Graph neural network with multimodal fusion. Each patient is represented as a heterogeneous graph in which biomarker, speech, gait and demographic nodes are joined by learned weighted edges, so that relationships between modalities are modelled directly. Developed by AlgoMD, Pune.
Non-autonomous clinical decision support for adults with mild cognitive impairment, estimating risk to inform a clinician’s decision on specialist referral. Design target: EU MDR Class IIa SaMD.
Out of scope:diagnosing Alzheimer’s disease; use without clinician oversight; use in people without MCI; use in populations not represented in the evaluation data.
Developed and evaluated on ADNI, with a second held-out evaluation on Bio-Hermes-001. Final validation on DementiaBank is planned; access requires company incorporation and is on hold.
| Metric | ADNI test set | Bio-Hermes-001 test set |
|---|---|---|
| Amyloid positivity AUC | 0.8897 | 0.9071 |
| Cognitive decline RMSE | 1.804 pts/yr | — |
| Dementia progression C-index | 0.651 | — |
Results from held-out splits of research datasets. They are not clinical performance claims and do not show that NFAD is accurate in clinical practice or in any population outside these cohorts.
Version 0.1, November 2026. Questions or corrections: regulatory@algo-md.com.