Highly accelerated MR parametric mapping by undersampling the k-space and reducing the contrast number simultaneously...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Magnetic Resonance Imaging

Background:

• Accelerated MR parametric mapping is crucial for efficient clinical workflows.
• Current methods often struggle to balance speed and image quality.
• Simultaneous undersampling and contrast reduction present a significant challenge.

Purpose of the Study:

• To introduce RG-Net, a deep learning framework for accelerated MR parametric mapping.
• To enable simultaneous k-space undersampling and reduction of acquired contrasts.
• To achieve high-quality parametric maps at accelerated scan times.

Main Methods:

• Developed RG-Net, comprising a reconstruction and a generative module.
• Reconstruction module uses undersampled k-space data and a data prior.
• Generative module synthesizes multi-contrast images, implicitly incorporating an exponential model.
• T1 mapping data from 8 volunteers were used for training and testing at acceleration rates of 17.

Main Results:

• RG-Net successfully generated high-quality T1 maps at an acceleration rate of 17.
• The framework demonstrated superior performance in T1 value analysis compared to k-space undersampling alone.
• Regional T1 analysis in cartilage and brain confirmed the method's efficacy.

Conclusions:

• RG-Net enables highly accelerated MR parametric mapping with excellent reconstruction quality.
• The simultaneous k-space undersampling and contrast reduction strategy is effective.
• The generative module is adaptable and can enhance other fast MR mapping techniques.