Improving quantitative MRI using self-supervised deep learning with model reinforcement: Demonstration for rapid T1...

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Artificial Intelligence (AI)
• Medical Imaging

Background:

• Quantitative MRI (qMRI) reconstruction is crucial for medical diagnostics.
• Traditional methods often require extensive training data and are computationally intensive.
• Accelerating qMRI parameter mapping is essential for clinical applications.

Purpose of the Study:

• To introduce RELAX-MORE, a novel self-supervised learning framework for accelerated qMRI reconstruction.
• To leverage model reinforcement for efficient and accurate MR parameter map generation.
• To enable subject-specific qMRI analysis without large datasets.

Main Methods:

• Developed RELAX-MORE, a self-supervised learning framework using model reinforcement.
• Utilized an optimization algorithm to integrate iterative model-based qMRI reconstruction into a deep learning framework.
• Applied the method to quantitative T1 mapping in brain, knee, and phantom data.

Main Results:

• RELAX-MORE generates high-quality MR parameter maps, correcting artifacts and reducing noise.
• The method significantly improves efficiency, accuracy, robustness, and generalizability compared to existing techniques.
• Demonstrated successful application on single-subject data for various anatomical regions.

Conclusions:

• RELAX-MORE offers a feasible and effective self-supervised learning approach for rapid MR parameter mapping.
• The framework is adaptable for clinical translation of qMRI.
• This method enhances the accessibility and practicality of qMRI studies.