Prediction of PD-L1 expression in nonsmall cell lung cancer using artificial intelligence models based on radiomics: a systematic review and meta-analysis

Clinical Snapshot

Key Findings

• P Value: Not reported in abstract

• Effect Size: Pooled AUC = 0.782 (TPS1 threshold); Pooled AUC = 0.798 (TPS50 threshold). Subgroup: machine learning superior for TPS50 prediction; deep learning superior for TPS1 prediction. PET/CT and contrast-enhanced CT outperformed non-contrast CT for TPS prediction.

• Primary Outcome: Diagnostic accuracy of AI-radiomics models for predicting PD-L1 expression in NSCLC, measured by AUC on validation sets

• Nnt Or Sensitivity: Sensitivity and specificity not reported in abstract. AUC of 0.782–0.798 indicates moderate-to-good discrimination but clinical utility requires paired sensitivity/specificity at defined operating thresholds.

• Confidence Interval: Not reported in abstract

Clinical Application

AI-radiomics pipelines require dedicated software infrastructure, standardised imaging acquisition protocols, and validated feature extraction workflows. These are not yet routinely available in most clinical radiology departments. Integration into clinical decision-making would require prospective validation in diverse, real-world cohorts and regulatory approval before deployment. The moderate AUC values (0.782–0.798) suggest these models could serve as adjuncts rather than replacements for tissue-based PD-L1 testing. In Australia, PD-L1 testing is required for PBS-subsidised access to pembrolizumab (Keytruda) as first-line therapy for advanced NSCLC (TPS ≥50% for monotherapy; TPS ≥1% in combination regimens), per TGA-approved indications and RACGP/TSANZ guidelines. Tissue biopsy remains the standard of care. AI-radiomics approaches are not currently TGA-approved or PBS-listed as diagnostic tools for PD-L1 assessment. Australian oncology centres with PET/CT infrastructure (available at major metropolitan cancer centres) would be best positioned to implement PET/CT-based radiomics pipelines if validated. The finding that PET/CT-based models outperform non-contrast CT is relevant given Australia's relatively good PET/CT access compared to many health systems. RACGP and Cancer Australia would need to endorse any future clinical translation pathway. Patients with confirmed or suspected NSCLC being evaluated for PD-L1 expression status prior to consideration of immune checkpoint inhibitor therapy, particularly where tissue biopsy is technically challenging, insufficient, or carries unacceptable procedural risk

Abstract

BACKGROUND: Since 1992, the incidence of lung cancer has shown an overall downward trend, with survival rates significantly improving in recent years due to the application of immune checkpoint inhibitors (ICIs). Assessing programmed death ligand 1 (PD-L1) expression status in lung cancer patients is crucial for predicting the efficacy of ICIs and formulating treatment plans. Therefore, there is an urgent need to assess PD-L1 expression status in lung cancer patients, particularly those with nonsmall cell lung cancer (NSCLC). While numerous studies have demonstrated promising results using artificial intelligence (AI) models based on radiomics, current understanding in this area remains incomplete. METHOD: We conducted a search of medical databases and screened studies that meet the criteria. We performed feature extraction and data extraction from these studies, analysed them using R software, and generated forest plots. We conducted subgroup analyses based on imaging modality and algorithm model. RESULTS: A total of 35 studies were included in the analysis. For the Tumor Proportion Score (TPS)1 group, the pooled area under the curve (AUC) for the validation set was 0.782. For the TPS50 group, the pooled AUC for the validation set was 0.798. Subgroup analysis revealed that machine learning performed better when predicting TPS50, while deep learning was superior for predicting TPS1. When predicting TPS, prediction models constructed using 2-fluoro-2-deoxy-d-glucose (18F) positron emission tomography/computed tomography (CT) and contrast-enhanced CT outperformed those based on noncontrast CT. Conversely, when predicting a TPS, the noncontrast-CT-based prediction model significantly outperformed the other two groups. CONCLUSIONS: Imaging modality and algorithm model are key factors influencing AI prediction models for PD-L1 expression in NSCLC patients.

References

1. 1.Zhang, P., Zhang, B., Liu, N., Huang, D., Zhong, H., Wu, J., Wang, Y., & Ren, Q. (2026). Prediction of PD-L1 expression in nonsmall cell lung cancer using artificial intelligence models based on radiomics: a systematic review and meta-analysis. European Respiratory Review, 35. https://doi.org/10.1183/16000617.0299-2025