Wavelet packet denoising of magnetic resonance images: Importance of Rician noise at low SNR

Abstract

Wavelet packet analysis is a mathematical transformation that can be used to post-process images, for example, to remove image noise ('denoising'). At a very low signal-to-noise ratio (SNR <5), standard magnitude magnetic resonance images have skewed Rician noise statistics that degrade denoising performance. Since the quadrature images have approximately Gaussian noise, it was postulated that denoising would produce better contrast and sharper edges if performed before magnitude image formation. Signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and edge blurring effects of these two approaches were examined in synthetic, phantom, and human MR images. While magnitude and complex denoising both significantly improved SNR and CNR, complex denoising yielded sharper edges and better low- intensity feature contrast.