基于超声影像组学在预测BI-RADS 4类肿瘤良恶性中的价值  

Value of Ultrasound-Based Radiomics in Differentiating Benign and Malignant BI-RADS Category 4 Tumors

作　　者：李春萍 杨瑞敏[2] 王义成[2] 王聪 张月[4] 崔书君[5] 张力维[2] Li Chunping;Yang Ruimin;Wang Yicheng;Wang Cong;Zhang Yue;Cui Shujun;Zhang Liwei(Hebei North University,Zhangjiakou,Hebei 075000,China;Department of Ultrasound,The Frist Affiliated Hospital of Hebei North University,Zhangjiakou,Hebei 075000,China;Department of Radiotherapy,The Frist Affiliated Hospital of Hebei North University,Zhangjiakou,Hebei 075000,China;Department of Breast Surgery,The Frist Affiliated Hospital of Hebei North University,Zhangjiakou,Hebei 075000,China;Department of Medical Imaging,The Frist Affiliated Hospital of Hebei North University,Zhangjiakou,Hebei 075000,China)

机构地区：[1]河北北方学院,河北省张家口市075000 [2]河北北方学院附属第一医院超声医学科,河北省张家口市075000 [3]河北北方学院附属第一医院放射治疗科,河北省张家口市075000 [4]河北北方学院附属第一医院乳腺外科,河北省张家口市075000 [5]河北北方学院附属第一医院医学影像部,河北省张家口市075000

出　　处：《中国超声医学杂志》2024年第3期274-277,共4页Chinese Journal of Ultrasound in Medicine

基　　金：河北省医学科学研究课题计划项目(No.20220595)。

摘　　要：目的 基于二维超声瘤内及含瘤周5 mm区域构建影像组学模型,判断其对乳腺影像报告与数据系统(BI-RADS)4类乳腺肿瘤良恶性的预测价值。方法 回顾性收集176例超声诊断为BI-RADS 4类且病理结果明确的女性患者的乳腺肿瘤超声图像,按照7∶3的比例随机分为训练集(123例)、测试集(53例)。在软件上勾画肿瘤区域(瘤内组),并自动适形外扩5 mm(含瘤周组),分别提取并筛选出最佳影像组学特征后建立瘤内组模型及含瘤周组模型。利用受试者工作特征(ROC)曲线下面积(AUC)、校准曲线和决策曲线来评价模型。结果 训练集中瘤内组及含瘤周组的AUC分别为82.3%、90.1%,测试集中瘤内组及含瘤周组的AUC分别为78.6%、87.1%,Delong检验P<0.05;灵敏度、特异度、准确度在训练集瘤内组分别为85.7%、50.0%、72.4%,含瘤周组分别为88.3%、71.7%、82.1%,在测试集分别为83.7%、46.9%、69.1%和83.7%、69.3%、78.1%。结论 基于超声的含瘤周区域影像组学模型能更好地对乳腺BI-RADS 4类肿瘤的良恶性进行预测。Objective The value of constructing radiomics model based on two-dimensional ultrasound in the intratumor and peritumor of 5 mm region to differentiate between benign and malignant tumors in the breast imaging reporting and data system(BI-RADS) category 4 breast tumors.Methods Ultrasound images of breast tumors from 176 female patients diagnosed with BI-RADS category 4 by ultrasonography and with definite pathology were retrospectively collected and assigned into a training set of 123 cases and a testing set of 53 cases according to a ratio of 7∶3 randomly.After outlining the tumor area(intratumor group) on the software and automatically adapting the tumor shape to expand 5 mm outside the tumor(peritumor-containing group),the intratumor group model and peritumor-containing group model were established after extracting and screening the best radiomics features respectively.The area under the curve(AUC) of the subject work characteristics(ROC),calibration curves,and decision curves were utilized to evaluate the model.Results The AUC for the intratumor and peritumor-containing groups in the training set were 82.3% and 90.1% respectively,and the AUC for the intratumor and peritumor-containing groups in the testing set were 78.6% and 87.1% respectively,with Delong test both P<0.05.The sensitivity,specificity,and accuracy were 85.7%,50.0%,and 72.4% for the training set intratumor group,88.3%,71.7%,and 82.1% for the peritumor-containing group,whereas they were 83.7%,46.9%,69.1% and 83.7%,69.3%,78.1% for the testing set,respectively.Conclusions Ultrasound-based radiomics model of peritumor-containing region provides better predictive value of benign and malignant BI-RADS category 4 tumors.

关 键 词：超声检查 影像组学 乳腺肿瘤 瘤周区域 

分 类 号：R445.1[医药卫生—影像医学与核医学] R737.9[医药卫生—诊断学]