Parallel MRI Reconstruction Using Broad Learning System.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Signal Processing

Background:

• Parallel magnetic resonance imaging (pMRI) reconstruction accelerates imaging by interpolating k-space data from phased-array coils.
• Deep learning (DL) enhances pMRI quality but requires extensive, diverse training data and long training times.
• Data variability between training and testing sets poses a challenge for DL-based pMRI reconstruction.

Purpose of the Study:

• To propose a novel broad learning system (BLS) for parallel MRI reconstruction.
• To leverage the approximation capabilities of a single-layer neural network through network broadening.
• To improve the efficiency and effectiveness of pMRI reconstruction compared to existing methods.

Main Methods:

• Developed a broad learning system specifically for pMRI reconstruction.
• Utilized a one-layer neural network with an expanded structure (broadening) to approximate k-space data.
• Evaluated the proposed BLS method against conventional pMRI reconstruction techniques.

Main Results:

• The proposed broad learning system demonstrated effective noise suppression in pMRI reconstruction.
• Experimental results indicate superior performance compared to conventional pMRI reconstruction methods.
• The BLS approach offers a potentially more efficient alternative, mitigating the need for extensive training data.

Conclusions:

• The broad learning system is a viable and effective approach for parallel MRI reconstruction.
• This method shows promise in reducing noise and improving the quality of accelerated MRI scans.
• The BLS framework offers advantages over traditional deep learning methods by potentially reducing data requirements and training time.