DEEP-DISORDER: Motion Correction in 3D MRI via Segment Reconstruction and Registration

Area of Science:

• Medical Imaging
• Neuroimaging
• Image Reconstruction

Background:

• 3D MRI acquisition is lengthy and prone to patient motion, causing artifacts and necessitating reacquisition.
• Intrascan motion degrades image quality and can compromise diagnostic accuracy in volumetric brain MRI.

Purpose of the Study:

• To develop and validate a modular framework for retrospective correction of intrascan motion in 3D brain MRI.
• To enable motion correction without requiring active motion tracking or external devices.

Main Methods:

• Utilized a distributed and incoherent sampling scheme (DISORDER) with a fast network for rapid, undersampled reconstructions.
• Employed groupwise registration to estimate rigid motion parameters from approximate reconstructions.
• Applied estimated motion parameters to reconstruct data, reducing motion-induced artifacts.

Main Results:

• Achieved high accuracy in motion parameter estimation (0.06 mm, 0.13°).
• Significantly improved image quality, with Structural Similarity Index Measure (SSIM) increasing from 0.942 to 0.992 for retrospective scans.
• Demonstrated substantial improvements for prospective scans, from 0.915 to 0.936 (gradual motion) and 0.764 to 0.923 (extreme motion).

Conclusions:

• The proposed framework successfully estimates and corrects 3D motion within 3D brain MRI scans retrospectively.
• The method enhances volumetric MRI tolerance to motion, leading to vastly improved image quality without external tracking.
• This approach offers a robust solution for motion artifacts in time-intensive MRI acquisitions.