摘要
目的探讨肺癌CT影像组学特征用于鉴别小细胞肺癌(small cell lung cancer, SCLC)和非小细胞肺癌(non-small cell lung cancer, NSCLC)的价值。方法回顾性分析经2010-01/2020-12年来自首都医科大学电力教学医院和首都医科大学大兴教学医院的89例肺癌患者的临床和影像资料的病理证实,其中SCLC 24例,NSCLC 65例。采用Python语言PyRadiomics软件包提取影像组学特征,使用最大相关-最小冗余(max-relevance and minredundancy, mRMR)算法和最小绝对收缩和选择算法(least absolute shrinkage and selection operator, LASSO)回归分析筛选影像组学特征。利用多变量Logistic回归构建预测模型,应用受试者工作特征曲线(receiver operating characteristic curve, ROC)评价预测模型的诊断效能,联合影像组学标签与临床特征构建列线图。结果从所提取的788种影像组学特征中,最终筛选了其中8种重要特征用于构建影像组学模型。训练组、测试组影像组学模型鉴别SCLC与NSCLC的ROC曲线下面积(area under thecurve, AUC)分别为0.93(95%CI:0.85~1.00)、0.92(95%CI:0.80~1.00)。联合诊断模型在测试组的诊断效能(AUC=0.93,95%CI:0.84~1.00)高于临床预测模型(AUC=0.73,95%CI:0.52~0.93)和影像组学模型(AUC=0.92,95%CI:0.80~1.00)。结论基于CT平扫的影像组学模型鉴别SCLC与NSCLC有较高的诊断效能,结合患者临床资料的联合模型较影像组学模型有更高的诊断效能。
Objective To investigate the association between radiomic features and the lung cancer histological subtypes, and establish a nomogram for the classification of small cell lung cancer(SCLC) and non-small cell lung cancer(NSCLC).Methods The clinical and imaging data of 89 patients with lung cancer confirmed by pathology from January 2010 to December 2020 were retrospectively analyzed in the capital medical university electric teaching hospital and capital medical university daxing teaching hospital. Among the patients, 24 cases were SCLC and 65 cases were NSCLC.The radiomic features were extracted by using the Python language PyRadiomics software package.The max-relevance and min-redundancy(mRMR) and the least absolute shrinkage and selection operator(LASSO) were used for radiomics signature building. Multivariable logistic regression analysis was used to develop the predicting model. The receiver operating characteristic curve(ROC) was used to verify the diagnostic performance of the predicting model. A nomogram based on the radiomics signature and clinical features was constructed.Results Eight of 788 radiomics features were screened as important features for establishing the radiomics model. The area under the curve(AUC) of radiomics model of the training group and the test group to distinguish SCLC and NSCLC were 0.93(95%CI:0.85-1.00) and 0.92(95%CI:0.80-1.00). In the test group, the combined model had better diagnostic performance(AUC=0.93,95% CI:0.84-1.00) than the clinical model(AUC=0.73,95% CI:0.52-0.93) and the radiomics model(AUC=0.92,95% CI:0.80-1.00).Conclusion The unenhanced CT radiomics model performed well in the classification of SCLC and NSCLC. The combined model based on the radiomic features and clinical data has better diagnostic performance for the classification of SCLC and NSCLC than simple application of the radiomics model.
作者
王晓瑞
WANG Xiaorui(Department of Radiology,Capital Medical University Electric Teaching Hospital,Beijing100073,China)
出处
《华南国防医学杂志》
CAS
2021年第3期188-192,共5页
Military Medical Journal of South China
关键词
肺肿瘤
小细胞肺癌
非小细胞肺癌
体层摄影
X线计算机
影像组学
Lung neoplasms
Small cell lung cancer
Non-small-cell lung cancer
Tomography
X-ray computed
Radiomics
