MRF denoising with compressed sensing and adaptive filtering

Abstract

The recently proposed Magnetic Resonance Fingerprinting (MRF) technique can simultaneously estimate multiple parameters through dictionary matching. It has promising potentials in a wide range of applications. However, MRF introduces errors due to undersampling during the data acquisition process and the limit of dictionary resolution. In this paper, we investigate the error source of MRF and propose the technologies of improving the quality of MRF with compressed sensing, error prediction by decision trees, and adaptive filtering. Experimental results support our observations and show significant improvement of the proposed technologies.