Research Of MRI Reconstruction Based On Compressive Sensing Adaptive Sampling

Magnetic resonance imaging(MRI)is one of the most important imaging techniques in medical diagnosis by its virtues of non-invasive,non-ionizing radiation and high imaging contrast.Dynamic magnetic resonance imaging is a kind of magnetic resonance imaging which recording the changes of a certain part of the body with time.It is more accurate than the description of the lesion by static magnetic resonance imaging.Fast and high quality of imaging are the two important needs of clinical medicine,MRI based on compression sampling can greatly improve the speed of sampling,but the existing compression magnetic resonance imaging technology ignores the similarity of the energy distribution among frames in the same set of dynamic MRI data.At the same time,the "new data" in the non-sampling position generated during the iteration period is not fully utilized and the different contributions of the different frequency data are neglected,which limits the further improvement of the reconstructed image quality.Based on the above considerations,this paper proposes an adaptive sampling method and reconstruction method for dynamic magnetic resonance imaging using the gradient descent method,the distribution of K-space data and the similarity between dynamic magnetic resonance images.The main works are as follows:1.Propose a method of magnetic resonance image reconstruction based on frequency division of non-sampling region data.In order to improve the quality of image reconstruction,we first analyze the "new data" which generated by the image non-sampling area during the reconstruction process.Second,according to its low-frequency,high-frequency image informations containing different contributions for reconstruction,we divide the "new data" into two parts and propose two penalty items.Finally,a new objective function and reconstruction method are proposed,which can improve the quality of static magnetic resonance image reconstruction.2.Propose an adaptive sampling method based on dynamic magnetic resonance images energy distribution.Considering that the traditional sampling mechanism uses the same sampling template or random sampling template for each frame of the dynamic K-space data.There is a drawback that the energy distribution of the current K-space can't be predicted,so traditional sampling mechanism can't collect the most important K-space data.First,thecertain dynamic K-space data of a specific body part is presented with a unique energy distribution.Second,through the previous frame sampling data energy and position relationship,we can adjust the current sampling probability function.Finally,the sampling template of the current frame is generated by the newly obtained sampling probability function.Through this adaptive sampling method,the traditional sampling method can be improved,and the more reasonable observation data in the current K-space can be collected to obtain the data which can best reflect the energy distribution,so as to enhance the reconstruction quality of the dynamic images.3.Propose a reconstruction method based on dynamic magnetic resonance images similarity.Considering that the adjacent two frames of the dynamic magnetic resonance images are similar.First,mark the non-sampling position of the current frame and obtain the Fourier coefficient matrix of the previous frame image during the iteration,so,the non-sampling location data of the current frame can be extracted.The position corresponds to the data of the matrix serves as a reference for un-sampled data.Second,according to the similarity of adjacent frames,a new target function is obtained by combining the “new data” construct constraint term in the reconstruction process of the current frame.Finally,realize the reconstructing method based on dynamic magnetic resonance images similarity,then reconstruction results shows the quality of dynamic images have been greatly improved.