Validation of direct recording of electrically evoked cortical auditory evoked potentials through a cochlear implant system

Abstract

Cochlear implants (CI) are one of the most successful treatments available to enable individuals with severe to profound hearing loss to regain access to the world of sound. This is accomplished through the electrical stimulation of the auditory nerve using electrodes implanted inside the cochlea. The use of subjective user feedback makes the process of fitting these devices much more challenging in cases where users are not able to actively or accurately report their experience (e.g. pediatrics), making an objective measurement that reflects the accuracy or effectiveness of a program quite attractive. We recorded one objective measure, the electrically-stimulated cortical auditory evoked potential (eCAEP), non-invasively using the CI in response to a simulated speech sound in seven adult participants and compared it to their eCAEP recorded using a scalp EEG set-up. The eCAEPs recorded with CI electrodes were comparable to scalp recorded eCAEPs (grand mean cross-correlation of r = 0.83, individual mean cross-correlations ranged from 0.13 to 0.70). Evoked potential peaks P1, N1 and P2 showed no significant latency difference based on if the eCAEP was recorded on the scalp or using the CI. The eCAEP waveforms recorded via the CI appear to converge in a distinct P1-N1-P2 waveform by as early as 130 sweeps. In conclusion, in this study we show the feasibility of recording the eCAEP directly through the CI system, which could potentially be used to guide CI fitting and track auditory cortex development in response to CI use.

Introduction

The cochlear implant (CI) is a remarkable technology that has transformed the lives of many individuals with severe to profound hearing loss. By directly stimulating the auditory nerve, CI’s provide users with access to sound, enabling speech perception and improved communication abilities^1,2. However, despite their efficacy, there remains variability in outcomes among CI users, particularly in speech intelligibility, especially in challenging listening environments^3,4.

In order to ameliorate difficulties for listeners with cochlear implants and optimize their performance, many strategies have been used. This includes the development of new electrode designs, stimulation methods, and surgical techniques have been developed which improve listening quality provided by CI’s in both quiet^2,5 and in noisy conditions^1,6,7. Even with these improvements, one contributing factor to this variability is the complexity of optimizing the settings of the CI to suit individual user needs, a process known as “fitting”⁸.

Traditional CI fitting relies heavily on behavioral feedback from users, making it time-consuming and challenging. This fitting is particularly challenging for individuals who cannot provide reliable feedback, such as infants, young children or those with cognitive impairments. Furthermore, the fitting process often starts from default settings, with only limited adjustments made based on behavioral responses, and these settings may not be updated frequently enough to accommodate changes in physiology, anatomy, cortical development or subjective preferences over time.

To address these challenges, there is growing interest in using objective measures, particularly those derived from electroencephalography (EEG), to improve the CI fitting process. EEG offers several advantages, including high temporal resolution necessary for the auditory system, non-invasive nature, and relatively low cost and complexity. Recent research has focused on developing electrophysiological-based objective measures relevant to CI fitting, such as estimating auditory thresholds^9–12, comfortable loudness levels^9,13–16, and neural survival in the cochlea¹⁷.

While current clinical practice primarily utilizes short-latency evoked potentials, such as the electrically evoked compound action potential (eCAP) and electrically evoked auditory brainstem response (eABR), which target peripheral components of the auditory pathway, there is growing recognition of the importance of measuring longer-latency responses associated with higher-level auditory processing. One of these responses, known as cortical auditory evoked potentials (CAEP), has been shown to be a reliable biomarker of cortical auditory development in children^18–22. The CAEP presents as a single positivity (the P1) in young children between 100 and 200 ms after presentation of an auditory stimuli, and by the age of 7–10 years of age the latency of this positivity moves earlier after the stimulation and is followed by a negativity (the N1) and then a second positivity (the P2); which collectively are described as obligatory CAEP responses^23–25.

The P1 has been shown to be a useful tool in describing cortical auditory maturation. In CI users, the CAEP P1 has been shown to be correlated with length of CI use where longer use is reflected by P1s that are closer to that of hearing children²⁶. Conversely, the CAEP P1 was used with early and late implanted children to show that there is a critical period for cortical auditory development in children with cochlear implants²⁷. Following this initial response, the N1 is a negative deflection that occurs just after the P1. In adult CI users, the N1 has been shown to relate to speech perception in CI subjects^28–30.

Recording EEG in the clinic, however, can be challenging due to the need for skin preparation, application of conductive gel, and the requirement for participants to remain still during recording to minimize motion artifacts. The CI itself generates a large artifact in the auditory EEG recordings which can obliterate the auditory P1 and N1 components³¹. Additionally, traditional clinic settings do not replicate real-world listening environments since all testing is done in a sound attenuated booth. Mobile EEG recording systems offer some advantages but are still too bulky and cumbersome for frequent use. Consequently, there are very few locations where CAEP tests are performed. Recording evoked potentials on CI users in the clinic presents additional complexity due to the need for CI programing devices, availability of scalp EEG recording systems and synchronization devices in between the EEG and CI units.

A novel alternative approach is to leverage the implanted electrodes of the CI for EEG recording. While current CI systems have limited telemetry capabilities for recording short-latency responses, recent advances have demonstrated the potential of recording EEG from implanted electrodes, targeting both subcortical (electrically evoked ABR) and cortical (implant recorded CAEP) responses^32–35. Somers et al.³⁴ did so in CI patients with percutaneous connector and used external amplifiers. The backward-telemetry systems embedded in the implant itself can be utilized to achieve a direct recording from the CI. However, in most devices this system is limited by a small recording buffer and thus the response must be stitched together to perceive a useful CAEP response^32,33, hence these recordings were time consuming to collect and potentially prone to error.

Attias et al.³⁵ were able to expand the continuous recording time through the use of the Advanced Bionics cochlear implant system with custom research software while recording CAEP responses in bimodal recipients. Bimodal listeners have access to acoustic hearing in one ear and electrical hearing from a CI in the other ear. In this study, acoustic stimuli were presented to the ear with residual hearing, with simultaneous recordings made from the CI on the contralateral side and a traditional EP machine. In most cochlear implants, the back telemetry system is designed specifically to record eCAP responses with a recording window of approximately 3 ms or less. However, because of the design of the recording and telemetry system of Advanced Bionics implants it is possible to record for longer, giving a possible recording window of even up to 1–2 s. This study effectively validated the use of a commercial cochlear implant system to record full length CAEPs in a CI population. However, it does not assess the response to electrical stimulation provided by the cochlear implant.

The present study represents a significant step toward bringing this measure to clinical practice by showing the feasibility and efficiency of recording electrically-stimulated cortical auditory evoked potential (eCAEP) in adults with bilateral cochlear implants. In this study, eCAEP responses were recorded directly through the intracochlear electrode array and implant, in response to a stimulated speech stimulus presented using the CI electrodes. We examine the similarity of the eCAEP to simultaneous recordings using a traditional electroencephalography (EEG) method. Additionally, this study examines how efficiently and reliably the eCAEP may be recorded via the cochlear implant and how it may be achieved in a clinically feasible manner.

Results

eCAEP waveforms

Data was collected successfully from all 7 participants, and no participants asked to end the testing before completion of the study. The recording procedure took approximately 25 min per run of 800 sets of stim and null sweeps. We successfully recorded sets of the simulated /uh/ sound presented at M-level for all participants. After the M-level recording, we did ask participants to verify their perceived sound perception level and all seven participants confirmed that they still perceived the stimuli to have been presented at their most-comfortable level.

Figure 1 shows the grand average eCAEP of all participants directly recorded from the CI, in green, and recorded from the scalp, in purple. When considering the similarity of the grand average waveforms, we used cross-correlation to quantify it. While the waveform figures presented here are low-pass filtered at 30 Hz to aid in the visualization of peaks as is standard in cortical potential literature, cross-correlations were conducted on 100 Hz low-pass filtered data. Cross-correlation was used to provide a statistically robust, objective measure that is used to compare the overall similarity of the two recorded signals: the scalp recorded eCAEP and the CI recorded eCAEP. The maximum cross-correlation for the grand average CI eCAEP waveform compared to the scalp eCAEP waveform was 0.83, with a lag of 0 ms. This high cross-correlation is consistent with the visual similarity that can be seen in the figure.

Fig. 1.

Shows the grand average electrically-stimulated cortical auditory evoked potential (eCAEP) in response to a simulated /uh/ stimuli, recorded using traditional EEG at the scalp (in purple) and recorded using the BEEP software through the cochlear implant (in green). Waveforms are lowpass filtered at 30 Hz for display. Labels have been added for the P1, N1, and P2 eCAEP components in the grand average waveforms.

Figure 2 depicts the M-level eCAEP response for each participant recorded from the CI array (green) compared to with the eCAEP using the traditional EEG setup, with the waveform from Cz referenced to the joint mastoid shown.

Fig. 2.

Individual electrically-stimulated cortical auditory evoked Potential (eCAEP) averages in response to a simulated /uh/ stimuli. Recordings for individual participants are plotted together with the scalp eCAEP in purple and CI eCAEP in green. Waveforms are lowpass filtered at 30 Hz for display. Scales are set to be ± 120% of the maximal rectified amplitude between 50 and 500 ms to avoid the potential influence of the CI artifact. Obligatory eCAEP P1, N1, and P2 peaks are indicated in each figure, and remaining implant stimulus artifact present in figures are indicated with “CI*” with color indicating device on which the artifact was still present.

To examine the similarity of the paired participant waveforms, we again examined the scalp and CI eCAEP M-level waveforms using cross correlation in individual subjects. Maximum cross-correlations, with associated lag in parentheses, were: S1 = 0.70 (− 16 ms); S2 = 0.61 (6 ms); S3 = 0.48 (8 ms); S4 = 0.51 (31 ms); S5 = 0.44 (− 48 ms); S6 = 0.13 (16 ms); S7 = 0.70 (1 ms). From this, the mean cross-correlation in individual subjects is 0.51, with a standard deviation of 0.19. The mean lag was − 0.29 ms, with a standard deviation of 25.42 ms. Note: a positive lag means that the maximum cross-correlation was achieved when the CI eCAEP waveform is shifted later in time, and a negative lag means that the maximum cross correlation was achieved when the CI eCAEP is shifted earlier in time in comparison to the scalp eCAEP waveform.

eCAEP peak analysis

The evoked potentials and morphology for each participant was examined by two experienced CAEP researchers, authors DBS & AS. In all participants, the morphology was consistent with what we would expect from adult listeners with years of experience in sound. This is to say that each waveform’s morphology contained an early positive deflection at a latency of between 25 and 55 ms post stimulus; followed by a negative deflection between 65 and 115 ms post stimulus; and finally, a second positive deflection between 140 and 210 ms post stimulus. This morphology is consistent with the obligatory CAEP responses described by Ponton et al.²⁴ and Gilley et al.²⁵. Obligatory CAEP peaks (the P1, N1, and P2 components) were measured for each participants’ CI and scalp eCAEP. Peak latency and amplitude comparisons can be seen in Fig. 3. For amplitudes, it is important to note that the baseline period was overlapping time recorded by both systems, not the pre-stimuli interval, as this does have the effect of occasionally causing the P1 (the first positive deflection in the waveform) to have a slightly negative amplitude.

Fig. 3.

Comparison of peak latencies and amplitudes by recording device. Peak latencies (on the left) and amplitude (on the right) are compared directly. Measures from scalp recordings using traditional EEG are on the horizontal (X) axis; measures recorded from the cochlear implant are on the vertical (Y) axis. In these plots, color is used to indicate which peak the measure is coming from, important as eCAEP P1 vs P2 amplitudes show a different pattern. Observations from the same participant are shown in all three plots using the same shape.

To examine the difference in waveform peak latencies and amplitudes quantitatively, we used repeated-measures ANOVA, with the two recording methodologies by three picked peaks within-subjects design. Mean peak latencies and standard deviations can be seen in Table 1. In the repeated-measures ANOVA the main effect of the recording device on peak latency, when controlling for which peak was measured, was not significant. Post-hoc Wilcoxon Signed-Ranks tests, using a false discovery rate correction, were utilized to examine if there were differences in latencies at each of the peaks separately. None of the three obligatory peaks showed significant differences in latency between CI and scalp recordings.

Table 1.

Mean peak latencies by recording method for the obligatory CAEP P1, N1 and P2 peaks, with standard deviation of the amplitude measures within this sample in parentheses.

Recording method	P1 mean latency (SD)	N1 mean latency (SD)	P2 mean latency (SD)
CI eCAEP	36.6 (7.4)	86.2 (9.7)	174.6 (20.6)
Scalp eCAEP	40.7 (11.6)	84.7 (13.1)	182.4 (19.9)

A second repeated-measures ANOVA was conducted to examine peak amplitude by peak and recording method. Mean peak amplitude and standard deviations for each peak and recording method can be seen in Table 2. The main effect of recording method on peak amplitude, when controlling for which peak was measured, was not significant in this ANOVA model. While there appears to be a difference in P1 amplitudes in Table 2, this difference may have been influenced in part by implant artifacts cooccurring with the P1 time window for adult CAEP responses. Post-hoc Wilcoxon Signed-Ranks tests, using a false discovery rate correction, were utilized again to examine if there were differences in amplitudes at each of the peaks separately. None of the three obligatory peaks showed significant differences in amplitude between CI and scalp recordings.

Table 2.

Mean peak amplitudes by recording method for the obligatory CAEP P1, N1 and P2 peaks, with standard deviation of the amplitude measures within this sample in parentheses.

Recording method	P1 mean amplitude (SD)	N1 mean amplitude (SD)	P2 mean amplitude (SD)
CI eCAEP	3.34 (2.70)	− 3.17 (1.30)	2.13 (1.16)
Scalp eCAEP	0.76 (1.20)	− 2.63 (1.76)	1.93 (1.07)

Stability of the eCAEP within a single recording

To examine the effect of the number of sweeps, electrically-stimulated cortical auditory evoked potential (eCAEP) waveforms for the CI recorded responses were created at with a varying number of sweeps, using sweeps from the beginning up to that number of sweeps to imitate the response in a clinical test. The grand average eCAEPs with different numbers of sweeps per subject can be seen in Fig. 4. As can be seen on the left side of Fig. 4, the general form of the eCAEP waveforms for the grand average can be generally identified even with as little as about 150 sweeps. As seen on the right side of Fig. 4, for an individual the general waveform is present by 150 sweeps, but the peaks are still somewhat obscured by noise present in the recording system.

Fig. 4.

Cochlear implant (CI) eCAEP waveforms by number of trials, lowpass filtered at 30 Hz. Left panel shows the convergence of the eCAEP ERP in the grand average for all 7 participants; right panel shows an exemplar participant (S2) demonstrating that for an individual the eCAEP waveform emerges in the 150 to 250 trial range.

In order to quantify recommendations for the number of trials, we performed cross-correlations for our eCAEP waveforms for each subset number of trials, in increments of 10 up to 800 sweeps. This cross-correlation was conducted using the full 800-sweep eCAEP per participant as the signal of comparison. Maximal cross-correlations and lags can be seen in Fig. 5.

Fig. 5.

Maximum cross-correlation and associated lags of the waveforms based on the number of trials included. Grand Average waveform cross-correlations by different numbers of trials per participant on the left; individual participant waveform cross-correlations on the right. In all cases, the standard being compared to in the cross-correlation is the full 800 trials.

In the grand average, shown on the left side of Fig. 5, the cross correlation swiftly rises within the first couple hundred trials, while after that there appears to be minimized benefit for adding more trials. Indeed, by 130 trials for each participant in our grand average, the cross-correlation has surpassed a level of 0.8 in comparison to the full 800-sweep CI eCAEP. However, in the individual participants, shown on the right side of Fig. 5, the results show more variability. The median cross-correlation at 130 trials is 0.60. The median cross-correlation does not reach 0.8 until 250 trials. It is also worth noting that after 130 trials per participant the lags for the maximal cross-correlation converge to 0, and at no point after this does any participant have a lag of more than 1 ms in comparison to their full 800-sweep eCAEP.

Replicability and stimulus perceptibility

As time and participant willingness permitted, we collected data for exploratory conditions in addition to the primary M-level recordings. For 2 participants, S1 and S2, with the participant’s consent, we collected a second run of 800 sets of stimuli to show the replicability for a measurement. Replication runs were taken during the same session, immediately following the initial runs. Replication run waveforms can be seen in Fig. 6.

Fig. 6.

Individual eCAEP waveforms for replication runs. Top shows replication on the scalp eCAEP recording, bottom shows replication on the CI eCAEP recording. On the left is S1, on the right S2. Obligatory eCAEP P1, N1, and P2 peaks are indicated in each figure, and remaining implant stimulus artifact present in figures are indicated with “CI*” with color indicating device on which the artifact was still present.

Replication eCAEP waveforms show a very similar pattern between the first and second runs. However, it can be observed visually that the CI eCAEP recordings retain their similarity to the scalp eCAEP recordings even to a degree in the places in which they deviate from the initial run. For example, for both participants, the P2 was slightly earlier in the replication run than the original run. This shift is present in both the scalp eCAEP waveform as well as in the CI eCAEP waveform.

For two participants, S6 and S7, we collected a sub-threshold recording in addition to the M-level recording. These waveforms can be seen in Fig. 7. The M-level recording is the same as above, and the sub-threshold recording is with the same simulated /uh/ sound being presented through the implant at a presentation level the participant reported not being able to perceive.

Fig. 7.

Individual M-level eCAEP waveforms compared to sub-threshold data. Top shows comparison on the scalp eCAEP recording, bottom the CI eCAEP recordings. On the left is S6, on the right S7. Obligatory eCAEP P1, N1, and P2 peaks are indicated in each figure, and remaining implant stimulus artifact present in figures are indicated with “CI*” with color indicating device on which the artifact was still present.

In the recordings for S6, amplitudes for the eCAEP were generally smaller than for other participants in both the scalp and CI recordings. However, the P1 and N1 components can be seen in both M-level recordings, the solid lines on the left of Fig. 7. Due to the scale compared to the noise floor, the P2 is more difficult to discern for the CI M-level recording. However, while there is a stimulus artifact at the beginning of the sub-threshold recording, seen in the dotted lines on the left of Fig. 7, there is no discernable cortical response represented in the sub-threshold recording.

Participant S7 shows this pattern even more clearly. In the sub-threshold recordings for S7, dotted line on the right of Fig. 7, the only activity above the noise floor is the stimulus artifact at the beginning of the recording. On the other hand, the M-level recordings for both scalp recording and CI recording methods showed a large cortical response typical of the P1-N1-P2 complex, shown in the solid lines on the right of Fig. 7. The P1 amplitude in the CI eCAEP recording is potentially artificially high as it is cooccurring with the CI stimulus artifact, but it is still clearly present and differentiable from the sub-threshold response.

Discussion

The purpose of the current study was to show the feasibility of recording an electrically- stimulated cortical auditory evoked potential (eCAEP) using the backward-telemetry system present in a cochlear implant (CI). We recorded the eCAEP in adult bilateral CI users by stimulating and recording the cortical response using only the CI’s that participants already had, without the need for a specialized external evoked potential (EP) or electroencephalography (EEG) device. Specifically, we found that (1) CI eCAEPs were similar to scalp eCAEPs when examining the grand average or individual recordings using cross-correlation; (2) eCAEPs peaks did not significantly differ based on whether they were recorded in the CI or on the scalp; and (3) CI eCAEPs waveforms including the P1-N1-P2 morphology emerged by 150–250 sweeps for each participant.

There was a clear similarity between the scalp and CI eCAEP waveforms. Given the differences between the recording techniques and electrode placements, the similarity between the CI and scalp recordings is remarkable. When looking at the average of all seven participants in Fig. 1, the cross-correlation is 0.83 indicating that there is a strong correlation between these two signals. Beyond the group average, all seven participants showed a positive cross-correlation point-by point between their CI eCAEP and scalp eCAEP, and similarities are visible in Fig. 2. This is consistent with previous literature that showed the ability to record auditorily evoked CAEPs using CI electrodes^32–35. The current study is a step forward in showing the ability to record electrically- stimulated cortical auditory evoked potential (eCAEP) waveforms that do not depend on external stimulation or communication of timing to an external device. By combining the equipment required for stimulation and recording of the eCAEP, the procedure is simplified making it more easily translatable to clinical use.

Obligatory P1-N1-P2 peaks were identifiable in all subjects in the CI eCAEP and these peaks showed similarities to the same peaks identified in the scalp eCAEP. The average latencies, in Table 1, for each peak did not vary significantly based on where they were recorded. Looking at the left side of Fig. 3, the latencies fall close to the line of equivalent latency between the scalp and CI eCAEP. This demonstrates that the CI eCAEP component peaks are at nearly the same latency as the scalp eCAEPs. However, we would need a larger sample size to definitively say at what level they are statistically correlated.

However, even with these similarities there were some instances in which the morphology of the waveforms did appear to differ between the CI and scalp eCAEP. For example, S4 and S6 appear to have a smaller and earlier P2 component. The variability in the lags for the cross-correlation is another example of the differences between scalp and CI eCAEP recordings. One potential reason could be differences in the location and directionality of the generators of these later components compared to a participant’s implant case and electrode positions, which are known to vary from patient to patient. Ponton et al.³⁶ used dipole source modeling to show that the waveform morphology of the CAEP originated from a dipole oriented tangentially to the auditory cortex. This is to say that it would have a dipole extending vertically from each auditory cortex and so it can be inferred that we should see this when recording from the CI. Future research in a larger sample using dipole source analysis will examine if the dipole orientation in reference to the sensor placement is indeed related to the observed lags between these recorded signals.

It has been further shown though that whereas the source of the CAEP P1 is reflecting primary auditory cortex processing and processing at the level of the cortex, the N1 and P2 components are reflective of higher-level cortical processing including secondary auditory cortex and cortico-thalamic feedback^37,38. Further research with a sample appropriate for dipole analysis is needed to examine if this difference in generators or their positions relative to participants’ implants is contributing meaningful variance in the CI eCAEP.

Waveform amplitude did vary noticeably within this sample, for both scalp and CI recordings of the eCAEP, as shown in Table 2 and the right side of Fig. 3. There was a trend for the CI recorded peaks to be slightly larger than the scalp recorded peaks, primarily in the early P1 peak, although the voltages were not significantly different statistically. Somers et al.³⁴ observed a similar trend in their data which they attributed to the recording electrodes being in closer proximity to the neural generators. However, both the Somers et al.³⁴ finding and the current study are limited by small sample size and thus these amplitude differences should not be regarded as conclusive.

Even with the observed variability in the eCAEP amplitudes, the CI recorded eCAEPs presented in this study represent an important step forward because it allows us to infer that the eCAEP recordings obtained represent cortical perception of the stimulus. First, the replication runs (Fig. 6) show remarkable stability; because of the stability, the eCAEP is a cortical auditory potential that can be utilized for clinical testing. Given the small sample in the current study, future research will be needed to examine test–retest reliability in more detail. Second, the presence of the eCAEP in response to most-comfortable level (M-level) stimuli and its absence in the presence of sub-threshold measurements (Fig. 7) show that the eCAEP is recording cortical perception of the stimuli, not simply noise in the recording system. For the CI eCAEPs, this difference is most notable for the later components, which have less overlap with the CI recording artifact, but even the P1 shows a difference from the M-level to sub-threshold level recordings.

Finally, this study shows that the number of sweeps necessary for the eCAEP morphology to emerge can be obtained in a clinically viable timeframe. In most participants, the eCAEP morphology emerges by 150 sweeps (Fig. 4). This is to say that by 150 sweeps the eCAEP has a clear P1, N1, and P2 pattern present in the waveform, a median cross-correlation with the full 800-sweep eCAEP of above 0.6 with a lag of 1 ms or less. However, participants with smaller eCAEP amplitudes (specifically participants S3, S5, and S6) showed noticeable noise at 150 sweeps but by 250 sweeps the eCAEP morphology can be clearly seen; this is consistent with the cross-correlations (Fig. 5) which shows these three individuals take longer for the eCAEP to emerge. Therefore, 250 sweeps will most reliably show the full eCAEP morphology but for most individuals 150 sweeps will be sufficient.

Clinical implications

The ability to record these eCAEP responses in a clinically feasible time period would be particularly useful in monitoring pediatric and adult patients with cochlear implants. Previous CAEPs collected in response to acoustic stimulation in CI users required the use of an external recording system^19,39–42 or at least a separate system to present auditory stimuli³⁵. However, for bilateral CI users, the BEEP software with the associated hardware used in this study allows all presentation and recording to be done in a unified environment that could be implemented in clinical practice. In addition, as the recording is done using specialized processors that do not transmit microphone input, there would be less concern about interference from competing auditory signals.

By a clinically feasible time-period, we mean that the eCAEP morphology emerges and can be measured by 150–250 sweeps, taking 5–8.3 min. Abbas and Brown⁴³ have shown that the CAEP is more useful than eCAP at predicting behavioral thresholds of perception. Thus, the eCAEP may be a useful measure in fitting and tracking perceptual thresholds in CI users, even in patients, such as small children, for whom it would traditionally be difficult to measure behavioral thresholds.

In pediatric patients, the eCAEP could be a powerful tool track auditory cortex development. In young children, the CAEP P1 occurs as late as 300 ms post stimulus and gradually decreases to adult like latencies, as seen in the current study, by the age of 12–16 years^24,44,45. This well-established developmental pattern also leads the CAEP to be a useful tool in addition to conventional audiological testing for tracking if the auditory cortex is responding to proper stimulation in a hearing aid or cochlear implant patient^{21,22,27,46–56}. Given that the CI recording artifact only extends for the first 25 to 30 ms post stimuli, as can be seen in Fig. 2, pediatric eCAEP P1’s should be clear and unaffected by the CI recording implant, implying that this recording technique for the eCAEP shows promise as a clinical tool.

Limitations

The limited sample size is an important limitation of this study. In this field, case studies and studies with smaller sample sizes have previously been used to show the effect of auditory development(cf.^20,42,46) and neuroplasticity (cf.^47,49) in implanted individuals. Further research with a larger sample size is needed to more reliably estimate how much of the variability in latencies, amplitudes, and cross-correlations is due to the different recording devices and how much is simply individual differences.

The current study also was only able to include participants with bilateral Advanced Bionics cochlear implants. While it is possible that there are some technological advances or signal processing methods that may eventually enable unilateral recording, at the current time we were only able to create usable eCAEP recordings by using one CI to present the stimuli and the contralateral CI to record the eCAEP. However, this is not likely a major hinder for pediatric use, as most pediatric populations currently are bilaterally implanted.

Another limitation of this study is that for adults the early P1 component present in the eCAEP is close in time as to potentially be co-occurring with the CI recording artifact which can be observed (Fig. 2) to occur in some participants between 20 and 50 ms after the eCAEP stimuli. However, as mentioned above, the CAEP P1 in pediatric hearing controls has a typical latency later than the CI rerecording artifact was observed^46,57,58. Thus, while the early CI recording artifact overlaps in time with the eCAEP P1 in some adult participants, thus interfering with the ability to definitively identify eCAEP P1’s for adults, we expect that future research will show that for pediatric patients that the CI recording artifact would have little to no effect on finding the eCAEP P1 for use as a marker of auditory cortex development.

Conclusions

In conclusion, this study demonstrates a feasible way to record electrically-stimulated cortical auditory evoked potentials (eCAEP) using cochlear implants. These evoked potentials can be recorded in a clinically viable time of about 5–8 min and do not require specialized evoked potential or electroencephalography devices. Thus, these eCAEP responses are a tool that could be collected clinically to have an objective measure to augment traditional audiological measures to better track cortical auditory development and improve fitting to ensure appropriate auditory access.

More work will need to be done to verify that this recording process produces comparable results in pediatric patients and in a broader sample to show usable norms. This would provide an innovative methodological technique that audiologists and clinicians could use as a hearing diagnostic, potentially toward a closed-loop fitting system, and providing a way to track auditory cortex development in response to sound after receiving a cochlear implant.

Methods

Participants and ethics declarations

A total of seven adult bilateral CI users (four females and three males) participated in this study. Participants ranged in age from 19 to 82 years of age. The participants were all bilaterally implanted with an Advanced Bionics CI device with a fully inserted electrode array. Table 3 details the demographic data, age of implantation, and type of implanted device. Note that there was a lack of uniformity in the etiology of hearing loss, although only one participant (S2) reported being deaf from birth. All experimental procedures were approved by the relevant ethical review board, the Institutional Review Board at the University of Colorado Boulder (Boulder, CO), and were carried out in accordance with their guidelines and regulations. Participants provided informed consent to participate in the study prior to all testing and were free to terminate their participation at any point during the study. Prior to analysis, all data were de-identified and managed with appropriate concern for the confidentiality of information obtained during testing.

Table 3.

Demographic data for each participant, including age, gender, age of implantation, and preferred CI ear.

Subject no	Age (yrs)	Gender	Age (yrs) at implantation	Left ear device	Right ear device
S1	52	M	42	HiRes 90K Advantage	HiRes 90K Advantage*
S2	54	F	50	HiRes Ultra 3D*	HiRes Ultra 3D
S3	81	F	78	HiRes 90K Advantage	HiRes Ultra 3D*
S4	58	M	53	HiRes Ultra 3D	HiRes Ultra 3D*
S5	19	F	9	HiRes 90K Advantage	HiRes 90K Advantage*
S6	64	F	54	HiRes Ultra 3D*	HiRes 90K Advantage
S7	82	M	64	HiRes 90K Advantage	HiRes 90K*

*Denotes the preferred ear indicated by the participant. The participant’s preferred CI was used to present the stimuli that evoked eCAEP responses. The contra-lateral CI was used to record eCAEP responses.

Experimental setup

Experimental setup is shown in Fig. 8 and shows the recording system for both scalp recorded eCAEPs and eCAEPs recorded via the implanted electrodes in the cochlear implant.

Fig. 8.

Experimental setup: the participant was sitting in an electromagnetically and acoustically shielded chamber. Simulated auditory stimuli were presented through the participant’s preferred cochlear implant (CI), and then recorded from the contralateral CI, using the BEEP research software. The BEEP software (Advanced Bionics Corp.) communicated with each CI using Advanced Bionics’ clinical programing interface (CPI-3) and triggering devices (Advanced Bionics research trigger dongle) to communicate timing information to create concurrent recordings. Simultaneous recordings were taken using traditional scalp EEG using EGI’s Net Station software (Electrical Geodesics, Inc.). A transistor-transistor logic (TTL) triggering from the trigger dongle to the EGI Net Amps amplifier was used to match the timing between the recording modalities.

CI eCAEP recording setup

A research interface developed by Advanced Bionics, Bionic Ear Evoked Potential (BEEP) research software version 1.0.1.1, was used to connect to the cochlear implants, using clinical programing interface (CPI-3), research trigger dongles, and Naida CI Q90 processors. The BEEP software controlled the presentation of the stimulus to one CI and the recording of the electrically-stimulated cortical auditory evoked potential (eCAEP) from the contra-lateral CI. The stimulus was presented directly to the ipsilateral CI via the CPI-3, and a trigger time locked to the end of the stimulus presentation was delivered through the research trigger dongle to the contralateral CI, which was then used to record the eCAEP response. In addition, this trigger was also used to mark a simultaneous event trigger in the standard EEG recording system.

The stimulus used in this study was a 20 ms simulation of a speech sound /uh/ that was modeled after a simple continuous interleaved sampling (CIS) strategy with a flat electrical map (same stimulation level for each electrode) and presented using the first (apical) 8 CI electrodes. The use of these 8 electrodes is a limitation of the current version of the research software, in which the apical half of the electrodes are reserved for presenting stimulus as they are more comfortable for listening while the 8 electrodes toward the base of the cochlea are reserved for recording. Prior to testing, the experimenter first needed to determine the most comfortable listening level (M-level) for the participant. To do this, the initial stimuli level was set at the minimum presentation level and the stimulus was repeatedly presented. The participants were asked to indicate their comfort level using a loudness scaling chart that ranges from 0 indicating no sound perceived, 6 indicating most comfortable, and 10 being too loud. The output stimulation in the cochlear implant was gradually raised until the participant indicated that it was at their M-level, at which point the stimulus was presented at that level for at least an additional 30 s, and then the participant was asked to confirm that they still perceived this to be at their most comfortable listening level.

In addition to the M-level recordings with the simulated speech sound which were recorded for all participants, additional recordings were taken as time permitted. In participants S1 and S2, we recorded an additional set of the simulated speech sound at M-level to show replicability of recording. In participants S6 and S7, we recorded a set with the simulated speech sound presented but at a stimulation level below the lowest level at which the participant reported being able to perceive the stimulus.

For the BEEP recordings, data was recorded at a sampling rate of 1000 Hz for 500 ms following the presentation of the stimuli. The data in this study were all recorded from electrode number 12 (towards basal side) of the implant as the active electrode, using the case electrode as the reference. The case electrode is situated in the CI housing, typically approximately 13 cm from the recording electrodes for an adult user. After recording, the data for each sweep was transferred back by the RF telemetry to the BEEP software for signal processing and recording.

Scalp eCAEP recording set-up

Standard EEG recordings for the eCAEP responses were measured using 128-channel high-density EEG (GSN-Hydrocell, Electrical Geodesics, Inc.). These scalp measurements of the eCAEP were recorded using NetStation 5 software (Electrical Geodesics, Inc.) at a sampling rate of 1000 Hz with a band-pass filter set at 0.1–200 Hz. Events from the BEEP interface were recorded in the continuous EEG data to be used for the extraction of segments to create the evoked potentials.

Throughout the eCAEP testing, participants were seated in an electromagnetically and acoustically shielded chamber. Participants were given a full explanation of the procedure prior to starting measurement. After setting up the BEEP and EEG systems, the comfort level of the participant in response to the stimuli was determined as described in the above section. Once this most comfortable level was established, eCAEP recordings were made using 800 repetitions of the simulated speech sound at this presentation level. For each repetition, or measurement “sweep”, BEEP software administrated a “stim” recording in which the stimulus was presented, and a subsequent “null” recording for the same length of time without a stimulus being presented. Each repetition took approximately two seconds to record, including stimulation (20 ms), stim recording (500 ms), null recording (500 ms), and any extra processing time that was needed to communicate with the CI. During testing, participants watched a muted video to maintain participants’ alertness and to minimize eye-blink and head movement artifacts.

EEG processing

CI eCAEP analysis

The CI eCAEP (using BEEP) was visualized real-time during recording and extracted for additional analysis for this paper. First, similar to the BEEP real-time visualization, we subtracted the "null” recording from the "stim” recording for each repetition. The “null” recording captures the intrinsic recording artifact from the CI system going from stim mode to record mode (the “recording start-up artifact”). This artifact is particular to each recording configuration and electrode-tissue interface, but is not dependent on stimulation level. The “stim–null” subtraction cancels out the recording start-up artifact, leaving the rest of the recorded signal being biologically evoked potential and any artifact that is due to the stimulus. To match the filter settings of the scalp EEG, the CI EEG data was then filtered using a finite impulse responses filter with a bandpass of 1 Hz to 100 Hz, matching the filter settings for the scalp EEG. After filtering, all sweeps were averaged to create an evoked potential. At the end, for all the plots, because stimulus duration was 20ms and recording only started at the end of the stimulus, 20 ms of empty data points was added before the recorded CI eCAEP data, so that time 0ms represents the onset of the stimulus and both sets of eCAEP responses are relative to the stimulus onset.

Scalp eCAEP analysis

Prior to comparison of the high-density scalp recorded EEG to the BEEP data, we first pre-processed the scalp recorded data to reduce the influence of line noise, head movements, and cochlear implant artifacts as follows. Data was processed using the EEGLAB toolbox⁵⁹ for MATLAB (The MathWorks®, Inc.). Initial steps were completed using the PREP pipeline⁶⁰, a standardized early-stage EEG processing pipeline. The PREP pipeline was used to reduce the influence of 60-Hz line-noise, identify and interpolate bad channels, and calculate a robust average reference. After processing using the PREP pipeline, the EEG signal was high-pass filtered at 1 Hz, and epochs were extracted. For the evoked eCAEP potentials epochs were extracted around each stimulus from 100 ms before the stimuli to 500 ms post-stimuli. Epochs with motion artifacts or electrode artifacts were automatically rejected using a joint probability filter and then by manual inspection. Finally, channels rejected by the PREP pipeline were replaced using a spherical interpolation to remove undue influence of noisy channels on the subsequent analyses.

Finally, independent components analysis (ICA) was performed using the extended Infomax algorithm⁶¹. This ICA decomposition was then used to remove independent components that consisted of eye-blinks or movement, channel noise, or muscle noise. Additionally, artifacts caused by the cochlear implant were minimized using ICA to remove as much of the artifacts as possible using the method outlined by Gilley et al.³¹. In one of the seven participants (S2), it was not possible to completely remove the implant artifact, but the artifact was minimized to enable a comparison to the BEEP data.

After artifact reduction, the data was then referenced to use the apical electrode (Cz) with a reference of the average of the two mastoid electrodes. This montage was used for the eCAEP as it is a typical CAEP layout that has previously been shown to create a waveform that is a reliable biomarker of auditory pathway maturation (cf.^18–20). Prior to averaging, the EEG data was filtered using a finite impulse responses filter with a bandpass of 1–100 Hz. The apical electrode (Cz) data was then averaged across trials to create the scalp eCAEP.

Following offline processing and the creating of averaged evoked potentials, two experienced CAEP researchers (authors DBS & AS) examined the waveforms marked results to compute latencies and amplitudes for the obligatory P1, N1, and P2 CAEP components. Waveforms were limited to the shared time-window and left unlabeled to blind researchers to recording device and participant number during peak picking. Peak latencies and amplitudes were computed marked at the center of the peaks. The absolute amplitude of the P1, N1 and P2 components were measured relative to a baseline consisting of the average of all points in the time-window shared between the two recording methods, i.e. 20–500 ms post stimulus.

Statistical analysis

Statistical analysis was performed using R statistical software (v 4.3.3) and the RStudio interface (v 2024.4.1). Similarities of the scalp and CI signals were quantified using a cross-correlation between the scalp recorded and CI recorded average signal. Cross-correlation has been shown to be a robust measure of mutual information between separate EEG signals^62–64. Due to differences in the length of data points collected, cross-correlations were conducted using the data from the time window that included the obligatory P1, N1 and P2 peaks with zero padding to match the length of the scalp eCAEP. In order to avoid matching with subsequent peaks, e.g. the P1 to the P2, a maximum lag of 50 ms was considered. Cross-correlation was examined to compare CI signals and scalp eCAEP EEG signals for the grand-averaged data for the entire group of participants as well as for individual participants.

Differences in peak latencies and amplitudes between the two recording methods were analyzed using repeated measure ANOVA, with the two recording methodologies by three picked peaks within-subjects design. For statistical analysis, the N1 peak amplitudes were rectified so that larger peak amplitudes were more positive. Due to the small sample size, we are unable to reliably determine if the latencies or amplitudes come from a gaussian distribution. Therefore, in addition to the ANOVA test, a post-hoc test was conducted using a Wilcoxon Signed-Ranks Test⁶⁵ to examine if there are differences between the latencies or amplitudes for each of the obligatory peaks. A false discovery rate correction was used on these to protect these post-hoc test against type I errors without being overly conservative^66,67.

Further, the efficiency of the recording of the eCAEP was examined by comparing the final CI eCAEP with all trials included to CI eCAEPs with fewer trials included. The comparison CI eCAEPs were created using increments of 10 up to the maximum (800 sweep) by taking only number of sweeps from the beginning of the recording to the number of increments. We have chosen the initial sweeps instead of randomly sampling the sweeps to simulate what a clinician would see and avoid any attenuation effects. For each comparison CI eCAEP the cross-correlation was computed compared to the complete 800 sweep final CI eCAEP to examine at what point adding more sweeps no longer provides meaningful change in the outcome signal.

Author contributions

Conceptualization: A. Sharma., A. Spahr, C.C. & D.B.S.; methodology: C.C. & DBS; formal analysis: D.B.S.; data curation: D.B.S. & C.C.; writing—original draft preparation, D.B.S. & C.C.; writing—review and editing: D.B.S., C.C., A. Sharma. & A. Spahr.; visualization: D.B.S.; supervision: A. Sharma.; project administration: A. Sharma.; funding acquisition: A. Sharma. & A. Spahr. All authors have read and agreed to the published version of the manuscript.

Data availability

Deidentified versions of the datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The lab of author Anu Sharma is funded by the National Institutes of Health. This study was funded in part by Advanced Bionics Corporation. Authors Chen Chen and Tony Spahr are current employees of Advanced Bionics, USA. Author Don Bell-Souder declares no potential conflict of interest.