Tolga Çukur and colleagues have published “Learning Fourier-Constrained Diffusion Bridges for MRI Reconstruction” in IEEE Transactions on Medical Imaging (March 2026).
Diffusion models, a type of generative AI, are increasingly used to reconstruct MRI images from fast, undersampled scans, but they usually start from random noise, which differs from the structured artifacts of undersampled MRI. The paper introduces the Fourier-constrained diffusion bridge, which instead learns a direct path from undersampled to fully sampled data. The approach aims for clearer images from faster scans.
Authors: M. Usama Mirza, Onat Dalmaz, Hasan A. Bedel, Gokberk Elmas, Yılmaz Korkmaz, Alper Güngör, Salman Ul Hassan Dar, Kader K. Oguz, Tolga Çukur