和瘤周超声影像组学特征对乳腺癌LVI状态有较高的术前预测效能,联合临床因素建立的列线图模型可进一步提升诊断效能。

Objective To explore the predictive value of intratumoral and peritumoral ultrasound radiomics models for lymphovascular invasion(LVI)of breast cancer.Methods A total of 553 patients with breast cancer confirmed by pathology from April 2018 to March 2024 in Lishui Central Hospital were enrolled,and all patients received routine ultrasound examination before surgery.Patients were randomly divided into a training set(387 cases)and a validation set(166 cases)at a ratio of 7∶3.The largest layer of the lesion on the two-dimensional ultrasound image was selected to outline the region of interest,which was automatically expanded outward by 5 mm;the radiomics features of the intratumoral,peritumoral,and intratumoral+peritumoral regions were extracted,respectively.The Mann-Whitney test and the least absolute shrinkage and selection algorithm were used to screen radiomics features.The support vector machine algorithm was used to construct intratumoral,peritumoral and intratumoral+peritumoral models,respectively.Univariate and multivariate logistic regression were used to screen clinical risk factors of LVI of breast cancer.Subsequently,a nomogram model was constructed by combining radiomics features and clinical factors.The predictive performance of the models for LVI status of breast cancer was evaluated through receiver operating characteristic(ROC)curve analysis,and the area under the curve(AUC),sensitivity,specificity,and accuracy were calculated.Results Among the clinical characteristics,menstrual status,posterior echo,and ultrasound-reported lymph node status were risk factors of LVI in breast cancer.In both the training set and validation set,the intratumoral+peritumoral radiomics model for predicting LVI of breast cancer showed the highest value,with AUCs of 0.875 and 0.823 respectively.The nomogram model of radiomics model combined with clinical risk factors had AUC,sensitivity,specificity,and accuracy of 0.901,0.836,0.858,and 0.850 in the training set,and 0.847,0.776,0.843,and 0.819 in the validation set,respectively.Conclusion The intratumoral and peritumoral ultrasound radiomics features have a high predictive performance for LVI of breast cancer.The nomogram model established in combination with clinical factors can further improve the diagnostic performance and help guide clinical decision-making.