颈部淋巴瘤与淋巴结结核CT平扫图像影像组学鉴别可行性研究

Feasibility study on the radiomics identification of cervical lymphoma and lymph node tuberculosis with CT plain scan images

目的基于CT平扫影像组学对于颈部淋巴结结核与淋巴瘤的鉴别研究鲜有报道。文章探讨基于CT平扫厚层图像影像组学的鉴别诊断价值。方法回顾性分析2017年3月至2021年12月40例颈部淋巴瘤患者和56例颈部淋巴结结核患者CT平扫厚层图像。采用ITK-SNAP软件分别勾画62枚淋巴瘤淋巴结和70枚结核淋巴结,通过python提取影像组学特征,并进行统计学检验,联合应用最小绝对收缩与选择算子算法(LASSO)和主成分分析法(PCA)筛选和降维组学特征,最后通过逻辑回归建立分类器。进行5折交叉验证,使用ROC曲线及灵敏度、特异度等评价分类器性能,并与高低年资医生分类准确性进行比较。结果从CT平扫厚层图像中共提取838个影像组学特征,LASSO筛选出28项组学特征,PCA降维后得到24项特征。Logistic regression建立的分类器在训练组的ROC曲线下面积(AUC)为0.965,敏感度为0.920,特异度为0.893,在测试组的ROC曲线AUC为0.874,敏感度为0.786,特异度为0.846。结论以CT平扫图像和logistic为基础建立的分类器在鉴别颈部淋巴结结核与淋巴瘤方面具有一定的准确度和临床应用价值。

Objective There are few reports on the identification of cervical lymph node tuberculosis and lymphoma based on plain CT radiomics.This study explores the value of radiomics based on plain CT thick-slice images for differential diagnosis.Methods The plain CT thick slice images of 40 patients with neck lymphoma and 56 patients with neck lymph node tuberculosis from March 2017 to December 2021 were analyzed retrospectively.ITK SNAP software was adopted to respectively generate the outline of 62 lymphoma lymph nodes and 70 tuberculosis lymph nodes.The radiomics features were extracted by Python and statistical testing were performed.The Least Absolute Shrinkage and Selection Operator(LASSO)and Principal Component Analysis(PCA)were jointly used to screen and reduce the dimensionality of the radiomics features.Finally,a classifier was established via logistic regression.A 5-fold cross-validation was performed,and the performance of the classifier was evaluated by ROC curve,sensitivity and specificity.It was also compared with the classification accuracy of doctors with high and low seniority.Results A total of 838 radiomics features were extracted from the CT plain scan thick layer images.28 radiomics features were screened by lasso,and 24 features were obtained after PCA dimensionality reduction.Of the training group,the area under the ROC curve(AUC)of the classifier established by logistic regression was 0.965,the sensitivity was 0.920 and the specificity was 0.893.Of the testing group,the AUC was 0.874,the sensitivity was 0.786 and the specificity was 0.846.Conclusion The classifier based on CT plain scan images and logistic has certain accuracy and clinical application value in distinguishing neck lymph node tuberculosis and lymphoma.