Magnetic Resonance Fingerprinting (MRF)-based Virtual Contrast-enhanced MRI Synthesis using Deep Transfer Learning

Motivation: The use of gadolinium-based contrast agents in MRI poses risks of renal damage and nephrogenic systemic fibrosis. Current virtual-contrast-enhanced deep-learning synthesis methods depend on qualitative and multiparametric MRI, leading to inter-facility variation and prolonged scanning time. Goal(s): We aimed to develop an MRF-based DL model for virtual contrast-enhanced (VCE) MRI synthesis to improve consistency and efficiency. Approach: To tackle the challenge of limited MRF training data, we proposed a deep transfer learning method to synthesize contrast-enhanced T1-weighted images from contrast-free MRF. Results: The MRF-based model effectively reconstructed contrast and spatial features, achieving performance comparable to conventional T1w/T2w-based models while reducing acquisition time. Impact: MRF-based VCE-MRI has the potential to enhance patient safety and streamline clinical workflows by eliminating the need for contrast agents. Offering comparable synthetic accuracy to conventional T1w/T2w-based models, the MRF-based approach features a more efficient single-sequence acquisition.