Nyquist Ghost Correction Without Reference Scan In Echo Planar Imaging Based On Deep Neural Network

Magnetic resonance imaging(MRI)can provide high-resolution anatomical images and functional images,which makes it one of the most significant techniques in modern clinical diagnosis.To obtain different anatomical or physiological characteristics to help clinical diagnosis,the image contrast can be weighted flexibly through adjusting scan parameters.However,conventional MRI acquisition generally require long scan time.To address this issue,many methods have been brought up to accelerate acquisition.Echo planar imaging(EPI),which was proposed by Sir Mansfield in 1977,gains the most popularity for its rapid acquisition and friendly implementation.However,during EPI acquisition,the quick switch of pulse field gradients of opposite polarities causes misalignment between odd lines and even lines in k space,leading to the Nyquist ghost.Though many methods have been proposed to remove the Nyquist ghost,they are generally based on reference scan or iteration correction,which would take more time or reduce temporal resolution.Deep neural network is an important branch in machine learning,and presents a great protential in areas such as speech recognition,image vision and natural language processing.This thesis focuses on the correction of Nyquist ghost in EPI without reference scan based on deep neural network.The main contents of this thesis are as follows:1.Firstly,the background and principle of MRI are introduced.Then,two kinds of EPI methods are expounded.The encoding and decoding processes of spin signal are interpreted,along with the derivation of signal formula.The reason for the occurrence of Nyquist ghost and several commonly used Nyquist ghost correction methods are elaborated.Finally,relevant theories of deep neural network are introduced,together with convolutional neural network which is used in this thesis.2.We explore Nyquist ghost correction method without reference scan in single-channel EPI based on deep neural network.For most exsiting correction methods,the modification of echo chain for reference scan or iteration operation is required,which increases the acquisition time and decrease the temporal resolution.To tackle the problem,the deep neural network is utilized to learn the characteristics of phase errors of EPI images with Nyquist ghosts.The phase error maps of experimental data can then be obtained and used for data post-processing.In this way,the Nyquist ghost can be eliminated without the need of reference scan,thus achieving desirable correction results.The validation of the proposed method is demonstrated through simulation and experimental results.3.We further evaluate the feasibility of the proposed method by correcting Nyquist ghost in multi-channel EPI data.By utilizing the characteristic of parallel imaging,we obtained the needed dataset effectively to extract the phase error map of EPI images and optimized the loss function.The comparison of phase error map generated by our method with that provided by conventional method verifies the effectiveness of the proposed method on obtaining phase error map.The feasibility of the proposed method for Nyquist ghost correction on multi-channel EPI images is demonstrated through simulation and experimental results.