SNR Enhancement for Multi-TE MRSI Using Joint Low-Dimensional Model and Spatial Constraints

Area of Science:

• Medical Imaging
• Neuroimaging
• Spectroscopy

Background:

• Magnetic Resonance Spectroscopic Imaging (MRSI) is crucial for non-invasive metabolite quantification in the brain.
• Acquiring high-quality multi-echo (multi-TE) MRSI data is challenging due to noise and limited acquisition time.
• Improving the signal-to-noise ratio (SNR) is essential for accurate metabolite analysis and clinical applications.

Purpose of the Study:

• To develop and validate a novel method for enhancing SNR in multi-TE ¹H-MRSI data.
• To improve the accuracy and reproducibility of metabolite quantification using the proposed technique.
• To enable faster and/or higher-resolution multi-TE ¹H-MRSI for clinical applications.

Main Methods:

• A deep complex convolutional autoencoder (DCCAE) was employed to learn a nonlinear low-dimensional representation of multi-TE ¹H spectroscopy signals.
• A reconstruction formulation integrated the spatiospectral encoding model, the learned DCCAE model, and spatial constraints for SNR enhancement.
• The method was evaluated using numerical simulations and in vivo human brain MRSI experiments.

Main Results:

• The proposed method demonstrated superior denoising performance compared to alternative techniques, both qualitatively and quantitatively.
• In vivo experiments showed improved metabolite quantification reproducibility and accuracy with the developed SNR-enhancing reconstruction.
• The learned low-dimensional model effectively reduced data dimensionality, acting as a constraint for noise reduction.

Conclusions:

• The novel SNR-enhancing reconstruction method significantly improves data quality in multi-TE ¹H-MRSI.
• This technique holds promise for advancing brain MRSI applications by enabling faster scans and higher resolutions.
• The improved metabolite quantification accuracy and reproducibility could benefit various clinical neuroimaging applications.