Correction Of Motion Artifacts In Chemical Exchange Saturation Transfer Magnetic Resonance Imaging

Chemical exchange saturation transfer?CEST?magnetic resonance imaging?MRI?is a new type of molecular imaging technology that can perform in vivo imaging of proteins and sugars containing exchangeable protons in molecules,and has great clinical application value.However,the time required for CEST MRI is relatively long,leading to motion artifacts.This thesis aims to research an image registration method suitable for CEST MRI motion artifact correction,which will help improve the accuracy of CEST analysis.At present,the image registration method based on traditional gradient descent is used to correct local CEST MRI motion artifacts,which is easy to fall into local optimization,resulting in low registration accuracy.For this problem,two solutions are given:?1?A CEST MRI image registration method based on adaptive stochastic gradient descent is proposed.Since the traditional gradient descent registration method uses a relatively fixed learning step size,it is easy to fall into a local optimum under the condition of multiple extremes,resulting in a low registration accuracy.In response to this shortcoming,this thesis adopts an adaptive stochastic gradient descent optimization strategy.Based on the inner product of the approximate gradients in the last two iteration optimizations,the learning step size is adaptively predicted to avoid falling into a local optimum and improve registration accuracy.Using the mouse 2-dimensional brain CEST Z-spectrum data as an example,different registration experiments were performed to simulate rigid body motion.The experimental results show that compared with the traditional gradient descent registration method,the MTR_asym?3.5ppm?image,Z-spectrum and pixel frequency offset spectrum of the proposed method have higher similarity with the original static data.Moreover,the correlation coefficient is higher and the root mean square error is lower.These verify that the proposed method has more accurate and robust advantages for the motion artifact correction of CEST MRI.?2?A hybrid registration method based on Fourier transform?FT?image correlation and traditional gradient descent is proposed.From another perspective,this method solves the problem that traditional gradient descent is easy to fall into local optimization leading to low registration accuracy:using FT image correlation and traditional gradient descent registration methods,registration is performed in a variety of combinations and multiple results are obtained,and then select the one with the largest correlation between the reference image as the final registration result.This method can provide a suitable initial spatial transformation before traditional gradient descent registration according to different motion situations,or perform finer correction after its registration,thereby reducing the impact of traditional gradient descent falling into local optimum and improving registration accuracy.Similarly,using the mouse 2-dimensional brain CEST Z-spectrum data,different registration experiments were performed to simulate rigid body motion.The experimental results show that,compared with the traditional gradient descent registration method,the registration result of this method is more similar to the original static data,which verifies that the method is also more accurate and robust in the correction of CEST MRI motion artifacts.