About
Greeting
Models that help people check and correct faster
MEDAI Lab studies AI that delineates organs in CT slices.
A radiologist may read thousands of slices a day. Rather than a model that replaces that work, we want one that helps people check and correct faster. Our models return a map of where they hesitated, and we study how to pick the slices a person should see first.
Low-dose scans are noisy, which makes segmentation harder, so we also restore low-dose images while keeping their structures — in the same pipeline. The three topics are three steps a slice goes through.
We only release data and code built from our own phantoms or cleared for release. Patient data stays inside joint studies with ethics review and de-identification.
Hajin Do, Principal investigator
What we hold to
Results people can check
Every mask ships with an uncertainty map.
Reproducible work
We release phantom data and versioned code.
Respect for data
Patient data only inside reviewed joint studies.