A New Open-Source Tool for EEG Source Reconstruction in Infants

Thursday, May 12, 2016: 11:30 AM-1:30 PM
Hall A (Baltimore Convention Center)
C. O'Reilly1, M. Elsabbagh2 and T. B. Team3, (1)Blue Brain Project, École Polytechnique Fédérale de Lausanne, Geneva, Switzerland, (2)McGill University, Montreal, PQ, Canada, (3)Birkbeck, University of London, London, United Kingdom
Background: EEG studies in autism have been so far limited to scalp signals even though this kind of analysis has known limitations (e.g., noisier patterns, impossibility to relate observations to brain structures). Tools for performing cortical source estimation are now widely available. However, limited availability of structural recordings of the infant’s head is among the factors hindering the adoption of latter techniques.

Objectives: With increasing availability of EEG recordings from infants at-risk for autism, our goal was to develop novel tools allowing progress in this area. We developed a population-averaged template allowing source reconstruction in one year-old infants without the need for individual participant MRI.

Methods: We used the BrainVisa pipeline to estimate a mesh surface of the cortical ribbon and some deep structures (e.g., the thalami) from the dataset of Shi et al. (2011) [PLoS One, 6(4), e18746] which contains an MRI template (average from 90 infants recorded at 1 year), tissue probability maps, and brain parcellation [Tzourio-Mazoyer, et al., (2002). NeuroImage, 15(1), 273-289]. Poor gray/white matter discriminability (typical for MRI at this age) and fuzziness due to inter-participant averaging of MRI volumes required additional manipulations to guide BrainVisa. Thus, a gray/white matter mask was computed using a Python toolbox (NiBabel) to merge the information from the gray and white matter probabilistic maps (voxels intensity < 25% of maximal intensity were set to 0; remaining voxels were classified as gray or white matter depending on the respective maps where their intensity was higher). The MRI-space Tzourio-Mazoyer parcellation was propagated to the cortical and subcortical mesh by coregistering every vertex with the corresponding voxel. BrainVisa provided a poor skull reconstruction, so Brainstorm was used for that purpose. As suggested by experimental results showing skull thickness of 1.5 to 4 mm in one year-old infants [Li et al., (2015) PLoS One, 10(5), e0127322], an 2.75 mm skull thickness was used for reconstructing the scalp, outer skull interface, and inner skull interface with boundary-element method.

Results: The resulting template is available online in the Matlab Brainstorm package and can freely be used for EEG source analysis of high-density EEG recordings.

Conclusions: Availability of these new tools will promote analyses of topologically-resolved functional connectivity in the study of autism through the estimation of EEG sources. Moreover, as data from large scale MRI studies of infants at-risk becomes available, it will be possible to improve validity of these EEG tools for infants across the first three years of life.