Non-invasive laminar inference with MEG: Comparison of methods and source inversion algorithms

Abstract

Magnetoencephalography (MEG) is a direct measure of neuronal current flow; its anatomical resolution is therefore not constrained by physiology but rather by data quality and the models used to explain these data. Recent simulation work has shown that it is possible to distinguish between signals arising in the deep and superficial cortical laminae given accurate knowledge of these surfaces with respect to the MEG sensors. This previous work has focused around a single inversion scheme (multiple sparse priors) and a single global parametric fit metric (free energy). In this paper we use several different source inversion algorithms and both local and global, as well as parametric and non-parametric fit metrics in order to demonstrate the robustness of the discrimination between layers. We find that only algorithms with some sparsity constraint can successfully be used to make laminar discrimination. Importantly, local t-statistics, global cross-validation and free energy all provide robust and mutually corroborating metrics of fit. We show that discrimination accuracy is affected by patch size estimates, cortical surface features, and lead field strength, which suggests several possible future improvements to this technique. This study demonstrates the possibility of determining the laminar origin of MEG sensor activity, and thus directly testing theories of human cognition that involve laminar- and frequency-specific mechanisms. This possibility can now be achieved using recent developments in high precision MEG, most notably the use of subject-specific head-casts, which allow for significant increases in data quality and therefore anatomically precise MEG recordings.SectionAnalysis methods.ClassificationsSource localization: inverse problem; Source localization: other.

Cortical source imaging of resting-state MEG with a high resolution atlas: An evaluation of methods

“…However, because the forward mapping has a unique solution for given source dynamics, (cross-validated) variance explained r 2 CV between the forward mapped source solution and the measured MEG data can be viewed as a proxy for accuracy of the source space solution. Indeed, Bonaiuto et al (2018) found high correlations (0.98-1) be-tween cross-validated sensor space errors and source space free energy for simulated dipoles. Sensor space variance explained is particularly useful for quantifying the quality of source reconstruction of resting-state data, since it makes no assumptions on the number or locations of active sources (i.e.…”

“…Another crucial methodological decision was choice of methods used to compare different algorithms. Previous studies have compared algorithms for source localization -identifying the origin of a small number of sources (Bai et al, 2007;Hassan et al, 2014;Bradley et al, 2016;Finger et al, 2016;Barzegaran and Knyazeva, 2017;Hassan et al, 2017;Hincapié et al, 2017;Bonaiuto et al, 2018;Pascual-Marqui et al, 2018;Seeland et al, 2018;Anzolin et al, 2019;Halder et al, 2019), such as known networks during task or simulated dipoles. These methods are not directly generalizable to resting-state data, where activity is not a point source but is distributed widely across the cortex.…”

“…Much effort has been made to assess the quality of source reconstruction algorithms in the literature, which mainly focuses on source localization, i.e. identifying the origin of a small number of sources, for example evoked potentials with ground truth based on known task-relevant activity (Bai et al, 2007;Hassan et al, 2014;Seeland et al, 2018;Halder et al, 2019), or simulated activity at a small number of dipoles (Bradley et al, 2016;Finger et al, 2016;Barzegaran and Knyazeva, 2017;Hassan et al, 2017;Hincapié et al, 2017;Bonaiuto et al, 2018;Pascual-Marqui et al, 2018;Anzolin et al, 2019;Halder et al, 2019).…”

Estimates of cortical column orientation improve MEG source inversion

“…In future work, the curvature of cortical columns could be approximated using sequences of straight vectors computed from laminar equivolumetric surfaces (Waehnert et al, 2014;Wagstyl et al, 2018). If each vector was tangential to the corresponding segment of the actual (curved) cortical column, this would result in a piecewise linear estimate of column shape, which may allow more precise source localization (Bonaiuto et al, 2018b(Bonaiuto et al, , 2018aTroebinger et al, 2014a). This development would benefit from higher resolution (e.g.…”

“…We performed source inversion, and compared the resulting model fits in terms of relative free energy compared to that of the downsampled surface normal model (the current most commonly used method). This was repeated using source space models restricted to the pial surface, white matter surface, and combined pial -white matter surface (Bonaiuto et al, 2018a(Bonaiuto et al, , 2018b. In this case the combined pial-white model had double the number of sources and these sources could be arranged with identical orientations on each surface (link vectors); or different orientations (downsampled surface normals, original surface normals, and variational vector field)..…”