Reconstructing and forecasting disease trajectories of patients with Alzheimer's disease using routine data in resource-constrained settings

In three linesGNOVA, a framework combining GRU, Neural ODE, and VAE, predicts cognitive trajectories (CDR-SB, MMSE) in 1,727 Alzheimer patients (ADNI, 10 years) without neuroimaging. MAE of 1.35 and 2.28 respectively. Enables interpolation/extrapolation and calibrated uncertainty estimation for resource-constrained settings.

Summary generated by Claude — human-verified