Deep Learning Methodology for Quantification of Normal Pancreas Structures.

Abstract

Histopathologic evaluation plays a crucial role in assessing morphological tissue alterations in disease models and toxicology studies. Identifying small quantitative shifts in specific substructures of organs can be challenging due to the subjective nature of visual assessment and the pathologist's reliance on categorical measurements rather than continuous ones. The emergence of digital pathology and artificial intelligence (AI) provides the ability to quantify different organ substructures using automated methods. Here, we employed a deep learning method to integrate normal pancreatic substructures into an algorithm. We also included areas of abnormal pancreas in the deep learning model. Once the image analysis pipeline was developed, we tested its effectiveness on a disease model and a toxicity study. The quantitative measurements clearly differentiated between control animals and those in the disease model or treated with a test article. In the toxicity study, we observed a distinct dose-dependent change. This approach could be applied to other organs and different species.