Coordinate-based neural representation enabling zero-shot learning for fast 3D multiparametric quantitative MRI.

Area of Science:

• Neuroimaging
• Biophysics
• Medical Physics

Background:

• Quantitative magnetic resonance imaging (qMRI) provides valuable tissue-specific physical parameters for neuroscience and clinical applications.
• Current 3D multiparametric qMRI acquisition methods suffer from lengthy scan times, limiting their clinical feasibility.

Purpose of the Study:

• To introduce SUMMIT, an innovative imaging methodology for simultaneous multiparametric qMRI acquisition and unsupervised reconstruction.
• To overcome the limitations of long scan times in 3D multiparametric qMRI.

Main Methods:

• SUMMIT encodes multiple quantitative properties into highly undersampled k-space.
• It utilizes implicit neural representation with a physics model for unsupervised reconstruction of multiparametric maps without external training data.
• The method reconstructs co-registered T1, T2, T2*, and quantitative susceptibility mapping (QSM).

Main Results:

• SUMMIT demonstrated high accuracy in simulations, phantom, and in vivo brain imaging.
• The technique successfully identified microstructural alterations in patients with white matter hyperintense lesions with high sensitivity and specificity.
• The unsupervised reconstruction approach enables zero-shot learning for multiparametric imaging.

Conclusions:

• SUMMIT significantly reduces scan times for multiparametric qMRI while maintaining high accuracy.
• This methodology offers a promising tool for neuroscience research and clinical practice, particularly for detecting white matter abnormalities.
• The developed unsupervised reconstruction paradigm has broad applicability across various medical imaging modalities.