Elastic maps brain changes in Alzheimer’s: Machine learning reveals sex differences in neural damage and inflammation

An investigative team from the Stevens Institute for Neuroimaging and Informatics and the Keck School of Medicine at the University of Southern California succeeded in developing a high-throughput imaging pipeline that produces precise spatial maps of Alzheimer’s disease pathology and glial inflammation in post-mortem brain tissue, using pixel-level machine-learning algorithms. The study, published in the journal Scientific Reports, examined tissue sections from the hippocampus and medial temporal lobe of 65 cases classified as low, medium or high levels of disease-related neurodegenerative changes, in an effort to overcome the limitations of conventional manual assessment and to generate accurate quantitative measurements that combine protein aggregates with neuroimmune responses.

The analytical pipeline relied on tissue staining for three primary biomarkers: IBA1 protein to detect microglia, the 4G8 antibody to identify beta-amyloid protein, and a Gallyas stain to trace neurofibrillary tangles. The team employed the open-source machine-learning tool Elastic to perform pixel-accurate image analysis, which enabled classification of amyloid plaques into three structural types, dense plaques, diffuse plaques, and intracellular amyloid-related protein, and morphological classification of glial cells into three functional phenotypes: ramified, amoeboid and ameboid.

The results showed that diffuse plaques exhibited the most pronounced increase with disease progression, localizing specifically to the subiculum and the molecular layer of the dentate gyrus.In contrast, dense plaques and intracellular protein accumulated in the inner entorhinal cortex and surrounding areas. The analyses documented a morphological shift in glial cells with advancing neurodegeneration, characterized by a reduction in ramified structures and an increase in ameboid features associated with acute inflammation in advanced cases. Moreover, amoeboid glia in specific hippocampal regions were linked to amyloid accumulation in later stages, and to tau protein and tangles in early stages.

The data also revealed subtle biological differences between sexes and genetic traits; males exhibited a higher burden of neurofibrillary tangles than females, especially in cortex adjacent to the hippocampus, and their cognitive decline correlated with those tangles, whereas females showed a greater burden of amyloid plaques. At the genetic level, memory deterioration was more strongly associated with amyloid accumulation in individuals who do not carry the epsilon-4 allele of the apolipoprotein E gene, confirming that the tissue response to the disease is shaped by distinct biological and genetic determinants.

This methodological shift has practical implications for medical-research teams and university hospitals in the Gulf, Egypt and the Arab Levant, especially centers that are building biobanks and studying neurodegenerative diseases. Moving from subjective visual assessment to pixel-level machine-learning pipelines provides these institutions with a precise means to reduce reliance on costly proprietary software suites and to obtain standardized, reproducible histological measurements across different laboratories and clinical labs.

Implementing this pipeline will require redesigning training programs for digital pathology specialists and medical data scientists, to link immunohistochemical staining skills with pixel-level computer-vision algorithms.It also obliges those conducting local clinical trials and biomarker studies to stratify data by sex and genetic profiles, and not to treat neurodegeneration pathways as a single physiological trajectory for all patients.