Modified POCS Based Reconstruction for Compressed Sensing in MRI

Compressed Sensing usage in Magnetic Resonance Imaging greatly enhances the utility and effectiveness of the data acquisition process. Magnetic resonance images can be safely reconstructed from the sparse randomly under-sampled data, using a non-linear recovery technique. There are several reconstruction algorithms used in Compressed Sensing to obtain a Magnetic resonance image from highly under-sampled data e.g. conjugate gradient, separable surrogate function, projection on convex sets and many more. This research work aims to review some of these compressed sensing reconstruction algorithms, analyze and evaluate their performance and to propose a similar technique that satisfies the requirements of compressed sensing and outperforms the methods already in use. A modified Projection over convex set technique has been proposed which involves the application of discrete cosine transform to standard algorithm. The proposed technique is evaluated using the actual data obtained from the Magnetic resonance imaging scanner at St. Mary's Hospital London. It can be concluded from the results that the proposed method exhibits better performance in terms of improved artifact power, signal-to-noise ratio and is computationally less intensive as compared to other reconstruction algorithms.