Joint MAPLE: Accelerated joint T1 and T2∗ mapping with scan-specific self-supervised networks

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging (MRI)
• Quantitative MRI

Background:

• Quantitative MRI is crucial for clinical and research studies but often requires long scan times.
• Accelerated MR parameter mapping techniques aim to reduce acquisition duration.
• Existing methods face challenges with encoding intensity and scan time.

Purpose of the Study:

• To propose an accelerated joint T1, T2*, frequency, and proton density mapping technique.
• To synergistically combine parallel imaging, model-based, and deep learning approaches for faster parameter mapping.
• To introduce scan-specific self-supervised network reconstruction for enhanced speed and accuracy.

Main Methods:

• Developed the Joint MAPLE framework integrating parallel imaging, signal modeling, and data consistency.
• Optimized these blocks jointly using a combined loss function.
• Embedded scan-specific self-supervised reconstruction leveraging multi-contrast data from multi-echo, multi-flip angle acquisitions.

Main Results:

• Joint MAPLE reduced reconstruction error by approximately two-fold on average at R=16 acceleration.
• Outperformed existing parallel reconstruction techniques by up to four-fold with challenging sub-sampling masks.
• Demonstrated robust performance at extreme acceleration rates (R=25, R=36) with <20% error.

Conclusions:

• Joint MAPLE enables high-fidelity parameter estimation at high acceleration rates.
• Synergistic combination of parallel imaging, model-based mapping, and multi-contrast data is key.
• Scan-specific self-supervised reconstruction improves parameter estimation without extensive training datasets.