Purpose : Quantitative synthetic magnetic resonance imaging (MRI) enables the synthesis of various contrast-weighted images based on simultaneous relaxometry. Herein, we developed a deep learning algorithm to generate magnetic resonance angiography (MRA) from three-dimensional (3D) synthetic MRI data. Materials and Methods : Eleven healthy volunteers underwent time-of-‰ight (TOF) MRA sequence and 3D synthetic MRI sequence, i.e., 3D-QALAS. Five raw 3D-QALAS images were used as inputs for deep

more »

... rning (DL-MRA). A simple linear combination model was prepared for comparison (linear-MRA). Three-fold cross-validation was performed. The peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) were calculated for DL-MRA and linear-MRA against TOF-MRA. The overall image quality and branch visualization were scored on a 5-point Likert scale by a neuroradiologist blinded to the data. Results : The PSNR and SSIM were signiˆcantly higher for DL-MRA that those of linear-MRA. Overall image quality and branch visualizations were comparable for DL-MRA and TOF-MRA. Conclusion : Deep learning based on 3D-synthetic MRI enabled the generation of MRA with quality equivalent to that of TOF-MRA.