Development and Evaluation of a Deep Learning-Based Tool for Interpretable Abnormality Detection in Preclinical Toxicology Whole Slide Images

Study Background

Before human testing, therapeutics must undergo testing in mammalian models to identify potential toxic effects to the liver, kidney, and other organs. However, manual evaluation of preclinical toxicology specimens is often time-consuming and inconsistent.

Artificial intelligence (AI)-based digital pathology models have the potential to improve manual assessments in toxicologic pathology through increasing speed, efficiency, accuracy, and reproducibility.

Here, we developed an AI digital pathology tool, leveraging a pathology foundation model and providing interpretable overlays, for the identification of histological abnormalities in whole slide images (WSIs) of rat specimens treated with experimental compounds.

Conference

Society for Toxicologic Pathology 2025