Improving Brain Metabolite Detection with a Combined Low-Rank Approximation and Denoising Diffusion Probabilistic...

Area of Science:

• Neuroimaging
• Biophysics
• Computational Neuroscience

Background:

• In vivo proton magnetic resonance spectroscopy (MRS) is crucial for noninvasive brain metabolite monitoring.
• Low signal-to-noise ratio (SNR) in MRS often requires lengthy scan times, limiting clinical utility.
• Conventional noise reduction like signal averaging is time-consuming and can cause discomfort.

Purpose of the Study:

• To develop a hybrid denoising strategy integrating low-rank approximation and denoising diffusion probabilistic models (DDPM).
• To enhance MRS data quality and reduce scan times for improved clinical applicability.
• To enable more precise and rapid monitoring of neurochemical changes in the brain.

Main Methods:

• Applied Casorati SVD (low-rank approximation) and DDPM to ¹H MRS datasets from 15 subjects.
• Utilized publicly available datasets including baseline and functional data during a pain stimulation task.
• Compared the hybrid method's performance against conventional signal averaging.

Main Results:

• The hybrid denoising strategy significantly improved SNR, outperforming or matching averaging over 32 signals.
• Achieved highly consistent metabolite measurements and accurately tracked temporal glutamate changes during pain stimulation.
• Demonstrated correlation between glutamate levels and pain intensity ratings.

Conclusions:

• The developed hybrid denoising approach enhances MRS data quality and efficiency.
• This method offers a viable alternative to conventional techniques, potentially shortening acquisition times.
• The findings support the integration of advanced denoising for faster, more precise brain metabolite analysis in real-time.