基于超声影像组学鉴别小儿肝移植后淋巴组织增生性疾病与淋巴结反应性增生  

Ultrasound-based Radiomics Analysis for Differentiating Post-transplant LymphoproliferativeDisorders from Reactive lymph Node Hyperplasia After Pediatric Liver Transplantation

作　　者：袁瑶 唐缨[1] 牛宁宁[1] 张国英[1] 王明阳 Yuan Yao;Tang Ying;Niu Ningning;Zhang Guoying;Wang Mingyang(Department of Ultrasound,Tianjin First Central Hospital,Tianjin 300192,China)

机构地区：[1]天津市第一中心医院超声科,天津市300192

出　　处：《中国超声医学杂志》2023年第5期540-544,共5页Chinese Journal of Ultrasound in Medicine

基　　金：国家自然科学基金(No.82172031);天津市科技计划项目(No.21JCYBJC01800)。

摘　　要：目的 基于超声影像组学特征,建立机器学习模型鉴别小儿肝移植后淋巴组织增生性疾病(PTLD)与淋巴结反应性增生。方法 回顾性分析小儿肝移植后经病理证实的112例PTLD及93例淋巴结反应性增生患者颈部增大淋巴结的二维超声图像。提取影像组学特征构建随机森林、支持向量机、决策树及逻辑回归模型。比较常规超声与4种模型的诊断效能。结果 每个淋巴结共提取118个影像组学特征,筛选7个最优特征建立4种机器学习模型。其中随机森林模型的诊断效能最好,优于常规超声[受试者操作特征(ROC)曲线下面积:0.816 vs 0.613,Z=5.991,P<0.05],模型的灵敏度、特异度及准确度分别为95.7%、68.6%及86.0%。结论 基于超声影像组学的随机森林模型对小儿肝移植后PTLD与淋巴结反应性增生有较好的鉴别诊断价值。Objective To establish machine learning models and distinguish post-transplant lymphoproliferative disorders(PTLD)from reactive lymph node hyperplasia after pediatric liver transplantation based on ultrasound radiomics.Methods The two-dimensional ultrasound imaging data of 112patients with PTLD and 93patients with reactive lymph node hyperplasia confirmed by pathology after pediatric liver transplantation were retrospectively analyzed.Radiomics features were extracted to construct random forest,support vector machine,decision tree and logistic regression models,and the diagnostic efficacy was compared between conventional ultrasound and the 4models.Results 118radiomics features were extracted from each lymph node,and 4machine learning models were established using the 7optimal features obtained after feature screening.The random forest model had the best diagnostic performance,which was better than conventional ultrasound[The area under the receiver operating characteristic(ROC)curve:0.816vs 0.613,Z=5.991,P<0.05].The sensitivity,specificity and accuracy of the random forest model were 95.7%,68.6%and 86.0%respectively.Conclusions The random forest model based on ultrasound radiomics has a good discrimination value for PTLD from reactive lymph node hyperplasia after pediatric liver transplantation.

关 键 词：移植后淋巴组织增生性疾病 淋巴结反应性增生 超声 影像组学 机器学习 

分 类 号：R445.1[医药卫生—影像医学与核医学] R726.5[医药卫生—诊断学]