Real-Time Filtering with Sparse Variations for Head Motion in Magnetic Resonance Imaging

Area of Science:

• Neuroimaging
• Signal Processing
• Biomedical Engineering

Background:

• Estimating time-varying signals like head motion in functional magnetic resonance imaging (fMRI) is complicated by concurrent temporal dynamics, such as functional activation.
• Accurate motion estimation is crucial for reliable fMRI data analysis and interpretation.

Purpose of the Study:

• To develop and evaluate a new Kalman filter-like framework for real-time estimation of head motion during fMRI acquisition.
• To improve the accuracy of prospective motion correction by incorporating a sparse residual term into the measurement model.

Main Methods:

• A novel Kalman filter-like framework is proposed, featuring a sparse residual term in the measurement model.
• An iterative augmented Lagrangian algorithm, akin to the alternating direction method of multipliers, is employed for the filter's update step.
• The method's accuracy, convergence rate, and parameter sensitivity are evaluated using simulated fMRI data, testing both small and large motion scenarios.

Main Results:

• The proposed iterative method demonstrates accurate real-time motion estimation suitable for prospective motion correction.
• Experiments on simulated fMRI data show improvements in the maximum Youden's J index for time series analysis by 2-3% compared to retrospective motion correction alone.
• Combining prospective and retrospective correction significantly increases the sensitivity index from 4.3 to 5.4.

Conclusions:

• The novel Kalman filter framework effectively addresses the challenge of estimating time-varying signals in the presence of other dynamics, such as head motion during fMRI.
• This approach enables robust real-time motion estimation, enhancing the quality and reliability of fMRI studies.
• The integration of this method with existing techniques offers a substantial improvement in motion correction efficacy for neuroimaging applications.