MR fingerprinting provides quantitative parameter maps from a single acquisition, but it also has the potential to reduce exam times by replacing traditional protocol sequences with synthetic contrast-weighted images. We present an empirical "artifact noise" model that makes it possible to train neural networks that successfully transform noisy and aliased MRF signals into parameter maps, which are then used to synthesize contrast-weighted images. We also demonstrate that a trained neural network can directly synthesize contrast-weighted images, bypassing incomplete simulation models and their associated artifacts.
This abstract and the presentation materials are available to members only; a login is required.