超声影像组学模型鉴别肝郁痰凝型与冲任失调型乳腺癌中医证型的价值  

Value of ultrasound-based radiomics model in differentiating liver stagnation and phlegm coagulation type and thoroughfare vessel and conception vessel disharmony type of breast cancer

作　　者：许荣[1] 刘磊磊[1] 林晴[1] 张陆婷 苏林丘[1] 刘琛[1] 欧阳秋芳[1] XU Rong;LIU Leilei;LIN Qing;ZHANG Luting;SU Linqiu;LIU Chen;OUYANG Qiufang(The Second Affiliated Hospital of Fujian University of Traditional Chinese Medicine,Fuzhou 350003,Fujian,China)

机构地区：[1]福建中医药大学附属第二人民医院,福建福州350003

出　　处：《现代中西医结合杂志》2024年第18期2496-2500,2539,共6页Modern Journal of Integrated Traditional Chinese and Western Medicine

基　　金：福建省自然科学基金面上项目(2023J01815)。

摘　　要：目的探讨超声影像组学模型鉴别肝郁痰凝型与冲任失调型乳腺癌中医证型的价值,寻求有助于中医辨证分型的客观指标。方法回顾性收集2018年1月—2021年12月于福建中医药大学附属第二人民医院行术前超声检查并经病理确诊且有完整中医辨证资料的231例(247个病灶)乳腺癌患者资料,获取术前超声图像,用ITK-SNAP软件勾画感兴趣区(ROI),利用Pyradiomics 3.0软件提取超声影像组学特征,病灶按7∶3比例分为训练集(175个)和测试集(72个),使用最小绝对收缩与选择算法(LASSO)进行特征降维筛选,采用支持向量机(SVM)构建影像组学模型,通过受试者工作特征(ROC)曲线评估模型的鉴别诊断效能。结果纳入病例中肝郁痰凝型107例115个病灶,冲任失调型124例132个病灶。LASSO算法筛选出17个可鉴别肝郁痰凝型和冲任失调型乳腺癌的超声影像组学特征,其中前7个特征权重系数较大,分别为Dependence Variance,Correlation,Sphericity,Center Of MassIndex2,Bounding Box5,Large Dependence High Gray Level Emphasis和Short Run Emphasis。利用上述17个有效特征构建的影像组学模型在训练集和测试集上的曲线下面积(AUC)分别为0.797(95%CI:0.730~0.864)和0.775(95%CI:0.666~0.883)。该模型训练集的敏感度、特异度、准确率分别为72.3%(60/83)、73.0%(65/89)、72.7%(125/172);该模型测试集的敏感度、特异度、准确率分别为75.5%(37/49)、65.4%(17/26)、72.0%(54/75)。结论超声影像组学模型能够有效鉴别肝郁痰凝型与冲任失调型乳腺癌,筛选出的影像组学特征可作为乳腺癌中医辨证分型的微观指标。Objective It is to explore the application value of ultrasound-based radiomics model in differentiating liver stagnation and phlegm coagulation type and thoroughfare vessel and conception vessel disharmony type of breast cancer,and to seek objective indicators that is helpful for the differentiation and typing of Chinese medicine.Methods A total of 247 lesions were retrospectively collected from 231 pathologically confirmed breast cancer patients with complete Chinese medicine diagnostic data of syndrome of liver stagnation and phlegm coagulation or thoroughfare vessel and conception vessel disharmony,and their preoperative ultrasound images were obtained.The region of interest was outlined with ITK-SNAP software,and the ultrasound radiomics features were extracted with pyradiomics software.The lesions were divided into a training set(n=175)and a test set(n=72)in a ratio of 7∶3.The least absolute shrinkage and selection operator(LASSO)was used to perform feature dimensionality reduction screening,and support vector machine(SVM)was used to construct the radiomics model,and the differential diagnostic efficacy of the model was evaluated by the ROC curve.Results The enrolled cases included 107 cases of syndrome of liver stagnation and phlegm coagulation with 115 lesions and 124 cases of syndrome of thoroughfare vessel and conception vessel disharmony with 132 lesions.The LASSO algorithm selected 17 ultrasound-based radiomics features related to identifying syndrome of liver stagnation and phlegm coagulation and syndrome of thoroughfare vessel and conception vessel disharmony of breast cancer,among which the top 7 features had larger weight coefficients,including DependenceVariance,Correlation,Sphericity,Center Of MassIndex2,Bounding Box5,Large Dependence High Gray Level Emphasis and Short Run Emphasis,respectively.The area under the curve of the radiomics model constructed by the aforementioned 17 effective features was 0.797(95%CI:0.730-0.864)and 0.775(95%CI:0.666-0.883)on the training and test sets,respectively.

关 键 词：超声 影像组学 乳腺癌 肝郁痰凝 冲任失调 中医证型 

分 类 号：R737.9[医药卫生—肿瘤]