Anatomically Veridical On-Scalp Sensor Topographies

Abstract

When working with sensor-level data recorded using on-scalp neuroimaging methods such as electroencephalography (EEG), it is common practice to use two-dimensional (2D) representations of sensor positions to aid interpretation. Positioning of sensors relative to anatomy, as in the classic 10-20 system of EEG electrode placement, enables the use of 2D topographies that are familiar to many researchers and clinicians. However, when using another increasingly popular on-scalp neuroimaging method, optically pumped magnetometer-based magnetoencephalography (OP-MEG), bespoke sensor arrays are much more common, and these are not prepared according to any standard principle. Consequently, polar projection is often used to produce individual sensor topographies that are not directly related to anatomy and cannot be averaged across people simply. Given the current proliferation of OP-MEG facilities globally, this issue will become an increasing hindrance when visualising OP-MEG data, particularly for group studies. To address this problem, we adapted and extended the 10-20 system to build a flexible, anatomical projection method applied to digitised head shape, fiducials and sensor positions. We demonstrate that the method maintains spatially veridical representations across individuals improving on standard polar projections at varying OPM sensor array densities. By applying our projection method, the benefits of anatomically veridical 2D topographies can now be enjoyed when visualising data, such as those from OP-MEG, regardless of variation in sensor placement as in sparse or focal arrays.

Keywords: 10–20; EEG; data visualisation; magnetoencephalography; optically pumped magnetometer; topography.

FIGURE 1

An overview of the 10–20 system transformation from 3D head shape to 2D polar grid. (a) Fiducials are marked and connected by lines crossing the scalp surface. Anatomical references (NAS, INI, LPA, RPA and Cz) are shown along with electrode placements that define the circumference of the polar grid shown in (b). Note, paired marks on lines indicate equidistance. (b) An illustration of the topography representing anatomical reference points. (c) Positions on the scalp (S1, S2) and their intersection with the circumference (C1, C2) are defined by a line from the vertex (Cz) to the scalp position, as marked. (d) This process is repeated on the 2D polar grid to represent scalp positions on the topography. NAS = nasion, INI = inion, LPA = left preauricular, RPA = right preauricular.

FIGURE 2

Flowchart showing a high‐level overview of the anatomical projection method. This protocol requires only the input of fiducial, head and sensor positions, with no further manual steps.

FIGURE 3

Visual explanation of how to measure across a convex mesh surface using a plane intersection method. (a) A plane is defined based on three points: two points on the mesh surface (yellow and magenta dots) and the origin (white dot). The normal of these three points (red arrow) defines a plane, shown by the grey gradient. Points where this plane intersects with the mesh are shown by the black dots. (b) In order to measure along the mesh from yellow to magenta, the positions must be indexed. This is achieved by measuring the polar angle (θ) to one of the points—in this case yellow—orthogonal to both the plane normal and a vector from the origin to each point. With the points indexed, a cumulative sum of distances between points from yellow to magenta (orange dashed line), following a left‐hand rule, can be taken. (c) An example of applying this method of measuring across the surface to find the distance from Cz to a position on the scalp. The dotted line represents the circumference of the head.

FIGURE 4

Comparison of anatomical and polar projection using varying numbers of OPM sensors on the head. The top panel (anatomical projection) shows that highlighted sensors are consistent across all conditions. However, in the lower panel (polar projection), highlighted sensor positions vary in each sensor count condition, producing unreliable results.

FIGURE 5

Performance of anatomical projection versus polar projection with varying number of channels under sparse and focal conditions. (a) An example of the data generated for this simulation based on the left cuneus ROI (LH_cuneus). Data based on a single subject's head geometry and sensor positions are shown having been used to produce a layout for 16 channels either randomly selected (sparse) or selected based on proximity to the channel with maximum leadfield magnitude (focal). (b) A summary of the mean t‐value (across ROIs and topographical grid) by number of channels under the conditions shown. Anatomical projection consistently outperformed polar projection, regardless of sparsity/focality and number of channels.