One model to unite them all: Personalized federated learning of multi-contrast MRI synthesis

Onat Dalmaz, Muhammad Usama Mirza, Gökberk Elmas, Muzaffer Özbey, Salman Ul Hassan Dar, Emir Ceyani, Kader Karli Oguz, Salman Avestimehr, Tolga Çukur

Abstract: Highlights•A novel personalized federated learning method for multi-contrast MRI synthesis.•A novel generator equipped with personalization blocks to improve model specialization.•Partial network aggregation to improve communication efficiency and personalization.•State-of-the-art performance in MRI synthesis for common and variable tasks across sites.