Predicting PD-L1 Expression in Non-Small Cell Lung Cancer Using Radiomics and Habitat Imaging Models
10.3969/j.issn.1005-5185.2025.09.005
- VernacularTitle:基于影像组学及生境成像模型预测非小细胞肺癌PD-L1表达水平
- Author:
Qi YAO
1
;
Qifeng LIU
1
;
Peng CHEN
1
;
Zhimin DING
1
Author Information
1. 皖南医学院弋矶山医院放射科,安徽 芜湖 241001
- Publication Type:Journal Article
- Keywords:
Carcinoma,non-small-cell lung;
Tomography,X-ray computed;
Radiomics;
Programmed death ligand 1;
Immunotherapy;
Habitat imaging;
Forecasting
- From:
Chinese Journal of Medical Imaging
2025;33(9):920-928
- CountryChina
- Language:Chinese
-
Abstract:
Purpose To evaluate the value of arterial-phase CT-based radiomics and habitat imaging models in predicting programmed death ligand 1(PD-L1)expression levels in non-small cell lung cancer(NSCLC).Materials and Methods Clinical and imaging data from 258 pathologically confirmed NSCLC patients at Yijishan Hospital of Wannan Medical College from April 2022 to May 2024 were retrospectively analyzed.Patients were randomly divided into training(n=207)and validation(n=51)sets at a 4∶1 ratio.Whole-lesion radiomic features were extracted from arterial-phase CT images.Subregional habitats were generated using local feature clustering,and their radiomic features were fused to derive habitat analysis features.Dimensionality reduction identified features for constructing whole-lesion radiomic and habitat analysis models.Logistic regression algorithms were used to build models and develop nomograms.Model performance was evaluated using the area under the receiver operating characteristic curve(AUC),and clinical utility was assessed via decision curve analysis.Results Two independent clinical risk factors(tumor location and necrosis presence),14 whole-lesion radiomic features and 16 habitat analysis features were selected.The clinical model achieved AUCs of 0.685(training)and 0.682(validation).The habitat analysis model(AUC:0.776 training,0.761 validation)outperformed the whole-lesion radiomic model(AUC:0.701 training,0.647 validation).The combined model integrating clinical,whole-lesion and habitat analysis features demonstrated superior performance(AUC:0.838 training,0.826 validation)and the highest clinical net benefit on decision curve analysis.Conclusion Habitat imaging features derived from arterial-phase CT effectively predict PD-L1 expression in NSCLC.Combining clinical characteristics with whole-lesion and habitat analysis features further enhances predictive performance.
