Hypercomplex magnetic resonance spectrum reconstruction method based on deep learning

Abstract

The invention discloses a supercomplex magnetic resonance spectrum reconstruction method based on deep learning, and relates to a magnetic resonance spectrum. The method comprises the following stepsof generating a full-sampling time domain signal of the supercomplex magnetic resonance spectrum according to the magnetic resonance spectrum signal index model; performing under-sampling on the full-sampling time domain signal according to an under-sampling template, and performing zero setting on the positions of data points which are not collected to obtain an under-sampling time domain signalof the supercomplex magnetic resonance spectrum; converting the under-sampling time domain signals and the corresponding full-sampling time domain signals into complex time domain signals, respectively performing Fourier transform to obtain under-sampling and full-sampling magnetic resonance spectra of corresponding frequency domains, and generating a training set for supercomplex spectrum reconstruction; constructing a deep learning network for supercomplex magnetic resonance spectrum reconstruction, and training the deep learning network by adopting the obtained training set to obtain trained network parameters for supercomplex magnetic resonance spectrum reconstruction; and reconstructing the under-sampling time domain signal of the supercomplex magnetic resonance spectrum by using thetrained network to obtain a complete supercomplex magnetic resonance spectrum.

Description

technical field [0001] The present invention relates to magnetic resonance spectroscopy, in particular to a method for reconstructing hypercomplex magnetic resonance spectroscopy based on deep learning. Background technique [0002] Magnetic Resonance Spectroscopy (MRS) can clearly elucidate the molecular structure and has been widely used in the fields of medicine, chemistry and biology. In undersampled magnetic resonance spectroscopy reconstruction, how to reduce the sampling time while ensuring the quality of the reconstructed spectrum is an important technical problem. [0003] Traditional MRI reconstruction methods mainly use the mathematical properties of MRI time or frequency signals to reconstruct the spectrum. Qu Xiaobo et al. (Qu X, Mayzel M, Cai J, Chen Z, Orekhov V. Accelerated NMRspectroscopy with low-Rank reconstruction [J]. Angewandte Chemie International Edition, 2015, 54(3): 852-854.) proposed a The MRI spectral reconstruction method based on the low-rank ...

AI Technical Summary

Problems solved by technology

However, the singular value decomposition in the iterative process of this kind of low-rank Hankel matrix reconstruction method consumes a lot of time, thus resulting in a long spectral reconstruction time

Guo Di et al. (Guo D, Lu H, Qu X.A fast low rank Hankelmatrix factorization reconstruction method for non-uniformly sampled magneticresonance spectroscopy[J].IEEE Access,2017,5:16033-16039.) successfully decomposed the low rank matrix And introduce parallel computing to avoid singular value decomposition with high time complexity, but it still takes a certain amount of time

Images