A new machine learning model has been developed that can predict the occurrence of dementia up to nine years before diagnosis with 80% accuracy. The model focuses on changes in the brain’s default mode network and has been tested on a large sample of fMRI scans from participants who later developed dementia.