A Joint Group Sparsity-based deep learning for multi-contrast MRI reconstruction.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computer Vision

Background:

• Multi-contrast MRI enhances diagnostic information but requires longer acquisition times.
• k-space undersampling accelerates MRI but introduces artifacts requiring advanced reconstruction.
• Existing joint sparsity methods are iterative and require manual hyperparameter tuning, while deep learning methods still have reconstruction errors.

Purpose of the Study:

• To develop an efficient and accurate deep learning-based reconstruction method for multi-contrast MRI.
• To address the limitations of traditional iterative methods and improve upon existing deep learning approaches for undersampled MRI data.
• To reduce reconstruction errors and accelerate the acquisition of high-quality multi-contrast MRI scans.

Main Methods:

• Proposed a Joint Group Sparsity-based Network (JGSN) that unrolls the iterative process of joint sparsity algorithms.
• Incorporated data consistency modules, learnable sparse transform modules, and joint group sparsity constraint modules.
• Implemented shared weights across different contrasts in the transform module to reduce network parameters.

Main Results:

• The JGSN method demonstrated superior performance in reconstructing multi-contrast MRI compared to state-of-the-art methods.
• Consistent improvements were observed across different undersampling patterns for both brain and knee in vivo data.
• The network effectively removed undersampling artifacts while preserving rich diagnostic information.

Conclusions:

• The proposed JGSN offers an effective deep learning solution for accelerated multi-contrast MRI reconstruction.
• This method significantly reduces reconstruction errors and outperforms existing techniques.
• JGSN has the potential to improve the efficiency and diagnostic capabilities of MRI examinations.