Self-supervised contrastive learning using CT images for PD-1/PD-L1 expression prediction in hepatocellular carcinoma

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Programmed cell death protein-1 (PD-1) and PD-L1 expression predict response to immune checkpoint inhibitors (ICIs) in hepatocellular carcinoma (HCC).
• Current methods rely on invasive immunohistochemistry (IHC), necessitating non-invasive predictive tools for clinical decision support.

Purpose of the Study:

• To develop and validate a novel deep learning model for non-invasively predicting PD-1 and PD-L1 expression status in HCC patients.
• To assess the model's performance against existing deep learning and machine learning approaches.

Main Methods:

• A cohort of 87 HCC patients was analyzed using 3094 computed tomography (CT) images.
• A deep learning model, Contrastive Learning Network (CLNet), was developed using self-supervised contrastive learning for feature extraction from CT images.
• The model was trained and validated for predicting PD-1 and PD-L1 expression.

Main Results:

• CLNet achieved an Area Under the Curve (AUC) of 86.56% for PD-1 prediction and 83.93% for PD-L1 prediction.
• The model demonstrated superior performance compared to other deep learning and machine learning models evaluated.

Conclusions:

• A non-invasive deep learning model (CLNet) can accurately predict PD-1 and PD-L1 expression status in HCC.
• This predictive capability may enhance precision treatment strategies, particularly for patients receiving ICIs.