Differential contrast enhancement using conditional deep learning for Gadolinium dose reduction in brain MR

In this work, we propose a novel deep learning (DL)-based method to reduce the dose of Gadolinium-based contrast agents administered in brain MRI examinations. In contrast to recent DL approaches, we explicitly focus on accurately predicting contrast enhancement signals and synthesizing realistic images, leveraging contrast signals from subtraction images of pre- and post-contrast T1-weighted image pairs. By training our model to only extract and enhance contrast signals, and by conditioning its layers on relevant physical parameters, we demonstrate its effectiveness across diverse datasets, including data at different dose levels from various scanners, field strengths, and contrast agents.