影像组学模型对肺腺癌谱病变病理侵袭性的诊断价值
Diagnostic value of radiomics models for histological invasiveness of adenocarcinoma spectrum lesions of lung
孙希子 1周舒畅 1夏黎明1
作者信息
- 1. 430030 武汉,华中科技大学同济医学院附属同济医院放射科
- 折叠
摘要
目的:探讨针对肺腺癌谱病变的影像组学模型对其病理侵袭性的诊断效能.方法:回顾性分析我院经手术病理证实的 172 例肺腺癌谱病变(5～30 mm)患者的术前 CT 影像资料.采用 Pyra-diomics包提取术前CT图像病灶的影像组学特征,通过组间相关系数和带 L2 惩罚项的逻辑回归进行特征筛选,根据所选特征建立逻辑回归、随机森林、极致梯度提升分类器模型并绘制 ROC 曲线.由两位高年资放射科医生在不知道病理结果的情况下对结节侵袭性概率进行评分.采用 Delong 检验将三种分类器的诊断效能与年龄、ROI体积、高年资医生的诊断效能进行比较.结果:从每一个感兴趣区提取、筛选后得到 420 个影像组学特征.逻辑回归、随机森林和极致梯度提升分类器模型在测试集上的ROC曲线下面积分别为 0.921、0.956 和 0.958.年龄、ROI 体积和高年资放射科医生在测试集的 ROC曲线下面积分别为 0.620、0.863 和 0.896.Delong检验提示三种分类器间的诊断效能差异无统计学意义(P＞0.05),三种分类器与 ROI 体积、高年资医生的诊断效能差异亦无统计学意义(P＞0.05).结论:影像组学分类器模型进行肺腺癌谱病变的术前病理诊断具有较高的准确性,其诊断效能与 ROI体积以及高年资医生相当.
Abstract
Objective:To investigate the diagnostic performance of the radiomics model for path-ological invasiveness of adenocarcinoma spectrum lesions of the lung.Methods:Preoperative CT ima-ging data of 172 patients with pathologically proven adenocarcinoma spectrum lesions of the lung in our hospital were retrospectively analyzed.Radiomics features were extracted from the preoperative CT images via Pyradiomics package and selected via intra-class correlation and logistic regression with L2 penalty.Logistic regression,random forest and extreme gradient boosting(XGboost)classifier model were established according to the selected features,and receiver operating characteristic(ROC)curves were plotted.Two senior radiologists scored the probability of nodule invasiveness without knowing the pathological findings.Delong test was used to compare the diagnostic performance of the three classifiers with that of age,volume of region of interest(ROI)and the senior radiologists.Results:420 radiomics features were obtained from each ROI after feature extraction and selection.On test set,the area under the ROC curve of logistic regression,random forest and XGboost was 0.921,0.956 and 0.958,respectively,and that of age,ROI volume and the senior radiologists was 0.620,0.863 and 0.896,respectively.Delong test indicated that there was no significant difference among the three ra-diomics classifiers and no significant difference in diagnostic performance between the three classifiers and ROI volume or the senior radiologists either(P＞0.05).Conclusion:The radiomics classifier mod-els have high accuracy in preoperative pathological diagnosis of adenocarcinoma spectrum lesions of the lung,the diagnostic performance of which is comparable to that of ROI volume and senior radiolo-gists.
关键词
肺腺癌/肺结节/影像组学/体层摄影术,X线计算机/Delong检验/人机竞赛Key words
Lung adenocarcinoma/Lung nodule/Radiomics/Tomography,X-ray computed/Delong test/Man-machine competition引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据