Accelerating Multiparametric Quantitative MRI Using Self-Supervised Scan-Specific Implicit Neural Representation With...

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Artificial Intelligence in Medical Imaging
• Quantitative Imaging

Background:

• Accelerated MRI techniques are crucial for reducing scan times and improving patient comfort.
• Multiparametric quantitative MRI (qMRI) provides rich tissue contrast but often requires long acquisition times.
• Developing efficient reconstruction methods for accelerated qMRI is essential for clinical translation.

Purpose of the Study:

• To develop a self-supervised, scan-specific deep learning framework for reconstructing accelerated multiparametric quantitative MRI (qMRI).
• To enhance the accuracy and efficiency of qMRI reconstruction using advanced AI techniques.

Main Methods:

• Proposed REFINE-MORE, integrating implicit neural representation (INR) with a model reinforcement module enforcing MR physics.
• Employed an unrolled optimization scheme for data consistency and a low-rank adaptation strategy for computational efficiency.
• Evaluated on accelerated multiparametric quantitative magnetization transfer imaging for simultaneous estimation of relaxation, proton fraction, and exchange rate.

Main Results:

• Achieved superior reconstruction quality on in vivo data under 4× and 5× accelerations, outperforming baseline and state-of-the-art methods.
• Demonstrated lowest normalized root-mean-square error and highest structural similarity index.
• Phantom experiments showed strong agreement with reference values, confirming robustness and generalizability.

Conclusions:

• REFINE-MORE enables accurate and efficient scan-specific multiparametric qMRI reconstruction.
• Offers a flexible solution for high-dimensional, accelerated qMRI applications.
• Highlights the potential of deep learning for advancing quantitative MRI techniques.