基于超声图像的肿瘤亚区域特征评估浸润性乳腺癌HER2表达异质性的价值  

Evaluation of HER2 Expression Heterogeneity in Invasive Breast Cancer Based on Ultrasound Image of Tumor Sub-Region Features

作　　者：吴林永 魏达友[1] 林艳[1] 李松桦 何敏诗 Wu Linyong;Wei Dayou;Lin Yan;Li Songhua;He Minshi(Department of Ultrasound Medicine,Maoming People's Hospital,Maoming,Guangdong 525000,China;Department of Pathology,Maoming People's Hospital,Maoming,Guangdong 525000,China)

机构地区：[1]茂名市人民医院超声医学科,广东省茂名市525000 [2]茂名市人民医院病理科,广东省茂名市525000

出　　处：《中国超声医学杂志》2024年第7期749-752,共4页Chinese Journal of Ultrasound in Medicine

基　　金：广东省颐养健康慈善基金会资助(No.2023CSM001)。

摘　　要：目的 探讨基于超声图像的肿瘤亚区域特征评估浸润性乳腺癌(IBC)人表皮生长因子受体2(HER2)瘤内表达异质性的价值。方法 回顾性分析90例HER2阳性和39例HER2阴性表达IBC患者的超声临床资料,并分为训练(n=90)和验证(n=39)队列。采用K均值共识别聚类方法将肿瘤整体区域划分为不同肿瘤亚区域,提取整体和亚区域特征。特征选择采用最小绝对收缩和选择算子方法,并提交给光梯度提升机(Light GBM)和梯度增强机器学习算法开发评估IBC-HER2阳性表达模型。使用受试者工作特性曲线下面积(AUC)评估模型效果。结果 基于肿瘤整体区域划分为A和B两个肿瘤亚区域。基于亚区域特征的机器学习模型均优于基于肿瘤整体区域特征,特别是B亚区域,Light GBM模型在训练和验证队列中表现最优,AUC分别为0.87和0.92,准确度分别为81%和85%。结论 本研究揭示了基于超声图像的肿瘤亚区域特征相对于肿瘤整体区域特征可更好地解析IBC-HER2表达的瘤内异质性,具有潜在实现非侵入性成像特征表征肿瘤分子异质性,达到精准诊疗目标的价值。Objective To explore the value of ultrasound image based on tumor sub-region features in evaluating the heterogeneity of human epidermal growth factor receptor 2(HER2) expression in invasive breast cancer(IBC).Methods Ultrasound images and clinical pathological data of 90 patients with HER2 positive expression and 39 patients with HER2 negative expression of IBC were retrospectively analyzed,which were divided into the training(n=90) cohort and the validation(n=39) cohort.The K-means co-identification clustering method was adopted to divide the tumor overall region into different tumor sub-regions,and radiomics features were extracted from the overall and sub-regions.The least absolute shrinkage and selection operator(LASSO) regression was used for feature selection,and the features were submitted to light gradient boosting machine(Light GBM) and gradient-boosting machine learning algorithms for development of IBC-HER2 positive expression models.The area under the curve(AUC) of the receiver operating characteristic was applied to evaluate the effectiveness of the models.Results Based on the overall tumor region,it was divided into two tumor sub-regions:A and B.Machine learning models based on sub-region features were superior to those based on overall tumor region features,especially in sub-region B.The Light GBM models performed the best in both the training and the validation cohorts,with AUC of 0.87 and 0.92,accuracy of 81% and 85%,respectively.Conclusions This study revealed that tumor sub region features could better analyze the intra-tumoral heterogeneity of IBC-HER2 expression compared to tumor overall region features based on ultrasound images,which had the potential to achieve the value of characterizing tumor molecular heterogeneity in non-invasive imaging features and achieving precise diagnosis and treatment goals.

关 键 词：浸润性乳腺癌 异质性 HER2 肿瘤亚区域 超声 

分 类 号：R737.9[医药卫生—肿瘤] R445.1[医药卫生—临床医学]