Making AI-supported radiology more transparent and reliable
The TRUST-RAD project developed AI tools for radiology that are not only powerful but, above all, reliable and transparent. The focus was on the development of RadVLM, an innovative AI model for analysing chest X-rays. The model can explain findings, draft reports, answer clinical questions and highlight relevant areas in the image.
To advance the development of trustworthy medical AI systems, the team published a comprehensive dataset with over 1.1 million image-instruction pairs. Thanks to a computing time grant of 500,000 GPU hours from the Swiss AI Initiative, the ‘3D vision-language model for radiology’ will be further developed and expanded from 2D X-ray images to 3D imaging such as CT scans with the help of AI.
TRUST-RAD was funded in the third Rapid Action Call.