MP-PCA denoising for diffusion MRS data: promises and pitfalls

Area of Science:

• Magnetic Resonance Imaging
• Neuroimaging
• Spectroscopy

Background:

• Diffusion-weighted magnetic resonance spectroscopy (DW-MRS) has a lower signal-to-noise ratio (SNR) than conventional MRS due to diffusion attenuation.
• Noise reduction is crucial for improving DW-MRS data quality.

Purpose of the Study:

• To evaluate the effectiveness of Marchenko-Pastur principal component analysis (MP-PCA) for denoising DW-MRS data.
• To assess the impact of different MP-PCA strategies on data quality and metabolite quantification.

Main Methods:

• MP-PCA denoising was applied to Monte Carlo simulations and in vivo DW-MRS data (rat brain at 9.4 T, human brain at 3 T).
• Two denoising strategies were compared: denoising the entire data matrix versus using a sliding window approach.
• Analysis included apparent SNR, rank selection, noise correlation, and quantification of metabolite concentrations and diffusion coefficients.

Main Results:

• MP-PCA denoising significantly increased the apparent SNR and improved B0 drift correction.
• Metabolite concentrations and diffusion coefficients estimated from denoised data were comparable to raw data.
• No spectral residuals were observed on individual shots, but noise correlations across shells were introduced, mitigated by the sliding window approach.

Conclusions:

• MP-PCA is a valuable tool for denoising DW-MRS data, enhancing SNR and B0 correction.
• The sliding window approach effectively mitigates introduced noise correlations.
• MP-PCA offers improved DW-MRS data quality for neuroimaging applications.