Magnetic Resonance Fingerprinting Reconstruction Using Recurrent Neural Networks

Area of Science:

• Medical Imaging
• Biophysics
• Machine Learning

Background:

• Magnetic Resonance Fingerprinting (MRF) provides unique tissue signals.
• Accelerated MRF acquisition necessitates efficient reconstruction methods.
• Current template matching reconstruction is time-consuming.

Purpose of the Study:

• Investigate Recurrent Neural Networks (RNNs) for MRF reconstruction.
• Address the computational bottleneck in MRF data processing.
• Compare RNN performance against existing deep learning methods.

Main Methods:

• Utilized Recurrent Neural Networks (RNNs) for MRF signal reconstruction.
• Explored the temporal correlation of MRF signals for model design.
• Evaluated model performance on in-vivo imaging data.

Main Results:

• RNN models demonstrated significantly improved reconstruction results.
• RNNs outperformed Convolutional Neural Networks (CNNs) in MRF reconstruction.
• The proposed RNN approach is effective for in-vivo MRF data.

Conclusions:

• RNNs are a viable and superior alternative for MRF reconstruction.
• This deep learning strategy accelerates MRF analysis.
• RNNs enhance the efficiency and accuracy of medical imaging techniques.