MIT Researchers Release AI Model for Early Alzheimer's Detection
A team at MIT's Picower Institute has released FINGERS-7B, an AI model designed to identify Alzheimer's disease up to a decade before symptoms appear. The model achieves four times more accurate diagnosis than previous methods by analyzing lifestyle, genetic, and protein data simultaneously.
FINGERS-7B will be presented at the International Conference on Learning Representations on April 27 in Rio de Janeiro.
How It Works
The model reads multiple types of biological data at once - lifestyle factors, clinical records, genetic markers, and protein profiles - rather than analyzing each separately. This combined approach reveals patterns invisible to single-source analysis.
Adrian Noriega, MIT-Novo Nordisk AI Fellow and co-lead of the FINGERPRINT project, said: "Each of us carries a biological fingerprint, basically a unique combination of signals that reveal disease risk and, if properly understood, could enable prevention and treatment of Alzheimer's disease."
The model was trained on data from tens of thousands of at-risk individuals across the global WW-FINGERS network, which spans 40 countries and 30,000 participants.
Performance and Accuracy
On existing datasets, FINGERS-7B delivers 130% better accuracy in predicting who will respond to specific interventions like dietary changes or new drugs. The model also produces personalized risk assessments, predicting an individual's disease timeline and which treatments are most likely to work for them.
Li-Huei Tsai, director of the Picower Institute, said: "Even as Alzheimer's research labs have gained the capability to generate huge volumes of data, including genetic, epigenetic and proteomic profiles from human tissue samples, we've faced the challenge of truly integrating all of it to gain a comprehensive view of individuals' risk, prognosis and likely treatment response."
Open Access and Deployment
The model's weights, training code, and evaluation pipelines are publicly available. Any research group can apply FINGERS-7B to its own patient cohort and contribute results back to the broader research community.
The model is deployed through the AD Workbench, a secure cloud environment operated by the Alzheimer's Disease Data Initiative. Researchers in over 40 countries already use this platform, allowing them to run analyses without moving sensitive patient data.
MIT's Aging Brain Initiative seeded the project with a $100,000 grant in June. Within ten months, the team trained the model, deployed it, and opened it for external use.
What "At Risk" Actually Means
An at-risk diagnosis from FINGERS-7B does not mean a person will definitely develop Alzheimer's. The model identifies the preclinical stage - a window of 10 or more years when biological changes begin but memory remains intact.
This window is when lifestyle interventions and early therapeutics have the strongest chance of preventing disease onset. The model aims to shift Alzheimer's from an inevitable condition to one that can be managed or prevented.
Current and Future Use
Currently, FINGERS-7B is available to researchers and clinicians through the AD Workbench. Because it is open source, the developers designed it for eventual integration into standard healthcare systems, allowing doctors to monitor brain health using routine check-up data and blood tests.
The Davos Alzheimer's Collaborative and the FINGERS Brain Health Institute announced a partnership in February to use FINGERPRINT for prevention research across diverse global populations. Partners include Novo Nordisk, the Broad Institute, Yale University, Imperial College London, and Brigham and Women's Hospital.
Learn more about AI for Science & Research or explore AI Research Courses to deepen your understanding of machine learning applications in medicine.
Your membership also unlocks:
