Model-based reconstruction for looping-star MRI.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Signal Processing

Background:

• Conventional gridding and model-based methods for gradient echo (GRE) image reconstruction often sacrifice signal-to-noise ratio (SNR) and spatial resolution.
• Accurate modeling of signal acquisition and reconstruction is crucial for high-quality MRI.

Purpose of the Study:

• To develop an advanced reconstruction method for GRE images that fully models acquisition signals.
• To enhance image quality in terms of SNR and spatial resolution compared to existing techniques.

Main Methods:

• Developed a novel approach modeling spoke trajectories and overlapping echoes (echo-in/echo-out mixture).
• Employed two system matrices to represent overlapping echoes.
• Utilized the conjugate gradient algorithm (CG-SENSE) with nonuniform fast Fourier transform (NUFFT) for image reconstruction optimization.

Main Results:

• Demonstrated effectiveness in phantom and in-vivo volunteer studies for 3D high-resolution T2*-weighted imaging and functional MRI (fMRI).
• The high-resolution protocol showed improved spatial resolution and reduced signal loss due to less intra-voxel dephasing compared to gridding.
• fMRI tasks revealed reduced artifacts and blurring, with more stable and prominent time courses using the model-based method.

Conclusions:

• The proposed model-based reconstruction method significantly improves spatial resolution and reduces artifacts in GRE imaging.
• Enhanced time series and activation maps in fMRI tasks were observed, attributed to the mitigation of overlapping echoes and undersampling artifacts.