Diagnostic accuracy of ¹⁸F-FDG PET/CT radiomics for non-invasive prediction of PD-L1 expression in non-small cell lung cancer: A systematic review and meta-analysis.
Authors
Affiliations (3)
Affiliations (3)
- Research Center of Thoracic Oncology (RCTO), National Research Institute of Tuberculosis and Lung Diseases (NRITLD), Shahid Beheshti University of Medical Sciences, Tehran, Iran.
- Western Michigan University Homer Stryker M.D. School of Medicine, Kalamazoo, MI, USA.
- Research Center of Thoracic Oncology (RCTO), National Research Institute of Tuberculosis and Lung Diseases (NRITLD), Shahid Beheshti University of Medical Sciences, Tehran, Iran. [email protected].
Abstract
To evaluate the diagnostic performance, methodological quality, and clinical feasibility of ¹⁸F-FDG PET/CT-based radiomics machine learning models for predicting PD-L1 expression in non-small cell lung cancer (NSCLC). Systematic searches of PubMed, Scopus, Web of Science, Embase, and IEEE Xplore were conducted up to July 2025. Eligible studies developed radiomics-only models from ¹⁸F-FDG PET/CT for pre-biopsy or pre-operative PD-L1 prediction, with immunohistochemistry (IHC) as the reference standard (tumor proportion score ≥ 1%). Study quality was assessed using QUADAS-2 and METRICS. Pooled area under the curve (AUC), sensitivity, and specificity, with 95% confidence intervals (CI), were measured via a bivariate random-effects model. Eleven studies met the inclusion criteria; eight were included in the meta-analysis (n = 1,053). The pooled AUC was 0.83 (95% CI: 0.79-0.86), sensitivity 0.75 (95% CI: 0.64-0.84), and specificity 0.77 (95% CI: 0.64-0.87). Subgroup analyses revealed higher accuracy with semi-automatic segmentation, larger training cohorts, and biopsy-only specimens. QUADAS-2 identified high bias risk in the index test domain, mainly owing to the absence of segmentation validation and unclear blinding. METRICS scores averaged 58.04% (range: 41-64.7%), indicating moderate methodological quality. ¹⁸F-FDG PET/CT-based radiomics models show promise for non-invasive PD-L1 prediction in NSCLC, but their clinical translation is limited by methodological heterogeneity, absence of multi-center design, lack of external validation, and variable segmentation practices. Future work should focus on multi-center datasets, standardized workflows, and rigorous validation to enable reliable real-world applications.