Iterative Reconstruction Algorithm to Remove Noise and Artifacts in MRI

Volume 4 Issue 3 July - September 2017

Research Paper

Iterative Reconstruction Algorithm to Remove Noise and Artifacts in MRI

C. Anjanappa*, H. S. Sheshadri**

* Assistant Professor, Department of Electronics and Communication Engineering, The National Institute of Engineering, Mysuru, Karnataka, India.

** Professor, Department of Electronics and Communication Engineering, PES College of Engineering, Mandya, Karnataka, India.

Anjanappa, C. and Sheshadri, H.S. (2017). Iterative Reconstruction Algorithm to Remove Noise and Artifacts in MRI. i-manager’s Journal on Image Processing, 4(3), 8-15. https://doi.org/10.26634/jip.4.3.13918

Abstract

Compressive technique is also known as compressive sampling in signal processing domain and its recent algorithm can be integrated into the MRI reconstruction pipeline for acquiring the complete MR data. The proposed work is based on the recursive iterative adaptive filtering technique based on every injection of random noise in the unobserved portion of the spectrum. The proposed volumetric filter attenuates the noise and relevant features from the incomplete k-space measurements which is required for the clinical practice. The proposed algorithm is based on stochastic approximation method with regularization parameter enabled by a spatial adaptive block-wise volumetric filter. The effectiveness of the proposed reconstruction algorithm is compared with CT reconstruction algorithm radon inversion from sparse projections, spiral, radial, and limited angle tomography. From the reconstruction algorithm it is observed that it achieves exact reconstruction from phantom data even at small projections. The accuracy of this method is to compete with the compressed sensing field. The proposed algorithm is tested on different sampling trajectories, especially on non- Cartesian data and reconstruction of volumetric phantom data with non-zero phase from noisy and incomplete Fourierdomain (k-space) measurements. Experimental results demonstrate that its performance is evaluated by PSNR, SSIM, and execution time.