Multi-level pooling encoder-decoder convolution neural network for MRI reconstruction

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Magnetic Resonance Imaging (MRI) reconstruction is crucial for image quality and speed.
• Traditional parallel imaging and reconstruction techniques accelerate MRI by undersampling data.
• Existing methods face limitations in balancing speed and image fidelity.

Purpose of the Study:

• To develop a novel, lightweight deep learning network for accelerated MRI reconstruction.
• To improve the quality and acceleration factor of MR image reconstruction.
• To reduce the computational burden and training time for MRI reconstruction.

Main Methods:

• A novel lightweight deep neural network, the Multi-Level Pooling Encoder-Decoder Net (MLPED Net), was proposed.
• The MLPED Net was trained end-to-end using undersampled data from the fastMRI knee dataset.
• The network reconstructs high-quality MR images from undersampled k-space data.

Main Results:

• The MLPED Net achieved high-quality MR image reconstruction with a high peak signal-to-noise ratio (PSNR).
• The network demonstrated superior performance compared to traditional encoder-decoder networks at 4-fold acceleration.
• Significant improvements were observed across all evaluation metrics.

Conclusions:

• The proposed MLPED Net offers a promising solution for accelerating MRI acquisition and reconstruction.
• Its lightweight architecture significantly reduces training time and computational cost.
• This deep learning approach enables higher acceleration factors without compromising image quality.