Matrix completion-based reconstruction for undersampled magnetic resonance fingerprinting data

Magn Reson Imaging. 2017 Sep:41:41-52. doi: 10.1016/j.mri.2017.02.007. Epub 2017 Mar 3.

Abstract

An iterative reconstruction method for undersampled magnetic resonance fingerprinting data is presented. The method performs the reconstruction entirely in k-space and is related to low rank matrix completion methods. A low dimensional data subspace is estimated from a small number of k-space locations fully sampled in the temporal direction and used to reconstruct the missing k-space samples before MRF dictionary matching. Performing the iterations in k-space eliminates the need for applying a forward and an inverse Fourier transform in each iteration required in previously proposed iterative reconstruction methods for undersampled MRF data. A projection onto the low dimensional data subspace is performed as a matrix multiplication instead of a singular value thresholding typically used in low rank matrix completion, further reducing the computational complexity of the reconstruction. The method is theoretically described and validated in phantom and in-vivo experiments. The quality of the parameter maps can be significantly improved compared to direct matching on undersampled data.

Keywords: MR Fingerprinting; Matrix completion.

MeSH terms

  • Algorithms
  • Artifacts
  • Brain / diagnostic imaging*
  • Calibration
  • Healthy Volunteers
  • Humans
  • Image Processing, Computer-Assisted / methods*
  • Magnetic Resonance Imaging / methods*
  • Models, Statistical
  • Phantoms, Imaging
  • Reproducibility of Results
  • Software