Semiblind Spectral Factorization Approach for Magnetic Resonance Spectroscopy Quantification.

Area of Science:

• Biomedical Engineering
• Spectroscopy
• Signal Processing

Background:

• Magnetic Resonance Spectroscopy (MRS) is crucial for quantifying metabolites, aiding in early disease diagnosis like brain tumors.
• Observed MR spectra contain metabolite signals, baseline, and noise, complicating accurate quantification.
• Existing methods face challenges in separating overlapping metabolite signals and removing baseline interference.

Purpose of the Study:

• To propose a novel semi-blind spectral factorization approach for Magnetic Resonance Spectroscopy (MRS) quantification.
• To develop a method for accurate baseline removal in frequency domain using Singular Spectrum Analysis (SSA).
• To improve metabolite separation and reconstruction by incorporating prior knowledge of metabolites.

Main Methods:

• A novel spectral factorization approach based on Singular Spectrum Analysis (SSA) was developed.
• Baseline removal was performed in the frequency domain using SSA.
• A new covariance matrix was proposed to incorporate prior knowledge of metabolites, exploiting correlations in the nuclear magnetic resonance signal.
• A novel cost function was used to reconstruct the metabolite of interest by combining extracted components.

Main Results:

• The proposed method effectively removed baseline from MRS signals.
• Underlying spectral components were successfully extracted using the novel covariance matrix.
• Metabolites of interest were accurately reconstructed, demonstrating the technique's effectiveness.
• Performance was validated using both synthetic and real MRS data.

Conclusions:

• The developed spectral factorization technique accurately quantifies metabolites in MRS.
• The method provides effective baseline removal and improved metabolite separation.
• This approach holds promise for enhancing early disease diagnosis through improved MRS analysis.