Premium
Direct and accelerated parameter mapping using the unscented Kalman filter
Author(s) -
Zhao Li,
Feng Xue,
Meyer Craig H.
Publication year - 2016
Publication title -
magnetic resonance in medicine
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.696
H-Index - 225
eISSN - 1522-2594
pISSN - 0740-3194
DOI - 10.1002/mrm.25796
Subject(s) - undersampling , imaging phantom , kalman filter , computer science , algorithm , extended kalman filter , unscented transform , nonlinear system , control theory (sociology) , ensemble kalman filter , artificial intelligence , physics , control (management) , quantum mechanics , optics
Purpose To accelerate parameter mapping using a new paradigm that combines image reconstruction and model regression as a parameter state‐tracking problem. Methods In T2 mapping, the T2 map is first encoded in parameter space by multi‐TE measurements and then encoded by Fourier transformation with readout/phase encoding gradients. Using a state transition function and a measurement function, the unscented Kalman filter can describe T2 mapping as a dynamic system and directly estimate the T2 map from the k‐space data. The proposed method was validated with a numerical brain phantom and volunteer experiments with a multiple‐contrast spin echo sequence. Its performance was compared with a conjugate‐gradient nonlinear inversion method at undersampling factors of 2 to 8. An accelerated pulse sequence was developed based on this method to achieve prospective undersampling. Results Compared with the nonlinear inversion reconstruction, the proposed method had higher precision, improved structural similarity and reduced normalized root mean squared error, with acceleration factors up to 8 in numerical phantom and volunteer studies. Conclusion This work describes a new perspective on parameter mapping by state tracking. The unscented Kalman filter provides a highly accelerated and efficient paradigm for T2 mapping. Magn Reson Med 75:1989–1999, 2016. © 2015 Wiley Periodicals, Inc.