AI driven pre-regulatory validation of PD-L1 analysis in lung cancer

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence

Background:

• The assessment of Programmed Death-Ligand 1 (PD-L1) expression in lung cancer using the Tumour Proportion Score (TPS) is crucial for immune-oncology treatments.
• Inter- and intra-pathologist variability in TPS scoring, particularly at clinical thresholds (<1% and ⩾50%), poses challenges for consistent treatment decisions.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) tool designed to assist pathologists in calculating TPS for lung cancer.
• To improve the accuracy and consistency of PD-L1 assessment, especially around critical clinical thresholds.

Main Methods:

• Development of a deep learning algorithm for TPS calculation.
• Utilized a validated multiplex immunofluorescence panel (PD-L1, CD68, cytokeratin) to establish ground truth values around clinical thresholds.
• The DL tool was designed to highlight cases near 1% and 50% thresholds for manual review and interpret intermediate scores.

Main Results:

• The DL tool assists in identifying lung cancer cases with PD-L1 TPS scores near critical clinical thresholds.
• The system facilitates more consistent manual review and direct interpretation of TPS values.
• Demonstrated the potential of DL tools to aid in the quantitative assessment of tissue-based biomarkers.

Conclusions:

• Deep learning tools can significantly assist in standardizing PD-L1 TPS assessment in lung cancer.
• This technology has the potential to enhance the reliability of immune-oncology biomarker quantitation.
• Assisted DL systems offer a promising pathway towards precise and reproducible clinical quantitation of tissue biomarkers.