Automated PD-L1 Scoring Using Artificial Intelligence in Head and Neck Squamous Cell Carcinoma

Area of Science:

• Oncology
• Immunotherapy
• Computational Pathology

Background:

• Immune checkpoint inhibitors (ICI) are a promising treatment for recurrent and metastatic head and neck squamous cell carcinoma (HNSCC).
• Patient selection for PD-1/PD-L1 inhibitor therapy relies on manual PD-L1 scoring (TPS, CPS, ICS), which is subject to inter-observer variability and potential biases.
• This variability can negatively impact treatment decisions and patient outcomes.

Purpose of the Study:

• To develop and validate a novel, fully automated method for PD-L1 scoring in HNSCC using sequential neural networks.
• To assess the reliability and reproducibility of automated PD-L1 scoring compared to manual scoring by human investigators.
• To provide insights into the performance and limitations of automated scoring for improving ICI patient selection.

Main Methods:

• A novel approach employing three sequentially applied neural networks for automated scoring of PD-L1 expression.
• Scoring included tumor proportion score (TPS), combined positive score (CPS), and tumor-infiltrating immune cell score (ICS).
• Validation was performed using whole slide images (WSIs) of HNSCC cases, with comparisons to manual scoring by multiple human investigators.

Main Results:

• The automated scoring system demonstrated high concordance with manual scoring, achieving inter-rater correlation (ICC) comparable to human-human agreement.
• ICC between human and machine was slightly higher for CPS and ICS compared to human-human correlations.
• A slightly lower ICC was observed for TPS between human and machine compared to human-human scoring.

Conclusions:

• Automated PD-L1 scoring using neural networks provides a reproducible and reliable method for HNSCC patient stratification for ICI therapy.
• The developed approach shows potential to overcome limitations of manual scoring, reducing variability and improving consistency.
• Further insights into automated scoring and its limitations can guide future improvements in ICI patient selection strategies.