Compressive manifold learning: estimating one-dimensional respiratory motion directly from undersampled k-space data.

Area of Science:

• Medical Imaging
• Machine Learning
• Biomedical Engineering

Background:

• Respiratory motion significantly impacts MRI quality, necessitating gating techniques.
• Traditional respiratory gating requires separate navigator acquisitions, increasing scan time.
• Manifold learning (ML) offers a promising approach for motion estimation from MR data.

Purpose of the Study:

• To develop and validate a manifold learning (ML)-based method for direct respiratory signal estimation from undersampled k-space data.
• To apply this method for respiratory self-gating in liver Magnetic Resonance Imaging (MRI).

Main Methods:

• Utilized compressive manifold learning (CML), combining compressed sensing with ML, to learn respiratory motion directly from undersampled k-space data.
• Applied CML to estimate the low-dimensional respiratory motion manifold from high-dimensional MR data.
• Validated the method using both simulated and prospective free-breathing abdominal MRI datasets.

Main Results:

• CML accurately estimated respiratory motion from highly retrospectively undersampled k-space data (up to 25-fold acceleration).
• Prospective 2D acquisitions demonstrated CML's feasibility for respiratory self-gating with up to 15-fold accelerated MR data.
• The method successfully estimated respiratory motion from accelerated free-breathing radial acquisitions.

Conclusions:

• The proposed CML method accurately estimates respiratory signals from highly undersampled k-space data.
• This technique is suitable for respiratory self-navigated 2D liver MRI, potentially reducing scan times and improving efficiency.