Machine Learning Classification of Alzheimer’s Disease Pathology Reveals Diffuse Amyloid as a Major Predictor of Cognitive Impairment in Human Hippocampal Subregions

Abstract

AbstractAnalyzing Alzheimer’s disease (AD) pathology within anatomical subregions is a significant challenge, often carried out by pathologists using a standardized, semi-quantitative approach. To augment traditional methods, a high-throughput, high-resolution pipeline was created to classify the distribution of AD pathology within hippocampal subregions. USC ADRC post-mortem tissue sections from 51 patients were stained with 4G8 for amyloid, Gallyas for neurofibrillary tangles (NFTs) and Iba1 for microglia. Machine learning (ML) techniques were utilized to identify and classify amyloid pathology (dense, diffuse and APP (amyloid precursor protein)), NFTs, neuritic plaques and microglia. These classifications were overlaid within manually segmented regions (aligned with the Allen Human Brain Atlas) to create detailed pathology maps. Cases were separated into low, intermediate, or high AD stages. Further data extraction enabled quantification of plaque size and pathology density alongside ApoE genotype, sex, and cognitive status.Our findings revealed that the increase in pathology burden across AD stages was driven mainly by diffuse amyloid. The pre and para-subiculum had the highest levels of diffuse amyloid while NFTs were highest in the A36 region in high AD cases. Moreover, different pathology types had distinct trajectories across disease stages. In a subset of AD cases, microglia were elevated in intermediate and high compared to low AD. Microglia also correlated with amyloid pathology in the Dentate Gyrus. The size of dense plaques, which may represent microglial function, was lower in ApoE4 carriers. In addition, individuals with memory impairment had higher levels of both dense and diffuse amyloid.Taken together, our findings integrating ML classification approaches with anatomical segmentation maps provide new insights on the complexity of disease pathology in AD progression. Specifically, we identified diffuse amyloid pathology as being a major driver of AD in our cohort, regions of interest and microglial responses that might advance AD diagnosis and treatment.