Undersampled Multi-Contrast MRI Reconstruction Based on Double-Domain Generative Adversarial Network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Image Reconstruction

Background:

• Multi-contrast magnetic resonance imaging (MRI) provides crucial diagnostic information but suffers from long acquisition times.
• Accelerating MRI scans is essential for clinical practice, with k-space subsampling being a common technique.
• Subsampling introduces artifacts and noise, necessitating advanced reconstruction methods.

Purpose of the Study:

• To develop an accelerated multi-contrast MRI acquisition method.
• To leverage shared information across different MRI contrasts for improved reconstruction.
• To reduce scan time while maintaining diagnostic image quality.

Main Methods:

• A cross-domain, two-stage generative adversarial network (GAN) was proposed for multi-contrast image reconstruction.
• The network integrates k-space estimation/completion and image-space refinement.
• It utilizes one fully-sampled contrast and undersampled data from other modalities to reconstruct all contrasts simultaneously.

Main Results:

• The proposed GAN method effectively reconstructed undersampled multi-contrast MRI images.
• Quantitative comparisons demonstrated superior performance over baseline methods.
• High acceleration rates were achieved while preserving texture details and reducing artifacts.

Conclusions:

• The developed GAN approach significantly accelerates multi-contrast MRI acquisition.
• This method offers a promising solution for efficient and high-quality medical imaging.
• It has the potential to improve the feasibility of multi-contrast MRI in daily clinical settings.