An efficient framework for denoising of diffusion-weighted mr brain images

Abstract

Improving the spatial resolution of diffusion weighted imaging (DWI) is quite significant and challenging task in recent days. Many researchers have made substantial attempts to face this challenge where the researchers aimed at accurate detection of elusive lesions and more faithfully solving fibre tracts of white matter (WM). In practice, signal-to-noise-ratio (SNR) restriction is the major pertain of high-resolution (HR) DWI, which is later counterbalance the benefits of high spatial resolution. In spite of the fact that the SNR of DWI data can be enhanced with the approach of denoising in the post-processing, the conventional approaches of denoising might possibly mitigate the anatomic solvable of HR imaging data. In addition, a signal bias that is caused by the non-Gaussian noise is also not possible to adjust every time. Therefore, this article presents an efficient framework for DW-MR brain image denoising that utilizes non-local Euclidean medians (NLEM) in non-subsampled contourlet (NSC) domain, which works based on multi-scale decomposition and directionality. Simulation results are tested with different kind of DW-MR images and the proposed NLEM-NSC approach shown the superiority over existing denoising approaches. Further, it is also provided that the qualitative analysis to disclose the robustness and effectiveness of NLEM-NSC method at lower strengths of noise.