Quantification of relaxation times in MR Fingerprinting using deep learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biophysics

Background:

• Magnetic Resonance Fingerprinting (MRF) enables rapid, simultaneous measurement of multiple tissue properties.
• Current MRF post-processing is slow and memory-intensive, hindering clinical application.
• Efficient MRF analysis is crucial for advancing quantitative MRI.

Purpose of the Study:

• To develop a rapid convolutional neural network (CNN) for MRF tissue property estimation.
• To accelerate the post-processing of MRF data.
• To validate the CNN's accuracy against traditional methods.

Main Methods:

• A convolutional neural network (CNN) was developed for MRF data analysis.
• The CNN was trained to estimate multiple tissue properties simultaneously.
• Performance was evaluated by comparing T1 and T2 values with pattern matching results.

Main Results:

• The CNN achieved rapid estimation of multiple tissue properties in just 0.1 seconds.
• Estimated T1 and T2 values in white matter and gray matter showed good agreement with pattern matching.
• The developed method significantly reduces MRF post-processing time.

Conclusions:

• A CNN-based approach offers a highly efficient solution for MRF post-processing.
• This method accelerates quantitative MRI, making MRF more clinically viable.
• Deep learning holds significant potential for optimizing complex medical imaging techniques.