MRI Reconstruction Based On Simultaneous Multislice Excitation And Parallel Imaging

The long scanning time has always been the bottleneck of magnetic resonance imaging,which limits its clinical application.Parallel imaging of magnetic resonance is a kind of fast imaging technology,which compensates the traditional gradient field encoding by utilizing the array coil with multiple channels,therefore,to shorten the scanning time through K-space undersampling.GRAPPA is one of the most used reconstruction methods for parallel imaging.The method fits the linear reconstruction kernel which is used to estimate the missing lines by the low-frequency harmonic signal in the center of K-space.Parallel imaging has the drawback of noise amplification during reconstruction.The acceleration factor of parallel imaging is limited by the number of coils and its signal-to-noise ratio is inversely proportional to R^1/2.Simultaneously Multi-Slice imaging（SMS）is is a fast imaging technology that scans multiple slices simultaneously by one excitation of composite RF pulse.The SMS reconstruction uses spatial encoding in the slice direction to separate each slice.Its signal-to-noise ratio is not affected by the slice-acceleration factor and can compensate the signal-to-noise ratio loss of R^1/2 in parallel imaging.Slice-GRAPPA is one of the most commonly used traditional reconstruction methods for SMS imaging.The approach of Slice-GRAPPA is similar to GRAPPA in principle,except that its reconstruction kernel is a fit from the aliased signal to each reference slice,and the reconstruction is more direct.When the slice-acceleration factor is higher,Slice-GRAPPA reconstruction has obvious noise amplification and low signal-to-noise ratio of reconstructed images.Virtual Conjugate Coil（VCC）technology generates additional virtual data for each channel using the characteristic of conjugate symmetry of complex data in K-space,which doubles the original dataset for all coils and improves the signal-to-noise ratio of parallel imaging reconstruction.The recently proposed Robust Artificial-neural-networks for K-space Interpolation（RAKI）is a non-linear reconstruction method based on artificial neural network.It is applied to SMS reconstruction and significantly improves the reconstruction quality compared with the linear Slice-GRAPPA method.At the same time,because a large number of training data sets of patients from other sources are not needed,the training time is significantly shortened compared with the reconstruction method based on general deep learning.Based on the advantages of VCC technology and the artificial neural network method of nonlinear reconstruction,this paper studies a better reconstruction method for SMS imaging,that is,VIRtual conjuGate coIls Neural-network InterpolAtion（VIRGINIA）.The network input and output of VIRGINIA are multi-coil magnetic resonance data in K-space.Specifically,the training data is constructed from the multi-coil low-resolution data of magnetic resonance during pre-scan process.The low-resolution multi-coil data in K-space is processed by the following procedure of zero filling,concatenation for all slices in image space and inverse fourier transform,and then it can be acquired of multi-coil K-space data of the concatenated slices which was undersampled by the slice-acceleration factor for acquiring the training input of network and the remain data as the training output of network;Next,the VCC technology is utilized to generate the virtual multi-coil data from the pre-scan multi-channel magnetic resonance low-resolution data and the virtual data undergoes the same process as the pre-scan multi-coil magnetic resonance low-resolution data to obtain another virtual input and virtual output for the training network;Then,The convolution neural network is trained with the two training data to obtain a network model;The aliased multi-coil SMS signal is feed into the trained model to obtain the multi-channel output data following by the combination with the input of network for obtaining the multi-channel K-space data of the multi-slice connected image,and the reconstructed image for each slice can be finally obtained through post-processing operation.The main contents of this paper are divided into the three parts as following:（1）The reconstruction method of Slice-GRAPPA using VCC technology in SMS is studied.By reconstruction from SMS data of the slice-acceleration factor of 3,4,5 and 7times,it was found that the introduction of VCC technology can reduce the local noise of reconstructed image of Slice-GRAPPA,thus imporove the reconstruction quality.（2）It is studied that the reconstruction performance improvement of RAKI and VIRGINIA methods based on neural network compared with Slice-GRAPPA method.Through reconstruction from the SMS data of the slice-acceleration factor of 2,3,4,5 and 15 times using the 3 methods,it was found that the reconstruction performance of RAKI and VIRGINIA methods based on neural network improved in terms of both reconstructed images and performance indexes of PSNR,SSIM and RMSE.The proposed VIRGINIA method also had quality improvement compared with RAKI method.（3）VIRGINIA method is proposed and tested.Compared with RAKI method,the quality of reconstructed images using VIRGINIA is better.Through comparison on noisy SMS data with slice-acceleration factors of 2,3 and 4 times,it was found that VIRGINIA reconstruction results are better than RAKI and are also robust to noise.In order to further verify the performance improvement of VIRGINIA method,64 pieces of noisy SMS data with slice-acceleration factor of 2,3 and 4 times are reconstructed respectively,and the improvement effect of VIRGINIA method is found to be stably better than RAKI method through analysis of performance indexes.To sum up,the proposed VIRGINIA method based on VCC technology and convolutional neural network in this paper has better SMS imaging reconstruction quality compared with traditional Slice-GRAPPA method and RAKI method.RAKI takes second place and the reconstruction quality of Slice-GRAPPA method is the worst.