Deep Learning-Based Auto-Navigation for Free-Breathing Golden-Angle Radial MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Free-breathing abdominal MRI is challenging due to respiratory motion.
• Accurate motion tracking is crucial for high-quality dynamic MRI reconstruction.
• Existing methods like Principal Component Analysis (PCA) have limitations.

Purpose of the Study:

• To develop a deep learning-based auto-navigation technique for free-breathing golden-angle radial MRI.
• To introduce Respiratory Auto-Navigator for Golden-angle Radial (RANGR) for enhanced abdominal MRI.

Main Methods:

• RANGR computes a 1D respiratory motion signal from k-space data.
• Retrospective sorting of k-space data into motion states for reconstruction with Movienet.
• Training and validation against PCA using phantom and in vivo abdominal MRI data.

Main Results:

• RANGR achieved sub-millimeter motion tracking accuracy in phantom studies.
• Outperformed PCA in qualitative image quality assessments by expert radiologists.
• Demonstrated significantly faster motion estimation compared to PCA on GPU.

Conclusions:

• RANGR offers a robust deep learning-based auto-navigation solution for free-breathing MRI.
• The technique is effective for golden-angle radial MRI acquisition in abdominal imaging.
• RANGR generalizes well, even in challenging cases where PCA fails.