Edge-enhanced dual discriminator generative adversarial network for fast MRI with parallel imaging using multi-view...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Magnetic Resonance Imaging (MRI) is crucial for clinical medicine but suffers from slow data acquisition.
• Existing MRI reconstruction methods often prioritize holistic reconstruction over edge detail enhancement.

Purpose of the Study:

• To develop a fast MRI reconstruction technique that excels at preserving and enhancing edge information.
• To introduce a novel deep learning approach for accelerated multi-channel MRI acquisition.

Main Methods:

• A parallel imaging coupled dual discriminator generative adversarial network (PIDD-GAN) was developed.
• The PIDD-GAN incorporates multi-view information and a dual discriminator for holistic and edge-focused reconstruction.
• An improved U-Net generator with residual learning and frequency channel attention blocks was utilized.

Main Results:

• The PIDD-GAN achieved high-quality reconstructed MR images with well-preserved edge details.
• Single-image reconstruction time was reduced to below 5 milliseconds.
• The method demonstrated superior performance compared to state-of-the-art reconstruction techniques on public brain MR datasets.

Conclusions:

• The proposed PIDD-GAN offers a significant advancement in fast and accurate MRI reconstruction.
• This technique effectively enhances critical edge information, improving diagnostic utility.
• The rapid reconstruction speed meets the demands for efficient clinical workflows.