HYDRA: Hybrid deep magnetic resonance fingerprinting

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Quantitative MRI

Background:

• Magnetic resonance fingerprinting (MRF) traditionally uses dictionary matching for quantitative parameter mapping.
• Dictionary matching in MRF is computationally intensive and introduces discretization errors.
• Existing MRF methods face challenges with speed, accuracy, and memory requirements.

Purpose of the Study:

• To introduce HYDRA, a HYbrid Deep magnetic ResonAnce fingerprinting approach.
• To overcome the limitations of dictionary matching in MRF, including discretization errors and computational complexity.
• To develop a faster and more accurate MRF method for quantitative tissue parameter estimation.

Main Methods:

• HYDRA employs a two-stage process: model-based signature restoration and learning-based parameter restoration.
• Low-rank de-aliasing techniques are used for signal restoration.
• A deep nonlocal residual convolutional neural network is utilized for parameter restoration, trained on Bloch-simulated MRF data.

Main Results:

• HYDRA significantly improves inference speed by eliminating dictionary matching.
• The approach alleviates discretization errors by providing continuous-valued parameters.
• Memory requirements are reduced due to the elimination of large dictionary storage.

Conclusions:

• HYDRA offers superior inference speed compared to conventional MRF techniques.
• The method demonstrates enhanced accuracy and reduced storage needs.
• HYDRA represents a significant advancement in quantitative MRI parameter mapping.