Abstracts

Rationale: The recent publication (Frauscher et al., Brain 2018; 141, 1130-44) of an open-source atlas of intracranial electroencephalography (iEEG) makes publicly available for the first time a repository of normative iEEG data from the majority of cortical areas of the human brain. We term this atlas the Montreal Neurological Institute Atlas (MNIA). In this work we present a method for the construction of a novel neurophysiological ‘brain map’ from the MNIA data. Methods: The data were downloaded from https://mni-open-ieegatlas.research.mcgill.ca, mean-subtracted, high-pass filtered above 1 Hz, normalized to unit variance and divided into successive 10s segments. The amplitude spectral density was computed in the 1-100 Hz frequency range, smoothed and resampled on a logarithmic (base 2) frequency scale. The resulting profile was fit to a five-component Gaussian mixture (GM) model (Kalamangalam et al., Clin Neurophysiol 2016; 32, 331-40). Each Gaussian was defined by its amplitude A, mean µ and standard deviation σ, so that procedure yielded the multicomponent model pertaining to each data epoch as the 15-dimensional vector of parameters [A_δ,µ_δ,σ_δ; A_θ,µ_θ,σ_θ; A_α,µ_α,σ_α; A_β,µ_β,σ_β; A_γ,µ_γ,σ_γ]. The full 15-dimensional representation was simplified to a 5-dimensional vector of just the amplitudes of the individual Gaussians. The 5-dimensional vector was converted into an RGB map, with intensity and hue variations. Each of the 38 MNIA regions was brought into correspondence with the Destrieux (Destrieux et al., Neuroimage 2010; 53, 1-15) and the subcortical Harvard-Oxford atlases (Desikan et al., Neuroimage 2006; 31, 968-80), and the RGB map plotted on the ICBM152 brain model using the correspondence. Results: All MNIA brain areas were fit to a high degree of accuracy by the GM modelling procedure. Statistical analysis showed that the brain regions, when classified into lobar territories, were statistically separable. The figure shows pial and inflated views of the ICBM cortical surface, colored to reflect the relative contributions of the Gaussian amplitudes to the iEEG spectral model from that area. Conclusions: Our method classifies the first public normative database of human intracranial EEG into a low-dimensional Fourier spectral-based formalism that yields a novel neurophysiological brain map visualized on a colorized cortical surface. Our method summarizes all the data of MNIA with high accuracy and proves the statistical individuality of each sampled brain region. On a neurobiological level, the uniform applicability of the model to all the data across the entire bandwidth suggests a corresponding uniformity in brain rhythm generating processes in the cerebral cortex, i.e., a ‘continuum hypothesis’ for iEEG generation. Implications of the findings, for both the study of normal brain function and neurological disease, will be discussed. Funding: This work was partially supported by the National Institute of Neurological Disorders and Stroke (5K23NS079900 to GPK) and the Wilder family endowments to the University of Florida.