New AI tool developed by ICR and Queen’s University Belfast enhances accuracy of lung cancer biomarker analysis

New AI tool developed by ICR and Queen’s University Belfast enhances accuracy of lung cancer biomarker analysis

(IN BRIEF) Researchers from The Institute of Cancer Research and Queen’s University Belfast have developed an AI tool to improve the accuracy and consistency of PD-L1 scoring in lung cancer diagnosis. The system uses deep learning to assist pathologists in evaluating tumour samples, particularly around critical thresholds that influence immunotherapy decisions. By reducing variability and highlighting borderline cases, the tool has the potential to improve treatment selection and patient outcomes. While not yet approved for clinical use, the technology represents a significant step toward integrating AI into diagnostic workflows and advancing precision medicine in oncology.

(PRESS RELEASE) LONDON, 27-Apr-2026 — /EuropaWire/ — Researchers from The Institute of Cancer Research and Queen’s University Belfast have developed a new artificial intelligence (AI) tool designed to support pathologists in evaluating complex lung cancer cases more consistently and accurately. The system focuses on improving the measurement of PD-L1 Tumour Proportion Score (TPS), a critical biomarker used to determine patient eligibility for immunotherapy treatments.

The study, published in Scientific Reports, demonstrates how AI can enhance diagnostic precision in oncology by reducing variability in how pathologists interpret test results. PD-L1 scoring plays a key role in guiding treatment decisions for patients with non-small cell lung carcinoma (NSCLC), the most common form of lung cancer.

PD-L1 is a protein expressed on certain tumour cells that helps cancers evade the immune system by interacting with PD-1 receptors on T-cells. Therapies that block this interaction have transformed treatment outcomes for many patients. However, the effectiveness of these therapies depends on accurately determining PD-L1 expression levels, typically measured as a percentage of tumour cells expressing the protein.

Assessing PD-L1 TPS is a complex and often subjective process. Even experienced pathologists can reach different conclusions when analysing the same tissue samples, particularly around clinically important thresholds such as 1% and 50%, which directly influence treatment pathways. Small differences in scoring can lead to significantly different clinical decisions, including whether a patient receives immunotherapy.

To address this challenge, the research team developed a deep-learning system trained to assist with PD-L1 scoring. The model was built using carefully annotated datasets and validated against robust “ground truth” measurements established through multiplex immunofluorescence techniques. This approach minimizes inconsistencies caused by staining quality and sample preparation.

Rather than replacing human expertise, the AI tool is designed to support decision-making by identifying borderline cases near key thresholds and providing consistent quantification for mid-range scores. By flagging ambiguous results, it allows pathologists to focus their attention where clinical judgment is most critical.

The findings indicate that the system can significantly improve consistency in PD-L1 evaluation, reducing both underestimation and overestimation of expression levels. This has important implications for patient care, as more reliable scoring could lead to better-aligned treatment decisions and improved outcomes.

The tool is currently in a pre-regulatory phase and is not yet approved for clinical use. However, researchers suggest that AI-assisted workflows could become particularly valuable in high-volume healthcare settings or regions with limited specialist resources, helping to standardize diagnostics and reduce workload pressures.

Manuel Salto-Tellez, lead author of the study and a senior figure in molecular pathology at both institutions, emphasized that the goal is to enhance—not replace—clinical expertise. He noted that improving consistency in PD-L1 scoring is a meaningful step toward more reliable and personalized cancer care, particularly in cases where treatment decisions hinge on narrow diagnostic margins.

The research reflects a broader global movement toward integrating AI into pathology, where digital tools are increasingly being used for tumour analysis, biomarker measurement, and predictive modelling. As validation and regulatory processes progress, such technologies could play a central role in the future of precision oncology.

Media Contact:

Tel: 0203 437 3502
email: mediaoffice@icr.ac.uk

SOURCE: The Institute of Cancer Research