A paper by the TRUST-RAD research team, in collaboration with the Medical Data Science group at ETH Zurich, and the computational infrastructure supported by the Swiss AI Initiative and the Swiss National Supercomputing Centre (CSCS) under project ID a02 on Alps, has been accepted for the ICML 2025 Workshop on Multi-modal Foundation Models and Large Language Models for Life Sciences.
It presents a method that enables AI systems to learn robust and meaningful representations from medical image data. Instead of relying on large, elaborately labeled data sets, the model uses the existing structure of the data: It processes frontal and side views as related pairs of images and learns from them to recognize important information, entirely without text or label data.
The result is an approach that is particularly suitable for clinical applications: adaptable and can also be used under real-life conditions with limited data availability.