A Physics-Informed Deep Learning Model for MRI Brain Motion Correction.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Magnetic resonance imaging (MRI) is crucial for brain imaging but is limited by motion artifacts due to long scan times.
• These artifacts can significantly degrade image quality and hinder accurate diagnosis.

Purpose of the Study:

• To develop and evaluate PI-MoCoNet, a novel physics-informed neural network for robust motion artifact correction in high-resolution brain MRI.
• To enhance diagnostic reliability by preserving image fidelity without explicit motion parameter estimation.

Main Methods:

• PI-MoCoNet utilizes a dual-network framework: one for motion detection (U-net with spatial averaging) and another for correction (U-net with Swin Transformer blocks).
• The correction network incorporates reconstruction, LPIPS, and data consistency losses, leveraging both spatial and k-space information.
• Simulated motion artifacts were applied, and the method was validated on IXI and MR-ART datasets against baseline models using PSNR, SSIM, and NMSE.

Main Results:

• PI-MoCoNet significantly outperformed baseline methods across all artifact levels on both datasets.
• For instance, on the IXI dataset with heavy artifacts, PSNR improved from 27.99 dB to 36.01 dB, and SSIM increased from 0.75 to 0.97.
• Ablation studies confirmed the benefit of combining data consistency and perceptual losses, yielding approximately 1 dB PSNR gain.

Conclusions:

• PI-MoCoNet offers a robust, physics-informed solution for mitigating motion artifacts in brain MRI.
• The framework's ability to integrate spatial and k-space data enhances image quality, showing strong potential for clinical applications.
• The open-source code facilitates further research and development in motion-corrected MRI.