Hermann Sonntag | The effect of uncertainty in MEG-to-MRI co-registrations on MEG inverse problems

For high precision in source reconstruction of magnetoencephalography (MEG) or electroencephalography data, high accuracy of the coregistration of sources and sensors is mandatory. Usually, the source space is derived from magnetic resonance imaging (MRI). In most cases, however, no quality assessment is reported for sensor-to-MRI coregistrations. If any, typically root mean squares (RMS) of point residuals are provided. It has been shown, however, that RMS of residuals do not correlate with coregistration errors. We suggest using target registration error (TRE) as criterion for the quality of sensor-to-MRI coregistrations. TRE measures the effect of uncertainty in coregistrations at all points of interest. In total, 5544 data sets with sensor-to-head and 128 head-to-MRI coregistrations, from a single MEG laboratory, were analyzed. An adaptive Metropolis algorithm was used to estimate the optimal coregistration and to sample the coregistration parameters (rotation and translation). We found an average TRE between 1.3 and 2.3mm at the head surface.

The uncertainty in MEG-to-MRI coregistrations propagates via the forward model to uncertainty in source estimates of MEG data. However, the common tools for source reconstruction in MEG or EEG analysis do not account for that source of uncertainty and usually only the variance of the magnetic noise is considered in the assessment of the standard error or covariance of source parameters. For realistic head models, the computational costs of source estimates are unfeasible for straightforward Monte Carlo simulations. To overcome this problem, a polynomial expansion of source estimates is constructed as a function of the coregistration parameters. Statistical measures like mean and variance of source estimates for uncertain MEG-to-MRI coregistrations are approximated in closed form from the polynomial expansion.

Poster