Task 1: Computer face inversion task
We first replicated the well-characterised face inversion effect in a standard laboratory setting. Linear mixed effects modelling was used to test for changes in power for upright and inverted faces, showing that EEG low frequency power over posterior electrodes was significantly greater for inverted faces compared to upright faces (mean difference = 0.404, t(1427) = 8.27, p < 0.0001; Figure 3A). Plotting EEG power across frequencies further illustrates that low frequency power differences peaked between 5 and 12 Hz (Figure 3B). Further, we calculated power differences for each electrode, showing greater power for inverted compared to upright faces that were primarily located at posterior central and posterior lateral electrodes (Figure 3C).
Task 2: mEEG + photos
Our next analysis asked whether face inversion effects could be seen in a more naturalistic setting using mobile EEG while participants viewed upright and inverted pictures of faces placed on the walls. Following the analyses used for the computer-based task, we first performed a linear mixed effects analysis of EEG power at posterior electrodes. This revealed a significant effect of face inversion, with greater power for inverted faces compared to upright faces (mean difference = 0.190, t(597) = 2.38, p = 0.017; Figure 3). Across the scalp, inversion effects were greatest over posterior central electrodes and frontal electrodes (Figure 3D). This shows that face inversion effects are detectable using mobile EEG in a natural indoor setting.
Task 3: mEEG + AR
An important issue for studies using mobile EEG in natural settings, is the ability to manipulate the environment for the purposes of the experiment. Here, we combine mobile EEG with a head-mounted AR system which enables us to present virtual objects embedded within the real environment. Upright and inverted virtual heads were placed at various locations along the corridor and in the adjoining rooms, and participants repeatedly viewed the faces and pressed a button when they were fixating on the face. A linear mixed effects model of EEG power averaged over posterior electrodes showed a significant effect of face inversion, with greater power for inverted compared to upright faces (mean difference = 0.170, t(828) = 2.54, p = 0.0112; Figure 3A). Across the scalp, the inversion effect was maximal over posterior central electrodes (Figure 3E), and were partially overlapping with those seen in the computer-based and Mobile+photos tasks, with similar effect sizes in both mobile tasks. Through the combination of mobile EEG and head-mounted AR, this analysis establishes a feasible approach to studying cognitive processes in natural, real environments in which the participant is immersed.
Figure 3. EEG results. A) Face inversion effect sizes for each experimental task. Red x indicates mean inversion effect with individual subjects shown by grey circles. B) Spectrogram showing group mean power between 4 and 35 Hz for upright and inverted conditions, and the difference between them. C-E) Topographies showing mean power difference for inverted-upright faces between 5 and 15 Hz for the computer task (C), mEEG + photos (D) and mEEG + AR (E).
Participant motion and EEG:
Finally, we quantified how participant motion, measured by the accelerometers within the LiveAmp 64 amplifiers, related to EEG signal amplitudes. To do this, the continuous data (EEG channels plus 3 accelerometer channels) were split into 2 second non-overlapping chunks, and the root mean square (RMS) calculated for each channel. The RMS was averaged across the three accelerometer channels, creating one value per 2 second period. These values were then binned into low, medium and high motion RMS groups, before averaging the EEG RMS according to the motion RMS bins. This resulted in an EEG RMS value for each electrode and each of the low, medium and high RMS bins. A linear mixed effects model was used to test if the EEG RMS values related to the levels of participant motion (as defined by the motion RMS bins).
Our results suggest EEG RMS signals appear stable over the 3 motion levels for both mobile tasks (Figure 4AB), with no significant effects of motion level on EEG RMS (Mobile + photos: estimate -0.003, t(1534) = -0.12, p = 0.90; Mobile + AR: estimate 0.022, t(1534) = 0.79, p = 0.43).
Figure 4. EEG RMS during low, medium and high participant motion for A) the Mobile + photos task and B) the Mobile + AR task. Red x indicates mean RMS over electrodes and participants with boxplots showing the distribution of RMS values across electrodes. Topographies show electrode RMS values for each motion RMS bin.