Full-spectrum comparison of denoising algorithms for real-time magnetic resonance imaging acoustics

• Published in: The Journal of the Acoustical Society of America Vol. 142; no. 4; p. 2542
• Main Authors: Klock, William, Barlaz, Marissa
• Format: Journal Article
• Language: English
• Published: 01-10-2017
• ISSN: 0001-4966; 1520-8524
• DOI: 10.1121/1.5014290

Using real-time MRI acoustic data, we employ two methods of signal denoising (DLWP and CS-SNG) to conduct a preliminary comparison between noisy, denoised, and noiseless data. The acoustic data collected in the MRI serve as the noisy, “baseline” group. Data collected from the same speakers in a sound-attenuated environment served as the noiseless, “ground truth.” We calculate acoustic power across the frequency spectrum in 32, 64, 128, and 256 bin experiments and perform k-means clustering on the first three principal components to compare the output of the denoising algorithms to the ground-truth and noisy data. Results show a quantitative difference between the denoising methods, through their different affinities for clusters associated with reference group labels. The groupings indicate that the CS-SNG data are better suited for establishing a map between the visual data from the MRI and the acoustic output because of its association with the noiseless data and its distinction from the noisy group. Since both denoising algorithms form independent clusters, there are potential differentiating features that could drive future improvement of these methods.