Anatomical Information in Beamformer Images

Abstract

Event Abstract Back to Event Anatomical Information in Beamformer Images Claire Stevenson1*, Matthew Brookes1, Peter Morris1 and Gareth Barnes2 1 University of Nottingham, Athena project team, United Kingdom 2 UCL, Athena project team, United Kingdom A metric for assessing the amount of anatomical information contained in a source power map would prove a useful tool, both for validating and comparing source localization techniques. Here we extend previous work [1] to determine both the amount of anatomical information in an estimated current distribution image and the scale of the anatomical information present. MEG data were collected in 4 subjects for a uni-lateral finger movement task. A beamformer was used to create unthresholded 1mm isotropic Pseudo-T images, comparing movement and rest in the 15-30Hz and 180-200Hz frequency bands. Pial surface meshes were extracted from MR anatomicals and their associated spherical harmonic components computed. Meshes were smoothed (3mm Gaussian kernel) and coregistered to the MEG measurement space. Surrogate brains were created by randomly rotating sets of adjacent spherical harmonic components at different spatial scales. The Pearson correlation coefficients between the MEG images and the true, as well as the surrogate, anatomy were computed. The correlation between the MEG image and the true brain was consistently greater than that of the MEG image and surrogates for the 15-30Hz image. For an example subject, mean MEG to surrogate correlation rose from R2=0.0358+/-0.004 to R2=0.0429+/-0.002 for harmonic distortions 8-14 and 26-32 respectively, whereas the correlation between the MEG image and the true anatomy was R2=0.0478. For the 180-200Hz image, true brain correlation was much smaller, R2=0.00383, and within the errors for correlation with surrogates computed across all L values (Lrot 26-32 R2=0.00334+/-0.00052 and Lrot 8-14 R2=0.00325+/-0.0008). This is a useful metric for assessing the accuracy of source localization techniques and demonstrates that significant anatomical information on a fine spatial scale is present in MEG beamformer images.