PROMISE-AD: AI Model Predicts Alzheimer's Progression from Cognitive Normal to Dementia
Researchers have developed a new survival framework called PROMISE-AD to predict how Alzheimer's disease progresses. It leverages irregular clinical visits and data from the ADNI/TADPOLE studies. The model turns pre-index visits into tokens that include standardized measurements, masks for missing data, longitudinal changes, time-normalized slopes, visit timing, and non-diagnostic categories. A temporal Transformer combines various representations to determine a progression score and latent discrete-time mixture hazards. The training involves survival likelihood, focal risk loss tailored by horizon, progression ranking, and hazard smoothing. This framework addresses issues like irregular visit patterns and diagnostic leakage, effectively predicting transitions from cognitive normalcy to mild cognitive impairment and from MCI to Alzheimer's dementia. The research is accessible on arXiv (2604.28055) and classified as cross-type.
Key facts
- PROMISE-AD is a survival framework for Alzheimer's disease progression prediction.
- It uses irregular visits and tabular data from ADNI/TADPOLE.
- The model predicts conversion from CN to MCI and from MCI to AD dementia.
- It employs a temporal Transformer with global, attention-pooled, and latest-visit representations.
- Training includes survival likelihood, horizon-specific focal risk loss, progression ranking, and hazard smoothing.
- The framework is leakage-safe and avoids diagnostic leakage.
- It provides calibrated horizon risks.
- The paper is available on arXiv with ID 2604.28055.
Entities
—