Generalizable Prediction of Alzheimer Disease Pathologies with a Scalable Annotation Tool and an High-Accuracy Model

Area of Science:

• Neuropathology
• Computational Biology
• Artificial Intelligence

Background:

• Characterizing Alzheimer disease (AD) neuropathology is labor-intensive and prone to variability.
• Current methods hinder the analysis of large datasets linking AD clinical features to molecular pathogenesis.

Purpose of the Study:

• To develop a high-throughput, unbiased computational pipeline for automated neuropathological analysis of AD.
• To overcome limitations of traditional methods in assessing AD brain tissue pathology.

Main Methods:

• Designed an annotation tool and computational pipeline using Mask Regional-Convolutional Neural Network (Mask R-CNN) for image analysis.
• Trained Mask R-CNN on expert-annotated digital microscopic images of postmortem AD brain tissue.
• Utilized QuPath platform for ground truth annotation and consensus building among experts.

Main Results:

• The automated pipeline achieved high accuracy (94.6%) in identifying amyloid pathology in independent samples.
• Quantitative measurements of amyloid pathology correlated significantly with expert neuropathological assessments (CERAD) and cognitive decline (CDR).
• The system demonstrated generalizable pathology feature detection.

Conclusions:

• The developed computational pipeline enables rapid, unbiased, and quantitative neuropathological analysis of large AD tissue collections.
• This approach facilitates integration with clinical and multi-omic data for a comprehensive understanding of AD.
• The tool significantly reduces time and cost associated with traditional neuropathology assessments.