Multi-band- and in-plane-accelerated diffusion MRI enabled by model-based deep learning in q-space and its extension...

Area of Science:

• Magnetic Resonance Imaging
• Diffusion MRI
• Computational Neuroscience

Background:

• Accelerated diffusion MRI scans are essential for high-resolution imaging.
• Reconstructing data with combined in-plane and multi-band acceleration presents significant aliasing challenges.
• Existing methods struggle with the high acceleration factors required for efficient diffusion imaging.

Purpose of the Study:

• To propose a novel reconstruction method for combined in-plane and multi-band accelerated diffusion MRI data.
• To leverage a q-space prior from the qModeL framework for improved data recovery.
• To enable efficient and accurate reconstruction of diffusion MRI scans with high acceleration.

Main Methods:

• A model-based iterative reconstruction incorporating a pre-learned q-space prior derived from the qModeL framework.
• Utilizing an incoherent under-sampling pattern in the k-q domain to maximize joint reconstruction power.
• Extending the learning framework to the spherical harmonic basis for rotational invariance.

Main Results:

• The qModeL joint reconstruction method successfully unaliased and recovered diffusion MRI data with high accuracy across various datasets and field strengths.
• Reconstruction errors for 18-fold accelerated multi-shell datasets were less than 3%.
• Derived microstructural maps showed reasonable accuracy in both healthy and abnormal tissues, comparable to traditional methods.

Conclusions:

• The qModeL framework facilitates joint recovery of combined in-plane and multi-band accelerated diffusion MRI data using deep learning.
• This method offers an efficient and accurate solution for reconstructing highly accelerated diffusion MRI scans.
• The approach demonstrates potential for improving the speed and quality of diffusion MRI acquisition and analysis.