SNR-enhanced diffusion MRI with structure-preserving low-rank denoising in reproducing kernel Hilbert spaces

Area of Science:

• Medical Imaging
• Neuroimaging
• Signal Processing

Background:

• Diffusion MRI (dMRI) is essential for mapping neural pathways but is susceptible to noise.
• Improving signal-to-noise ratio (SNR) in dMRI is critical for accurate analysis and diagnostic capabilities.
• Existing denoising methods may struggle to preserve subtle signal details.

Purpose of the Study:

• To introduce and evaluate a novel denoising technique for dMRI data.
• To leverage nonlinear data redundancy for enhanced SNR boosting.
• To ensure preservation of essential signal information during the denoising process.

Main Methods:

• Kernel Principal Component Analysis (KPCA), a nonlinear extension of PCA, was employed to exploit data redundancy.
• A Gaussian kernel was utilized, with parameters automatically selected based on noise statistics.
• The KPCA technique was validated using Monte Carlo simulations and in vivo human brain dMRI data, including multi-coil acquisitions.

Main Results:

• KPCA achieved SNR improvements up to 2.7x in vivo, surpassing the 1.8x gains from linear PCA (MPPCA).
• KPCA demonstrated lower normalized root mean squared error compared to MPPCA against gold-standard references.
• Statistical analysis confirmed that KPCA preserves anatomical information and noise characteristics, improving diffusion parameter estimation.

Conclusions:

• Kernel Principal Component Analysis (KPCA) effectively utilizes nonlinear redundancy in dMRI signals for superior noise reduction.
• KPCA offers significant SNR improvements over traditional methods like Marchenko-Pastur PCA (MPPCA) without compromising signal integrity.
• The developed KPCA technique enhances the quality of dMRI data, leading to more accurate downstream analyses.