Domain Adaptation-Based Deep Learning for Automated Tumor Cell (TC) Scoring and Survival Analysis on PD-L1 Stained...

Area of Science:

• Computational pathology
• Artificial intelligence in oncology
• Cancer biomarker discovery

Background:

• Non-small cell lung cancer (NSCLC) treatment with immune checkpoint inhibitors (ICIs) efficacy varies significantly among patients.
• Accurate patient stratification is crucial for optimizing anti-PD-L1 therapy response prediction.
• Current methods for assessing PD-L1 expression, such as Tumor Cell (TC) scoring, can be subjective and labor-intensive.

Purpose of the Study:

• To develop and evaluate deep learning (DL) systems for stratifying NSCLC patients receiving anti-PD-L1 therapy.
• To assess the performance of DL systems in predicting patient survival based on digital histopathology images.
• To reduce reliance on manual annotation and traditional scoring methods in PD-L1 assessment.

Main Methods:

• Two DL-based decision systems were developed using functional and morphological properties of epithelial regions in whole slide images.
• The first system replicated pathologist-assessed Tumor Cell (TC) scores (cut-point at 25%).
• The second system directly learned patient stratification from survival data, bypassing TC scoring assumptions.
• Both systems utilized a novel unpaired domain adaptation deep learning solution for segmentation, minimizing the need for extensive manual annotations.

Main Results:

• The first DL system demonstrated high concordance with pathologist TC scoring (Lin's concordance 0.93-0.96) on independent cohorts.
• Both DL systems achieved similar and significant patient stratification power compared to pathologist TC scoring (HR ~0.53, p < 0.01).
• The DL approach effectively stratified patients treated with anti-PD-L1 therapy, indicating potential for improved survival prediction.

Conclusions:

• Deep learning systems can accurately stratify NSCLC patients for anti-PD-L1 therapy response.
• These AI tools offer a robust and potentially more efficient alternative to traditional PD-L1 scoring methods.
• The developed DL approach holds promise for enhancing personalized treatment strategies in NSCLC.