Ultrafast magnetic resonance spectroscopic imaging using SPICE with learned subspaces

Area of Science:

• Magnetic Resonance Imaging
• Spectroscopy
• Biomedical Engineering

Background:

• Subspace-based methods like SPICE have advanced Magnetic Resonance Spectroscopic Imaging (MRSI).
• Achieving ultrafast, high-resolution 1H-MRSI remains a challenge, often requiring subject-specific calibration data.

Purpose of the Study:

• To develop a novel subspace learning method for ultrafast 1H-MRSI of the brain.
• To enhance the SPICE approach by eliminating the need for subject-dependent navigator data.

Main Methods:

• A novel strategy was formulated to learn a low-dimensional subspace representation of MR spectra from training data.
• Physics-based models and training data were integrated to learn empirical distributions of molecule-specific spectral parameters.
• High-resolution MRSI acquisitions with ultrashort echo time/repetition time, sparse sampling, and no water suppression were performed.

Main Results:

• The developed method accurately produced high-resolution 3D 1H metabolite maps and high-quality spatially resolved spectra (2.4 × 2.4 × 3 mm³ in 5 minutes).
• Phantom and in vivo studies validated the accuracy of the learned subspace and the method's capabilities.
• Eliminating water suppression allowed extraction of valuable information from water signals for data processing and mapping tissue properties.

Conclusions:

• The proposed method enables ultrafast 1H-MRSI of the brain using a learned subspace.
• It eliminates the need for subject-dependent navigator data, a key limitation of the original SPICE technique.
• This represents a significant advancement toward practical, high-resolution MRSI applications.