Integrative radiomics and habitat imaging models for predicting PD-L1 expression in non-small cell lung cancer.

Objective: To evaluate the feasibility of an arterial-phase computed tomography (CT)-based radiomics and habitat imaging model for noninvasive prediction of programmed death-ligand 1 (PD-L1) expression in non-small cell lung cancer (NSCLC). Methods: This retrospective study included 246 patients with pathologically confirmed NSCLC who underwent arterial-phase CT between January 2021 and March 2024. Patients were randomly divided into a training cohort (n = 172) and a validation cohort (n = 74). Whole-tumor radiomics features were extracted from manually segmented lesions, while tumor habitats were generated using texture-based clustering to derive habitat imaging features. After feature selection, whole-tumor radiomics, habitat imaging, and combined models incorporating clinical variables were constructed using logistic regression. Model performance was assessed using the area under the receiver operating characteristic curve (AUC) and decision curve analysis. Results: Tumor maximum diameter and intratumoral necrosis were identified as independent clinical predictors of PD-L1 expression. The habitat imaging model outperformed the whole-tumor radiomics model, achieving AUCs of 0.774 and 0.758 in the training and validation cohorts, respectively. The combined model demonstrated the best performance, with AUCs of 0.840 in the training cohort and 0.828 in the validation cohort, and showed superior net clinical benefit on decision curve analysis. Conclusion: An arterial-phase CT-based habitat imaging model provides effective noninvasive prediction of PD-L1 expression in NSCLC. Integration of radiomics and clinical features further improves predictive performance, highlighting its potential utility for pre-immunotherapy evaluation.