Wavelet-denoising of complex magnetic resonance images

Wavelet and wavelet packets may be used to "denoise" magnetic resonance images by finding image basis functions that preferentially concentrate signal relative to background Gaussian noise. At very low SNR (<5), however, standard magnitude MRI images have skewed Rician noise statistics which degrade denoising performance. We hypothesized that wavelet-packet denoising techniques would yield better edge preservation if performed on the raw real and imaginary images prior to rectification. To test this hypothesis, synthetic, phantom, and volunteer cardiac images were denoised either in the complex or magnitude domains. The techniques were compared with regard to signal-to-noise, signal-to-background, contrast-to-noise and edge blurring. RESULTS: While magnitude and complex denoising both significantly improved SNR, SBR, and CNR, complex denoising yielded sharper edge resolution and feature extraction. Wavelet packet denoising of MRI images should be performed prior to signal rectification in very low SNR applications.