EEG beyond the classical laboratory environment has the potential to change our understanding of how the brain works by providing insights into neuronal processes in everyday life. Long term EEG recordings in everyday life, however, will remain limited for practical considerations in the number of electrodes that can be placed. Ear-EEG may provide a viable middle ground between transparency (Bleichner et al. (2015); Bleichner and Debener (2017)) and signal sensitivity. To make optimal usage of the small area in- and around the ear, and the relatively small number of electrodes that can be placed here, we need a better understanding of sensor-source relationships. Simulations as the ones provided in this study aim to elucidate this relationship. This will help developing better sensor technology, identifying possible ear-EEG applications and understanding inter-individual differences in ear-EEG recordings.
We addressed the sensor-source relationship in EEG by showing how the amplitudes that can be recorded around the ear are dependent on the orientation, position and depth of the source of interest relative to the recording electrodes. These factors influence, first, whether it can be expected to reliably capture signals from a cortical location with ear-electrodes at all, and, second, which ear-electrodes are most sensitive for a given source. In terms of source orientation and position, our results demonstrate that while one electrode configuration may be highly sensitive to one source, it may be insensitive to another (see Fig. 2a1–3). While this point is trivial for experienced EEG researchers, it is an important factor when working with ear-EEG and a limited number of electrodes in and around the ears. From simulations in the depth panel in Fig. 2, it becomes apparent that deeper sources are harder to detect. As discussed in Kappel et al. (2019), the proximity of the source to the recording electrodes is one of several factors determining signal amplitude. That is, in an EEG recording, even with ideal orientation and position of a source, its signal still may not be observable at sensor level, due to the source’s distance to the electrodes and the resulting low signal amplitudes. Another influence that follows the same principle as the depth of the source and its distance to the electrodes is the signal strength. Even a source close to a channel might not be detectable if the electric potential difference is not strong enough (i.e. because of a too small population of neurons firing synchronously). However, a strong neural source at a large depth could still emit a signal with an amplitude high enough to be detected.
In summary, it should be recognized that orientation, position and depth are not fully independent properties of a source, but instead their complex combination determines what amplitudes can be recorded. These considerations are what should drive the placement of electrodes. While it is important to use a sufficiently high number of electrodes that are arranged in a way that captures as many different source orientations as possible, high user comfort, low visibility demands and technical constrains set some limitations.
Comparison of Ear- and Cap-EEG
We quantified the expected loss in signal amplitude for three forms of ear-EEG to compare their capability in recording from different cortical areas. For the cEEGrid, our simulations of the signal loss relative to cap-EEG show that temporal areas exhibit the lowest signal loss, both for the patches in Fig. 3 and the more detailed mapping of signal loss in Fig. 5. This is an indication that one can measure neural activities from these areas with a quality comparable to classical EEG. For a direct comparison between ear-EEG results and cap-EEG results (e.g. as reported in Bleichner et al. (2016)), only a small difference in amplitude can be expected for these areas. This is likely due the cEEGrid covering space below as well as above the ear with a relatively high density of electrodes. These findings are reassuring for ear-EEG targeting auditory processes located within the temporal lobe, like the auditory cortex (Woldorff et al. (1993); Grady et al. (1997); Hine and Debener (2007)). However, for areas further away, one can expect lower amplitudes in comparison to what is described in the literature for cap-EEG. In this context, one should not pay too much attention to the exact ranking of the patches in Fig. 3a and b in terms of signal loss, as this measure is dependent on the random parcelling of the cortex. This indication was confirmed by the fine-grained mapping of signal loss in Fig. 5. It can be seen that in the left temporal lobe, for two adjacent areas, the cEEGrid showed very pronounced differences in sensitivity. In fact, the region with the cEEGrid’s highest sensitivity on the cortex was within high proximity to areas with signal losses of over 50%. These variations in the fine-grained map are due to differently oriented sources, which underlines that the sensitivity of ear-EEG (or cap-EEG) is not only a matter of distance but an interplay of several factors. This observation highlights the necessity of individualized head models for any kind of source modeling.
For Fig. 5, note that the signal losses for medial areas must be interpreted with caution: here, the signal losses for the cEEGrid were low on large parts of the brain, indicating that these sources can be recorded well. Yet, the ventral views of both the sensitivity map of the high-density EEG and the cEEGrid (see Supplementary Information for the sensitivity maps) show that low amplitudes were recorded from these regions for both devices. Therefore, low signal losses here do not indicate a high performance of the cEEGrid, but only that neither cap- nor ear-EEG captured signals from medial areas very well, i.e. the cEEGrid is equally unsuitable as the cap. Besides the medial sources, another aspect in Fig. 6 (see Supplementary Information) must be mentioned. As can be seen in the lateral views of the cortex, there is a peak sensitivity in the left temporal lobe for both the cap-EEG and the cEEGrid that is not present for the contralateral hemisphere. The authors attribute this difference to an anatomical characteristic that can be seen in the ventral view (middle column) of the specific brain used here: in the left temporal region (here displayed on the right side) there is a bulge that is not present on the right side. We assume that the increased proximity of this cortical patch to the electrodes is the reason for this one-sided peak. This observation shows exemplary that the exact results of our simulation approach are not to be taken as generalizable to the entire population. On the contrary, while more global patterns can be derived from this detailed sensitivity map, it highlights that individual anatomical differences will influence what can or cannot be recorded with ear-EEG.
Arrangement of Ear-Electrodes
Manipulating the arrangement of electrodes leads to different amplitude recordings for various sources, in particular when done within a small area. Comparing the three different ear-EEGs, our results indicate that the bipolar channels have a higher signal loss relative to cap-EEG. Interestingly, while the patches with higher amplitude loss (being the ones far away from the electrodes that all uniformly measure amplitudes near zero) were comparable for both bipolar channel orientations, in the quarter with the lowest signal loss for the cEEGrid (areas 40–50), there were highly different signal losses (up to 68.98% difference in patch 46) for the two bipolar channels: in accordance with the results from Sect. Source orientation, when one orientation of a channel is suited well to capture a high amplitude from a source-patch, the orthogonal pair is most likely not. In the most extreme case, a cortical source that is captured well by one pair of electrodes will be “invisible” to another. For practical considerations, whether a single bipolar channel suffices to capture the signal of interest is difficult to answer a priory for an individual. Depending on the individual brain anatomy it may be sufficient for some people but not for others. The presence or absence of an (ERP) effect may hence be simply due to anatomical but not functional differences between people. A multi-channel setup will be less susceptible to this problem as the optimal channel configuration could be used for each participant and thereby reduce seemingly large inter-individual differences.
Besides the higher sensitivity to cortical sources when using several channels (instead of having only a single bipolar channel), multi-electrode setups are advantageous for pre-processing and analysis steps like artefact rejection and source localization. The quality of several algorithms depends highly on the spatial coverage and the number of electrodes used. Therefore, even if the optimal channel orientation and position for the detection of a signal was known a priori, a multi-channel EEG is still recommendable for high-quality measurements.
Our simulations show the limits of using only a single bipolar channel; a multi-channel setup will always provide a better sensitivity to a variety of neural sources. For some applications, however, a reduced number of electrodes may be wanted. It is therefore interesting to examine which electrodes could be discarded, if a minimal number of electrodes is of paramount importance. Therefore, in Fig. 4, the number of times a channel recorded the highest amplitude for a patch was counted. It can be seen that for the cEEGrid, no electrodes were redundant for optimally capturing differently oriented sources, since every electrode was at least once part of the channel that recorded the highest amplitude from a source. Yet, vertical channels (e.g. L2 and L7) are selected more frequently than horizontal ones (e.g. L8 and L5). The likely explanation is that vertically oriented electrode combinations have larger between-channel distances due to the ellipsoid shape of the cEEGrid. In general, a larger between-channel distance will lead to a higher amplitude, as discussed in Mirkovic et al. (2016), Bleichner et al. (2016) and Bleichner and Debener (2017).
Implications for ear-EEG
In conclusion, the number and the C-shaped arrangement the cEEGrid electrodes allow to capture neural activity from a wide range of orientations. Nevertheless, it seems likely that the setup would benefit from a perfectly round shape to capture orientations more precisely. To summarize the implications of our findings and to provide recommendations for the design of ear-EEG solutions, there is a sufficient theoretical basis to measure ear-EEG from areas with low signal loss, namely in the temporal lobe and adjacent areas. Regarding the arrangement of the electrodes, a simulation in Bleichner and Debener (2017) hinted towards ear-EEG sensitivity being dependent on both source- and channel orientation. The present paper advances these findings with more distinct simulations of different source- and channel properties and demonstrates in a systematic way that a C-shaped multi-electrode setup will be sensitive to more cortical sources than a single, bipolar channel-setup and arguably allows to account better for inter-individual differences, which are known to exist in source orientations.
Using only a single electrode pair will therefore reduce the number of cortical sources that can be recorded. Of course, determining to which areas ear-EEG is sensitive to must be found in applied research, yet with simulating the sensitivity of an electrode arrangement to dipole sources, there is a clearer guideline for experimental decisions. Regarding the number of electrodes in the case of the cEEGrid, one has to weigh between a less obtrusive, unilateral and a more visible bilateral design that can record higher amplitudes. The cEEGrid in particular meets some of the favourable conditions addressed in Sects. Source orientation, Source position, Source depth, including sensitivity to different source orientations and high proximity to temporal sources, which is especially useful for measuring auditory evoked potentials. In this context, from previous research carried out with the cEEGrid, we already know of some effects that can be measured reliably, as stated in the introduction of this paper. Consequently, we expect a high chance of similar source amplitudes with low signal loss to be reliably found in a real-life setting. For research with a limited number of electrodes, our approach is made in a way that is generalizable to other forms of ear-EEG for additional simulations. This will help to relate existing EEG results obtained with cap-EEG to ear-EEG.
The simulation of cortical activity can produce useful models for EEG research. To clarify some of the basic structures that are important in this regard, we included several factors that are known to influence the recording of a signal into our simulations, such as precise head and brain geometry, different electrode setups and several different source properties. While revisiting some of the fundamental principles of EEG, we nevertheless aimed to keep our work as simple and illustrative as possible. Yet, our simulations can be extended to account for higher complexity: first, in our forward model we seed activity either in one vertex or in a group of neighbouring vertices. For the rest of the brain, no activity is assumed. Obviously, neural activity is never confined to only one part of the brain. So, despite the fact that our model indicates some degree of sensitivity for a given region, this activity may be masked by activity from other regions that is simply larger in magnitude. It may be the case that certain areas, despite having a good sensitivity in our simulation, are not recordable with the cEEGrid in reality. A solution for this may be to add noise to the model and see how robust the measurement of a given activity is against noise.