The global prevalence of lung cancer calls for innovative methods to improve diagnosis and treatment, particularly in developing nations where there is a critical shortage of onco-pathologists. This systematic review examines the role of artificial intelligence (AI) in lung cancer diagnosis, focusing on machine learning and deep learning ap-proaches as potential solutions to this challenge. The evaluation utilized PRISMA guidelines to include 14 studies for conducting a meta-analysis that measured AI-based tool effectiveness in lung cancer pathology diagnosis. Key performance metrics ana-lyzed included sensitivity, specificity, and predictive values. Advanced AI architec-tures, particularly convolutional neural networks (CNNs) and generative adversarial networks (GANs), were identified as instrumental in enhancing diagnostic accuracy and enabling large-scale screening. The review also investigates AI-based identification of promising biomarkers which enhance medical diagnosis while modernizing clinical procedures for improved workflow success. AI has demonstrated success in reducing clinical workloads, thereby optimizing healthcare operations. This review further ad-dresses critical barriers to implementation, including limited generalizability across diverse populations and ethical concerns related to clinical deployment. Unlike pre-vious reviews, this work evaluates AI technologies across multiple imaging modalities, including histopathology, computed tomography (CT), and X-rays. The findings sug-gest that AI has significant potential as a transformative tool for achieving more accu-rate diagnoses, facilitating personalized treatment strategies, and ultimately improving patient outcomes worldwide.
8 June 2026 in British Journal of Biomedical Science
Semi-quantitative scoring of PD-L1 expression on tumor cells (TCs) and/or infiltrating immune cells (ICs) is complex and requires expert pathologist training to reduce inter- and intra-reader variability. We conducted a 1- or 2-day training of 751 pathologists world-wide from 63 countries over a period of 2 years (2016 and 2017). Pathologists read microscopic slides or fully digitized slides stained with PD-L1 immunohistochemistry 22C3 pharmDx assay with expression levels enriched around the clinically relevant cutoffs of 1% and 50% in non-small cell lung cancer (NSCLC). The overall inter-reader agreement (OPA) for PD-L1-stained NSCLC was 95.6% for TPS ≥1% and 87.3% for TPS ≥50% cut-off. The corresponding intra-reader agreement was 95.9% for the ≥1% and 91.4% for the ≥50% cut-off. The inter-reader negative percent agreement (NPA) for TPS ≥1% was 85.6% and 91.9% for the TPS ≥50% cut-off, and the positive percent agreement (PPA) was 97.6% (TPS ≥1%) and 81.0% (TPS ≥50%). The observed high inter- and intra-reader agreements are promising given the difficulty of reproducible scoring tumor cells with heterogeneous distribution of PD-L1 staining combined with the varied professional training, expertise, and experience of the participants. The results may reflect the utility of the expert led standardized protocol used to train pathologists for scoring PD-L1 staining in NSCLC specimens. The digital image-led training approach has the additional advantage of providing a computer-assisted scoring system and allows for remote training of pathologists worldwide.