LAPNet: Non-Rigid Registration Derived in k-Space for Magnetic Resonance Imaging

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging (MRI)
• Computational Imaging

Background:

• Physiological motions like cardiac and respiratory cycles introduce artifacts in Magnetic Resonance (MR) images.
• Existing motion correction methods for thoracic scans require accurate motion estimation from undersampled data, which is challenging.
• Image-based registration methods can be hindered by aliasing artifacts inherent in undersampled motion-resolved reconstructions.

Purpose of the Study:

• To develop a novel deep learning approach for fast and accurate non-rigid motion estimation directly in k-space.
• To improve motion correction in undersampled, motion-resolved MR imaging, particularly for thoracic applications.
• To evaluate the performance of the proposed k-space registration method against conventional image-based techniques.

Main Methods:

• Proposed LAPNet, a deep learning model for non-rigid registration in k-space, based on the Local All-Pass (LAP) technique.
• Utilized undersampled, motion-resolved 3D MR images acquired with different acceleration factors and sampling strategies.
• Compared LAPNet against traditional and deep learning image-based registration methods on a cohort of patients and healthy subjects.

Main Results:

• LAPNet demonstrated consistent and superior performance compared to image-based registration approaches.
• The method proved effective across various undersampling trajectories and acceleration factors.
• The proposed k-space registration effectively mitigated artifacts caused by physiological motion in MR images.

Conclusions:

• Performing non-rigid registration directly in k-space using deep learning (LAPNet) is a viable and effective strategy for motion correction in MR imaging.
• LAPNet offers a significant advancement over traditional image-based methods, especially in scenarios with undersampled data and motion artifacts.
• This approach holds promise for enhancing the quality and reliability of thoracic MR scans for diagnosing conditions like liver and lung metastases.