Teaching machines to read the body.
We build models that turn EEG, ECG, muscle and imaging data into clinical answers, and test them on patients they have never seen.
Research
Six kinds of data, one standard of evidence.
Our standard is to test models on patients, hospitals or devices they have never seen, because that is where clinical AI usually breaks.
Sleep and biosignals
Sleep staging from overnight recordings, including newborns in intensive care.
Brain imaging
Alzheimer's, stroke and hemorrhage from scans, validated per patient, not per slice.
Wearables
Stress and emotion from body-worn sensors, and how the hardware changes the answer.
Neurodegeneration
Earlier detection and staging from imaging and physiology.
Language models for health
Models that reason over signals, records and scans together.
Population health
Risk factors of substance use from population-scale data.
Illustrations generated with ChatGPT for the AI4Health Lab. They depict research themes, not study data.
Track record
A record built over 25 years.
Lab director Abdulhamit Subasi's career record, from his CV.
Recent publications
What we published lately.
Join us
Work on models that have to hold up in the clinic.
We welcome students and collaborators interested in machine learning for health. Write to the lab director with a short note on your background and the problems you want to work on.