Zero-DeepSub: Zero-shot deep subspace reconstruction for rapid multiparametric quantitative MRI using 3D-QALAS.

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Quantitative MRI
• Biomedical Engineering

Background:

• Accurate and rapid T1 and T2 mapping are crucial for quantitative MRI.
• Conventional methods often face limitations in speed and image fidelity.
• Interleaved Look-Locker acquisition sequences with T2 preparation pulse (3D-QALAS) offer potential for time-resolved imaging.

Purpose of the Study:

• To develop and evaluate a low-rank subspace method for reconstructing 3D-QALAS time-series images.
• To enhance the fidelity of subspace QALAS using deep learning and subspace modeling.
• To enable accurate and rapid T1 and T2 mapping with improved image quality.

Main Methods:

• Proposed a low-rank subspace method for 3D-QALAS (subspace QALAS).
• Introduced a zero-shot deep-learning subspace method (Zero-DeepSub) for reconstruction.
• Evaluated accuracy and reproducibility using an ISMRM/NIST system phantom and in vivo scans.
• Compared performance against conventional QALAS at high acceleration factors (up to nine-fold).

Main Results:

• Subspace QALAS demonstrated good linearity and improved precision/reduced bias compared to conventional QALAS, particularly for T2 maps.
• In vivo results showed subspace QALAS with Zero-DeepSub yielded better g-factor maps, reduced blurring, noise, and artifacts.
• Achieved whole-brain T1, T2, and PD mapping at 1mm isotropic resolution in under 2 minutes with nine-fold acceleration.

Conclusions:

• Subspace QALAS combined with Zero-DeepSub enables high-fidelity, rapid multiparametric quantification.
• The developed methods significantly improve whole-brain imaging capabilities.
• This approach advances time-resolved quantitative MRI techniques.