Research On MRI Reconstruction Based On Image Domain And Fourier Domain Fusion

In recent years,MRI has become an extremely important medical image analysis tool because it can provide more adequate pathological information.However,due to the relatively slow imaging speed of current magnetic resonance detectors,which may affect the final medical diagnosis,so increasing the imaging speed of MRI has become a hot research topic.At present,there are two main ways to speed up the speed of magnetic resonance imaging:the first is to achieve rapid imaging by transforming the NMR hardware;the second is to directly reduce the way of sampling data in K-space,and later by design reconstruction algorithm to reconstruct the complete MR image from partial K-space data.Among them,the second type of method not only costs less than the first type but also ensures high image quality.Therefore,the design of MRI reconstruction algorithm has become the research hotspot of current MRI accelerated imaging.Especially with the rapid development of deep learning in recent years,MRI reconstruction algorithms based on deep learning have attracted much attention.Although some achievements have been achieved,there are still many challenges.In this paper,based on the shortcomings of current MRI reconstruction algorithms,an MRI reconstruction method based on space-frequency domain fusion is proposed.The main innovations are as follows:Firstly,the current mainstream deep learning-based MRI reconstruction method generally uses a real convolutional network model,which does not effectively extract complex-valued features.This paper uses a complex convolutional network to learn more complex representations of complex-valued K-spaces and improve MRI reconstruction performance.Secondly,the current MRI reconstruction method generally transforms the original data from K-space(ie,frequency domain data)to the image domain(ie,spatial domain data),and then reconstructs according to the image representation model,and does not effectively exploit and utilize the valid information of the K-space data itself.In this paper,considering the importance of the original K-space data,the intrinsic structure of K-space data is mined by the powerful nonlinear fitting ability of the deep learning model.At the same time,combined with the image domain reconstruction process,a deep cascading network with dual domain fusion is proposed.The model can adaptively mine the data information implied in the K space and the image domain,complement each other's advantages,improve the MRI reconstruction performance,and provide a new model for MRI reconstruction.This paper verifies on a complex MR image dataset.Experiments show that the reconstruction model using complex convolution is better than the reconstruction model using real convolution.The reconstruction performance of the dual domain reconstruction model in this paper has a comparative advantage compared with the classical algorithm,and the reconstruction performance is significantly improved compared to the single domain reconstruction model.