At Worcester Polytechnic Institute, a small research team has trained an artificial intelligence model to spot brain changes linked to Alzheimer’s disease, and they say it is clocking in at nearly 93% accuracy in tests. Led by assistant research professor Benjamin Nephew, with PhD student Senbao Lu and alumnus Bhaavin Jogeshwar, the group compared MRI scans from older adults with normal cognition, mild cognitive impairment and Alzheimer’s to tease out anatomical patterns too subtle for the human eye. WPI reports that the system also turned up age and sex specific signatures that could tighten the window for earlier diagnosis and more tailored treatment.
Peer‑reviewed paper and methods
The work is published in the journal Neuroscience and went online in December 2025, with the PubMed entry listing the paper under the title “Neuroanatomical-based machine learning prediction of Alzheimer's Disease across sex and age.” For the study, researchers measured volumes in 95 brain regions from 815 MRI scans drawn from the Alzheimer’s Disease Neuroimaging Initiative and used those measurements to train a Random Forest algorithm. According to the paper, the model achieved 92.87% accuracy when distinguishing Alzheimer’s disease from mild cognitive impairment and normal cognition. More technical details, including the DOI, are available through PubMed.
What the algorithm flagged
Across different age groups and between men and women, the model consistently ranked volume loss in the hippocampus, amygdala and entorhinal cortex as the strongest predictors of Alzheimer’s. In younger members of the cohort, ages 69 to 76, the right hippocampus stood out, while women in the study showed changes in the left middle temporal region. The researchers suggest those patterns could reflect interactions with sex hormones.
WPI also leans on this project as a showcase for interdisciplinary training, noting that students from biology, computer science and data science all worked on building and testing the models, as described by WPI. For students, it is a crash course in how lab code can inch its way toward real clinical tools.
Patients and clinicians weigh in
For people already living with early-onset cognitive changes, shaving time off the diagnostic process can be a very big deal. Marc Ehrlich, diagnosed with Alzheimer’s at 64, told local reporters that “it’s better to know than not to know.”
At the same time, Nephew has stressed that no one is handing the keys to the clinic over to an algorithm. Local coverage quoted him underscoring that “this type of prediction wouldn’t be used to make clinical decisions” on its own, but that it “could be used to inform” clinicians’ judgment. That careful mix of hope and restraint came through in reporting by NBC Boston.
Part of a larger push
WPI’s results land in the middle of a broader rush to use AI to catch Alzheimer’s earlier, from tools that listen for subtle changes in speech to systems that infer protein markers from routine tests. Researchers at Boston University, for instance, have developed a complementary AI method that predicts amyloid and tau markers using standard clinical data, suggesting different teams are converging on a similar goal of earlier detection, as reported by Boston University.
That work is unfolding against a sobering backdrop. More than 7 million Americans age 65 and older are living with Alzheimer’s, according to the Alzheimer's Association, which is why accessible, scalable screening tools remain a public health priority.
Next steps and local impact
Nephew and his students say they are now experimenting with deep learning models and exploring how clinical factors such as diabetes might influence prediction performance. WPI presents the project as both a scientific step forward and a training ground that links campus talent with Worcester’s growing biotech ecosystem.
The team and the paper also emphasize that the Alzheimer’s Disease Neuroimaging Initiative is a research cohort, not an everyday clinic population, and that the model will need validation in more diverse clinical settings before hospitals consider adopting similar systems, as the university notes in its release.
The study shows how large public imaging datasets can be transformed into decision support tools, but the real world payoff will depend on replication, clinical trials and regulatory sign off. For Worcester, it is another example of how local universities can turn AI research into jobs, startups and care innovations that stay rooted in the region.
