DeepSENSE: Learning coil sensitivity functions for SENSE reconstruction using deep learning.

Area of Science:

• Medical Imaging
• Machine Learning
• Neuroimaging

Background:

• SENSE reconstruction in brain imaging relies on accurate coil sensitivity functions.
• Limited auto-calibration signals (ACS) pose a challenge for precise sensitivity estimation.
• Improving sensitivity function estimation is crucial for robust SENSE-based MRI.

Purpose of the Study:

• To develop a deep learning approach for improved coil sensitivity function estimation.
• To address the challenge of limited ACS data in SENSE reconstruction.
• To enhance the quality of brain imaging using SENSE.

Main Methods:

• Proposed a deep learning model using deep convolutional neural networks.
• Trained the network to map initial sensitivity to high-resolution counterparts.
• Implemented sensitivity alignment to mitigate geometric variations.
• Validated the method using cross-validation and iterative SENSE reconstruction.

Main Results:

• The deep learning method yielded superior sensitivity estimates and SENSE reconstructions.
• Demonstrated significant improvements in aliasing and noise suppression with limited ACS data.
• Showcased the transferability of the learned network across different MRI sequences (GRE, spin-echo, MPRAGE).

Conclusions:

• A novel deep learning-based method effectively improves coil sensitivity function estimation.
• The proposed approach shows significant potential for enhancing SENSE reconstructions, particularly with scarce ACS data.
• Validated the feasibility and utility of deep learning for brain imaging sensitivity estimation.