Machine-learning-based image analysis algorithms improve interpathologist concordance when scoring PD-L1 expression...

Area of Science:

• Oncology
• Pathology
• Medical Imaging

Background:

• Programmed death ligand 1 (PD-L1) expression is a key predictive biomarker for immunotherapy in non-small-cell lung cancer (NSCLC).
• Accurate interpretation of PD-L1 expression is challenging, potentially impacting treatment decisions.

Purpose of the Study:

• To evaluate if machine-learning-based image analysis tools can enhance inter-pathologist concordance in assessing PD-L1 expression in NSCLC.
• To determine if these tools improve the consistency of classifying PD-L1 expression levels for clinical decision-making.

Main Methods:

• Five pathologists independently scored 13 NSCLC biopsies for PD-L1 (SP263 clone) expression, both with and without a machine-learning image analysis algorithm.
• Inter-pathologist concordance was measured using Fleiss' kappa and intraclass correlation coefficients.

Main Results:

• The image analysis tool significantly improved inter-pathologist concordance (Fleiss' kappa 0.886 vs 0.613; ICC 0.954 vs 0.837).
• Fewer cases were inconsistently classified into clinical categories (negative/weak/strong) when the algorithm was used (15% vs 38%).

Conclusions:

• Machine-learning image analysis tools demonstrably improve the consistency of PD-L1 expression assessment among pathologists.
• These tools hold potential for more reliable prediction of immunotherapy response in NSCLC patients.