FFVN: An explicit feature fusion-based variational network for accelerated multi-coil MRI reconstruction

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biophysics

Background:

• Magnetic Resonance Imaging (MRI) offers high soft-tissue contrast without radiation but is limited by long scan times.
• Traditional acceleration techniques like Compressed Sensing (CS) and Parallel Imaging (PI) are enhanced by Deep Learning (DL).
• Existing DL variational networks may not fully exploit MR feature information for accelerated reconstruction.

Purpose of the Study:

• To develop an accelerated multi-coil MRI reconstruction method using a novel variational network.
• To enhance the exploitation of MR features through explicit feature fusion.
• To improve reconstruction speed and quality in accelerated MRI.

Main Methods:

• A variational network incorporating explicit feature fusion was developed.
• The method integrates Compressed Sensing (CS), Parallel Imaging (PI), and Deep Learning (DL) principles.
• The approach was evaluated on a public multi-coil brain dataset with 5-fold and 10-fold acceleration.

Main Results:

• The proposed method achieves satisfying performance comparable to state-of-the-art techniques.
• Explicit feature fusion effectively leverages additional information for improved reconstruction.
• The method demonstrates efficiency with minimal computational overhead.

Conclusions:

• The developed variational network with explicit feature fusion accelerates multi-coil MRI reconstruction.
• This approach efficiently utilizes MR features, outperforming previous methods.
• The technique shows promise for faster and more effective MRI diagnostics.