Sparse Modeling for Prospective Head Motion Correction

In this talk, I present recent work exploiting sparse modeling to correct prospectively for head motion in functional magnetic resonance imaging (fMRI). I begin by motivating the problem of head motion and describe existing approaches for correcting for such motion, both prospectively (during a scan) and retrospectively (after the scan, during the data analysis). Existing prospective correction often relies on side information to estimate motion. Instead, the proposed method estimates motion from the fMRI data directly with a combination of sparse modeling and Kalman-like filtering. I depict statistical correlation analysis of simulated data, performed with and without the proposed prospective motion correction applied in real time. I demonstrate that not only is the improvement in sensitivity and specificity notable, it remains so even when the motion and noise model parameters are estimated from the data. In the near term, I will apply my algorithms to fMRI experiments with real human subjects.

This research is supported by an NIH NRSA Postdoctoral Fellowship (NIH F32 EB015914).