Model-based iterative reconstruction for direct imaging with point spread function encoded echo planar MRI.

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Image Reconstruction
• Medical Imaging

Background:

• Echo Planar Imaging (EPI) is a rapid MRI technique susceptible to reconstruction artifacts.
• Point Spread Function (PSF)-encoded EPI reduces distortion but increases scan time.
• A novel Model-Based Iterative Reconstruction (MBIR) framework is proposed for direct imaging with PSF-EPI.

Purpose of the Study:

• To introduce an MBIR framework for accelerated PSF-EPI.
• To improve image quality and acceleration potential in PSF-EPI.
• To enable direct, distortion-free imaging with faster scan times.

Main Methods:

• Developed an MBIR platform utilizing subspace representation and Local Low-Rankness (LLR) regularization.
• Employed variable splitting for efficient iterative reconstruction.
• Compared MBIR against standard reconstructions, Homodyne, and explored partial Fourier acceleration and random PSF sampling.

Main Results:

• MBIR consistently yielded superior image quality compared to standard and Homodyne reconstructions.
• Arbitrary sampling of the PSF space was enabled, with nonuniform sampling and partial Fourier showing best performance.
• MBIR reconstructed high-quality images from sub-minute scan datasets and was rated superior in neuroradiologists' assessments.

Conclusions:

• Introduced a novel MBIR framework for direct imaging with PSF-EPI.
• The framework allows arbitrary PSF space sampling.
• Diagnostic-quality images can be reconstructed from highly accelerated PSF-encoded EPI data.