基于平扫CT影像组学Rad-score鉴别肾上腺乏脂性腺瘤和转移瘤的价值

VALUE OF RAD-SCORE BASED ON UNHANCED CT RADIOMICS FOR DIFERENTIATION BETWEEN ADRENAL LIPID-POOR ADENOMAS AND METASTASES

目的探讨基于CT平扫图像建立的影像组学Rad-score对鉴别肾上腺乏脂性腺瘤和转移瘤的诊断价值。方法回顾性搜集2015年1月—2021年12月于唐山市人民医院行胸部或腹部CT平扫的150例肾上腺乏脂性腺瘤和130例肾上腺转移瘤患者,并随机分成训练集(196例)和测试集(84例)。利用3D-slicer软件从CT平扫图像提取影像组学特征,采用最小绝对收缩与选择算子算法(LASSO)和皮尔森相关系数检验进行特征降维、筛选得到有鉴别诊断价值的影像组学特征。利用多因素Logistic回归分析得到风险因素构建Rad-score,利用受试者操作特征曲线(ROC)及曲线下面积(AUC)评价其在训练集和测试集中诊断效能。结果共提取1316个影像组学特征,经LASSO降维并筛选出17个组学特征,后去掉强相关特征,得到10个影像组学参数。利用多因素Logistic回归分析得到4个风险因素并构建了Rad-score。在训练集中,Rad-score鉴别诊断乏脂性腺瘤和转移瘤的最佳临界值为0.552,AUC为0.934,敏感度、特异度和准确率分别为0.817、0.918和0.868;在测试集中,Rad-score的最佳临界值为0.373,AUC值为0.927,敏感度、特异度和准确率分别为0.842、0.905和0.875。结论基于CT平扫图像的影像组学Rad-score在训练集和测试集中对鉴别肾上腺乏脂性腺瘤和肾上腺转移瘤有较高的诊断效能,具有较高的临床价值。

Objective To investigate the value of Rad-score constructed by unhanced CT-based radiomics for differentiating adrenal lipid-poor adenomas from metastases.MethodssFrom January2015to December 2021,totally one hundred and fifty cases of adrenal lipid-poor adenomas and one hundred and thirty cases of adrenal metastases underwent chest or abdominal CT plain scan in Tangshan People's Hospital were collected retrospectively.And the patients were randomly divided into a training set(n=196)and testing set(n=84).Radiomics features were extracted from unhanced CT images by using 3D-slicer software,and radiomics features with differential diagnosis value were selected by the least absolute shrinkage and selection operator regression and pearson correlation coefficient test The Rad-score was developed by using the risk factors obtained from multivariate logistic regression analysis,and its diagnostic effectiveness in the training and testing sets were evaluated by the receiver operating characteristic curve(ROC)and the area under receiver operating characteristic curve(AUC).Results A total of 1316 radiomics features were extracted and 17 features were selected by LASSO dimensionality reduction,and then 10 radiomics parameters were obtained after removing strong correlation features.Finally,Rad-score was constructed based on four risk factors obtained by multiva riate logistic regression analysis.In the training set,the best cutoff value of Rad-score for differentiating adrenal lipid-poor adenomas from metastases was 0.552,and the AUC,sensitivity,specificity and accuracy were 0.934,0.817,0.918 and 0.868,respectively;In the testing set,the best cutoff value of Rad-score was 0.373,and the AUC,sensitivity,specificity and accuracy were 0.927,0.842,0.905 and 0.875,respectively.Conclusion Rad-score based on unhanced CT radiomics shows high diagnostic efficiency and clinical value in differentiating adrenal lipid-poor adenomas from metastases both in training and testing sets.