Optical flow approaches to the identification of brain dynamics

Abstract

The superior temporal resolution of magneto- and electroencephalography (MEEG) provides unique insight into the dynamics of brain function. The analysis of the spatial dimensions of MEEG recordings can take a multiplicity of approaches: from the original scalp recordings to the identification of their generators through localizing or imaging techniques. Overall, both MEEG native or imaging data may be considered as multidimensional structures with potentially dense information contents. Quantitative analysis of the spatiotemporal flow of information conveyed by MEEG begins with a feature-extraction problem, thereby leading to improved insight in the multidimensional structure of brain dynamics. In this contribution, we approach this endeavor by suggesting that brain dynamic features from local through global spatial scales may be identified using the previously-introduced technique of surface-based optical flow. We illustrate this assertion by the quantitative analysis of time-resolved sequences of brain activity through the identification of episodes of relative topographical stability. In that respect, we revisit the concept of brain microstates with a new approach and distinct operational hypotheses. Local dynamic features from a variety of brain systems may also be explored through this methodology, as illustrated by experimental data on fast responses in the visual system as revealed by MEEG source imaging.

Spatiotemporal dynamics of optogenetically induced and spontaneous seizure transitions in primary generalized epilepsy

“…Directions of propagation of LFP waves were extracted using analysis based on the Horn-Schunck method (Horn and Schunck 1981), a computer vision algorithm classically used to extract the optical flow, or apparent motion, between two consecutive frames. Optical flow methods (Horn-Schunck or similar) have been used previously to study the spatiotemporal dynamics of neural signals (Lefèvre and Baillet 2009;Mohajerani et al 2013;Slater et al 2012). Briefly, we normalized LFPs, interpolated spatially, applied a mask at a negative threshold, ran the Horn-Schunk algorithm, and took the average optical flow, yielding a directionality vector for each time.…”

Increasing the accuracy of electromagnetic inverses using functional area source correlation constraints

“…The associated FACE procedure we presented for the main set of simulations and the analysis of the real data assumed a L2‐minimum‐norm uninformed procedure, this choice was mainly motivated by the fact that this method is widely used in the neuroscience community as emphasized by the increasing number of associated articles published in the last years [Busse et al,2009; Cottereau et al2011a,b; Jerbi et al,2007; Lefevre and Baillet,2009; Schupp et al,2007; Sergent et al,2005]. However, our priors on the matrix R can be easily introduced into other reconstruction techniques if they include the correlation matrix of the cortical current distributions in the inverse procedure.…”

Mapping dynamical properties of cortical microcircuits using robotized TMS and EEG: Towards functional cytoarchitectonics

“…Significant spatial differences in the LSA map across stimulated areas were assessed over time using the Skillings-Mack test (Chatfield and Mander, 2009). This test is derived from the Friedman test (non-parametric equivalent of the repeated measures ANOVA test) and can handle missing data.…”