Fast image reconstruction with L2-regularization.

Area of Science:

• Medical Imaging
• Computational Science
• Signal Processing

Background:

• Iterative L1- and L2-regularization methods are standard for Magnetic Resonance Imaging (MRI) reconstruction.
• These iterative approaches can be computationally intensive, limiting real-time applications.
• Developing faster reconstruction algorithms without sacrificing image quality is crucial for advancing MRI.

Purpose of the Study:

• Introduce novel L2-regularized reconstruction algorithms with closed-form solutions.
• Achieve substantial computational speed-up compared to current state-of-the-art iterative methods.
• Maintain similar image quality across diverse MRI applications.

Main Methods:

• Compared fast L2-based methods against iterative L1- and L2-regularization algorithms.
• Evaluated performance on numerical phantoms and in vivo data.
• Tested in three key applications: Quantitative Susceptibility Mapping (QSM), Magnetic Resonance Spectroscopic Imaging (MRSI) lipid artifact suppression, and Diffusion Spectrum Imaging (DSI).

Main Results:

• Demonstrated two to three orders of magnitude speed-up in all tested applications.
• Achieved similar reconstruction quality compared to state-of-the-art algorithms.
• Specific examples include 3D QSM reconstruction under 5 seconds, MRSI lipid suppression under 1 second, and DSI under 30 seconds on a standard workstation.

Conclusions:

• Closed-form L2-regularization provides a faster alternative to iterative methods for MRI reconstruction.
• This speed advantage is achieved without compromising image quality.
• The proposed methods are suitable for various MRI applications, enhancing computational efficiency.