UMC Utrecht deep-learning auto-contouring for radiotherapy

par UMC Utrecht

AI Application

Faible confiance

Self-developed deep-learning software of the Radiotherapy department of UMC Utrecht that proposes contours (delineations) of organs at risk, such as the heart and lungs, on the CT and/or MRI scans used to plan radiation treatment. UMC Utrecht says the department built it with deep-learning experts, radiation therapy technologists, clinical informaticians and physicians, and that by July 2023 radiation therapists had made 10,000 automatic contours with it over about three and a half years. The algorithm proposes a contour that a radiation therapy technologist checks; one technologist says the proposal is often very good, sometimes needs a small adjustment and occasionally falls short. UMC Utrecht reports about 20 minutes saved per patient and more consistent contours. It says the software was implemented in a certified development environment by clinical informaticians who are externally certified to develop medical software for clinical use, which is why the contours may be used clinically. Where the software runs and which planning system receives the contours are not stated.