S-Detect技术联合临床特征信息基于BI-RADS分类对乳腺肿块良恶性的诊断价值  

The diagnostic value of S-Detect technology combined with clinical characteristics based on BI-RADS classification for benign and malignant breast masses

作　　者：李文肖 张玉瑞 刘文[1] 闫晓君[1] 石丽楠 曹春莉 王思睿 李军[1] 成静[1] LI Wenxiao;ZHANG Yurui;LIU Wen;YAN Xiaojun;SHI Linan;CAO Chunli;WANG Sirui;LI Jun;CHENG Jing(Department of Ultrasound,the First Affiliated Hospital of Shihezi University,Xinjiang Shihezi 832008,China)

机构地区：[1]石河子大学第一附属医院超声医学科,新疆石河子832008

出　　处：《现代肿瘤医学》2025年第4期607-613,共7页Journal of Modern Oncology

基　　金：国家自然科学基金(编号:82060318,82460353,81860498,81560433);天山英才科技创新团队:中亚地区高发疾病防治应用研究创新团队(编号:2023TSYCTD0020);兵团科技计划项目(编号:2022CB002-04);石河子大学自然科学基金(编号:ZZZC2023035,ZZZC2023040,ZZZC201955A);石河子大学第一附属医院青年基金项目(编号:QN202126,QN202107)。

摘　　要：目的:评估S-Detect技术在乳腺肿块四个不同切面的诊断效能,以及联合乳腺影像学报告及数据系统(BI-RADS)分类、临床特征信息建立乳腺肿块的诊断模型并评估其效能。方法:前瞻性收集120个肿块的常规超声、临床特征信息、S-Detect四个切面检查结果,根据术后病理结果分为良性组与恶性组。构建肿块四个不同切面的ROC曲线,并比较其诊断效能。通过肿块四个切面的S-Detect识别的结果,重新定义BI-RADS分类。通过单因素及多因素Logistic回归分析,筛选出乳腺癌的危险因素,建立诊断模型并进行一致性检验。结果:120个肿块中,病理结果证实为恶性肿块45个,良性肿块75个。乳腺肿块水平横切面、垂直纵切面、最大长轴切面、最长轴横切面S-Detect诊断乳腺癌的灵敏度分别为76.923%、63.385%、80.769%、80.769%,特异度分别为89.655%、91.379%、96.552%、89.655%,阳性预测值分别为76.923%、77.273%、91.304%、77.778%,阴性预测值分别为89.655%、88.484%、91.803%、91.228%,准确率分别为85.714%、83.333%、91.667%、86.904%,ROC曲线下面积(AUC)分别为0.824、0.792、0.887、0.852。单因素分析结果显示,是否触及肿物、临床症状、绝经与否、左右径、前后径、原始BI-RADS、调整BI-RADS比较两组间差异具有统计学意义,多因素结果显示:左右径大于1.75 cm、已绝经、原始BI-RADS分类4a类及以上、调整BI-RADS分类4a类及以上是乳腺癌的独立危险因素(P<0.05),其AUC分别为0.867、0.676、0.779、0.891,联合以上特征建立诊断模型,诊断模型的AUC为0.978,大于所有单一特征参数,一致性检验结果表明该诊断模型具有较好的诊断效能(Kappa值=0.719)。结论:在乳腺肿块四个不同切面中,最长轴横切面S-Detect诊断乳腺癌具有较好的诊断效能。人工智能S-Detect技术联合临床特征信息对乳腺肿块良恶性鉴别诊断具有重要意义。Objective:To evaluate the diagnostic efficacy of S-Detect technology in four different sections of breast mass,and to establish a diagnostic model of breast mass combined with the classification of the Breast Imaging Reporting and Data System(BI-RADS)and clinical characteristics.Methods:The results of routine ultrasound,clinical features and S-Detect were prospectively collected in 120 lumps.They were divided into benign group and malignant group according to postoperative pathological results.ROC curves of four different sections of the tumor were constructed and their diagnostic efficiency was compared.The BI-RADS classification was redefined by the results of S-Detect identification of the four sections of the mass.Through univariate and multivariate Logistic regression analysis,the risk factors of breast cancer were screened.The diagnostic model was established and the consistency test was carried out.Results:Among the 120 lumps,45 lumps were malignant and 75 lumps were benign.The sensitivity of S-Detect in the diagnosis of breast cancer in horizontal transverse section,vertical longitudinal section,maximum long axis section and longest axis transverse section was 76.923%,63.385%,80.769%and 80.769%,respectively.The specificity was 89.655%,91.379%,96.552%,89.655%.The positive predictive value was 76.923%,77.273%,91.304%,77.778%,and the negative predictive value was 89.655%,88.484%,91.803%,91.228%,respectively.The accuracy rates were 85.714%,83.333%,91.667%and 86.904%,respectively.The area under ROC curve(AUC)was 0.824,0.792,0.887 and 0.852,respectively.The results of univariate analysis showed that the difference between the two groups was statistically significant in whether the tumor was touched,clinical symptoms,menopause,left and right diameter,anterior-to-posterior diameter,original BI-RADS,and adjusted BI-RADS.The results of multivariate analysis showed that the left and right diameter greater than 1.75 cm,menopause,original BI-RADS classification 4a and above,and adjusted BI-RADS classification 4a and a

关 键 词：超声 乳腺肿块 S-Detect技术 临床特征 诊断效能 

分 类 号：R737.9[医药卫生—肿瘤]