A deep error correction network for compressed sensing MRI.

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Compressed sensing for magnetic resonance imaging (CS-MRI) reconstructs images from limited k-space data by exploiting sparsity.
• Current CS-MRI methods often produce structural reconstruction errors due to imperfect inverse imaging models.
• Addressing these errors is crucial for enhancing CS-MRI reconstruction quality.

Purpose of the Study:

• To develop a novel deep learning framework for improving CS-MRI reconstruction.
• To introduce a method that compensates for structural errors inherent in existing CS-MRI algorithms.
• To enhance the fidelity and accuracy of MRI reconstructions using deep learning.

Main Methods:

• A deep error correction network (DECN) framework was proposed for CS-MRI.
• The DECN comprises three modules: a template module, an error correction module, and a data fidelity module.
• Existing CS-MRI algorithms can function as the template, guiding a convolutional neural network (CNN) to correct reconstruction errors.

Main Results:

• Experimental results demonstrate that the DECN framework significantly improves upon existing CS-MRI inversion algorithms.
• The error-correcting CNN effectively maps k-space data to adjust for template image reconstruction errors.
• The proposed DECN CS-MRI framework shows considerable improvements in reconstruction quality.

Conclusions:

• The DECN framework allows integration of any off-the-shelf CS-MRI algorithm for template generation.
• A deep neural network is employed within the framework to compensate for reconstruction errors.
• Experimental validation confirms the effectiveness and utility of the proposed deep error correction framework for CS-MRI.