Image distortion correction method for echo planar imaging without reference scanning under inhomogeneous magnetic field

Abstract

The invention discloses an echo planar imaging no-reference scanned image distortion rectification method under nonuniform magnetic field, and belongs to the field of nuclear magnetic resonance imaging reconstruction method. According to the method, a single scanning echo planar imaging sequence is adopted for coding and sampling of signals, after one time of 90 DEG pulse excitation, a series of echo gradient sampling are carried out so as to obtain the signals in the whole k space, influences of nonuniform magnetic fields are accumulated in the whole sampling process. Singles obtained via sampling are subjected to two dimensional Fourier transform of image reconstruction, the real part and the imaginary part are taken as inputs of residual network. Neural network training is realized based on simulated data sets. At first random generation of labels is carried out, and then the image of distortion corresponding to the labels is achieved based on sampling conditions; batch generation of a plurality of groups of data is realized, and the obtained data is adopted for network training; network hyper-parameter is adjusted to ensure convergence of training error is realized; and at lastafter network training, network parameters are introduced, and input of actual measurement single scanning echo planar imaging is carried out so as to obtain orthoscopic images.

Description

technical field [0001] The invention relates to a nuclear magnetic resonance imaging reconstruction method, in particular to a residual neural network-based echo plane imaging non-reference scanning image distortion correction method under an inhomogeneous magnetic field. Background technique [0002] Because Magnetic Resonance Imaging (MRI) has the characteristics of less damage to soft tissue and higher resolution, it plays an important role in functional MRI (fMRI) and clinical disease diagnosis. [1,2] . In some traditional sequences such as multi-scan gradient echo, multiple RF pulse excitations are required to obtain a complete k-space signal [3] . Due to the long time interval between two consecutive scans, the total scan time is long, causing discomfort to the patient, and movement may occur during the scan, resulting in motion artifacts in the image. Therefore, the development of ultra-fast imaging technology is of great significance, especially in functional magn...

AI Technical Summary

Problems solved by technology

These methods have some objective limitations that lead to unsatisfactory correction results. For example, the point spread function method is sensitive to regular parameters, the nonlinear registration method only corrects shape distortion and the correction effect depends on the registration algorithm, and the PSF method cannot correct the frequency dimension. Distortion

Images