CT-based multi-regional radiomics for predicting radiation pneumonitis in lung cancer patients
10.3969/j.issn.1005-202X.2025.08.005
- VernacularTitle:多区域CT影像组学预测肺癌放射性肺炎
- Author:
Binghua LIANG
1
;
Jianwei SUN
;
Honglin CHEN
;
Tao ZHANG
;
Heng ZHANG
;
Xinye NI
Author Information
1. 南京医科大学附属常州第二人民医院放疗科,江苏 常州 213000
- Publication Type:Journal Article
- Keywords:
radiotherapy;
radiomics;
lung cancer;
radiation pneumonitis;
target-to-lung ratio;
prediction model
- From:
Chinese Journal of Medical Physics
2025;42(8):1011-1017
- CountryChina
- Language:Chinese
-
Abstract:
Objective To establish a reliable prediction model for radiation pneumonitis(RP)based on multi-regional radiomics analysis of localizable CT images.Methods A retrospective analysis was conducted on 185 patients who received radiotherapy from January 2021 to June 2023 in the Department of Radiotherapy,Xuzhou Cancer Hospital.Patients were classified as having RP or not based on imaging combined with clinical diagnosis.Three regions of interest(ROI)were defined in the localizable CT images:Lung,Lung-PTV and PTV,and their radiomics features were extracted.After feature screening using methods such as Mann-Whitney Utest,recursive feature elimination,and Lasso,a prediction model was established using support vector machine classification algorithm.The model performance was validated using 6 evaluation metrics:the area under the receiver operating characteristic curve(AUC),accuracy,specificity,sensitivity,positive predictive value,and negative predictive value.Results The prediction model consisted of 7 radiomics features.The clinical model of target-to-lung ratio,PTV model,Lung model,and Lung-PTV model achieved AUC values of 0.535,0.801,0.672,and 0.706 in the test set,respectively.The AUC value and accuracy of PTV model reached 0.843 and 0.775 in the training set,while 0.801 and 0.750 in the test set.PTV model was superior to Lung model,Lung-PTV model,and clinical model in predictive performance.The AUC values of the combined PTV+(Lung-PTV)model in the training and test sets were 0.867 and 0.806,respectively,higher than those of PTV model and Lung-PTV model.Conclusion The predictive ability of the prediction models constructed from radiomics features in different ROI for symptomatic RP varies.The radiomics prediction model using PTV as ROI exhibits superior predictive performance,and the combined multi-regional radiomics model can further improve the predictive ability for RP.
