Biomedical Signal Processing and Control 94 (2024) 106264

Contents lists available at ScienceDirect

Biomedical Signal Processing and Control

journal homepage: www.elsevier.com/locate/bspc

Automated ABR and MMN extraction using a customized headband for

hearing screening
Rathin K. Joshi a, Manu K. S. a, Hari R. S. b, Ajay Krishnan A b, M. Jayachandra c,
Manjunath Dandinarasaiah d, Hardik J. Pandya a, e, *
a
Department of Electronic Systems Engineering, Division of EECS, Indian Institute of Science, Bangalore, India
b
Easescan MedTech Private Limited, Raman Building, CPDMed TBI, Indian Institute of Science, Bangalore, India
c
Center for BioSystems Science and Engineering, Indian Institute of Science, Bangalore, India
d
Department of ENT, Karnataka Institute of Medical Science Hubli, India
e
Centre for Product Design and Manufacturing, Division of Mechanical Sciences, Indian Institute of Science, Bangalore, India

A R T I C L E I N F O A B S T R A C T

Keywords: Objective: Stimuli-elicited EEG responses, known as Event-Related Potentials (ERPs), reflect the health status of
Biopotentials underlying electrophysiological processes and are frequently used for scanning sensory pathways. Current ERP
Electroencephalography extraction systems are expensive, complex, and bulky. This study aims to overcome these limitations by
Brain computer interface
developing and validating a bimodal auditory ERP extractor headband.
Event-related potentials (ERPs)
Auditory brainstem response (ABR)
Methods: An affordable, portable bimodal auditory ERP extractor was developed and validated for ABR (Auditory
Mismatch negativity (MMN) Brainstem Response) and MMN (Mismatch Negativity) response interpretation. Auditory stimuli were created
using Presentation software. Adaptive filtering-based extraction was performed in EEGLAB (MATLAB) to derive
the neural inferences. Extracted responses from n = 5 young adults were validated against CE-certified acqui­
sition systems used in clinical practice.
Results: Validation results showed the grand average responses of ABR and MMN for n = 5 subjects matched.
Furthermore, with and without stimuli ERP analysis showed significant differences for identified features (p =
0.00794 for ABR wave-V amplitudes, p = 1.22 * 10− 5, and p = 2.62 * 10− 5 for MMN peak latencies and area
under the curves), ensuring that the response was due to the presented auditory stimuli.
Conclusion: The results confirm the competency of the developed system to obtain MMN and ABR from young
adults. A clear contrast between latency-amplitude scatter plots further ensures the competency of a developed
system to acquire ABR and MMN.
Significance: The novelty of the study is an easy-to-operate, affordable ERP extractor with adaptive filtering,
wearable design, and configurable stimuli. This system can be a potential solution for large-scale hearing
screening, providing detailed neural insights, including averaged responses and inter-trial variabilities.

1. Introduction brain, heart, eyes, muscles, and nerve cells. The brainwaves, signatures
for neurophysiological processes, are obtained by EEG, ECoG, and sEEG
Electrical discharges, captured at the different anatomical levels of (stereo EEG) [2]. Considering the ease of operation and non-
the human body, convey important information about the underlying invasiveness, EEG is preferred as a preliminary diagnostic measure for
biological processes. These electrical discharges, broadly known as the assessment of neurological conditions.
biopotentials, are used to assess the health status of respective organs EEG-based Brain-Computer Interface (BCI) applications can be
[1]. Clinically used biopotentials are Electroencephalography (EEG), broadly classified into two categories: (i) Free-running multichannel
Electrocorticography (ECoG), Electrocardiography (ECG), Electromy­ EEG applications involving epilepsy screening and classification [3],
ography (EMG), and Electrooculography (EOG). EEG, EcOG, ECG, EMG, sleep staging [4,5], and assessment of several neurological disorders,
and EOG provide accurate real-time information about the status of the including dementia[6] and depression[7]; (ii) stimuli-evoked potential-

* Corresponding author at: Department of Electronic Systems Engineering, Division of EECS, Indian Institute of Science, Bangalore, India.
E-mail address: [email protected] (H.J. Pandya).

https://doi.org/10.1016/j.bspc.2024.106264
Received 8 August 2023; Received in revised form 22 January 2024; Accepted 20 March 2024
Available online 4 April 2024
1746-8094/© 2024 Elsevier Ltd. All rights reserved.
R.K. Joshi et al. Biomedical Signal Processing and Control 94 (2024) 106264

based applications, in which a strategically generated auditory [8,9], brainstem [32]. CAEPs are used for subjective auditory threshold esti­
visual [10] or somatosensory[11] stimulus is presented, and the stim­ mation [33]. Additionally, CAEPs have been used to study the matura­
ulus elicited EEG is acquired and analyzed. tion of auditory pathways in infants, children, and adults [34].
Robust acquisition and detailed interpretation of free-running EEGs Mismatch Negativity (MMN) is an obligatory cortical ERP generated in
lead to a plethora of clinical applications. EEGs have been frequently response to any change in auditory stimulus parameters, including in­
used for seizure detection, classification, and prediction [3,12–14]. tensity, duration, timbre, and spatial location following an oddball
Alkan et al. utilized EEG power spectra as an input for two fundamen­ paradigm. As MMN is an obligatory and the earliest observed ERP
tally different classifiers [12]. The study concluded that the multi-layer (present at the time of birth), it promises to prove a potential pattern
perceptron neural networks showed promising receiver operating signature to screen newborns for deafness [35 36].
characteristic curves compared to the conventional logistic regression- The brainstem and cortical auditory evoked responses collectively
based counterpart [12]. Additionally, Saab et al. concluded that CNNs provide crucial neurophysiological information in clinical neuro-
with weak annotations are more effective in seizure detection models otology [26,30,33]. Therefore, developing a robust, cost-effective ERP
[15]. However, the approach required a large training dataset with la­ extraction system is essential. However, current ABR systems are
bels for optimal classification of ictal, pre-ictal, and interictal EEG expensive and require skilled professionals for experimentation and
segments. response interpretation, making large-scale hearing screening chal­
EEG-based interpretation is an important subjective measure of un­ lenging in developing countries [37]. This study reports designing,
derstanding children with a specific disability. Sheshadri et al. per­ developing, and validating an affordable headband to extract a brain­
formed a case-control study to understand the differences in brain stem (ABR) and cortical (MMN) auditory evoked response. The simpli­
dynamics between dyslexic children and normal children. EEG inter­ fied experimental flow of the work is shown in Fig. 1.
pretation results of visual continuous performance tasks showed poor The present work is the first system development and validation
functional integrity and attention failure in the dyslexic group [16]. In study integrating ABR and MMN in a single system. It discusses the
another research involving children with learning, the research group experimental flow from biopotential acquisition to response extraction
classified children with learning disabilities with 96 % accuracy [17]. and analysis. The obtained response was compared with commercially
Another important usage of free-running EEG is for sleep stage classifi­ available systems.
cation [4,5,18]. Sharma et al. demonstrated an iterative filtering-based The developed, affordable bimodal auditory ERP extraction system
approach for sleep staging using single-channel EEG with promising includes several vital contributions, including (i) robust adaptive
results for wearable sleep scoring device conversion [18]. Moreover, a filtering-based auditory brainstem response (ABR) extraction, (ii)
recent study illustrated a novel spectral graph wavelet representation to customized bi-material wearable, comfortable headband design with
extract features from EEG for emotion recognition [19]. Saideepthi et al. head size adjustability, and (iii) response interpretation strategies to
developed a novel post-processing step to enhance the performance of differentiate between extracted responses with and without auditory
EEGNet. This approach performed equally well not only in both within stimuli presentation. Additionally, the results showed a possible habit­
and cross-subject decoding frameworks but also for both healthy and uation for deviants presented during the MMN experiment following
stroke patients’ datasets [20]. oddball paradigm-based stimuli. Moreover, the system development and
These BCI systems are designed to gather specific information about validation study led to fundamental experimental parameters for audi­
underlying neurophysiological processes, including attention and tory stimulation, biopotential acquisition, neural response extraction,
working memory measures [21], and screening for neurological disor­ and interpretation approaches.
ders, including Schizophrenia [22], Tinnitus [23], and Alzheimer’s [24].
One widespread usage of evoked potential is checking the auditory and 2. Data
visual pathways [8,9,25]. Considering the random electrical fluctua­
tions in the human brain, the stimulus is presented repetitively for a Brainstem (ABR) and Cortical (MMN) responses were recorded from
finite number of trials, and the averaged time-locked response of pre- 5 participants (Mean: 28.6 years, SD = 3 years, range: 23–31 years; two
processed EEG forms the final Event-Related Potential (ERP) [26]. females and three males) with normal hearing. None of the participants
Moreover, pattern interpretation with the extracted averaged ERPs’ had a history of auditory deficits or neurological or psychiatric condi­
time and frequency domain features conveys vital information about tions. The experiment followed institutional ethical approvals (IHEC No:
brain functionalities [27,28]. 03/15.06.2023). The experimental protocol was explained to each
Auditory Brainstem Response (ABR) is the gold standard for hearing participant, and written consent was obtained before the experiment.
screening [29]. Mathematically, ABR is a pattern with five to seven
waves, typically induced within 10 ms of auditory stimuli presentation 3. Methods
[9]. Clinically, ABR is an essential measure in neuro-otology to assess
the integrity of the auditory pathway from the cochlea to the auditory This section discusses the fundamental building blocks for this
thalamus [30]. Additionally, ABR is used intraoperatively to check bimodal (ABR and MMN) auditory ERP extraction system. The detailed
imminent damage to the cochlea, VIIIth cranial nerve, and brainstem, experimental protocol is also described, including stimuli parameters,
which can minimize damage to these vital structures [30]. Furthermore, cortical and brainstem extraction and interpretation parameters, system
ABR can be recorded from individuals who have been sedated, which design aspects, statistical analysis, and necessary information about the
allows for the screening of uncooperative individuals with hearing loss. participants.
[31].
Cortical Auditory Evoked Potentials (CAEPs) are the scalp-recorded
neuronal discharges elicited due to auditory stimuli during 50 ms to 3.1. Auditory stimuli generation and trigger transmission
400 ms post-stimulus. A typical CAEP response shows three peaks: P1-
N1-P2 [30]. Clinically, CAEPs are a comprehensive measure for a Two types of auditory stimuli were used to elicit brainstem and
hearing screening as they check the auditory pathways beyond the cortical responses. Initially, one unit trial of the auditory click stimulus
was generated using Audacity® software. Subsequently, these generated

2
R.K. Joshi et al. Biomedical Signal Processing and Control 94 (2024) 106264

Fig. 1. Simplified Experimental Flow of bimodal auditory ERP extraction system: Block Diagram of the methods for Auditory Brainstem Response (ABR) and
Mismatch Negativity (MMN) extraction. The obtained ABR and MMN responses from the developed system are validated using the reference system.

audio files were instantiated in the customized script developed in In contrast, the cortical auditory evoked potential (CAEP – MMN) is
Presentation software (Neurobehavioral Systems Inc.). generated within 400 ms post-stimulus presentation, with a typical
amplitude range from 1 µV to 10 µV. Hence, there is a need for an
3.1.1. ABR stimuli generation acquisition system with a higher sampling rate and the capability of
Nine hundred clicks were presented. Stimuli parameters for the capturing sub-microvolt signals, especially for ABR, making it a chal­
clicks were 0.1 ms duration and 33 ms inter-trial interval (ITI), with ~ lenging aspect of the ERP extraction system development. The circuit
70 dB intensity measured using Brüel & Kjær Type 2240 Integrating- diagrams and exploded view of the customized acquisition subsystem
averaging Sound Level Meter. for auditory evoked EEG (ERP) are shown in Fig. 2. Moreover, the cas­
ings were developed for acquisition systems, as shown in Fig. 2(b). The
3.1.2. MMN stimuli generation casing dimensions for the ABR acquisition system were 170 mm* 152
Mismatch Negativity (MMN) is a proven biomarker of neuronal mm* 72 mm, and for the MMN acquisition system were 71 mm* 94 mm*
electrical changes due to auditory stimuli variations. This acoustic 59 mm.
change can be in any stimuli characteristics, including frequency, Considering the difference in amplitude and latencies of ABR and
duration, timbre, intensity, spatial location, or type [30,40,41]. In the MMN, two different acquisition systems were developed to obtain
experiments, two different clicks were presented periodically for 100 brainstem response and cortical response, explained below.
trials. The deviant sound (100 trials; presentation duration: 1 ms) was
presented after four successive standard sounds (400 trials; presentation 3.2.1. ABR acquisition
duration: 5 ms), with an ITI of 600 ms. Recent progress in microelectronics research led to a higher sampling
rate of biopotential acquisition at a higher sensitivity. ADS 1299 is one
3.1.3. Transmission of generated triggers such low noise high-precision 24-bit Analog to Digital Converter (ADC);
Stimuli transmission is equally important as stimuli generation. it has eight channels for simultaneous acquisition using sigma-delta
Precise timing information of presented triggers helps in accurate ERP ADCs and a sampling rate of up to 16 kHz. ADC has a sub-microvolt
extraction. Hence, the detailed time log of these generated stimuli was (0.5 µV) resolution, making it appropriate for EEG/ERP acquisition.
transmitted from the configured USB port to the digital input of the Considering the low amplitude of the ABR, preamplifiers are used before
acquisition subsystem using serial to parallel communication. Moreover, the ADC [42,43]. The developed ABR acquisition system uses a dual-
a software log was also recorded using Presentation software to obtain stage voltage preamplifier with an effective voltage gain of 1000. The
ERPs. first stage preamplifier was designed with a low noise, high accuracy
instrumentation amplifier-AD620, with a voltage gain of 10. The first
stage was connected to the electrode at Fpz and an earlobe reference
3.2. Biopotential and stimuli acquisition system
point. A high-pass filter with a cut-off frequency of 30 Hz was used to
couple the first stage to the second and to eliminate non-neural lower
Higher sampling rate and sub-microvolt sensitivity are critical pa­
frequency signals. The second stage preamplifier was designed using an
rameters for the auditory-evoked response biopotential acquisition
INA114 precision instrumentation amplifier with a gain of 100. The
system design and electronic component selection. Brainstem response
amplified signals were acquired using the ADS1299 ADC at a sampling
(ABR) is typically generated within 10 ms after stimulus presentation.
rate of 16 kHz, controlled by a TMS320VC5507 DSP controller from
The amplitude range of the stimulated brainstem response is 0.1 to 1 µV.

3
R.K. Joshi et al. Biomedical Signal Processing and Control 94 (2024) 106264

Fig. 2. Biopotential acquisition module for auditory evoked response extraction: (a) Auditory Brainstem Response (ABR) acquisition circuit diagram, and Cortical
Auditory Evoked Potential (MMN) acquisition circuit diagram, and (b) Exploded view of a printed acquisition system for cortical and brainstem response extraction.

Texas Instruments. This acquired biopotential was transferred to the BCI Cyton Board with a sampling rate of 250 Hz. Open BCI Cyton is a 6 V
laptop via USB. The ADC and the controller were part of the battery-powered system with an analog front-end ADS1299, a low-noise,
ADS1299EEGFE performance demonstration kit (PDK) from Texas In­ 24-bit, 8-channel ADC commonly used for EEG and biopotential mea­
struments. The data from the DSP were collected using the customized surements [44]. The biopotentials from the scalp were sent to the ADC
user interface associated with the PDK. Generated Triggers were stored inputs (Channels 1 and 2) via an RC filter (R = 2.2 K, C = 1000pF) and an
in a log file from the Presentation software and ultimately used to extract ESD protection circuit based on transient voltage suppressor (TVS) di­
an evoked ABR. odes. The ADC was connected to a 32-bit PIC32MX microcontroller
through an SPI communication interface. The controller captured both
3.2.2. MMN acquisition the ADC output and the temporal information of the presented auditory
The cortical auditory evoked potentials were acquired using an Open stimuli and wirelessly transmitted the data using the RFD22301, a

4
R.K. Joshi et al. Biomedical Signal Processing and Control 94 (2024) 106264

Bluetooth Low Energy (BLE) 4.0 module. The PC receives the data brainstem response extraction, a flat snap dry electrode was replaced by
through the dongle containing the BLE 4.0 module and the FT232RL a wet electrode, resulting in less skin-electrode impedance and, hence,
(USB to serial UART converter). Moreover, the FT232RL module in bit more sensitive neuro-potential acquisition. The headband was made of
banging mode transmitted timing information of stimuli generated by two different materials: (i) Formlabs standard grey resin (shown in grey
the Presentation software to the microcontroller digital input pins of the in Fig. 3(a-c)) and (ii) Formlabs flexible 80A resin (shown in white in
Cyton board. Fig. 3 (a-c)). Stereolithography (SLA) 3D printing was used to print the
headband. An optimal proportion between these two materials was
vital, resulting in a comfortable and robust evoked EEG acquisition for
3.3. Headband design various head sizes. Finally, to acquire the evoked EEG, one soft velcro
strap was used to fasten the headband for optimum skin-electrode
A headband design is a vital aspect of all successful ERP studies. For interface. Thus, a subject could be prepared very quickly for bio­
the developed headband, a well-connected electrode skin interface, potential acquisition.
precise electrode placement, and midline coverage were primary design The designed headband has an important provision for swapping the
factors from a neural engineering perspective. Additionally, variabilities electrode arm, as shown in Fig. 3(b). A metal wire was placed inside the
in human head sizes, comfort, material selection, and optimization in swappable electrode arm to provide optimum stiffness and flexibility,
material proportion were the secondary factors while designing and ensuring proper skin-electrode contact. The electrode position could be
before 3D-printing a headband. modified by unmounting a center screw and connecting the electrode
The developed design, electrode swapping provision illustration, and arm in the opposite direction. Hence, depending on the type of ERP
engineering drawings are shown in Fig. 3. The exploded view of the experiments, the evoked response acquisition scalp positions can be
developed headband is shown in Fig. 3a. It shows three active elec­ selected whether to record from Fz and Cz or Cz and Pz. Usually, visual
trodes, including two spike electrodes (Fz and Cz) and one flat snap ERPs are acquired from parieto-occipital lobes, whereas auditory ERPs
electrode (Fpz). Two ear clip electrodes are also shown on either side of are obtained from fronto-central lobes [30,45]. The engineering
the head, which worked as a reference and ground. In the case of

Fig. 3. Auditory ERP Extraction Headband Design: (a) Exploded View of the developed headband, (b) Electrode Position Swapping Provision: (b1) Fronto-central
Configuration with Fz and Cz Electrodes, (b2) Parieto-central Configuration with Cz and Pz Electrodes, (c) Illustrative design of the developed headband, and the
headband on a mannequin, (d) Actual image of the headband and subject wearing it, and (e) Engineering drawings of the developed headband including top view,
side view, front view, and isometric view.

5
R.K. Joshi et al. Biomedical Signal Processing and Control 94 (2024) 106264

Fig. 4. Signal processing steps for ABR and MMN extraction with exemplary waveforms and spectrums:(a) ABR extraction pre-processing with a bandpass filter and
adaptive band-reject filters for non-neural line noise peaks, spectrum changes are shown after each stage of signal transformation, (b) Common ERP extraction
routine for ABR and MMN with respective parameters, time domain waveforms are shown after each step, and (c) MMN extraction filtering with a bandpass filter and
line noise removal, raw and filtered spectrum is shown in corresponding red boxes.

drawings with dimensions and isometric view are shown in Fig. 3(d). Related Potential (ERP). The simplified signal processing flow of the
ABR and MMN extraction, with a series of sequenced transformations, is
summarized in Fig. 4. The ERP (MMN and ABR) extraction was per­
3.4. ABR and MMN extraction and analysis formed using EEGLAB v2022.0 [46] and ERPLAB 8.30 [47] in MATLAB
2022b v9.9 (Institutional Academic License). Two different strategies
Post-evoked response acquisition, an ordered sequence of mathe­ were automated in the form of two customized MATLAB scripts to obtain
matical operations was performed to extract the auditory evoked Event-

6
R.K. Joshi et al. Biomedical Signal Processing and Control 94 (2024) 106264

cortical and brainstem responses. Table 1

Developed Auditory ERP Extraction System Parameters.
3.4.1. ABR extraction Parameter Brainstem Cortical Extraction
The initial step was to import the acquired single channel (Fpz) Extraction
biopotential to EEGLAB and import a log file for event generation. A Targeted Pattern ABR MMN
finite impulse response filter was applied with 100 Hz and 1000 Hz cut- Signature
off frequencies. All the filters are standard Hamming windowed sinc FIR Auditory Stimuli Parameters
filter of Firfilt plugin (Version 2.5.1) of EEGLAB. The filter order and Event Types 1 (Standard Clicks) 2 (Standard and Deviant)
Intensity of Auditory 70–75 dB 70 dB
transition bandwidth are automatically selected based on certain pa­ Stimuli
rameters, including cut-off and critical frequencies. Subsequently, the Inter-trial Interval 33 ms 600 ms
amplitude spectrum check was performed to capture non-neural peaks (ITI)
using ratios of consecutive peaks and peak prominence. Thresholds for Duration 0.01 ms 1 ms (for Standard)
5 ms (for Deviant)
the ratio of successive peaks and peak prominence were identified. Most
Number of 900 400 Standards and
of these identified frequencies were harmonics of power line fre­ Repetitions 100 Deviants
quencies. The frequency values of such non-neural frequency compo­ Trigger Presentation Log Serial to Parallel Communication
nents were stored, and band-reject filters of the same frequencies were Communication Files using FTDI
applied before rechecking the amplitude spectrum. This subroutine of Biopotential Acquisition Parameters
Sampling Rate 16000 Hz 250 Hz
amplitude spectrum-based non-neural frequency removal was termi­ Number of Electrode 1 2
nated when no peaks satisfied the peak ratio and the peak prominence (s)
criteria. This adaptive filtering (Fig. 4(a)) helped retain neural frequency Type of Electrode(s) Wet Dry
components. The remaining ABR response extraction steps involved five Electrode Position Fpz Fz, Cz
Test Duration 30 s 5m
conventional ERP extraction steps: (i) event list creation, (ii) epoch
Event Related Potential (ABR and MMN) Extraction and Interpretation Parameters
generation with baseline correction, (iii) artifact rejection, (iv) ERP Type of Filtering Adaptive Definite
generation using averaging, and (v) plotting (Fig. 4(b)). Allowable Frequency 100–1000 Hz 3–30 Hz
Range
3.4.2. MMN extraction Epoch Time − 2 ms to 12 ms − 50 ms to 400 ms
Specifications
Initially, biopotentials and timing information were imported to
Artifact Rejection ± 10 µV ± 50 µV
MATLAB. A hamming windowed sinc (Firfilt Plugin, EEGLAB) finite Criteria
impulse response band-pass filter with 3 Hz and 30 Hz cut-off fre­ Features Considered Wave V Latency and MMN Peak Latency, Amplitude,
quencies was applied to the acquired biopotentials. The amplitude Amplitude and Area Under the Curve
spectrum was monitored to avoid any non-neural frequency compo­
nents. Additionally, with the event timing information, epochs were
These strategies were realized using a customized script to draw the
generated and checked for artifacts for the ± 50 µV amplitude range.
following features: (i) Wave V latency and Wave V amplitude were
Accepted epochs were averaged, and the MMN trace was plotted for Fz
considered for ABR, and (ii) MMN peak Amplitude, MMN peak latency,
and Cz following the signal processing flow shown in Fig. 4(c).
and area under the curve (to quantify mismatch negativity) were
considered for MMN.

3.5. Evoked response (ABR and MMN) interpretation

3.6. Experimental parameters and protocol
Accurate interpretation of the evoked response provides essential
neural insights about the sensory pathway. Auditory Evoked ERPs are The central focus of this research was to develop and validate a
assessed based on peak latencies, peak amplitudes, and area under the neurotechnological system to extract auditory evoked responses. ABR
curve. Fig. 5 shows the feature extraction strategies for ABR and MMN. and MMN experiments were performed on each subject using both

Fig. 5. Response interpretation strategies for extracted ERPs using developed system: (a) Obtained ABR waveform from Fpz channel of one subject with wave V
latency and amplitudes as features, and (b) Obtained MMN waveforms from Fz and Cz channels of one subject with wave V latency and amplitudes as features.

7
R.K. Joshi et al. Biomedical Signal Processing and Control 94 (2024) 106264

systems to validate the response from the developed system. Both ex­ 4.1. ABR extraction results
periments were performed on the same day with an hour gap between
systems for a particular subject. The total weight of the developed Auditory Brainstem response was obtained from five subjects using
headband was 105 gm, providing ease of acquisition. Before the an ADS1299PDK-based developed system. The extracted results were
experiment, free-running EEGs were monitored for known physiological verified with commercially available CE-certified intelligent hearing
maneuvers, including eye blinks, jaw clenching, and eye movements. screening Brainstem Evoked Response Audiometry (BERA) test. Fig. 6
These checks assured an adequately coupled skin-electrode interface. (a1) shows the scalp location from where ABR is acquired. Fig. 6(a2, a3)
Subsequently, the experiment was performed in a silent room, keeping shows the grand average of five subjects using both systems. Fig. 6(a2)
only the necessary electronic apparatus on and the connecting interfaces shows the grand average obtained from the developed system, in which
shielded. Binaural balanced auditory stimuli were presented using the amplitude of the evoked response was significantly less due to the
Sennheiser wired CX80 S earphones, providing ambient noise isolation. exhaustive filtering performed to eliminate non-neural line noise. Even
Since ABR is a far-field response, the electrode was placed on the though the developed system response is low, all seven peaks were
forehead midline (Fpz), whereas for the cortical response, electrodes visible, indicated in blue Roman numerals, in Fig. 6(a2, a4, and b).
were placed at the midline of the frontal and central lobe (Fz, Cz) Additionally, wave V was prominent, as observed in the literature.[30]
[9,40,41]. The essential experimental modular parameters are summa­ Moreover, peak latencies from the developed system matched the
rized in Table 1. reference systems, shown in supplementary figure S1. Additionally,
Finally, it is important to note that both (ABR and MMN) experi­ subject-wise peak amplitudes acquired from the developed system are
ments were verified using a reference system. Hence, generated stimuli lower than corresponding values from the reference system due to
and all experimental parameters (shown in Table 2) remained consistent adaptive filtering, resulting in more perceptible peaks. Peaks are shown
for experiments with the developed and reference systems. Additionally, in Roman numerals in supplementary Figure S1.
both tests (ABR and MMN) were repeated without presenting the stimuli Additionally, we experimented by muting the stimuli presentation
to ensure that the response was evoked and not resulted due to any device, keeping electrical trigger transmission on. Expected peaks and
random non-neural variations. peak latencies were absent in the absence of auditory stimuli, thus
confirming the claim one more time in the case of brainstem response
3.7. Statistical analysis that the response obtained from the developed system is indeed due to
the auditory stimuli presented. Fig. 6(a4 and a5) shows obtained ABR in
The significant differences in the extracted patterns were evaluated both scenarios. Fig. 6(a4) shows the actual response elicited due to
for the identified ABR and MMN features. These compared extracted stimuli presentation, whereas Fig. 6(a5) shows the different responses
patterns were: (i) with and without stimuli ABR pattern parameters with without the auditory stimuli. Similar experiments were conducted for all
wave V peak amplitude and wave V peak latency as a parameter, (ii) five subjects, and results are shown in supplementary Figure S2.
with and without stimuli MMN pattern parameters with MMN peak Grand average or subjective response extraction is an offline measure
amplitude, MMN area under the difference curve and MMN peak latency to assess brainwaves responding to presented stimuli. Additionally, it is
as a parameter. A non-parametric Shapiro-Wilk test [48] was used to essential to notice the actual electrical discharges presented during the
check the grouped data’s normality before a single-factor ANOVA test. trials. We obtained an ERP image (averaged ERP plot and inter-trial
However, if the normality is rejected, The Mann-Whitney test [49] was responses in heatmap) for one subject, shown in Fig. 6(b).
performed considering a 5 % confidence interval for p-value calculation.

4. Results and analysis 4.2. MMN extraction results

A grand average response and comparison of extracted Event- The MMN experiment was performed with an Open BCI Cyton-based
Related Potentials were obtained for n = 5 young adults. Additionally, developed system and an FDA-approved CE-certified Enobio8 system for
verification was done to ensure that the generated response was due to n = 5 young adults. The same stimuli were presented while conducting
the presented stimuli, and ERP image plots were obtained for brainstem experiments with the developed and the reference system, and the same
and cortical responses. extraction steps were followed to obtain MMN in each subject. The
grand average of n = 5 young adults is shown in Fig. 7(a). The expected
timing range for MMN occurrence was highlighted in brown. The peak
latencies for deviant events in both channels are similar for both chan­
Table 2 nels in both systems. Extracted MMN traces for all five subjects (n = 5)
Statistical analysis between with and without stimuli response: Feature-wise from the developed and reference systems are shown in supplementary
Results. Figure S3.
ERP Feature No. of Significance Test Additionally, the MMN response extracted in the absence and pres­
Observations Results ence of auditory stimuli and the averaged extracted response are shown
ABR Wave V n1 = 5, n2 = 5 Significantly Mann- in Fig. 7(b). The experiment with auditory stimuli resulted in the ex­
Amplitude different Whitney U pected cortical auditory evoked response, including P1, N1, and P2, with
p = 0.00794, U = Test the negativity associated with deviant events (MMN). On the contrary, a
25 trace without any clear N1 P2 was observed in the case of absent
Wave V n1 = 5, n2 = 5 Significantly not Single Factor
Latency different ANOVA
auditory stimuli; there was no delineation between the two events,
MMN MMN Peak n1 = 10, n2 = Significantly Single Factor further ensuring the system’s competency of acquiring cortical auditory
Amplitude 10 different ANOVA evoked response. With and without stimuli, ERP results for all five
p = 1.22 * 10− 5, F subjects are shown in supplementary figure S4.
= 35.51
ERP image plots were generated for enhanced insight into the ERP
MMN AUC n1 = 10, n2 = Significantly Single Factor
10 different ANOVA response during the experiment. ERP image plots were shown for both
p = 2.62 * 10− 5, F events, including standards and deviants for Fz and Cz channels. A
= 31.29 decline in MMN peak latencies was observed in both channels, sug­
MMN Peak n1 = 10, n2 = Significantly not Single Factor gesting a possible habituation during the experiment, indicated by a
Latency 10 different ANOVA
white arrow in Fig. 7(c).

8
R.K. Joshi et al. Biomedical Signal Processing and Control 94 (2024) 106264

Fig. 6. Auditory Brainstem Response (ABR) Extraction Results: (a) Grand average comparison between the developed and the reference system (n = 5) and
confirmation on auditory evoked response: (a1) Scalp position illustration for electrode placement, (a2) Grand averaged of n = 5 subjects, extracted using the
developed system, (a2) Grand averaged of n = 5 subjects, extracted using the reference system, (a3) Extracted ABR in one subject with auditory stimuli presentation,
(a4) ABR without any auditory stimuli presentation, and (b) ERP plot and averaged extracted ABR trace for one subject, Dashed black box depicts the occurrence of
prominent wave I and prominent wave V, a colour bar on the right shows the amplitude of evoked ABR, within ± 0.16 µV range.

4.3. Extracted pattern interpretation results curve are significantly different, whereas latencies are similar. However,
Fig. 8 shows that peak latencies were dispersed compared to data points
Scattered plots were obtained to quantify the difference between when stimuli were presented. Furthermore, feature-wise box plots are
peak latencies and peak amplitudes for the response obtained in the shown in supplementary figure S5.
absence and presence of auditory stimuli. Fig. 8 displays the ABR and
MMN scattered plots extracted from all five subjects. The green color, 5. Discussion
including green dots and green dashed region, represents an instance
when stimuli were presented. In contrast, the red color, including red The section involves a comparative analysis of existing ERP extrac­
dots and red dashed region, suggests the experiment when stimuli were tion studies with the presented research. Additionally, the role of arti­
muted. ficial intelligence and machine learning is briefly discussed, along with
Three inferences are evident from the ABR and MMN scattered plots: cost analysis.
(i) Latencies are well localized when the stimuli are presented, whereas
in the absence of stimuli, all values are temporally dispersed, (ii) Peak
5.1. Comparative analysis with other ERP studies
amplitudes are significantly different under both situations for ABR and
MMN, and (iii) In the case of MMN, the area under the curve values
ERP extraction requires the customization of all fundamental
shows striking differences for absence and presence of auditory stimuli.
building blocks, including stimuli generation, biopotential acquisition,
Results of significant tests (p-values) are enlisted in Table 2.
experimental setup, headband design, and signal processing parameters
It is important to note that the peak amplitudes and area under the
for extraction and response interpretation. Several studies focused on

9
R.K. Joshi et al. Biomedical Signal Processing and Control 94 (2024) 106264

Fig. 7. Cortical Auditory Evoked Potential (MMN) Extraction Results: (a) Grand Average of extracted cortical auditory evoked potentials: (a1) Scalp Position for
electrode placement, (a2) Grand averaged of extracted MMN traces using the developed system (n = 5 subjects), with Fz on top and Cz below, and (a3) Grand
averaged of extracted MMN traces using the reference system (n = 5 subjects), with Fz on top and Cz below; Legends denotes standard and deviant events, (b)
Confirmation on Auditory change evoked response (b1) Scalp Position for electrode placement, (b2) Extracted Averaged MMN traces with auditory stimuli presented,
with Fz on top and Cz below, and (b3) Extracted Averaged MMN traces without auditory stimuli presented, with Fz on top and Cz below, and (c) ERP image plots and
averaged traces of Fz and Cz channel for Standard and Deviant Events: Color bar shows the amplitude of the extracted biopotential. (Range: ± 9 µV).

10
R.K. Joshi et al. Biomedical Signal Processing and Control 94 (2024) 106264

Fig. 8. Extracted ERP interpretation using scattered plots with green dot represents obtained features when the stimuli were presented and red dot represents
obtained features when the stimuli were not presented: (a) ABR ERP scatter plots with wave V peak Latency and Wave V peak amplitude as features, and (b) MMN
features scattered plots with MMN peak latencies, peak amplitudes and area under the curve as features.

improvising one aspect of the interrelated experimental flow, utilizing considering manufacturing overheads, we envisage converting the
the available resources for the remaining subsystems. Elberling et al. prototype into a product within 4000 USD, one-fifth of the actual cost of
have devised a novel chirp stimulus and concluded that a prominent clinically used systems, ultimately to take a step towards affordable
ABR response is acquired compared to conventional click stimuli [38]. healthcare.
Casson et al. summarize the recent growth in wearable EEG and beyond
the wearable system design to enable EEG monitoring to move out of the 6. Conclusion
lab setup [39]. Additionally, some recent studies discussed an auto­
mated, accurate recognition of ABR waveform to help neurologists make A cost-effective, portable, wearable bimodal ERP extractor headband
quicker diagnoses [27,28]. Table 3 shows a comparative analysis of ERP- was developed to obtain ABR and MMN. The ABR and MMN extraction
extraction-based studies with module-wise significant contributions to experiments used the developed headband with auditory stimuli gen­
ERP experimentation. eration and timing information transmission, biopotential signal
extraction and analysis, and additional electronic modules. The grand
5.2. Role of AI for EEG/ERP-based analysis averages of MMN and ABR were compared with respective CE-certified
systems currently used in clinical practice. The resultant grand average
This research presents a direct approach for ABR and MMN extrac­ pattern for ABR and MMN matched the reference system for n = 5
tion utilizing definite (MMN) or adaptive filtering (ABR) followed by subjects. ERP responses with and without auditory stimuli further
conventional ERP signal processing sequences. However, multiple confirmed that evoked response was generated due to auditory stimuli.
research works have attempted to leverage the benefits of advancement Additionally, striking contrast was observed while extracting features
in machine learning and artificial intelligence-based approaches for from the responses in the absence and presence of auditory stimuli (p =
extracting and interpreting neural responses [54–58]. As described, the 0.00794 for ABR wave-V amplitudes, p = 1.22 * 10–5, and p = 2.62 *
obtained averaged response can be interpreted using known spatio­ 10–5 for MMN peak latencies and area under the curves). Moreover, the
temporal parameters, including peak latencies, peak amplitudes, and subjective ERP imaging showed inter-trial responses during the experi­
area under the curves. These obtained features can be correlated with ment, suggesting a possible habituation for MMN and wave V domi­
underlying neural dynamics. However, machine learning models require nance for ABR.
a sizeable, labelled dataset to include all human EEG variabilities. The research explains a bimodal, affordable auditory ERP extraction
Adequate training and testing data, identical acquisition systems, and system development and validation for ABR and MMN, including cus­
consistent experimental protocol are some major challenges before tomization of all required subsystems. However, there is a scope for
employing AI-based automated neural information extraction studies in further enhancement: (i) The developed system relies on licensed soft­
clinical settings [54,59]. ware (MATLAB) for ABR and MMN extraction; a standalone extraction
software can improve the coverage of the system significantly with a
5.3. Cost analysis provision of online data processing, encryption, and sharing, and (ii)
Responses are extracted after experimentation, a real-time onboard
The developed system has several novel facets, including an adaptive signal processing can demonstrate the response in fewer trials and lead
filtering approach for ABR (sub-microvolt biopotential) acquisition. to test time reduction; however, it requires a higher computational
Additionally, the work discusses the design aspects of the novel power.
biomaterial comfortable headband. We further aim to convert the With minor modifications in the design and acquisition strategy, the
automated scripts to a standalone application for further ease of oper­ same system can be used for subjective screening/assessment of various
ation. The current cost of the brainstem and cortical extraction system is neurological conditions, including sensory deficits, detection, classifi­
19,000 USD. At the same time, the current bill of materials (BOM) cost of cation, and prediction of epilepsy seizures, attention, working memory,
the developed system was less than 3000 USD. Using customized stimuli sleep, and auditory threshold estimation. We envisage realizing trans­
and acquisition units, robust signal processing approaches, and an lational opportunities of this prototype by repeating the experiment for a
optimized experimental setup with minimum electrodes, and large cohort of subjects to screen young adults for hearing loss and to

11
 R.K. Joshi et al. Biomedical Signal Processing and Control 94 (2024) 106264

understand inter-subject EEG/ERP variabilities.

sounds, including speech and music,

method is introduced for robust ABR

Funding

for ABR and AMLR was developed.

Developed an adaptive filtering-based MATLAB Compared results against the FDA Additional checks were performed
R Programming language with several packages Validated with manual impressions An automated Annotation method
image processing method for ABR
Validated with manual impressions The study reported an automated

Neuronal processing of relevant

This work was supported by Science and Engineering Research Board

Validated with manual impressions Automated Recognition of ABR

The study showed that Chirp is
better than clicks for Auditory

This study checks Age-related

Adaptive Kalman Filter based

script to remove all non-neural noises for robust approved CE certified systems for n for with and without stimuli
(CRG/2020/004427); Department of Science and Technology (TDP/
BDTD/40/2021/General); and Abdul Kalam Technology Innovation
Additional Remarks

waves was achieved.

National Fellowship (INAE/121/AKF/49).
peak identification.

evoked potentials.

presentation tests.
changes in ABR

was discussed.
CRediT authorship contribution statement

extraction.
Rathin K. Joshi: Conceptualization, Methodology, Software, Vali­
dation, Formal analysis, Investigation, Data curation, Visualization,
Writing – original draft. K.S. Manu: Conceptualization, Investigation,
Writing – review & editing. R.S. Hari: Investigation, Methodology,
system/clinical impression?

Writing – review & editing. A. Ajay Krishnan: System Design, Visual­

Validation with reference

ization. M. Jayachandra: Conceptualization, Methodology, Writing –

review & editing. Manjunath Dandinarasaiah: Conceptualization,
of three clinicians

= 5 young adults
Resources, Funding acquisition. Hardik J. Pandya: Conceptualization,
Not Reported

Methodology, Visualization, Writing – review & editing, Supervision,

Not reported

Not reported

Not reported
of clinicians

of clinicians

Project Administration, Resources, Funding acquisition.

Declaration of competing interest

Biopotential Extraction and Interpretation

The authors declare the following financial interests/personal re­

lationships which may be considered as potential competing interests: The
ABR Wave identification using Image

Intelligent Hearing System SmartEP

authors declare the following financial interests/personal relationships

which may be considered as potential competing interests: Hardik J.
Pandya reports financial support was provided by Science and Engineering
ABR and MMN acquisition

Research Board. Manjunath Dandinarasaiah reports financial support was

Maico MB11 Software

provided by Department of Science and Technology. Hardik J. Pandya has

patent pending to Indian Design Office. Rathin K. Joshi has patent pending
Not reported

Not reported

to Indian Design Office. Hari R S has patent pending to Indian Design Of­
Processing

fice. Ajay Krishnan A. has patent pending to Indian Design Office.

MATLAB

Data availability

Data will be made available on request.

Wearable System Design

headband for comfortable

biopotential acquisition
Designed a biomaterial
Maico MB11 Headset

Neuroscan EEG data acquisition Neuroscan Headcap

Appendix A. Supplementary data

Not reported

Not reported

Not reported

Not reported

Not reported

Supplementary data to this article can be found online at https://doi.

org/10.1016/j.bspc.2024.106264.

References
Granson-Stadler (GSI) Audera

acquisition systems for MMN

Parameters reported, but the

[1] D. Prutchi, M. Norris, Design and Development of Medical Electronic

Maico MB11 with separate

Intelligent Hearing System

Intelligent Hearing System

Biopotential Acquisition

system was used for ABR

Comparison of Auditory Event-Related Potential Extraction Studies.

EAR3A Earphones and a

Instrumentation: A Practical Perspective of the Design, Construction, and Test of

Customized open-access
system is not specified

Medical Devices, John Wiley & Sons, 2005.

and ABR acquisition

[2] J. Parvizi, S. Kastner, Promises and limitations of human intracranial

electroencephalography, Nat. Neurosci. 21 (2018) 474–483, https://doi.org/
10.1038/s41593-018-0108-2.
Not reported
preamplifier
acquisition

[3] R.K. Joshi, V. Kumar, M. Agrawal, A. Rao, L. Mohan, M. Jayachandra, H.J. Pandya,
SmartEP

SmartEP

Spatiotemporal analysis of interictal EEG for automated seizure detection and

system

classification, Biomed. Signal Process. Control 79 (2023) 104086, https://doi.org/

10.1016/j.bspc.2022.104086.
[4] P. Chriskos, C.A. Frantzidis, C.M. Nday, P.T. Gkivogkli, P.D. Bamidis, C. Kourtidou-
Generated Auditory Stimuli for
Several complex sounds were

Parameters reported, but the

Papadeli, A review on current trends in automatic sleep staging through bio-signal

Introduced CE Chirp Stimuli
CE Chirp Stimuli generated

Stimuli Generation Module

Stimuli Generation Module

Intelligent Hearing System

Intelligent Hearing System

recordings and future challenges, Sleep Med. Rev. 55 (2021) 101377, https://doi.
using the Granson-Stadler

Stim System, Neuroscan

org/10.1016/j.smrv.2020.101377.
system is not specified
(GSI) Audera system

[5] E. Eldele, Z. Chen, C. Liu, M. Wu, C.-K. Kwoh, X. Li, C. Guan, An attention-based
Auditory Stimuli

deep Learning approach for sleep stage classification with single-channel EEG,
MMN and ABR

IEEE Trans. Neural Syst. Rehabil. Eng. 29 (2021) 809–818, https://doi.org/

10.1109/TNSRE.2021.3076234.
generated

[6] G. Henderson, E. Ifeachor, N. Hudson, C. Goh, N. Outram, S. Wimalaratna, C. Del

System

Percio, F. Vecchio, Development and assessment of methods for detecting dementia

using the human electroencephalogram, IEEE Trans. Biomed. Eng. 53 (2006)
1557–1568, https://doi.org/10.1109/TBME.2006.878067.
[7] F.S. de Aguiar Neto, J.L.G. Rosa, Depression biomarkers using non-invasive EEG: a
review, Neurosci. Biobehavioral Rev. 105 (2019) 83–93, https://doi.org/10.1016/
This work

j.neubiorev.2019.07.021.
Table 3

[8] T.W. Picton, S.A. Hillyard, H.I. Krausz, R. Galambos, Human auditory evoked
[50]

[38]

[51]

[52]

[53]

[27]

[28]
Ref

potentials. I: evaluation of components, Electroencephalogr. Clin. Neurophysiol. 36

(1974) 179–190, https://doi.org/10.1016/0013-4694(74)90155-2.

12
R.K. Joshi et al. Biomedical Signal Processing and Control 94 (2024) 106264

[9] D.L. Jewett, J.S. Williston, Auditory-evoked far fields averaged from the scalp of Front. Hum. Neurosci. 3 (2009), https://www.frontiersin.org/articles/10.3389
humans, Brain 94 (1971) 681–696, https://doi.org/10.1093/brain/94.4.681. /neuro.09.048.2009.
[10] D.A. Jeffreys, J.G. Axford, Source locations of pattern-specific components of [35] M. Cheour-Luhtanen, K. Alho, T. Kujala, K. Sainio, K. Reinikainen, M. Renlund,
human visual evoked potentials. I. Component of striate cortical origin, Exp Brain O. Aaltonen, O. Eerola, R. Näätänen, Mismatch negativity indicates vowel
Res. 16 (1972) 1–21, https://doi.org/10.1007/BF00233371. discrimination in newborns, Hear Res 82 (1995) 53–58, https://doi.org/10.1016/
[11] I.M. Muzyka, B. Estephan, Chapter 35 - Somatosensory evoked potentials, in: K.H. 0378-5955(94)00164-l.
Levin, P. Chauvel (Eds.), Handbook of Clinical Neurology, Elsevier, 2019: pp. [36] V. Fellman, M. Huotilainen, Cortical auditory event-related potentials in newborn
523–540. Doi: 10.1016/B978-0-444-64032-1.00035-7. infants, Semin. Fetal Neonatal Med. 11 (2006) 452–458, https://doi.org/10.1016/
[12] A. Alkan, E. Koklukaya, A. Subasi, Automatic seizure detection in EEG using j.siny.2006.07.004.
logistic regression and artificial neural network, J. Neurosci. Methods 148 (2005) [37] B.L. Therrell, C.D. Padilla, Barriers to implementing sustainable national newborn
167–176, https://doi.org/10.1016/j.jneumeth.2005.04.009. screening in developing health systems, Int. J. Pediatr. Adol. Med. 1 (2014) 49–60,
[13] U.R. Acharya, S.L. Oh, Y. Hagiwara, J.H. Tan, H. Adeli, Deep convolutional neural https://doi.org/10.1016/j.ijpam.2014.10.004.
network for the automated detection and diagnosis of seizure using EEG signals, [38] C. Elberling, M. Don, M. Cebulla, E. Stürzebecher, Auditory steady-state responses
Comput. Biol. Med. 100 (2018) 270–278, https://doi.org/10.1016/j. to chirp stimuli based on cochlear traveling wave delay, J. Acoust. Soc. Am. 122
compbiomed.2017.09.017. (2007) 2772–2785, https://doi.org/10.1121/1.2783985.
[14] M. Savadkoohi, T. Oladunni, L. Thompson, A machine learning approach to [39] A.J. Casson, Wearable EEG and beyond, Biomed. Eng. Lett. 9 (2019) 53–71,
epileptic seizure prediction using electroencephalogram (EEG) signal, Biocybernet. https://doi.org/10.1007/s13534-018-00093-6.
Biomed. Eng. 40 (2020) 1328–1341, https://doi.org/10.1016/j.bbe.2020.07.004. [40] R. Näätänen, E.S. Sussman, D. Salisbury, V.L. Shafer, Mismatch negativity (MMN)
[15] K. Saab, J. Dunnmon, C. Ré, D. Rubin, C. Lee-Messer, Weak supervision as an as an index of cognitive dysfunction, BrainTopogr 27 (2014) 451–466, https://doi.
efficient approach for automated seizure detection in electroencephalography, Npj org/10.1007/s10548-014-0374-6.
Digital Medicine 3 (2020) 1–12, https://doi.org/10.1038/s41746-020-0264-0. [41] R. Näätänen, P. Paavilainen, T. Rinne, K. Alho, The mismatch negativity (MMN) in
[16] N.P.G. Seshadri, B.K. Singh, R.B. Pachori, EEG based functional brain network basic research of central auditory processing: a review, Clin. Neurophysiol. 118
analysis and classification of dyslexic children during sustained attention task, (2007) 2544–2590, https://doi.org/10.1016/j.clinph.2007.04.026.
IEEE Trans. Neural Syst. Rehabil. Eng. 31 (2023) 4672–4682, https://doi.org/ [42] H. Wang, B. Li, Y. Lu, K. Han, H. Sheng, J. Zhou, Y. Qi, X. Wang, Z. Huang, L. Song,
10.1109/TNSRE.2023.3335806. Y. Hua, Real-time threshold determination of auditory brainstem responses by
[17] N.P. Guhan Seshadri, S. Agrawal, B. Kumar Singh, B. Geethanjali, V. Mahesh, R. cross-correlation analysis, iScience 24 (2021) 103285, https://doi.org/10.1016/j.
B. Pachori, EEG based classification of children with learning disabilities using isci.2021.103285.
shallow and deep neural network, Biomed. Signal Process. Control 82 (2023) [43] X. Wang, M. Zhu, O.W. Samuel, X. Wang, H. Zhang, J. Yao, Y. Lu, M. Wang, S.
104553, https://doi.org/10.1016/j.bspc.2022.104553. C. Mukhopadhyay, W. Wu, S. Chen, G. Li, The effects of random stimulation rate on
[18] R. Sharma, R.B. Pachori, A. Upadhyay, Automatic sleep stages classification based measurements of auditory brainstem response, accessed February 20, 2023, Front.
on iterative filtering of electroencephalogram signals, Neural Comput. Appl. 28 Hum. Neurosci. 14 (2020), https://www.frontiersin.org/articles/10.3389/fnhum.2
(2017) 2959–2978, https://doi.org/10.1007/s00521-017-2919-6. 020.00078.
[19] R. Krishna, K. Das, H.K. Meena, R.B. Pachori, Spectral graph wavelet transform- [44] V. Peterson, C. Galván, H. Hernández, R. Spies, A feasibility study of a complete
based feature representation for automated classification of emotions from EEG low-cost consumer-grade brain-computer interface system, Heliyon 6 (2020)
signal, IEEE Sens. J. 23 (2023) 31229–31236, https://doi.org/10.1109/ e03425.
JSEN.2023.3330090. [45] J. Polich, Updating P300: an integrative theory of P3a and P3b, Clin. Neurophysiol.
[20] P. Saideepthi, A. Chowdhury, P. Gaur, R.B. Pachori, Sliding window along with 118 (2007) 2128–2148, https://doi.org/10.1016/j.clinph.2007.04.019.
EEGNet-based prediction of EEG motor imagery, IEEE Sens. J. 23 (2023) [46] A. Delorme, S. Makeig, EEGLAB: an open source toolbox for analysis of single-trial
17703–17713, https://doi.org/10.1109/JSEN.2023.3270281. EEG dynamics including independent component analysis, J. Neurosci. Methods
[21] R.K. Joshi, K.S. Manu, R.S. Hari, M. Jayachandra, H.J. Pandya, Design, 134 (2004) 9–21, https://doi.org/10.1016/j.jneumeth.2003.10.009.
development and validation of a portable visual P300 event-related potential [47] J. Lopez-Calderon, S.J. Luck, ERPLAB: an open-source toolbox for the analysis of
Extraction system, in: In: 2022 IEEE Biomedical Circuits and Systems Conference event-related potentials, Front. Hum. Neurosci. (2014), https://doi.org/10.3389/
(BioCAS), 2022, pp. 409–413, https://doi.org/10.1109/ fnhum.2014.00213.
BioCAS54905.2022.9948657. [48] S.S. Shapiro, M.B. Wilk, An analysis of variance test for normality (complete
[22] G.-Y. Bae, C.J. Leonard, B. Hahn, J.M. Gold, S.J. Luck, Assessing the information samples)†, Biometrika 52 (1965) 591–611, https://doi.org/10.1093/biomet/52.3-
content of ERP signals in schizophrenia using multivariate decoding methods, 4.591.
Neuroimage: Clinical 25 (2020) 102179, https://doi.org/10.1016/j. [49] H.B. Mann, D.R. Whitney, On a test of whether one of two random variables is
nicl.2020.102179. stochastically larger than the other, Ann. Math. Stat. 18 (1947) 50–60, https://doi.
[23] S. Tegg-Quinn, R.J. Bennett, R.H. Eikelboom, D.M. Baguley, The impact of tinnitus org/10.1214/aoms/1177730491.
upon cognition in adults: a systematic review, Int. J. Audiol. 55 (2016) 533–540, [50] A. Majidpour, S.K. Jameel, J. Majidpour, H. Bagheri, T.A. Rashid, A. Nazeri, M.
https://doi.org/10.1080/14992027.2016.1185168. M. Aleaba, Detection of auditory brainstem response peaks using image processing
[24] R.M. Chapman, J.W. McCrary, M.N. Gardner, T.C. Sandoval, M.D. Guillily, L. techniques in infants with normal hearing sensitivity, Biomed. Signal Process.
A. Reilly, E. DeGrush, Brain ERP components predict which individuals progress to Control 86 (2023) 105117, https://doi.org/10.1016/j.bspc.2023.105117.
Alzheimer’s disease and which do not, Neurobiol. Aging 32 (2011) 1742–1755, [51] J.H. Grose, E. Buss, H. Elmore, Age-related changes in the auditory brainstem
https://doi.org/10.1016/j.neurobiolaging.2009.11.010. response and suprathreshold processing of temporal and spectral modulation,
[25] M. Taylor, D. McCulloch, Visual evoked potentials in infants and children, J. Clin. Trends Hearing 23 (2019), https://doi.org/10.1177/2331216519839615,
Neurophysiol. : Off. Publ. Am. Electroencephal. Soc. 9 (1992) 357–372, https:// 2331216519839615.
doi.org/10.1097/00004691-199207010-00004. [52] E. Skoe, N. Kraus, Auditory brain stem response to complex sounds: a tutorial, Ear
[26] S.J. Luck, An Introduction to the Event-related Potential Technique, second Hear. 31 (2010) 302, https://doi.org/10.1097/AUD.0b013e3181cdb272.
edition, MIT Press, 2014. [53] H. Zhang, M. Zhu, Y. Jiang, D. Wang, X. Wang, Z. Yang, W. Huang, S. Chen, G. Li,
[27] O. Manta, M. Sarafidis, N. Vasileiou, W. Schlee, C. Consoulas, D. Kikidis, E. Vassou, A robust Extraction approach of auditory brainstem response using adaptive
G.K. Matsopoulos, D.D. Koutsouris, Development and evaluation of automated Kalman filtering method, IEEE Trans. Biomed. Eng. 69 (2022) 3792–3802, https://
tools for auditory-brainstem and middle-auditory evoked potentials waves doi.org/10.1109/TBME.2022.3178550.
detection and annotation, BrainSci 12 (2022) 1675, https://doi.org/10.3390/ [54] R.A. Osman, H.A. Osman, On the use of machine Learning for classifying auditory
brainsci12121675. brainstem responses: a scoping review, IEEE Access 9 (2021) 110592–110600,
[28] C. Chen, L. Zhan, X. Pan, Z. Wang, X. Guo, H. Qin, F. Xiong, W. Shi, M. Shi, F. Ji, https://doi.org/10.1109/ACCESS.2021.3102096.
Q. Wang, N. Yu, R. Xiao, Automatic recognition of auditory brainstem response [55] H. Wimalarathna, S. Ankmnal-Veeranna, C. Allan, S.K. Agrawal, J. Samarabandu,
characteristic waveform based on bidirectional long short-term memory, Front. H.M. Ladak, P. Allen, Machine learning approaches used to analyze auditory
Med. 7 (2021), https://doi.org/10.3389/fmed.2020.613708. evoked responses from the human auditory brainstem: a systematic review,
[29] W. Team, Hearing screening: considerations for implementation, World Health Comput. Methods Programs Biomed. 226 (2022) 107118, https://doi.org/
Organization (n.d.). https://www.who.int/publications/i/item/9789240032767 10.1016/j.cmpb.2022.107118.
(accessed April 21, 2023). [56] N. Armanfard, M. Komeili, J.P. Reilly, J.F. Connolly, A machine Learning
[30] G.G. Celesia, Disorders of peripheral and central auditory processing 1: disorders of framework for automatic and continuous MMN detection with Preliminary results
peripheral and central auditory processing, Elsevier Health Sciences (2013). for coma outcome prediction, IEEE J. Biomed. Health Inform. 23 (2019)
[31] J.P. Pillion, G. Bibat, S. Naidu, Effects of sedation on auditory brainstem response 1794–1804, https://doi.org/10.1109/JBHI.2018.2877738.
in Rett syndrome, Pediatr. Neurol. 42 (2010) 331–334, https://doi.org/10.1016/j. [57] R.M. McKearney, R.C. MacKinnon, Objective auditory brainstem response
pediatrneurol.2010.01.003. classification using machine learning, Int. J. Audiol. 58 (2019) 224–230, https://
[32] G.M. Swords, L.T. Nguyen, R.A. Mudar, D.A. Llano, Auditory system dysfunction in doi.org/10.1080/14992027.2018.1551633.
Alzheimer disease and its prodromal states: a review, Ageing Res. Rev. 44 (2018) [58] D. Ahmedt-Aristizabal, T. Fernando, S. Denman, J.E. Robinson, S. Sridharan, P.
49–59, https://doi.org/10.1016/j.arr.2018.04.001. J. Johnston, K.R. Laurens, C. Fookes, Identification of children at risk of
[33] G. Lightfoot, Summary of the N1–P2 cortical auditory evoked potential to estimate schizophrenia via deep Learning and EEG responses, IEEE J. Biomed. Health
the auditory threshold in adults, Semin Hear 37 (2016) 1–8, https://doi.org/ Inform. 25 (2021) 69–76, https://doi.org/10.1109/JBHI.2020.2984238.
10.1055/s-0035-1570334. [59] J. Zhang, H. Wu, W. Su, X. Wang, M. Yang, J. Wu, A New Approach for
[34] S. Lippe, N. Kovacevic, R. McIntosh, Differential maturation of brain signal Classification of Epilepsy EEG Signals Based on Temporal Convolutional Neural
complexity in the human auditory and visual system, accessed January 7, 2023, Networks, in: 2018 11th International Symposium on Computational Intelligence
and Design (ISCID), 2018: pp. 80–84. Doi: 10.1109/ISCID.2018.10119.

13