Improving Auditory Steady-State Response Detection Using Independent Component Analysis on Multichannel EEG Data

A CNN-Based Deep Learning Approach for SSVEP Detection Targeting Binaural Ear-EEG

This paper discusses a machine learning approach for detecting SSVEP at both ears with minimal channels. SSVEP is a robust EEG signal suitable for many BCI applications. It is strong at the visual cortex around the occipital area, but the SNR gets worse when detected from other areas of the head. To make use of SSVEP measured around the ears following the ear-EEG concept, especially for practical binaural implementation, we propose a CNN structure coupled with regressed softmax outputs to improve accuracy. Evaluating on a public dataset, we studied classification performance for both subject-dependent and subject-independent trainings. It was found that with the proposed structure using a group training approach, a 69.21% accuracy was achievable. An ITR of 6.42 bit/min given 63.49 % accuracy was recorded while only monitoring data from T7 and T8. This represents a 12.47% improvement from a single ear implementation and illustrates potential of the approach to enhance performance for practical implementation of wearable EEG.

The identification of predominant auditory steady-state response brain sources in electroencephalography using denoising source separation

Different approaches have been used to extract auditory steady-state responses (ASSRs) from electroencephalography (EEG) recordings, including region-related electrode configurations (electrode level) and the manual placement of equivalent current dipoles (source level). Inherent limitations of these approaches are the assumption of the anatomical origin and the omission of activity generated by secondary sources. Data-driven methods such as independent component analysis (ICA) seem to avoid these limitations but only to face new others such as the presence of ASSRs with similar properties in different components and the manual selection protocol to select and classify the most relevant components carrying ASSRs. We propose the novel approach of applying a spatial filter to these components in order to extract the most relevant information. We aimed to develop a method based on the reproducibility across trials that performs reliably in low-signal-to-noise ratio (SNR) scenarios using denoising source separation (DSS). DSS combined with ICA successfully reduced the number of components and extracted the most relevant ASSR at 4, 10 and 20 Hz stimulation in group and individual level studies of EEG adolescent data. The anatomical brain location for these low stimulation frequencies showed results in cortical areas with relatively small dispersion. However, for 40 and 80 Hz, results with regard to the number of components and the anatomical origin were less clear. At all stimulation frequencies the outcome measures were consistent with literature, and the partial rejection of inter-subject variability led to more accurate results and higher SNRs. These findings are promising for future applications in group comparison involving pathologies.

Error Correction Regression Framework for Enhancing the Decoding Accuracies of Ear-EEG Brain-Computer Interfaces

Ear-electroencephalography (EEG) is a promising tool for practical brain-computer interface (BCI) applications because it is more unobtrusive, comfortable, and mobile than a typical scalp-EEG system. However, an ear-EEG has a natural constraint of electrode location (e.g., limited in or around the ear) for acquiring informative brain signals sufficiently. Achieving reliable performance of ear-EEG in specific BCI paradigms that do not utilize brain signals on the temporal lobe around the ear is difficult. For example, steady-state visual evoked potentials (SSVEPs), which are mainly generated in the occipital area, have a significantly attenuated and distorted amplitude in ear-EEG. Therefore, preserving the high level of decoding accuracy is challenging and essential for SSVEP BCI based on ear-EEG. In this paper, we first investigate linear and nonlinear regression methods to increase the decoding accuracy of ear-EEG regarding SSVEP paradigm by utilizing the estimated target EEG signals on the occipital area. Then, we investigate an ensemble method to consider the prediction variability of the regression methods. Finally, we propose an error correction regression (ECR) framework to reduce the prediction errors by adding an additional nonlinear regression process (i.e., kernel ridge regression). We evaluate the ECR framework in terms of single session, session-to-session transfer, and subject-transfer decoding. We also validate the online decoding ability of the proposed framework with a short-time window size. The average accuracies are observed to be 91.11±9.14%, 90.52±8.67%, 86.96±12.13%, and 78.79±12.59%. This paper demonstrates that SSVEP BCI based on ear-EEG can achieve reliable performance with the proposed ECR framework.

Improving the power of objective response detection of evoked responses in noise by using average and product of magnitude-squared coherence of two different signals

Objective response detection (ORD) techniques such as the magnitude-squared coherence (MSC) are mathematical methods tailored to detect potentials evoked by an external periodic stimulation. The performance of the MSC is directly proportional to the signal-to-noise ratio (SNR) of the recorded signal and the time spent for collecting data. An alternative to increasing the performance of detection techniques without increasing data recording time is to use the information from more than one signal simultaneously. In this context, this work proposes two new detection techniques based on the average and on the product of MSCs of two different signals. The critical values and detection probabilities were obtained theoretically and using a Monte Carlo simulation. The performances of the new detectors were evaluated using synthetic data and electroencephalogram (EEG) signals during photo and auditory stimulation. For the synthetic signals, the two proposed detectors exhibited a higher detection rate when compared to the rate of the traditional MSC technique. When applied to EEG signals, these detectors resulted in an increase of the mean detection rate in relation to MSC for visual and auditory stimulation of at least 25% and 13.21%, respectively. The proposed detectors may be considered as promising tools for clinical applications. Graphical Abstract.

Extractions of steady-state auditory evoked fields in normal subjects and tinnitus patients using complementary ensemble empirical mode decomposition

BACKGROUND

Auditory steady-state response (ASSR) induced by repetitive auditory stimulus is commonly used for audiometric testing. ASSR can be measured using electro-encephalography (EEG) and magnetoencephalography (MEG), referred to as steady-state auditory evoked potential (SSAEP) and steady-state auditory evoked field (SSAEF), respectively. However, the signal level of SSAEP and SSAEF are weak so that signal processing technique is required to increase its signal-to-noise ratio. In this study, a complementary ensemble empirical mode decomposition (CEEMD)-based approach is proposed in MEG study and the extraction of SSAEF has been demonstrated in normal subjects and tinnitus patients.

METHODS

The CEEMD utilizes noise assisted data analysis (NADA) approach by adding positive and negative noise to decompose MEG signals into complementary intrinsic mode functions (IMF). Ten subjects (five normal and five tinnitus patients) were studied. The auditory stimulus was designed as 1 kHz carrier frequency with 37 Hz modulation frequency. Two channels in the vicinities of right and left temporal areas were chosen as channel-of-interests (COI) and decomposed into IMFs. The spatial distribution of each IMF was correlated with a pair of left- and right-hemisphere spatial templates, designed from each subject's N100m responses in pure-tone auditory stimulation. IMFs with spatial distributions highly correlated with spatial templates were identified using K-means and those SSAEF-related IMFs were used to reconstruct noise-suppressed SSAEFs.

RESULTS

The current strengths estimated from CEEMD processed SSAEF showed neural activities greater or comparable to those processed by conventional filtering method. Both the normal and tinnitus groups showed the phenomenon of right-hemisphere dominance. The mean current strengths of auditory-induced neural activities in tinnitus group were larger than the normal group.

CONCLUSIONS

The present study proposes an effective method for SSAEF extraction. The enhanced SSAEF in tinnitus group echoes the decreased inhibition in tinnitus's central auditory structures as reported in previous studies.

Habituation of Auditory Steady State Responses Evoked by Amplitude-Modulated Acoustic Signals in Rats

Generation of the auditory steady state responses (ASSR) is commonly explained by the linear combination of random background noise activity and the stationary response. Based on this model, the decrease of amplitude that occurs over the sequential averaging of epochs of the raw data has been exclusively linked to the cancelation of noise. Nevertheless, this behavior might also reflect the non-stationary response of the ASSR generators. We tested this hypothesis by characterizing the ASSR time course in rats with different auditory maturational stages. ASSR were evoked by 8-kHz tones of different supra-threshold intensities, modulated in amplitude at 115 Hz. Results show that the ASSR amplitude habituated to the sustained stimulation and that dishabituation occurred when deviant stimuli were presented. ASSR habituation increased as animals became adults, suggesting that the ability to filter acoustic stimuli with no-relevant temporal information increased with age. Results are discussed in terms of the current model of the ASSR generation and analysis procedures. They might have implications for audiometric tests designed to assess hearing in subjects who cannot provide reliable results in the psychophysical trials.

CNT/PDMS-based canal-typed ear electrodes for inconspicuous EEG recording

OBJECTIVE

Current electroencephalogram (EEG) monitoring systems typically require cumbersome electrodes that must be pasted on a scalp, making a private recording of an EEG in a public place difficult. We have developed a small, user friendly, biocompatible electrode with a good appearance for inconspicuous EEG monitoring.

APPROACH

We fabricated carbon nanotube polydimethylsiloxane (CNT/PDMS)-based canal-type ear electrodes (CEE) for EEG recording. These electrodes have an additional function, triggering sound stimulation like earphones and recording EEG simultaneously for auditory brain-computer interface (BCI). The electrode performance was evaluated by a standard EEG measurement paradigm, including the detection of alpha rhythms and measurements of N100 auditory evoked potential (AEP), steady-state visual evoked potential (SSVEP) and auditory steady-state response (ASSR). Furthermore, the bio- and skin-compatibility of CNT/PDMS were tested.

MAIN RESULTS

All feasibility studies were successfully recorded with the fabricated electrodes, and the biocompatibility of CNT/PDMS was also proved.

SIGNIFICANCE

These electrodes could be used to monitor EEG clinically, in ubiquitous health care and in brain-computer interfaces.

Real-time detection of auditory steady-state responses

Detection of the auditory steady-state responses (ASSRs) is a difficult task, its main technical impediment is no other than the excessively lengthy recording time required for the estimation process due to the extremely low signal-to-noise ratio (SNR). To improve the detection rate of ASSRs, a new observer-based real-time ASSR detector is derived as an alternate solution to the existing methods. The ASSR detector has a simple structure, and provides fast and reliable signal detection. Simulation and experimental recorded data were used to evaluate the performances of the proposed approach. Compared with the conventional methods, the proposed method requires shorter recording time which could be proven as a valuable hearing screening or diagnostic tool.

Optimum principal components for spatial filtering of EEG to detect imaginary movement by coherence

Several techniques have been used to improve the signal-to-noise ratio to increase the detection rate of Event Related Potentials (ERPs). This work investigates the application of spatial filtering based on principal component analysis (PCA) to detect ERP due to left-hand index finger movement imagination. The EEG signals were recorded of central derivations (C4, C2, Cz, C1 and C3), positioned according to 10-10 International System. The optimal spatial filter was found by using the first principal component and the ERP detection was obtained by magnitude squared coherence technique. The best detection rate, by using original signal (without filtering), was obtained at C2 derivation, with 54.73% for significance level of 5%. For the same significance level, the detection rate of the filtered signal was drastically improved to 96.84%. Results suggest that spatial filter by using PCA might be a very useful tool in assisting the ERP detection for movement imagination for applications on brain machine interface.

Optimal electrode selection for multi-channel electroencephalogram based detection of auditory steady-state responses

Auditory steady-state responses (ASSRs) are used for hearing threshold estimation at audiometric frequencies. Hearing impaired newborns, in particular, benefit from this technique as it allows for a more precise diagnosis than traditional techniques, and a hearing aid can be better fitted at an early age. However, measurement duration of current single-channel techniques is still too long for clinical widespread use. This paper evaluates the practical performance of a multi-channel electroencephalogram (EEG) processing strategy based on a detection theory approach. A minimum electrode set is determined for ASSRs with frequencies between 80 and 110 Hz using eight-channel EEG measurements of ten normal-hearing adults. This set provides a near-optimal hearing threshold estimate for all subjects and improves response detection significantly for EEG data with numerous artifacts. Multi-channel processing does not significantly improve response detection for EEG data with few artifacts. In this case, best response detection is obtained when noise-weighted averaging is applied on single-channel data. The same test setup (eight channels, ten normal-hearing subjects) is also used to determine a minimum electrode setup for 10-Hz ASSRs. This configuration allows to record near-optimal signal-to-noise ratios for 80% of subjects.

A flexible research platform for multi-channel auditory steady-state response measurements

The possibilities of currently commercially available auditory steady-state response (ASSR) devices are mostly limited to avoid unintentional misuse and to guarantuee patient safety as such. Some setups, e.g. do not allow the application of high intensities or the use of own stimuli. Moreover, most devices generally only allow data collection using maximal two EEG channels. The freedom to modify and extend the accompagnying software and hardware is very restricted or inexistent. As a result, these devices are not suited for research and several clinically diagnostic purposes. In this paper, a research platform for multi-channel ASSR measurements is presented, referred to as SOMA (setup ORL for multi-channel ASSR). The setup allows multi-channel measurements and the use of own stimuli. It can be easily extended to facilitate new measurement protocols and real-time signal processing. The mobile setup is based on an inexpensive multi-channel RME soundcard and software is written in C++. Both hardware and software of the setup are described. An evaluation study with nine normal-hearing subjects shows no significant performance differences between a reference and the proposed platform. SOMA presents a flexible and modularly extensible mobile high-end multi-channel ASSR test platform.