Accelerated multi-shell diffusion MRI with Gaussian process estimated reconstruction of multi-band imaging

Area of Science:

• Magnetic Resonance Imaging
• Biophysics
• Neuroscience

Background:

• Diffusion-weighted MRI (dMRI) is crucial for mapping tissue microstructure.
• Acquisition speed limits the application of advanced multi-shell dMRI models.
• Current reconstruction methods struggle with high acceleration factors.

Purpose of the Study:

• To develop a robust dMRI reconstruction method that accelerates data acquisition by exploiting shared information across shells.
• To enable rapid tissue microstructure mapping through faster dMRI scans.

Main Methods:

• Extended the Diffusion Acceleration with Gaussian process Estimated Reconstruction (DAGER) method.
• Introduced a multi-shell covariance function for Gaussian Process modeling.
• Corrected for Rician noise in magnitude data during Gaussian Process fitting.
• Evaluated the method using both simulated and in vivo dMRI data.

Main Results:

• Demonstrated significant image quality improvement in reconstructed dMRI data at high acceleration (up to factor 12).
• Achieved superior performance compared to conventional k-only reconstruction methods.
• Enabled more robust diffusion model fitting, facilitating advanced multi-shell diffusion analysis.

Conclusions:

• The proposed method allows for highly accelerated multi-shell dMRI without compromising image quality.
• This acceleration significantly shortens scan times compared to conventional methods.
• Facilitates wider adoption of advanced dMRI models in neuroscience research and clinical applications.