Denoising MR spectroscopic imaging data with low-rank approximations

Area of Science:

• Medical Imaging
• Signal Processing
• Spectroscopy

Background:

• Magnetic Resonance Spectroscopic Imaging (MRSI) is crucial for analyzing metabolic information.
• Low signal-to-noise ratio (SNR) is a significant challenge in MRSI data acquisition and analysis.
• Existing denoising methods may compromise important spatial-spectral features.

Purpose of the Study:

• To develop an advanced denoising technique for MRSI data.
• To exploit inherent low-rank properties of MRSI data for noise reduction.
• To improve the quality of MRSI data across various SNR levels.

Main Methods:

• Utilized partial separability and linear predictability for low-rank structures.
• Arranged MRSI data into Casorati and Hankel matrix forms.
• Applied Singular Value Decomposition (SVD) for low-rank approximation.

Main Results:

• Successfully denoised MRSI data across a broad range of SNR values.
• Preserved essential spatial-spectral characteristics of the MRSI signals.
• Validated the method using both simulated and experimental MRSI datasets.

Conclusions:

• The proposed low-rank approximation method offers effective denoising for MRSI.
• This technique shows potential for improving other spatial-spectral and spatial-temporal imaging modalities.
• Enhanced MRSI data quality can lead to more accurate metabolic analysis.