Compressed Sensing-magnetic Resonance Image Reconstruction Based On Analysis Model

Magnetic resonance imaging (MRI)) using magnetic resonance principle to obtainimages inside the human body has a very important role in medicine. Reduce the dataacquisition time to improve the imaging speed, has been a hot topic in the field of MRItechnology.Recently proposed compressed sensing (CS) thory suggest that sparse orcompressible signals can be accurately reconstructed from a small sampling rate below theNyquist sampling data. Currently sparse model is divided into synthetic and analyticalmodels. Synthetic model has been very mature in mathematical theory and practicalapplication. However, the analytical model is still in its infancy.Therefore, with different prior information, this paper mainly research and discuss thefollowing three methods, which based on the analytical meodel:（1）GSTV reconstruction method. Based on CS thory, this model combines the TVnorm and group sparse effectively. Using traditional methods to recover the image as areference image and use the conjugate gradient method to get the target image. This is aniterative process, in which the reference picture is updated each time, and thus the priorinformation will be more accurate and more precise image reconcstruction followed.Simulation results show that this method has a higher image quality in varying degrees atdifferent sampling rates, comparing with other reconstruction methods.（2）The mixed difference method based on prior support information. Thisreconstruction model makes the support information and the mixed difference method intoone frame. The initial image was estimated by the traditional method. Using the initialimage to compute the support information.In this model, the separate Bregman algorithmwas used. Experimental results show that this method improve the quality of imagereconstruction, compared with the hybrid differential reconstruction method.（3）The second order total generalized variation (TGV) based on prior information.This model incorporates prior information and TGV, using the structural information of theimage, so that making the image reconstruction more accurately. First, get an initialestimated image by the traditional CS-MRI reconstruction method, and then select theappropriate support information. Experimental results show that this method reduce theamount of data effectively and improve the quality of image reconstruction.