Aykut Koç and Tolga Çukur and colleagues have published “Meta-Entity Driven Triplet Mining for Aligning Medical Vision-Language Models” in IEEE Journal of Biomedical and Health Informatics (March 2026).
Medical vision-language models read chest X-rays together with their radiology reports, and their accuracy depends on how well images and text are aligned. The paper introduces MedTrim, which uses details extracted from the reports, such as the location, size and severity of findings, to align the two more precisely. Better alignment can make AI tools that support radiologists with growing workloads more reliable.
Authors: Melih Yilmaz, Şaban Öztürk, Muti Kara, M. Talat Yavuz, Ekim Gümeler, Aykut Koç, Tolga Çukur