Deep learning magnetic resonance spectrum reconstruction method based on sparse representation

Abstract

The invention discloses a deep learning magnetic resonance spectrum reconstruction method based on sparse representation, and relates to a magnetic resonance spectrum reconstruction method. The methodcomprises the steps of (1) simulating to generate a fully-sampled time domain signal by utilizing an exponential function characteristic of a time domain signal of a magnetic resonance spectrum; 2) carrying out undersampling on the time domain signals, and establishing a training set comprising wave spectrums corresponding to the full-sampling time domain signals, the undersampling time domain signals and corresponding undersampling templates; 3) designing a deep learning network model based on sparse representation, a feedback function of the network and a loss function; 4) solving an optimal parameter of the deep learning network based on sparse representation by utilizing the training set obtained in the step 2); and 5) inputting the undersampled magnetic resonance time domain signal to be reconstructed into the network to reconstruct the magnetic resonance spectrum. The deep neural network is designed by constraining the sparsity of the magnetic resonance frequency domain signal and taking the traditional optimization method as guidance, so that the method has the characteristics of high reconstruction speed, high reconstruction quality and strong network interpretability.

Description

technical field [0001] The invention relates to a magnetic resonance spectrum reconstruction method, in particular to a deep learning magnetic resonance spectrum reconstruction method based on sparse representation. Background technique [0002] Magnetic resonance spectroscopy can provide atomic-level information on molecular structure and is an important analytical tool in the fields of medicine, chemistry and life sciences. In the magnetic resonance experiment, the sampling time of the signal increases with the increase of the resolution and the sampling dimension. Non-uniform sampling techniques are widely used to accelerate the acquisition of experimental data by acquiring partial data, but it requires advanced spectral reconstruction methods to obtain complete spectra. [0003] In spectral reconstruction, some researchers use the properties of magnetic resonance time and frequency domain signals to reconstruct the spectrum. One of the better ones utilizes the sparse c...

AI Technical Summary

Problems solved by technology

[0005] However, in the reconstruction of magnetic resonance spectroscopy, the reconstruction quality of existing methods still needs to be improved, and there is no method to use the sparse characteristics of frequency domain signals to establish a deep learning neural network to achieve fast and high-quality magnetic resonance spectroscopy reconstruction

Images