PD-L1 scoring agreement improves among expert pathologists using AI assistance

A new study in npj Precision Oncology demonstrates a novel method to build trust in the outputs of an AI-based immunohistochemistry (IHC) scoring algorithm for programmed cell death ligand 1 (PD-L1) from MindPeak. Expert pathologists examined in detail 35 challenging regions of interest (ROIs) from gastroesophageal biopsies in a multi-headed microscope to generate a combined positive score (CPS) by counting PD-L1 positive and negative tumor cells, and PD-L1 positive immune cells. CPS scores were determined both with and without MindPeak algorithm overlays with cell-level PD-L1 labels. Agreement among pathologists increased using AI assistance, with 31% more cases diagnosed as positive by all 11 pathologists. This is a promising step towards building trust in AI-based IHC scoring systems, where a robust ground truth is not available and unassisted interpretation shows high variability.

Read full study