Artificial Intelligence-driven image analysis for standardised programmed death-ligand 1 expression evaluation in...

Area of Science:

• Computational pathology
• Oncology
• Immunohistochemistry

Background:

• Accurate programmed death-ligand 1 (PD-L1) immunohistochemical (IHC) assessment is crucial for non-small cell lung cancer (NSCLC) immunotherapy.
• Current PD-L1 interpretation is subjective, time-consuming, and prone to inter-observer variability, potentially leading to patient mis-stratification.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) deep-learning model for automated quantification of PD-L1 expression in NSCLC.
• To improve the accuracy, speed, and consistency of PD-L1 assessment in digital pathology.

Main Methods:

• A deep-learning approach was developed using 1212 whole slide images (WSIs) from 706 NSCLC patients across three cohorts.
• Tumor regions were extracted, and a multi-granular multiple-instance learning method was used to capture patch-level morphological features.
• A multi-grained expression interpreter model aggregated features to stratify PD-L1 expression status.

Main Results:

• The AI model demonstrated strong interpretive ability and wide applicability across different specimen types (surgical, biopsy, metastases).
• High macro-average area under the receiver operating characteristic curve (AUC) values were achieved, ranging from 0.844 to 0.958 across specimen types and cohorts.

Conclusions:

• Deep learning offers significant potential for automated, rapid, and accurate PD-L1 expression inference from complex IHC images.
• AI frameworks can substantially enhance routine digital pathology workflows for PD-L1 detection.