One-Dimensional Deep Low-Rank and Sparse Network for Accelerated MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Accelerated magnetic resonance imaging (MRI) relies on deep learning for enhanced performance.
• Current deep learning MRI reconstructions predominantly use 2D convolutions, limiting training efficiency and generalization.
• 2D convolutions are standard due to the 2D nature of many MRI images and k-space data.

Purpose of the Study:

• To introduce a novel 1D convolutional deep learning approach for accelerated MRI reconstruction.
• To develop and evaluate the One-dimensional Deep Low-rank and Sparse network (ODLS) for improved MRI reconstruction.
• To demonstrate the efficacy of 1D convolutions in enhancing deep learning MRI reconstruction efficiency and robustness.

Main Methods:

• Development of the One-dimensional Deep Low-rank and Sparse network (ODLS) integrating 1D convolutions.
• Unrolling the iterative procedure of a low-rank and sparse MRI reconstruction model within the ODLS framework.
• Extensive validation using in vivo knee and brain MRI datasets under various undersampling scenarios.

Main Results:

• The proposed ODLS network demonstrates superior reconstruction performance compared to state-of-the-art methods, both visually and quantitatively.
• ODLS shows significant suitability for scenarios with limited training subjects.
• The method exhibits robustness to different undersampling patterns and training-test data mismatches.

Conclusions:

• 1D deep learning schemes are memory-efficient and robust for fast MRI applications.
• The ODLS network offers an effective alternative to 2D convolutional approaches in accelerated MRI.
• This work highlights the potential of 1D convolutions for improving deep learning-based MRI reconstruction.