A preliminary attempt to harmonize using physics-constrained deep neural networks for multisite and multiscanner MRI datasets (PhyCHarm)

Open Webpage

Gawon Lee, Dong Hye Ye, Se-Hong Oh

Published: 2025, Last Modified: 05 Nov 2025NeuroImage 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•PhyCHarm is a physics-informed harmonization framework for multisite and multiscanner MRI data.•PhyCHarm generates quantitative T1 and M0 maps along with harmonized T1-weighted images.•PhyCHarm demonstrated enhanced harmonization accuracy for 3T T1-weighted brain MRI.•PhyCHarm outperforms U-Net and Pix2Pix; performs comparably to CALAMITI and HarmonizingFlows on SSIM, PSNR, and Dice metrics.•PhyCHarm is extendable to other contrasts such as T2-weighted and FLAIR sequences.