Introduction to ear-eeg

Brain-computer interface (BCI)the system uses a variety of neuroimaging methods, including electroencephalography (EEG)it is the most commonly used brain-computer interface system due to its high temporal resolution and portability. in recent years, eeg around or inside the ear has been (ear-EEG)it is beginning to be used to develop practical brain-computer interface systems.

So far, the feasibility of ear-eeg-based brain-computer interfaces has been confirmed by many studies in terms of performance, setup time, and long-term use.

Previous brain-computer interfaces mostly used electroencephalograms induced by external stimulation, such as steady-state evoked potentials (SSEP)event-related potentials (ERP)。some ear-eeg-based bci studies have used self-regulated eeg in cognitive psychological tasks, such as motor imagery. (MI)and mental arithmetic (mental arithmetic, MA)。

In our recent study, we validated the use of ear-eeg to develop endogenous BCI the feasibility of the system, we will MA we also investigated ear-eeg-based BCI the optimal re-reference method for the system and confirm that the contralateral average information is most useful as a reference signal. in addition, we are currently implementing a real-time ear-eeg based BCI system to examine its feasibility in an online environment and to try to improve the ear-eeg based BCI system performance.

In our recent study, we validated the use of ear-brain electroencephalography to develop endogenous BCI feasibility of the system. in this study, we used the performance of scalp-eeg to MA we also studied the optimal re-reference method for the ear-eeg-based bci system and confirmed that the contralateral average information is the most useful as a reference signal. in addition, we are currently implementing a real-time ear-eeg-based bci system to test its feasibility in an online environment and trying to use deep learning algorithms to improve the performance of the ear-eeg-based bci system.

  Figure 2 classification accuracy of four regions of interest

Exist [2] in this study, we investigated the use of ear-eeg to develop BCI to investigate the feasibility of this approach, we recorded ear-eeg from behind both ears and compared it with conventional scalp-eeg. (figure 1. electrode positions for eeg )。in the experiment, 18 subjects performed MA and baseline tasks (BL， resting state )，we evaluated four regions of interest: frontal, middle, occipital, and auricular (ROI)the classification accuracy of (figure 1 )。as a result, the average classification accuracy of ear-eeg was 78.36%. although the relative performance of ear-eeg was lower than that of scalp eeg, it was comparable to that of scalp eeg and there was no significant difference. (figure 2 )。based on this result, we can demonstrate the feasibility of developing an endogenous bci system using ear-eeg.

Reference sites, such as the mastoid, earlobe, or nose, are often used to measure scalp-eeg because the potentials at these sites are relatively weak relative to the recording site. when recording ear-eeg, the reference sites are limited due to the miniaturization of the device structure. [2]we systematically study the effects of different re-referencing methods on the performance of ear-eeg based bci using the same dataset used in . we test 5 different re-referencing methods: (figure 3 )，the overall advantage of the contralateral mean reference method using the average of the contralateral eeg over the recorded eeg was confirmed; higher signal-to-noise ratio and classification performance were obtained using the contralateral mean method.

Based on previous research, we implemented an online brain-computer interface system based on ear-eeg. each subject conducted three experiments over three days. on the first day of the experiment, we used an offline experiment to conduct MA、BL and word association (WA)the subjects were divided into three intelligence tasks, and the optimal combination of two intelligence tasks was selected based on the classification results. on the second and third days of the experiment, the optimal intelligence task was selected for each subject to conduct online experiments to investigate the test-retest reliability of the implemented online brain-computer interface system. however, the average online performance was less than 70%, which was not good for the two categories. BCI this is an acceptable classification accuracy for the proposed approach, so we are currently using deep learning algorithms to perform pseudo-online analysis to improve the classification performance.

D. classification of eyes open and eyes closed

  Many studies have demonstrated the use of ear-eeg to develop BCI and showed that their performance is comparable to that of scalp eeg-based BCI in terms of experimental setup and portability, the ear-eeg developed BCI although more convenient than scalp eeg, the performance of ear-eeg is still lower than that of scalp eeg due to the limited information obtained around the ears. BCI future directions will be to use new ear-eeg specific features and classification algorithms to improve the classification performance.