AIMA 2026 projects ✦ Project VIII ✦ Neurology
Alzheimer’s disease before memory fails: early diagnosis & progression
Alzheimer’s disease starts in the brain years before the first symptoms. Blood tests can now detect it, and the first disease-modifying drugs are given at the earliest symptomatic stage. Two questions have become urgent: who is on this path, and how fast will each person progress? We will build models that answer both, with honest uncertainty, across cohorts.
In partnership with
The cascade, and one person’s future
Schematic after the hypothetical model of dynamic biomarkers1. Curves, timings and people are illustrative, not patient data.
Why now
- A p-tau217 blood test reached about 90% accuracy, against 61% for primary-care physicians and 73% for specialists2. The FDA cleared the first blood test to aid diagnosis in May 20253.
- Lecanemab and donanemab slowed decline in early symptomatic disease4,5, so finding people early now changes what can be offered.
- Multimodal AI on 51,269 people from nine cohorts reached AUROC 0.94, and neurologists did better with it6.
The gap we target
Forecasting is far from solved. In the TADPOLE challenge, 92 algorithms from 33 teams forecast diagnosis well, but none forecast a cognitive score (ADAS-Cog13) significantly better than random guessing7. Most models are trained on one cohort; real clinics lack PET or CSF; and people progress along different subtypes8.
Harmonize and screen
Harmonize ADNI, OASIS-3 and NACC. Baseline: predict amyloid status from MRI and routine data, as a cheap step before blood or PET tests.
Forecast each person
Sequence models for irregular visits and missing modalities that forecast cognition and time to MCI, with calibrated uncertainty.
Across cohorts
Train on one cohort, test on others; check whether progression subtypes replicate and whether they improve forecasts.
Data
- ADNI: MRI, PET, CSF, blood, genetics (data use agreement)
- OASIS-3: 1,378 participants, MRI and PET
- NACC: over 56,000 participants from Alzheimer’s centers
You will learn
- Longitudinal and time-to-event modelling
- Neuroimaging pipelines
- Multimodal fusion with missing data
- Calibration and cross-cohort validation
Good fit if you
- Like statistics and time series
- Want your model to survive a new cohort
- Care about neurology and ageing
Data
Access, harmonisation, visit alignment.
Imaging
MRI features and amyloid prediction.
Forecasting
Sequence models and uncertainty.
Evaluation
Cross-cohort tests and clinical framing.
The plan · 6 months
From first paper to submission.
Six months, from first paper to submission, with weekly group meetings and a monthly review with mentors.
Onboard
Read 10–15 key papers, set up data access and compute, write a one-page proposal.
Reproduce
Match a published baseline. Share a reproducibility report.
Propose
First version of the method, defended in a mentor review.
Iterate
Ablations, failure analysis, fix what the review found.
Validate
External data or pilot study; freeze results.
Submit
Paper, code release and a talk for the AIMA community.
Project VIII · Neurology
Want to work on
this question?
Apply through the form by November 7, 2026, choosing Project VIII, with an academic CV and a statement of purpose of up to 300 words. Interviews run November 9 to 13, and groups start around November 18. Mentors from Carnegie Mellon University · Stony Brook University.
References
- Jack et al. Tracking pathophysiological processes in Alzheimer's disease: an updated hypothetical model of dynamic biomarkers. Lancet Neurology 2013. DOI
- Palmqvist et al. Blood Biomarkers to Detect Alzheimer Disease in Primary Care and Secondary Care. JAMA 2024. DOI
- US Food and Drug Administration. FDA clears first blood test used in diagnosing Alzheimer’s disease. Press release, 16 May 2025. link
- van Dyck et al. Lecanemab in Early Alzheimer’s Disease. NEJM 2023. DOI
- Sims et al. Donanemab in Early Symptomatic Alzheimer Disease. JAMA 2023. DOI
- Xue et al. AI-based differential diagnosis of dementia etiologies on multimodal data. Nature Medicine 2024. DOI
- Marinescu et al. The Alzheimer's Disease Prediction Of Longitudinal Evolution (TADPOLE) Challenge: Results after 1 Year Follow-up. MELBA 2021. DOI
- Young et al. Uncovering the heterogeneity and temporal complexity of neurodegenerative diseases with Subtype and Stage Inference. Nature Communications 2018. DOI