How it works

A vision-language model was pretrained on nearly one million routine hospital neuroimaging studies, pairing scans with real clinical reports and notes.