Accelerating 7T Quantitative MRI with Self-Supervised Few-Shot Deep Learning

Motivation: Ultra-high field 7T quantitative magnetic resonance imaging (qMRI) of the brain is valuable for its superior soft tissue contrast and physiological insights but is hindered by long acquisition times, limiting its clinical adoption. Goal(s): To accelerate 7T qMRI acquisition while preserving image quality and quantitative accuracy using deep learning (DL) algorithms. Approach: We developed a self-supervised DL model that integrates consistency mechanisms with zero-shot and few-shot learning techniques to generate high-fidelity reconstructions from under-sampled 7T qMRI data. Results: The method demonstrated significant acceleration in 7T qMRI acquisition with minimal loss of image quality, achieving comparable quantitative metrics to fully sampled acquisitions. Impact: The proposed DL approach for accelerating 7T qMRI reduces scan times without compromising image quality, facilitating broader adoption of high-field MRI. Its generalizability with limited training data enhances advanced neuroimaging accessibility and efficiency, contributing to better clinical utility.