An artificial intelligence model predicts PD-L1 expression in Non-small-cell lung cancer from radiomics clinical and pan-inflammatory indicators

This study aimed to develop an artificial intelligence model integrating CT-based radiomic features, clinical parameters, and systemic inflammatory markers to predict PD-L1 expression in patients with non-small-cell lung cancer (NSCLC). However, non-invasive prediction of PD-L1 expression remains challenging in clinical practice. Clinical and imaging data were retrospectively collected from electronic medical records and the Picture Archiving and Communication System. Feature selection was performed using minimum redundancy maximum relevance, least absolute shrinkage and selection operator, and mutual information techniques, and predictive models were constructed using six machine learning classifiers. In 135 patients, models based on clinical-inflammatory variables and radiomic features achieved area under the curve (AUC) values ranging from 0.605 to 0.745. A feature-fused LightGBM model incorporating 40 features achieved an AUC of 0.776 (95% confidence interval: 0.645–0.908), demonstrating superior performance compared with models based on clinical-inflammatory variables, tumor radiomics, and peritumoral radiomics. External validation in 38 patients yielded an AUC of 0.90. Decision curve analysis demonstrated clinical benefit, with an optimal decision threshold between 0.30 and 0.60. These findings suggest that integrating radiomic, clinical, and inflammatory features improves non-invasive prediction of PD-L1 expression in NSCLC and may support patient stratification for immunotherapy.