High-definition motion-resolved MRI using 3D radial kooshball acquisition and deep learning spatial-temporal 4D...

Area of Science:

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

Background:

• Free-breathing lung MRI is crucial for high-definition imaging but challenged by motion artifacts.
• Current methods often require long scan times or compromise resolution.

Purpose of the Study:

• To develop a novel motion-resolved, high-definition (HD) 4D MRI technique for free-breathing lung imaging.
• To achieve 1.1 mm isotropic resolution with scan times under 5 minutes using deep learning reconstruction.

Main Methods:

• Utilized a 3D radial kooshball acquisition sequence with ultrashort echo times.
• Developed two HD-Movienet deep learning models (2D and 3D convolutional kernels) for reconstructing motion-sorted data.
• Applied motion-sorting via amplitude-binning on respiratory signals.

Main Results:

• HD-Movienet achieved reconstruction times under 6 seconds, significantly faster than XD-GRASP (>10 minutes).
• Maintained comparable image quality to XD-GRASP with reduced scan times (2-4 minutes).
• 3D-based HD-Movienet enhanced reconstruction quality at the cost of slightly longer reconstruction times (<11 seconds).

Conclusions:

• HD-Movienet enables feasible motion-resolved 4D lung MRI with 1.1 mm isotropic resolution.
• Achieved scan times of 2 minutes (4 motion states) and 4 minutes (10 motion states).
• Represents a significant advancement for clinical free-breathing high-definition lung MRI.