Dual-scale categorization based deep learning to evaluate programmed cell death ligand 1 expression in non-small cell...

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence

Background:

• Immune checkpoint blockade therapy is a novel precision oncology strategy.
• Programmed death ligand 1 (PD-L1) expression is a validated predictive biomarker for therapy response.
• Automating pathological image analysis is an unmet need.

Purpose of the Study:

• To develop and evaluate a deep learning method for PD-L1 expression assessment in digital pathology.
• To improve the accuracy and efficiency of PD-L1 evaluation in non-small cell lung cancer (NSCLC).

Main Methods:

• Proposed a Dual-scale Categorization (DSC)-based deep learning method using two VGG16 neural networks.
• Applied the DSC method to analyze PD-L1 immunohistochemistry images from 110 NSCLC patients.
• Compared the DSC method's concordance with pathologist evaluation against a single-scale method.

Main Results:

• The DSC-based deep learning method achieved 88% concordance with pathologist assessment.
• This concordance rate is higher than the 83% achieved by a single-scale categorization method.
• Demonstrated the potential of DSC for computer-aided diagnosis in digital pathology.

Conclusions:

• The DSC-based deep learning method enhances the application of AI in digital pathology for PD-L1 evaluation.
• This approach can facilitate more accurate and efficient computer-aided diagnosis in oncology.
• The method shows promise for improving patient stratification for immune checkpoint blockade therapy.