NeuroImage: Deep learning maps large-scale histological tau proteins to validate neuroimaging biomarkers for Alzheimer's disease

Abnormal tau inclusions are a hallmark of Alzheimer's disease and a predictor of clinical decline
.
Several tau PET tracers are available for the study of neurodegenerative diseases, opening avenues for molecular diagnostics

Lea T.
Grinberg et al developed a computational method to identify and segment particles of interest in multi-billion-pixel digital pathology images to generate quantitative 3D density maps
.

The convolutional neural network IHCNet for immunohistochemical samples is at the heart of the program
.

Research Process

The research team successfully processed and immunostained more than 500 slides from two intact human brains containing three phospho-tau antibodies (AT100, AT8 and MC1)
.

IHCNet Architecture

The study's artificial neural network estimated tau inclusions from brain images, with ROC AUCs of 0.
87, 0.
85, and 0.
91 for AT100, AT8, and MC1, respectively
.
An introspective study further evaluated the ability of the trained model to learn tau-related features

Block surface AT8, AT100, MC1 quantitative heat map (first column), MRI (second column) of cases 1 and 2
.
The third column describes the respective 3D rendering of each tau

original source

Deep learning for Alzheimer's disease: Mapping large-scale histological tau protein for neuroimaging biomarker validation