Rapid whole brain motion-robust mesoscale in-vivo MR imaging using multi-scale implicit neural representation

Area of Science:

• Neuroimaging
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• High-resolution in vivo MR imaging at mesoscale resolutions faces challenges like long scan times, motion artifacts, and low signal-to-noise ratio (SNR).
• Reconstructing accurate isotropic volumes from anisotropic scans is difficult due to lack of ground truth and inter-scan motion.

Purpose of the Study:

• To introduce Rotating-view super-resolution (ROVER)-MRI, an unsupervised framework for accurate recovery of fine anatomical details from multi-view thick-slice acquisitions.
• To address challenges in high-resolution whole-brain MRI, including scan duration, motion artifacts, and SNR.

Main Methods:

• Utilizing multi-scale implicit neural representations (INR) with coordinate-based neural networks to encode image structures continuously.
• Implementing an integrated registration mechanism for simultaneous anatomical continuity modeling and inter-view motion correction.
• Validating the framework on ex-vivo monkey and in-vivo human brain datasets.

Main Results:

• ROVER-MRI demonstrated substantially improved reconstruction performance compared to bi-cubic interpolation and regularized least-squares super-resolution reconstruction (LS-SRR).
• Achieved whole-brain in vivo T2-weighted imaging at 180μm isotropic resolution in 17 minutes with a 22.4% reduction in relative error compared to LS-SRR.
• Showcased improved SNR and enhanced image sharpness for various super-resolution factors (5 to 11).

Conclusions:

• ROVER-MRI offers a rapid, accurate, and motion-resilient solution for mesoscale neuroimaging.
• The framework enables high-resolution whole-brain imaging with reduced scan times and improved image quality.
• Highlights ROVER-MRI's potential for substantial advantages in neuroimaging studies.