Development and piloting of automation for analysis of the blood-brain barrier using deep neural networks

This project aims to develop and pilot a deep learning–driven methodology for quantifying astrocyte integrity and blood–brain barrier (BBB) disruptions in human striatal tissue under pathological conditions. Building on prior research highlighting basement membrane alterations in chronic alcohol use, we will focus on astrocytes, which contribute significantly to BBB maintenance. The study employs immunohistochemical techniques on formalin-fixed, paraffin-embedded samples (including Alzheimer’s and chronic alcoholism cases) to visualize astrocytes in both grey and white matter. Two complementary annotation strategies will facilitate the training of supervised deep neural network models. These models will be validated against expert manual counts, offering a comparative measure of reliability, efficiency, and reproducibility. The results are expected to improve diagnostic accuracy, reveal structural changes in astrocytes linked to neurodegeneration, and guide therapeutic interventions targeting BBB permeability. Longer-term, the approach could be extended to other complex cell types in neuropathology, thereby broadening its impact.