Deep Learning for Magnetic Resonance Fingerprinting: A New Approach for Predicting Quantitative Parameter Values from...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biophysics

Background:

• Magnetic Resonance Fingerprinting (MRF) enables quantitative parameter mapping.
• MRF relies on comparing acquired signals to a physical model.
• Current MRF methods use dictionary matching, which is computationally intensive.

Purpose of the Study:

• To evaluate deep learning methods for Magnetic Resonance Fingerprinting (MRF).
• To accelerate the computation of quantitative parameter maps in MRF.
• To demonstrate the potential of Convolutional Neural Networks (CNNs) in MRF.

Main Methods:

• Training a Convolutional Neural Network (CNN) on simulated MRF dictionary data.
• Utilizing pseudo-random excitation patterns to generate non-steady state signals.
• Comparing the performance of the CNN against traditional dictionary matching.

Main Results:

• The trained CNN implicitly encodes the MRF dictionary.
• The CNN effectively replaces the dictionary matching process.
• This deep learning approach significantly accelerates quantitative map computation.

Conclusions:

• Deep learning, particularly CNNs, offers a viable and faster alternative to dictionary matching in MRF.
• CNNs can streamline the generation of quantitative parameter maps from MRF data.
• This method holds promise for improving the efficiency of MR-based quantitative imaging.