基于CT平扫影像组学特征在预测胸腺上皮性肿瘤WHO简化病理分型中的价值  

Value of Simplified Pathological Classification of Thymus Epithelial Tumors Based on Plain CT Imaging Features

作　　者：陈杰[1] 洪悦[1] 王艳[1] Chen Jie;Hong Yue;Wang Yan(Department of Radiology and Medical Imaging,People's Hospital of Xinjiang Uygur Autonomous Region,Urumqi 830001,Xinjiang Uygur Autonomous Region,China)

机构地区：[1]新疆维吾尔自治区人民医院放射影像中心,新疆乌鲁木齐830001

出　　处：《中国CT和MRI杂志》2024年第1期71-73,共3页Chinese Journal of CT and MRI

摘　　要：目的探讨基于CT平扫的影像组学特征在预测胸腺上皮性肿瘤WHO简化病理分型中的应用价值。方法回顾性收集2010年1月-2022年3月由病理结果证实的胸腺上皮肿瘤(TETS)患者共57例,通过TETS的WHO简化病理分型,分为低危组(A、AB、B1型)23例、高危组(B2、B3、C型)34例,并按照8:2的比例,随机分成训练集和测试集。每个病灶均由两名放射科医生经过协商后,利用ITK-SNAP软件对兴趣区(ROI)进行勾画。使用Python v3.67提取放射组学特征,并使用Spearman相关系数和LASSO特征选择方法进行降维和筛选。在训练集中,应用支持向量机(SVM)、多层感知机(MLP)和逻辑回归(LR)构建术前诊断预测模型。采用受试者工作特征曲线(ROC)评估预测效果,并通过内部测试集验证预测模型。结果共提取了1649个影像组学特征参数,经过Spearman相关系数筛选得到221个差异特征,并通过LASSO方法将其降维至12个组学特征。在测试集中,基于SVM、MLP和LR构建的术前预测模型分别表现出AUC值为0.800、0.868和0.971,其中LR模型具有更好的预测效果。结论基于CT平扫影像组学特征构建的SVM、MLP和LR模型在预测胸腺上皮性肿瘤WHO简化病理分型方面展现出较好的预测潜力,其中LR模型具有更好的预测效果。Objective To investigate the utility of CT plain scan based radiomics features in predicting who simplified pathologic classification of thymic epithelial neoplasms.Methods a total of 57 patients with thymic epithelial tumors(TETS)confirmed by pathological findings from January 2010 to March 2022 were retrospectively enrolled and divided into low-risk group(types A,AB,and B1)23 patients,high-risk group(types B2,B3,and C)34 patients by who simplified pathological classification of TETS,and randomly divided into training set and test set according to the ratio of 8:2.Each lesion was delineated by a region of interest(ROI)using itk-snap software after consultation by two radiologists.Radiomics features were extracted using Python v3.67 and further feature dimensionality reduction,filtering was performed by using Spearman's correlation coefficient,least absolute shrinkage and selection operator method(lasso).In the training set,three machine learning methods including support vector machine(SVM),multi-layer perceptual machine(MLP),and logistic regression(LR)were selected to construct the diagnostic prediction model.The receiver operating characteristic curve(ROC)was employed to evaluate its predictive efficacy,and an internal test set was applied to validate the above prediction model.Results a total of 1649 radiomics feature parameters were obtained,221 differential features were obtained by using Spearman correlation coefficients,and lasso was reduced to 12 omics features.The AUC values of the preoperative prediction model built by SVM,MLP and LR in the test set were 0.800,0.868,0.971,respectively,among which LR had better prediction.Conclusion in predicting the WHO simplified pathological classification of thymic epithelial tumors,the SVM,MLP,and LR models constructed based on CT plain imaging omics features have predictive potential,among which the LR model has better predictive effect.

关 键 词：胸腺上皮肿瘤 WHO病理分型 影像组学 支持向量机 多层感知机 逻辑回归 

分 类 号：R736.3[医药卫生—肿瘤]