On the Keyhole Hypothesis: High Mutual Information between Ear and Scalp EEG

Kaare B. Mikkelsen, Preben Kidmose, Lars Kai HansenFrontiers in Human Neuroscience 2017 · DOI

Abstract

This paper proposes and tests the "keyhole hypothesis"—the idea that low-dimensional EEG measurements, such as those from ear-EEG, reflect a broadly distributed set of neural processes. By formulating this hypothesis in information-theoretical terms and utilizing linear predictive models, we demonstrate high mutual information between scalp and ear-EEG data. Our findings show that ear-EEG signals can be reliably predicted from scalp data and vice versa, with these mappings remaining stable across various mental states and long durations. This suggests that the ear-EEG "keyhole" provides a surprisingly wide view of global brain activity.

Introduction

Traditional Electroencephalography (EEG) is the gold standard for monitoring brain activity but remains impractical for long-term, daily use due to bulky caps and conductive gels. Ear-EEG—placing electrodes discreetly in the ear canal—offers a wearable alternative. However, skepticism exists regarding whether such a localized sensor can capture complex, distributed neural processes. We propose the Keyhole Hypothesis: due to volume conduction within the head, a small sensor in the ear acts as a keyhole through which much of the brain's global state can be observed.

Methodology: Modeling the Keyhole

To quantify the shared information between the scalp and the ear, we employed linear predictive modeling. We developed models to predict ear-EEG signals from scalp data ($x \to \hat{y}$) and scalp signals from ear data ($y \to \hat{x}$). The correlation coefficient ($\rho$) between predicted and measured signals serves as a lower bound for the Shannon communication capacity of the channel between these two regions.

Schematic of the forward models and the linear prediction models for scalp-to-ear and ear-to-scalp mapping.

To ensure the models captured genuine neural relationships rather than noise, we used "leave-one-paradigm-out" cross-validation. Models were trained on one task (e.g., eyes-closed) and tested on another (e.g., auditory stimuli). Additionally, a 13-hour longitudinal study tracked signal stability across wakefulness and sleep.

Results: Predictive Power and Stability

Our analysis of laboratory and "in the wild" datasets yielded several key insights:

  • High Predictive Accuracy: Correlation coefficients for predicting ear-EEG from scalp data typically ranged from 0.1 to 0.6. In many cases, the predicted signals showed a higher signal-to-noise ratio (SNR) than actual measurements by filtering out local ear-specific noise.
  • Stability Across States: The linear mapping between the scalp and ear remained remarkably stable across different mental tasks. The only significant shift occurred during sleep, likely due to changes in underlying neural sources rather than physical conduction properties.
  • Bidirectionality: The ear-EEG could reconstruct significant portions of the scalp EEG, particularly in temporal regions, with correlations often exceeding 0.6.
Comparison between measured ear-EEG (cyan) and predicted signals (red) showing high correlation across different time segments.

Furthermore, the models successfully reconstructed classic Event-Related Potentials (ERPs), such as auditory onset and mismatch negativity (MMN), from ear-EEG data, matching waveforms measured directly on the scalp.

Conclusion

The evidence strongly supports the Keyhole Hypothesis. Ear-EEG is not merely a "narrow" local measurement; it captures aggregated activity from across the cortex. While the relationship is sensitive to physical electrode placement, its stability across mental states suggests that ear-EEG is a robust tool for long-term brain monitoring. Future work will focus on decoding specific mental states, such as fatigue or focus, through this ear-EEG "keyhole."

Summary updated

Abstract

We propose and test the keyhole hypothesis-that measurements from low dimensional EEG, such as ear-EEG reflect a broadly distributed set of neural processes. We formulate the keyhole hypothesis in information theoretical terms. The experimental investigation is based on legacy data consisting of 10 subjects exposed to a battery of stimuli, including alpha-attenuation, auditory onset, and mismatch-negativity responses and a new medium-long EEG experiment involving data acquisition during 13 h. Linear models were estimated to lower bound the scalp-to-ear capacity, i.e., predicting ear-EEG data from simultaneously recorded scalp EEG. A cross-validation procedure was employed to ensure unbiased estimates. We present several pieces of evidence in support of the keyhole hypothesis: There is a high mutual information between data acquired at scalp electrodes and through the ear-EEG "keyhole," furthermore we show that the view-represented as a linear mapping-is stable across both time and mental states. Specifically, we find that ear-EEG data can be predicted reliably from scalp EEG. We also address the reverse view, and demonstrate that large portions of the scalp EEG can be predicted from ear-EEG, with the highest predictability achieved in the temporal regions and when using ear-EEG electrodes with a common reference electrode.
BibTeX
@inproceedings{mikkelsen2017keyhole,
  author = {Kaare B. Mikkelsen and Preben Kidmose and Lars Kai Hansen},
  title = {{On the Keyhole Hypothesis: High Mutual Information between Ear and Scalp EEG}},
  booktitle = {Frontiers in Human Neuroscience},
  year = {2017},
  doi = {10.3389/fnhum.2017.00341},
}
Figure