BCI-Walls: A robust methodology to predict if conscious EEG changes can be detected in the presence of artefacts

PLOS ONE RESEARCH ARTICLE BCI-Walls: A robust methodology to predict if conscious EEG changes can be detected in the presence of artefacts Bernd Porr ID☯*, Lucı́a Muñoz Bohollo☯ Biomedical Engineering, University of Glasgow, Glasgow, Scotland, United Kingdom a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Porr B, Bohollo LM (2023) BCI-Walls: A robust methodology to predict if conscious EEG changes can be detected in the presence of artefacts. PLoS ONE 18(8): e0290446. https://doi. org/10.1371/journal.pone.0290446 Editor: Anwar P.P. Abdul Majeed, XJTLU: Xi’an Jiaotong-Liverpool University, CHINA Received: November 1, 2022 Accepted: August 8, 2023 ☯ These authors contributed equally to this work. * Abstract Brain computer interfaces (BCI) depend on reliable realtime detection of conscious EEG changes for example to control a video game. However, scalp recordings are contaminated with non-stationary noise, such as facial muscle activity and eye movements. This interferes with the detection process making it potentially unreliable or even impossible. We have developed a new methodology which provides a hard and measurable criterion if conscious EEG changes can be detected in the presence of non-stationary noise by requiring the signal-to-noise ratio of a scalp recording to be greater than the SNR-wall which in turn is based on the highest and lowest noise variances of the recording. As an instructional example, we have recorded signals from the central electrode Cz during eight different activities causing non-stationary noise such as playing a video game or reading out loud. The results show that facial muscle activity and eye-movements have a strong impact on the detectability of EEG and that minimising both eye-movement artefacts and muscle noise is essential to be able to detect conscious EEG changes. Published: August 24, 2023 Peer Review History: PLOS recognizes the benefits of transparency in the peer review process; therefore, we enable the publication of all of the content of peer review and author responses alongside final, published articles. The editorial history of this article is available here: https://doi.org/10.1371/journal.pone.0290446 Copyright: © 2023 Porr, Bohollo. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: The data underlying the results presented in the study are available from the University of Glasgow’s repository for research data: https://researchdata.gla.ac.uk/1258/ The code underlying the results presented in the Introduction Brain computer interfaces (BCI) have broadly the task to turn brain activity (EEG) into actions, for example, to control a character in a video game or a wheelchair. In the simplest way this is achieved by testing if the EEG signal has reached a certain threshold [1]. However, it is well known that noise originating from muscle activity, eye-movements and other movement artefacts can interfere with the detection of conscious EEG changes [2]. Various techniques have been devised to minimise the effect of noise, in particular the independent component analysis (ICA) for offline processing [3] but for realtime closed-loop applications such as BCI one needs to resort to direct causal filtering techniques using bandpass filters, the short-time Fourier Transform, wavelet transform [4–6] or the derivative [7]. All these approaches are inferior to the offline noise removal techniques such as ICA and even after filtering the remaining noise will substantially interfere with the detection process. Given the limited effectiveness of noise reduction techniques for BCI applications, electrode measurements will be substantially contaminated with noise, in particular EMG, because of the overlap in EEG and EMG frequencies, no matter what kind of pre-filtering has been PLOS ONE | https://doi.org/10.1371/journal.pone.0290446 August 24, 2023 1 / 22 PLOS ONE BCI-Walls study is available from Zenodo: https://doi.org/10. 5281/zenodo.7852162. Funding: This work was supported in part by the School of Engineering, University of Glasgow. Competing interests: B.P. is CEO of Glasgow Neuro LTD which manufactures the Attys DAQ board. This does not alter our adherence to PLOS ONE policies on sharing data and materials. applied. Consequently, the detection process itself needs to cope with the noise and minimise its impact. The standard solution to maximise the signal and minimise the noise during detection is averaging which can be achieved in both the time domain and frequency domain: • Time-domain: In the “evoked potential” (ep) paradigm the EEG is stimulus-locked and assumes that EMG noise is uncorrelated to the stimulus repetition and averages out. ep½m� ¼ N 1X d½m þ n � N� N n¼0 ð1Þ where N is the number of stimulus repetitions and their responses registered as d[m + n � N]. The more repeated stimuli N are presented the more the EMG noise is reduced. For example in a P300 speller a subject looks at a flashing “A” and the EEG is then added over and over again until a threshold has been reached. The more repetitions of the letter “A” the better the signal-to-noise ratio (SNR) but the longer the time to bring it over a threshold to decide the user has looked at the flashing “A”. • Frequency-domain: Here, the idea is that the subject can consciously reduce (or sometimes increase) the power of a narrow frequency band. To detect this change the signal is analysed in the frequency domain. If the band-power of a frequency band reaches a certain threshold then an action can be triggered for example moving a cursor. Again, averaging takes place because the Fourier Transform or a bank of bandpass filters accumulate the correlation between sine/cosine-waves (e−j2πkn/N) and chunks of EEG d[n]: X½k� ¼ N 1 X d½n� � e j2pkn=N k ¼ 0; 1; 2; . . . ; N 1 ð2Þ n¼0 where N is the number of samples the averaging takes place, d[n] is the EEG and X[k] its spectrum. If the chunk of EEG is long then the frequency spectrum will deliver clear peaks in the band of interest and thresholding becomes more and more reliable. No matter if the detection process is performed in the time- or frequency-domain one needs to wait for N samples until a decision can be made so that a signal reaches a threshold. Fig 1A shows such a case where a cartoon signal is shown where a signal has to reach the threshold γ which then can be used to control for example a cursor in a BCI game. The threshold γ0 is chosen in a way that the noise with noise variance s20 does not reach the threshold but the desired conscious EEG signal does (indicated with the tick symbol). Fig 1. Effects of stationary versus non-stationary noise on signal detection. A) Signal detection with stationary noise, B) signal detection with non-stationary no (...truncated)