A quality control method for detecting and suppressing uncorrected residual motion in fMRI studies

Area of Science:

• Neuroimaging
• Data Analysis
• Machine Learning

Background:

• Motion correction is critical for fMRI data integrity.
• Current methods for handling motion artifacts lack objectivity.
• Excluding subjects due to motion is inefficient and costly.

Purpose of the Study:

• To develop an objective method for identifying and mitigating motion artifacts in fMRI data.
• To improve the efficiency of fMRI data analysis by reducing subject exclusion.
• To refine data quality through unsupervised learning and statistical suppression.

Main Methods:

• Utilized k-means clustering on mean square derivative (MSD) features before and after realignment.
• Refined classifications using correlation analysis of activation maps.
• Employed regression analysis to assess tasking and motion relationships.
• Identified residual motion using MSD and suppressed it with nuisance regressors.

Main Results:

• The proposed method objectively identifies fMRI data with motion artifacts.
• Statistical suppression of residual motion improved activation map correlation in affected subjects.
• Correlation between individual and group activation maps doubled for corrected subjects.
• Reduced need for subject exclusion, leading to more efficient data utilization.

Conclusions:

• The unsupervised learning approach provides an objective threshold for motion artifact identification in fMRI.
• Integrating nuisance regressors effectively suppresses residual motion, improving data quality.
• This method enhances fMRI data efficiency by minimizing subject exclusion and data loss.