A Pitch-level Sparse MRI Reconstruction Via Dictionary Learning

With the advantages of non-invasive of human body and can be imaged in any position, MRI technology becomes an important basis for clinical diagnosis gradually, especially for imaging that with high requirements of non-destructive to the body, like brain imaging, it’s widely used. But the large amount of data will lead to a long scan time, affecting the time of diagnosis. Applying compressed sensing theory to reduce the amount of data collected in the imaging process can reduce MRI scan time. Then reconstruct the image using some non-linear algorithms. That’s what called sparse MRI which received many interests of researchers in recent years.In this paper, First sample experiments show the sparsity of medical images which point out the feasibility of the idea that applying the sparse optimization techniques into magnetic resonance imaging. Second, a pitch-level sparse method with dictionary learning for magnetic resonance image reconstruction was proposed. In this method, Image sparse constrained the lo-norm problem will be relaxed by the lp-norm problem to make it easy to calculate and maintain the sparsity of the good situation at the same time. Then use a weighted the l1-norm problem to approximate the lp-norm problem, making it a convex optimization problem, which is easy to compute. Also in this method, an image will be split into thin pieces other than square blocks. And for dictionary learning, this method trains an adaptive dictionary considering characters of interest images.This paper discusses the earlier research work and conclusions of the author in later section.