From Microscope to Machine: A Practical Guide to PD-L1 Testing in NSCLC

Objective: PD-L1 immunohistochemistry (IHC) is an essential predictive biomarker test guiding immune checkpoint inhibitor (ICI) treatment in individuals with
non-small cell lung cancer (NSCLC). However, variability in antibody clones, scoring systems (Tumor Proportion Score (TPS), Combined Positive Score (CPS), Immune
Cell scoring (IC)), and pre- analytical/analytical conditions complicates interpretation and reproducibility—especially in small biopsies and cytological specimens
in NSCLC. To review current practices, challenges, and advances in PD-L1 testing in NSCLC, with emphasis on tumor heterogeneity, cytological limitations, and the
evolving role of artificial intelligence (AI)-based digital pathology tools. We also aimed to explore how multimodal approaches, including radiomics, may complement
tissue-based assessment and improve patient selection for ICI therapy.
Materials and Methods: A comprehensive literature review was performed, focusing on studies evaluating PD-L1 expression in NSCLC using validated clones (22C3,
28-8, SP263, SP142), cytology– histology concordance, pre-analytical factors, and AI-based PD-L1 scoring platforms. The search covered publications from January
2020 to June 2025. Data were synthesized thematically, addressing technical variables, interpretive variability, and emerging digital solutions.
Results: PD-L1 expression in NSCLC is affected by spatial heterogeneity and technical variables, leading to diagnostic inconsistency. Cytological specimens pose
unique challenges due to limited architecture and fixation artifacts. Inter-observer variability is highest in the 1– 49% TPS range. AI-assisted algorithms and digital
platforms have demonstrated improved reproducibility (κ up to 0.74), accuracy (up to 95%), and potential correlation with clinical outcomes. Commercial AI platforms,
such as Lunit SCOPE PD-L1 and HALO Lung PD-L1 AI, achieved up to 92% accuracy and reduced borderline misclassification rates by 18–30%. Radiomics using PETbased
imaging—incorporating SUVmax, metabolic tumor volume, and heterogeneity indices—shows promise as a non-invasive adjunct, particularly when tissue
sampling is limited.
Conclusions: Reliable PD-L1 testing requires clone-specific validation, adherence to standardized protocols, and awareness of sample limitations. Integration of AIbased
digital pathology and radiomics can enhance diagnostic precision, particularly in ambiguous or limited samples.