Clinical AI · Neurological diseases
Neurological disease staging and progression risk — from data hospitals already hold.
SynapSeek’s clinical AI reads six streams of patient data together to show where a patient sits on the disease course today — and how quickly they are likely to progress.
The problem
Neurological disease is found late — just as the drugs finally work.
Alzheimer’s pathology can build for 15 to 20 years before the first symptom. By the time symptoms bring a patient to a clinic, the window in which the newest treatments work is already closing.
The records that could show it sooner often already sit inside the health system — in separate places, each read one test at a time. No one is reading them together.
People living with a neurological disease — the world’s leading cause of illness and disability
Americans aged 85 and older living with Alzheimer’s dementia
How long Alzheimer’s pathology can build before the first symptoms appear
Global annual cost of dementia today, and projected by 2030
Disease-slowing treatment now exists
Lecanemab (Leqembi) and donanemab (Kisunla) slow Alzheimer’s — but are approved only for mild cognitive impairment or mild dementia. Finding disease late now has a direct cost.
The signal is in the combination
No single test catches the disease early. The signal appears only when several kinds of patient data are read together and weighed against one another.
AI can now read them together
Clinical AI can fuse many weak signals into one explainable answer — and do it across a whole population, not one patient at a time.
Sources — Lancet Neurology / WHO Global Burden of Disease (2024) · Alzheimer’s Association, Alzheimer’s Disease Facts and Figures · WHO dementia cost estimates · FDA prescribing information for Leqembi and Kisunla
The platform
One clinical AI platform. Six data streams. One explainable answer.
SynapSeek’s AI reads six streams of patient data together — with capability for more in development — and returns a result a clinician can interrogate: the stage of disease, how quickly it is likely to progress, and the findings behind the call.
Disease staging
Places a patient on a five-stage scale, from cognitively normal to Alzheimer’s disease, with the evidence behind the call shown.
Progression risk
Estimates how quickly a patient is likely to progress, as a probability and a range — a forecast, not only a reading of the present.
Treatment-response tracking
Following how a diagnosed patient responds to treatment over the years that follow, measured against their own earlier baseline rather than a population average.
Earlier detection
Once built, the longitudinal record is meant to help find disease earlier. It is where the platform is going — not a result we claim today.
How it works
From records a hospital already holds to a staged, explained result.
-
01
Ingest
Designed to read data from the systems a hospital already runs. No new machine, no new test.
-
02
Fuse
The six streams are read together and weighed against one another, because the early signal is invisible to any single test.
-
03
Explain
Every result carries its reasoning — which findings drove it, and where in the brain — together with a confidence level. No black box.
-
04
Decide
The result is decision support: a qualified clinician interprets it and makes the clinical decision. Follow-up against the patient’s own baseline is on the roadmap.
Results arrive from different departments, months or years apart. Each is read on its own, and each looks unremarkable. The patient is told they are fine, and the record sits unread.
The same records, read as one. Two answers come back: where the patient sits on the five-stage scale today, with the evidence shown, and how quickly they are likely to progress, as a probability and a range. A clinician reviews both and decides what happens next.
Evidence
Results, stated exactly as they stand today.
All results are retrospective — measured on research cohorts, not yet on live hospital patients. Every figure is available on request.
| Result | Measure | What it means |
|---|---|---|
| ~81% | Balanced accuracy across five stages, from cognitively normal to Alzheimer’s disease | Measured on held-out ADNI patients the model never saw during training. Class-balanced, so every stage counts, not only the most common. |
| 0.84 | C-index, predicting which people with mild cognitive impairment progress to dementia | ADNI, calibrated. Shown two people with mild cognitive impairment, the model identified who would progress to dementia sooner about 84 times in 100. On an independent cohort (OASIS-3) it retains 0.97–1.0 of its performance on shared inputs, in both directions — external dataset, internal analysis. |
| No collapse | Cross-cohort transfer, ADNI ↔ OASIS-3 | Staging and progression both carry over to an independent cohort. Staging retains 1.00 and 0.96 of its performance in the two directions, on a three-stage view both cohorts share (OASIS-3 does not use the five-stage scale); progression retains 0.97–1.0. External dataset, internal analysis. |
Source — SynapSeek internal retrospective analyses on the ADNI and OASIS-3 research cohorts · figures available on request
The team
The science and the engineering are on the founding team.
-
Aryan Durrani
Chief Executive Officer -
Dr. Johana C.M. Vega Leonel
Chief Operating Officer -
Varun Turlapati
Chief Business Officer -
Kevin Li
Lead Engineer
Advisors
Senior clinical and scientific oversight.
Neurosurgery
A senior academic neurosurgeon and former department chair, guiding clinical and surgical questions.
AI in medical imaging
An imaging engineer and researcher working on AI in medical imaging.
Neurological drug discovery
A chemist working in neurological drug discovery and clinical translation — biomarkers, substrates and drug interactions.
Advisor details available on request
Work with us
Build the record with us.
Become a founding data partner
We are looking for a small group of health systems and research institutions to be the first to test the platform on their own records, and to build the longitudinal record with us.
- A silent-mode pilot on de-identified records — results never reach patient care
- Performance measured on your own patients before any clinical use
- Joint publication of validation results
- Data custody and residency agreed in writing before any record moves
Open to investment discussions
SynapSeek is open to conversations with investors who take a long view of neurology — clinical evidence first, then the data platform that evidence makes possible.