An Improved Reconstruction Method for Compressively Sampled Magnetic Resonance Images Using Adaptive Gaussian Denoising

摘要

In this paper, an improved Compressively Sampled Magnetic Resonance Imaging (CS-MRI) method that suppresses reconstruction noisy artifacts is proposed. The algorithm involves random undersampling of the k-space data of an MR image followed by reconstruction of the k-space data coefficients in a wavelet sparsifying domain. The high frequency noise in the reconstructed coefficients is suppressed in the Fourier transform domain by an adaptive Gaussian low pass filter. The reconstructed MR image is finally obtained by Inverse Discrete Fourier Transformation (IDFT) of the denoised k-space data. Experimental results demonstrate the robustness of the proposed method to sub-Nyquist sampling associated artifacts in terms of terms of Structural SIMilarity (SSIM) index and Peak Signal to Noise Ratio (PSNR) assessments.