Motion correction in MR image for analysis of VSRAD using generative adversarial network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurodegenerative Diseases

Background:

• Voxel-based specific region analysis (VSRAD) uses MRI to measure hippocampal atrophy in Alzheimer's disease.
• Motion artifacts in MRI scans can distort VSRAD results, impacting diagnostic accuracy.

Purpose of the Study:

• To evaluate the effectiveness of the Pix2Pix deep learning network for motion artifact correction in MRI images used for VSRAD analysis.
• To compare the performance of Pix2Pix-corrected images with artifact-laden and original images in VSRAD.

Main Methods:

• Supervised deep learning using a Pix2Pix network was trained on MRI data with manipulated k-space images to generate motion-corrected images.
• VSRAD analysis metrics (VOI atrophy severity, GM atrophy extent, VOI atrophy extent) were compared between original, artifact, and Pix2Pix-corrected images.
• Image quality of Pix2Pix-generated images was compared against U-Net-generated images.

Main Results:

• Bland-Altman analysis indicated successful motion correction, with smaller limits of agreement for Pix2Pix-corrected images compared to artifact images.
• Spearman's rank correlation coefficients showed near-perfect agreement between original and Pix2Pix-corrected images for all VSRAD metrics.
• Pix2Pix demonstrated superior quantitative and qualitative image quality compared to U-Net for motion correction.

Conclusions:

• Pix2Pix-based motion correction is a valuable tool for improving the reliability of VSRAD analysis in Alzheimer's disease.
• This deep learning approach enhances the accuracy of measuring hippocampal atrophy from MRI scans.
• The findings support the clinical utility of Pix2Pix for artifact reduction in neuroimaging analysis.