摘要
目的:探讨基于增强CT影像组学特征鉴别头颈部木村病淋巴结病变和淋巴瘤的可行性,并验证建立的逻辑回归诊断模型。方法:回顾性分析经手术病理或穿刺活检证实的14例头颈部木村病(38枚肿大淋巴结)和27例淋巴瘤患者(37枚肿大淋巴结)的相关资料,所有患者均行头颈部增强CT扫描。将病灶所有显示层面的CT静脉期图像导入ITK-SNAP软件(www.itksnap.org),手动勾画立体感兴趣区(VOI),使用artificial intelligence kit软件提取纹理特征。按照7:3的比例将数据随机分为训练组与验证组。采用方差分析+秩和检验、一般线性模型和Lasso算法进行特征降维,并用最终筛选出的纹理特征构建逻辑回归模型并进行5折交叉验证。用验证组数据对模型进行验证,评价指标采用ROC曲线下面积(AUC)、敏感性和特异性。结果:从75个病灶中共提取396个特征,通过降维最终筛选出5个可用于鉴别两种病变的组学特征。逻辑回归模型在训练组中鉴别效能的AUC为0.987,特异度为0.958,敏感度为0.966;验证组的AUC为0.938,特异度为0.786,敏感度为1。结论:影像组学鉴别头颈部木村病淋巴结病变和淋巴瘤具有可行性,基于CT影像组学特征建立的逻辑回归模型具有较高的诊断效能。
Objective:To investigate the feasibility to differentiate of lymph node involved by Kimura disease from lymphoma of the head and neck based on radiomic features from contrast enhanced CT image,and to establish the logistic regression diagnosis model.Methods:Fourteen patients with head and neck Kimura disease(a total of 38 enlarged lymph nodes)and 27 lymphoma patients(a total of 37 enlarged lymph nodes)confirmed bypathology were enrolled in this study and analyzed retrospectively.After contrast enhanced CT scan,all slices of CT venous images with enlarged lymph nodes were imported into ITK-SNAP(www.itksnap.org)software with manually delineating of the volume of interest(VOI).The AK software was used to extract radiomic texture features.The data were randomly divided into training and validation group according to the ratio of 7 to 3.The feature reduction was performed by ANOVA,Wilcoxon,General Linear model(GLM),and LASSO algorithm.And then,a logistic regression model was established with the final selected texture features,which was evaluated by performing a 5-folds cross-validation and verified by the validation group data.The evaluation index of the model included the area under the ROC curve(AUC),sensitivity,and specificity.Results:A total of 396 radiomics features were extracted from 75 lesions.Finally,five dimensionality features were identified as the factors of distinguishing enlarged lymph node involved by Kimura disease from lymphomas of the head and neck.The AUC,specificity,and sensitivity of the logistic regression model were 0.987,0.958,and 0.966 in the training group and 0.938,0.786,and 1 in validation group respectively.Conclusion:Radiomics is feasible for differentiation of lymph node involved by Kimura disease from lymphomas in the head and neck,and the logical regression model from CT-based radiomics features has a high diagnostic efficiency.
作者
张力
于淑靖
张迎
马建楠
付兰
姚丽
ZHANG Li;YU Shu-jing;ZHANG Ying(Department of CT Diagnosis,Cangzhou Central Hospital,Hebei 061001,China)
出处
《放射学实践》
北大核心
2020年第2期159-164,共6页
Radiologic Practice
基金
沧州市重点研发计划指导项目(183302009).
作者简介
张力(1981-),男,河北沧州人,硕士,副主任医师,主要从事医学影像诊断工作。
