Dynamic MRI with Locally Low-Rank Subspace Constraint: Towards 1-Second Temporal Resolution Aided by Deep Learning

Area of Science:

• Radiology
• Medical Imaging
• Biomedical Engineering

Background:

• Magnetic Resonance Imaging (MRI) is crucial for high-risk breast cancer screening.
• Current MRI relies on qualitative morphology, with limited use of contrast kinetic information.
• Balancing spatial and temporal resolution is a key challenge in dynamic MRI.

Purpose of the Study:

• To develop a radial MRI reconstruction framework for Dynamic Contrast Enhanced (DCE) imaging.
• To address limitations in spatial and temporal resolution in dynamic MRI.
• To improve image quality and flexibility in DCE-MRI.

Main Methods:

• Proposed a radial MRI reconstruction framework utilizing a locally low-rank (LLR) subspace model.
• Integrated a neural network to enable flexible temporal resolution (down to 1 second).
• Leveraged tissue information for spatially localized dynamics representation.

Main Results:

• Demonstrated substantial improvements in Contrast-to-Noise Ratio (CNR) and noise reduction.
• Achieved flexible temporal resolution, enhancing image acquisition speed.
• Reduced undersampling penalties, leading to better image quality.

Conclusions:

• The proposed framework offers a joint solution to spatial and temporal resolution challenges in DCE-MRI.
• Shows significant potential for breast cancer screening and other applications like head and neck, and brain MRI.
• Presents a viable alternative for various DCE-MRI examinations.