Motion-Aware Neural Networks Improve Rigid Motion Correction of Accelerated Segmented Multislice MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Motion artifacts are a significant challenge in Magnetic Resonance Imaging (MRI), degrading image quality and potentially leading to misdiagnosis.
• Accurate motion correction is crucial for reliable diagnostic information from MRI scans, especially in segmented multislice acquisitions.

Purpose of the Study:

• To develop and evaluate a novel deep learning approach for fast, retrospective, intraslice rigid motion correction in segmented multislice MRI.
• To improve the quality of MRI reconstructions compared to existing model-based techniques and deep learning methods that do not incorporate motion estimates.

Main Methods:

• A hypernetwork architecture was employed, utilizing auxiliary rigid motion parameter estimates.
• The hypernetwork generates image-specific reconstruction networks tailored to the estimated motion parameters.
• The proposed method was compared against traditional model-based techniques and alternative deep learning strategies.

Main Results:

• The deep learning approach achieved higher quality MRI reconstructions.
• The method demonstrated superior performance compared to model-based techniques and networks lacking motion estimation.
• The strategy showed reduced sensitivity to inaccuracies in motion parameter estimation.

Conclusions:

• Deep learning, specifically using a hypernetwork with motion parameter estimation, offers a powerful solution for retrospective motion correction in MRI.
• This technique significantly enhances MRI image quality and robustness, paving the way for more reliable diagnostic imaging.