Concurrent Validity of the Menteve EEG System: Signal Quality, Spectral, and Evoked-Response Comparison with a Research-Grade Reference Device

Figure 2a.

Executive summary

Electroencephalography (EEG) is widely used across research and clinical settings to assess neural activity with high temporal resolution. As portable, cost-effective EEG systems extend EEG's applicability beyond traditional laboratory environments, systematic validation against established research-grade devices becomes essential before such systems can be reliably adopted for research or clinical use (Radüntz, 2018; Pontifex & Coffman, 2023).

This study validates the Menteve EEG-R device against the Neuroelectrics Starstim 32, a well-established research-grade EEG system, through simultaneous acquisition from both devices under identical experimental conditions. Participants underwent a resting-state session (eyes-open and eyes-closed) and a visual oddball task known to elicit robust event-related potentials (Polich, 2007), allowing signal quality to be assessed across frequency-domain, time-domain, and signal-to-noise measures.

Across 13 participants (n = 5 for the time-domain/ERP analysis), the Menteve EEG-R showed strong concurrent validity with the Neuroelectrics Starstim 32 across signal-quality, spectral, and time-domain measures. Resting-state spectra were closely matched between devices, with both systems capturing a distinct alpha peak near 10 Hz under eyes-closed conditions. The subject device showed a significantly steeper aperiodic (1/f) spectral decay than the benchmark in both eyes-open and eyes-closed conditions, consistent with a favorable, low-noise broadband signal. Canonical band power was generally comparable between devices, with only eyes-closed alpha power showing a statistically significant difference; Bland-Altman analysis confirmed small mean biases (0.75 µV² EC, 0.40 µV² EO) and narrow limits of agreement across all frequency bands. Individual alpha frequency showed the strongest cross-device agreement of all measures, with high correlation and concordance and low absolute error under both conditions. In the time-domain analysis, both devices produced a clear P300 response to oddball stimuli, with closely matched peak latencies (within 16 ms at the parietal midline) and comparable amplitudes across all eight matched electrode pairs, while standard stimuli produced no spurious deflection in either device. Together, these results indicate that the Menteve EEG-R reliably captures the spectral and event-related characteristics of resting-state and cognitive EEG activity relative to a research-grade reference system, with further validation warranted in larger and more diverse samples .

Methods

Participants
Thirteen participants (12 male, 1 female; age range 17-26 years, mean age 20.7 ± 2.5 years) were recruited from the student community at the Indian Institute of Technology Kanpur through a campus-wide email call for participation. Individuals with a history of or current neurological conditions were excluded. All participants provided informed consent prior to data collection.Resting-state and spectral analyses are based on all 13 participants; the time-domain/ERP analysis is based on data collected from five participants to date.

EEG Data Acquisition:
The study employs a cross-sectional validation design with simultaneous EEG recording from two devices. Both EEG systems record neural activity concurrently while participants perform identical experimental tasks. Simultaneous acquisition ensured both systems captured identical underlying neural activity under the same experimental conditions enabling direct comparison of signal features across devices.

EEG data was recorded simultaneously from the following two systems:

Menteve EEG-R: 32 channel research-grade device from Menteve, used as the subject device for this study. 8 channels were used for the purpose of this validation study, the channel locations used were: ['PO3', 'PO4', 'F2', 'F1', 'Pz', 'FCz', 'CP1', 'CP2']. The data acquisition was recorded using its companion application Neurostream. AFz was used as the ground electrode, while the ear clip electrode was used as the reference electrode at the left ear. The raw EEG data was recorded in .xdf as a standard format provided in the Neurostream application.

Neuroelectrics Starstim 32: A research-grade EEG-stimulation system with up to 32-channel EEG recording from Neuroelectrics , used as the benchmark device. Following channels were used from the Neuroelectrics device: ['P2', 'CPz', 'P1', 'FC2', 'Fz', 'FC1', 'Cz', 'POz']. The data acquisition was recorded using its companion application NIC2. The right ear-clip electrode provided with the Neuroelectrics system was used to serve for the reference and ground electrodes. The raw EEG data was recorded in .edf as a standard format provided in the NIC2 application.

Because the two systems could not share an identical electrode placement, the electrode placement was matched to anatomically proximate channel pairs for direct comparison (e.g., Menteve's Pz corresponds to Neuroelectrics' POz in this mapping). This layout configuration was adapted from Pontifex and Coffman (2023). Both devices used the same type of EEG electrodes (sintered Ag/Agcl electrodes), and for both devices the data acquisition was done at the sampling rate of 500 Hz.

Experimental Protocol
Both devices recorded simultaneously while each participant completed two tasks in a single session.

1. Resting state. Participants sat quietly for 2 minutes with eyes opened followed by 2 minutes with eyes closed.

2. Visual oddball task. The visual oddball paradigm is a well-established attention task that reliably elicits event-related potentials (ERPs), most notably the P300. The task had three blocks of 40 trials each (120 trials in total), with a 30-second rest after each block. Each trial began with a fixation cross shown for a random duration between 0.3 and 0.5 seconds, followed by a colored circle shown for 1 second. Red circles were targets (oddball stimuli) and blue circles were non-targets (standard stimuli). Participants pressed a designated key when a red circle appeared and made no response to blue circles.

Data Pre-processing
Raw EEG data from both devices were preprocessed using a standardized pipeline in MNE-Python.
Following steps were conducted to perform pre-processing of EEG data from both devices.

Figure 1. Experimental protocol. Top: Session sequence. Bottom left: Electrode layout for simultaneous recording (orange: subject device; navy: benchmark device). Bottom right: Oddball trial sequence, with targets on 20% of trials and non-targets on 80% in randomized order.

Figure 2b.

Analysis and Statistical Methods

1. Aperiodic spectral analysis

The aperiodic component of the EEG spectrum was characterized using the FOOOF/specparam approach (Donoghue et al., 2020). Each power spectrum was split into an aperiodic (1/f) component and oscillatory peaks using the FOOOF/specparam algorithm. The aperiodic component was modeled as:

log₁₀ P(f) = b − χ · log₁₀(f)

Here, P(f) is power at frequency f, b is the aperiodic offset, and χ is the aperiodic exponent.

2. Spectral Characteristics

These measures test whether the two devices capture the same oscillatory content: how power is distributed across the standard frequency bands, and where each participant's alpha rhythm peaks. Both are well-established resting-state markers, so agreement on them indicates the subject device preserves physiologically meaningful spectral information.

2.1 Canonical band power

Band power was calculated by integrating the PSD over the delta, theta, alpha, beta and gamma frequency ranges (Donoghue et al., 2022). Agreement between the subject and benchmark devices was assessed for each band using Bland–Altman analysis. For each participant, the between-device difference was plotted against the between-device mean. The mean difference represented the bias, and the 95% limits of agreement were calculated as bias ± 1.96 × SD of the paired differences (Bland & Altman, 1986).

2.2 Individual alpha frequency

Individual alpha frequency (IAF) was defined as the frequency of maximum power within the alpha band. It was determined for each participant, separately for each device, in both the eyes-closed (EC) and eyes-open (EO) conditions. Agreement between devices was assessed using Pearson correlation, the concordance correlation coefficient (CCC), mean absolute error (MAE) and root mean squared error (RMSE) (Corcoran et al., 2018).

3. Time-domain analysis:

ERP: To evaluate the performance of subject device across time-domain during the visual oddball task, epochs were extracted from -0.2 to 0.6 seconds, with 0 as the time of stimulus presentation event for the standard and oddball stimulus. These event locked data were further plotted as ERP for each of the stimulus conditions and across all channels. Further, the grand-average ERP waveforms from the subject and the benchmark device were also plotted (Polich, 2007).

Results

Before the quantitative comparisons, Figure 3 shows resting state raw EEG traces along with power spectral density (PSD) for the subject device. In the raw signal plot, rhythmic alpha activity (8–12 Hz) is clearly visible with eyes closed and is most prominent over the posterior channels (PO3, PO4). With eyes open, this activity is markedly attenuated, and eye blinks appear as large deflections over the frontal channels (Fp1, Fp2). PSD plots further confirms this pattern showing a distinct peak in the alpha band for the eyes-closed spectrum that is substantially reduced with eyes open, while the rest of the spectrum follows a similar profile in both conditions. This eyes-closed alpha enhancement, known as the Berger effect, is one of the most robust signatures of resting-state EEG. Its clear presence in both the raw signal and the spectrum shows the subject device captures physiologically meaningful brain activity.

Aperiodic Spectral Parameterization (FOOOF):

For each device, condition, and participant, FOOOF/specparam was fitted to the post-ASR PSD (1–40 Hz) to estimate the aperiodic offset and exponent. Participant-level aperiodic parameters were then averaged across participants for statistical comparison between devices. Participant PSDs were interpolated to a common frequency grid and averaged in linear power to obtain a grand-average spectrum, the 1/​f component was reconstructed as b−χlog10​(f) and plotted against the log-transformed grand-average spectrum, providing a representation of the mean EEG spectrum and its spectral decay. Further,differences between devices in aperiodic offset and exponent were assessed using the Shapiro-Wilk test for normality, followed by paired t-tests for normally distributed data and Wilcoxon signed-rank tests for non-normal distributions.

Figure 3: Resting-state EEG from the subject device. (Top-left) Raw EEG, eyes closed: prominent posterior alpha activity. (Top-Right) Raw EEG, eyes open: attenuated alpha activity and frontal blink artifacts. (Bottom-Left) & (Bottom-Right) Power spectral density for eyes-closed and eyes-open conditions, showing the alpha peak with eyes closed.

Canonical Bandpowers & Bland-altman analysis:

Average band-power values desired frequency bands were calculated separately for the Subject and Benchmark devices under EC and EO conditions and reported as mean ± SD. Between-device differences were assessed independently for each frequency band and condition. Normality of the paired differences was evaluated using the Shapiro-Wilk test; paired t-tests were performed for normally distributed differences, whereas Wilcoxon signed-rank tests were used for non-normally distributed differences. Further, Bland-Altman analysis was performed to assess the agreement between devices.

Figure 4: (Top-Left & Top-Right)Eyes-closed (EC) and Eyes-open (EO) averaged power spectra demonstrated consistent agreement between devices, featuring a distinct alpha peak near 10 Hz and comparable spectral decay. Figure 4C:(Bottom-Left & Bottom-Right) Spectral decay was higher for the subject device than the benchmark device in both EC-Eyes-Closed (1.62 ± 0.16 vs. 1.45 ± 0.15; Wilcoxon signed-rank test, p = .002) and EO-Eyes-Open (1.56 ± 0.14 vs. 1.47 ± 0.18; paired t-test, p = .034) conditions.

Figure 5:(Left & Right) Band power was generally higher for the subject device across frequency bands under both EC and EO conditions.However, a statistically significant increase was observed only for EC alpha power (11.41 ± 3.28 vs. 10.02 ± 2.94 µV²; Wilcoxon signed-rank test, p = .02); all other between-device differences were not statistically significant.

2. Time domain analysis:

Grand-average ERPs were computed for each device at eight anatomically matched electrode pairs, separately for standard and oddball stimuli, using ICA-cleaned data from all five subjects (n = 5; shading throughout = ±1 SEM across subjects).

For standard stimuli, neither device showed a sustained positive deflection within the canonical P300 measurement window (300–500 ms) at any of the eight channel pairs; both waveforms remained close to baseline through this interval (Figure 1). This is the expected pattern in an oddball paradigm, where the frequent, non-target stimulus does not elicit a P3b, and it indicates that neither device produces a spurious positive deflection in this window in the absence of a genuine oddball response.

Oddball stimuli, by contrast, elicited a clear, sustained positive deflection within the 300–500 ms window at every one of the eight matched channel pairs for both devices (Figure 2), consistent with a canonical P300 (P3b) response. The spatial pattern was broadly consistent between devices, with the largest deflections over parieto-occipital and centro-parietal sites (P1/PO3, P2/PO4, POz/Pz, CPz/CP1, Cz/CP2) and smaller, later-developing deflections over frontal sites (FC1/F1, FC2/F2, Fz/FCz). The subject device tracked the benchmark device's waveform shape and approximate peak timing at nearly all sites, with amplitude running modestly higher at fronto-central channels and closely matched at parieto-occipital channels. Across-subject SEM bands for the two devices overlapped at nearly every channel and time point within the P300 window, indicating that the two devices produce statistically indistinguishable group-level P300 waveforms despite the small sample.

Figure 7: (Left & Right) IAF showed strong cross-device agreement under both EC (r = 0.89, CCC = 0.89, bias = 0.05 Hz) and EO (r = 0.94, CCC = 0.92, bias = −0.13 Hz), with low MAE (0.30 and 0.29 Hz) and narrow 95% LoA (−0.83 to 0.93 Hz and −1.09 to 0.84 Hz, respectively).

Figure 6: (Left & Right) Bland-Altman analysis showed a small positive mean bias between devices for both EC (0.75 µV²) and EO (0.40 µV²). The 95% LoA were −3.24 to 4.75 µV² for EC and −3.44 to 4.24 µV² for EO, with most observations distributed around the mean bias and no clear systematic trend across the range of band power.

Individual Alpha Frequency (IAF):

IAF showed strong agreement between the Subject and Benchmark devices under both EC and EO conditions.

To examine the canonical single-site P300 report more closely, Figure 7 shows the grand-average ERP at the parietal midline (POz for the benchmark device, Pz for the subject device) - the classic electrode pair for P300 measurement. Standard stimuli produced no sustained deflection in the P300 window, while oddball stimuli elicited a clear positive peak for both devices, with SEM bands overlapping throughout. The subject device peaked at 396 ms and the benchmark device at 380 ms, a 16 ms difference that falls well within typical inter-trial and cross-device timing precision, part of this 16 ms difference likely reflects the two devices being compared at slightly different scalp locations (POz vs. Pz) rather than a device-specific timing effect. (P3b latency is known to shift systematically across nearby parietal sites (Polich, 2007) and both devices reached comparable peak amplitude (~6–7 µV), reinforcing the channel-wise result at the single site most commonly reported in P300 studies.

Summary Tables:

Figure 8: (Left) Channel-wise grand-average ERP, standard stimuli (n = 5; benchmark device in navy, subject device in orange; shading = ±1 SEM; gray band = 300–500 ms P300 window). (Right) Channel-wise grand-average ERP, oddball stimuli. Sustained positive deflection is present within the shaded window at every channel pair for both devices, consistent with a P300 response.

Discussion

The resting-state findings demonstrate comparable EEG characteristics between the Menteve EEG-R and the Neuroelectrics Starstim 32. The averaged spectra under both eyes-closed (EC) and eyes-open (EO) conditions showed similar overall spectral profiles and a clear alpha peak near 10 Hz, indicating that both devices captured the major features of resting-state EEG.

Menteve demonstrated significantly steeper spectral decay than the benchmark device under both EC and EO conditions. This indicates a stronger decline in power at higher frequencies with highly similar spectral profiles with a distinct alpha peak near 10 Hz, supporting Menteve’s ability to capture key resting-state spectral characteristics.

Canonical bandpower was generally higher for Menteve, although a significant difference was observed only for EC alpha power, this can be attributed to having more number of channels on the occipital area as compared to the benchmark device. Band-wise Bland-Altman analysis showed the closest agreement for gamma and beta, with relatively small differences for theta across both EC and EO conditions. Larger deviations were observed for delta across both conditions, while alpha showed a more pronounced difference under EC. The pooled analysis showed small mean biases of 0.75 µV² for EC and 0.40 µV² for EO, with no clear systematic trend across the range of power values. Overall, the results indicate good correspondence between devices, with the degree of agreement varying across frequency bands and conditions.

The strongest agreement was observed for individual alpha frequency. Individual Alpha Frequency (IAF) showed high correlation and concordance between devices under both EC and EO, with small biases and low errors. This suggests that Menteve reliably captures the frequency characteristics of individual alpha activity, alongside other spectral powers across the spectrum.

Both devices showed a clear P300 response, with larger positive deflections for oddball than for standard stimuli across all eight matched electrode pairs and at the parietal midline (POz/Pz), where the P300 is typically largest. Grand-average peak latencies agreed closely between devices, within about 5–20 ms in the 300–500 ms window, which indicates that the subject device captures the timing and topography of this cognitive ERP comparably to the benchmark device. The subject device showed somewhat larger P300 amplitudes and a more pronounced negative deflection before the P300 (about 150–250 ms). These differences will be examined further in the full-sample analysis.

Overall, the resting-state findings demonstrate good agreement between Menteve EEG-R and Neuroelectrics Starstim 32 across key spectral measures, with particularly strong correspondence in relative spectral characteristics and IAF. The observed differences in aperiodic measures, EC alpha power highlight device-specific characteristics while maintaining overall consistency in standard resting-state EEG measurements.

Limitations

This work was conducted as an initial assessment and validation of a subject device as compared to an established benchmark device. The reported study was conducted on a small sample size that may limit statistical power and generalizability. The resting-state protocol was limited to 2 minutes each of eyes-open and eyes-closed recording, and longer, repeated recordings would provide a stronger assessment of reliability. Additionally, the devices used different electrode layouts, with anatomically proximate rather than identical channels, which may contribute to differences in absolute power and aperiodic measures. Finally, further analyses using larger datasets and more rigorous statistical approaches are warranted to strengthen the assessment of cross-device agreement and generalizability.

Conclusion

Menteve EEG-R demonstrated strong agreement with the research-grade benchmark device across key signal quality, spectral and time-domain measures obtained during resting-state and evoked potential EEG data. Its steeper spectral decay further highlights favorable broadband signal characteristics. While some device-specific differences were observed, the overall findings support Menteve EEG-R as a reliable system for resting-state EEG characterization, with further validation warranted in larger and more diverse datasets.

References:

Bland, J. M., & Altman, D. G. (1986). Statistical methods for assessing agreement between two methods of clinical measurement. The Lancet, 327(8476), 307–310. https://doi.org/10.1016/S0140-6736(86)90837-8

Corcoran, A. W., Alday, P. M., Schlesewsky, M., & Bornkessel-Schlesewsky, I. (2018). Toward a reliable, automated method of individual alpha frequency (IAF) quantification. Psychophysiology, 55(7), e13064. https://doi.org/10.1111/psyp.13064

Donoghue, T., Haller, M., Peterson, E. J., Varma, P., Sebastian, P., Gao, R., Noto, T., Lara, A. H., Wallis, J. D., Knight, R. T., Shestyuk, A., & Voytek, B. (2020). Parameterizing neural power spectra into periodic and aperiodic components. Nature Neuroscience, 23(12), 1655–1665. https://doi.org/10.1038/s41593-020-00744-x

Donoghue, T., Schaworonkow, N., & Voytek, B. (2022). Methodological considerations for studying neural oscillations. European Journal of Neuroscience, 55(11–12), 3502–3527. https://doi.org/10.1111/ejn.15361

Polich, J. (2007). Updating P300: An integrative theory of P3a and P3b. Clinical Neurophysiology, 118(10), 2128–2148. https://doi.org/10.1016/j.clinph.2007.04.019

Pontifex, M. B., & Coffman, C. A. (2023). Validation of the g.tec Unicorn Hybrid Black wireless EEG system. Psychophysiology, 60, e14320. https://doi.org/10.1111/psyp.14320

Radüntz, T. (2018). Signal quality evaluation of emerging EEG devices. Frontiers in Physiology, 9, 98. https://doi.org/10.3389/fphys.2018.00098

Figure 9:Grand-average ERP at the parietal midline (POz for the benchmark device, Pz for the subject device -nearest matched pair), standard vs. oddball, n = 5. Conventions as in Figures 1–2. Peak latency is labeled only on the oddball trace, within the 300–500 ms window.

Canonical Bandpower Analysis

Aperiodic Analysis

Individual Alpha Frequency

Bland-Altman Analysis

1. Frequency Domain Analysis

Authors: Harsh Arora, Ajeetha R, Surya V S, Arjun Ramakrishnan

Menteve Health Pvt. Ltd., Indian Institute of Technology Kanpur

Study Information: This study was conducted at the Decision Lab, Department of Biological Sciences and Bioengineering, Indian Institute of Technology Kanpur, in collaboration with Menteve Health Private Limited. A peer-reviewed manuscript based on this work is in preparation.

Date: 11.September.2026

[View in Desktop mode only]

3A

3B