GABA+-Edited Magnetic Resonance Spectroscopy Deep Learning Quality Assessment Framework.

Area of Science:

• Neuroimaging
• Biophysics
• Spectroscopy

Background:

• Improving the quality of GABA+-edited magnetic resonance spectroscopy (MRS) is crucial for accurate neurochemical analysis.
• Traditional averaging methods like Equal-weighting and MSE-weighting have limitations in optimizing spectral data.
• Deep learning (DL) offers potential for advanced signal processing in MRS.

Purpose of the Study:

• To develop and evaluate a novel three-module framework utilizing deep learning to enhance transient averaging in GABA+-edited MRS.
• To hypothesize that a DL model can differentiate spectrum quality better than existing methods for improved averaging.
• To compare the DL-based approach against Equal-weighting and MSE-weighting algorithms.

Main Methods:

• A three-module framework was developed: automated quality labeling, a dual-domain DL model for quality assessment, and a DL-informed transient weighting algorithm.
• The quality labeling algorithm used MRS metrics to focus on retaining GABA+ peak shape in difference spectra.
• The DL model learned from quality labels to assign weights for transient pairs in the final average difference spectrum, with results compared to MSE-weighting and Equal-weighting.

Main Results:

• The DL-based framework improved spectral quality, confirmed by traditional and novel metrics, and visual assessment of GABA+ and Glx peaks.
• Application to in vivo scans showed improved fit quality (lower fit error) compared to Equal-weighting (4.759±1.545 vs. 4.877±1.762).
• The DL model achieved a higher signal-to-noise ratio (SNR) compared to MSE-weighting (18.758±2.392 vs. 18.004±2.68).

Conclusions:

• The proposed framework moderately improves data quality in GABA+-edited MRS through optimized transient averaging.
• This DL-based approach offers a promising avenue for enhancing MRS data analysis and opens opportunities for future research.