Deep learning for 3D MR fingerprinting: a dual pathway parameter mapping and reconstruction approach

摘要

The signature element of MR Fingerprinting (MRF), the matching of acquired signal evolutions to a precomputed dictionary, has proven to allow reliable and accurate parameter quantification. Its heavy memory and computational requirements, however, make it very inefficient. Compression algorithms~1 and deep learning approaches2 have hence emerged to accelerate MRF reconstruction. Inspired by their success, we address the inherent limitation of MRF by proposing a hybrid neural network architecture which is capable of:-Efficiently compressing the MRF data in the time domain.