SuperMRF: deep robust reconstruction for highly accelerated magnetic resonance fingerprinting

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Quantitative MRI

Background:

• Magnetic Resonance Fingerprinting (MRF) enables simultaneous mapping of T1 and T2 relaxation times.
• Conventional MRF reconstruction methods (pattern matching, low rank) have limitations in utilizing spatiotemporal data and computational efficiency.
• Deep learning, specifically 3D Convolutional Neural Networks (CNNs), offers potential for high-quality, rapid MRF reconstruction.

Purpose of the Study:

• To design and evaluate SuperMRF, a novel deep learning framework for MRF.
• To enable direct transformation of undersampled MRF data into quantitative T1 and T2 maps, bypassing traditional pattern matching.
• To assess the performance of SuperMRF against conventional and state-of-the-art reconstruction techniques.

Main Methods:

• SuperMRF utilizes a 3D CNN to exploit both spatial and temporal information in MRF data.
• Simulations were conducted using reference parameter maps, with testing on noisy data to evaluate robustness.
• Performance was compared against conventional pattern matching and iterative low rank reconstruction using metrics like SSIM, PSNR, and NMSE.

Main Results:

• SuperMRF achieved accurate T1 and T2 mapping with high acceleration (15x k-space undersampling, 20-fold frame reduction), showing low error (5% NMSE) and high resemblance (94% SSIM) to reference maps.
• SuperMRF demonstrated superiority over conventional and low rank methods in NMSE, SSIM, and noise robustness in simulations.
• Prospective real-world data showed SuperMRF provided comparable T1 and T2 maps to low rank MRF, with conventional MRF T2 values being significantly different.

Conclusions:

• SuperMRF enables rapid and robust MRF reconstruction, even with reduced frames and k-space undersampling.
• The deep learning framework outperforms traditional methods in simulations and yields comparable results to advanced techniques in real-world applications.
• SuperMRF offers a promising approach for efficient and accurate quantitative MRI.