Amplitude growth of intracochlear electrocochleography in cochlear implant users with residual hearing

Cochlear implants (CI) were initially developed to restore auditory perception for severe-to-profound hearing loss by mimicking the functionality of the peripheral auditory system (Loizou, 1998; Wouters et al., 2015). The CI delivers electrical stimulation to the auditory nerve that elicits sound sensations and thus enables speech perception. In the clinical routine of CI treatment, the telemetry capabilities of current CIs are used to examine the electrode-nerve interface (e.g., Brown et al., 2000; Hughes et al., 2000) by measuring electrically evoked compound action potentials (ECAP) generated by the auditory nerve in response to electrical stimulation. In CI users with residual acoustic hearing, the CI holds the potential to examine the electrode-nerve interface including the pathophysiological condition of the hearing system for combined electric acoustic stimulation (EAS). Furthermore, it can be used to objectively characterize EAS interaction at peripheral level (at the level of the hair cells and the auditory nerve).

Novel surgical techniques (Lenarz et al., 2009; von Ilberg et al., 1999) combined with the development of soft electrode arrays (Hochmair et al., 2015) allow the preservation of substantial low-frequency residual hearing during and after implantation (Fraysse et al., 2006; James et al., 2005; Kiefer et al., 2005; von Ilberg et al., 2011). Consequently, the CI implantation criteria have been expanded towards people with residual hearing in the implanted ear to benefit from combined EAS (Kiefer et al., 2005; von Ilberg et al., 2011; Skarzynski and Lorens, 2010). It has been shown that low frequency acoustic hearing significantly improves speech intelligibility with a CI, especially in noisy environments (Büchner et al., 2009; Gifford et al., 2013).

Auditory masking, the phenomenon by which one sound makes another sound less audible, has been widely investigated through psychoacoustic experiments in normal hearing listeners (e.g., Moore, 2008; Wegel and Lane, 1924; Zwicker and Schorn, 1978; Zwislocki, 1972). In EAS subjects, recent studies have also demonstrated interaction or masking between electric and acoustic stimulation (Imsiecke et al., 2018; Krüger et al., 2017; Lin et al., 2011). Lin et al. (2011) showed psychoacoustic masking of acoustic tones due to the presence of an electric masker stimulus in a single EAS subject. Krüger et al. (2017) characterized psychoacoustic interaction in a group of five EAS subjects. It was shown that acoustic stimuli were masked by electric stimuli and that electric stimuli were also masked by acoustic stimuli. An asymmetry between these two conditions was observed, probably because each condition measured interaction effects toward different directions in the cochlea in relation to the masker stimulus. In the electric masking experiment, the acoustic probes were presented to more apical cochlear locations than the electric masker stimulus. In contrast, in the acoustic masking experiment, the electric probe was presented at cochlear locations more basal than the acoustic masker stimulus. Therefore, the mechanisms involved in electric on acoustic and acoustic on electric may be different as EAS subjects only have active inner hair cells toward the apex of the cochlea. Furthermore, Krüger et al. (2017) showed that electric on acoustic masking decayed exponentially with increasing distance between the relative cochlear locations at which the electric and the acoustic stimuli were presented. It was suggested that this exponential decay in masking can be explained from peripheral mechanisms related to the voltage spread when current is injected into an intracochlear electrode. Imsiecke et al. (2018) confirmed the presence of interaction not only across locations but also across time using a forward masking experiment. The relation between EAS interaction and speech understanding for EAS users still needs to be investigated. However, promising results were shown by Imsiecke et al. (2019), where a significant decrease in performance was observed for subjects presenting large EAS interaction when the electric and the acoustic modality shared some frequency bandwidth.

Under the assumption that the origin of EAS interaction is at least partially peripheral (at the level of the hair cells or the auditory nerve) it should be feasible to demonstrate EAS interaction using electrophysiological measurements. Electrocochleography (ECochG) is a method to record electric potentials elicited by the hair cells and the auditory nerve in response to acoustic stimulation. It includes the cochlear-microphonic potentials (CM), the auditory nerve neurophonic potentials (ANN) responses, the action potentials (AP), and the summating potential (SP). However, the emphasis of the current study was on the CM and the ANN. Difference (CM/DIF) or summation (ANN/SUM) potentials from ECochG recordings in response to acoustic stimuli with alternating polarities were used to derive CM and ANN responses. However, the CM and the ANN cannot be isolated completely by the CM/DIF and the ANN/SUM, although the CM/DIF is dominated by the CM and the ANN/SUM is dominated by the ANN. As described by, e.g., Fontenot et al. (2017) or Forgues et al. (2014), the CM/DIF emphasizes odd harmonics of the stimulus frequency, which typically result in a spectral shape showing a peak at the acoustic stimulus frequency (first harmonic). This peak is dominated by the CM but could also include the biggest part of the ANN. The rectification and asymmetric characteristic of the ANN lead to additional even harmonics of the stimulus frequency that are emphasized by the ANN/SUM. In the presence of ANN responses, the spectral shape of the ANN/SUM shows the most dominant peak at the second harmonic of the stimulus frequency.

In EAS users, ECochG can be recorded without additional external or transtympanic electrodes using the intracochlear electrodes of the CI (e.g., Abbas et al., 2017; Koka et al., 2017a; Koka et al., 2017b; Koka and Litvak, 2017; Krüger et al., 2020). Therefore, the ECochG can serve to objectively investigate peripheral EAS interaction and to distinguish between CM and ANN interaction. Moreover, in combination with psychoacoustic EAS interaction experiments, it can be used to investigate the contribution of peripheral EAS interaction on psychoacoustic EAS interaction.

Previous studies have shown that ECochG responses can be used to monitor cochlear functionality during and after implantation (Dalbert et al., 2015b; Dalbert et al., 2015a). Koka et al. (2017b) showed that intracochlear ECochG recordings can be used to estimate hearing thresholds from CM/DIF and ANN/SUM responses in EAS users. They found a better correlation for CM/DIF responses to hearing thresholds than for ANN/SUM responses. Furthermore, ANN/SUM responses were not as present as CM/DIF responses and therefore could not be recorded for all subjects. A similar observation was shown by Abbas et al. (2017); for a subset of the tested subjects CM/DIF, responses grew with increased stimulation levels while ANN/SUM responses grew less or were not measurable. Consequently, intracochlear CM/DIF recordings seem more suitable to evaluate peripheral hearing mechanisms than ANN/SUM responses.

In Koka and Litvak (2017), EAS interaction was investigated through ECochG in response to acoustic stimuli in the presence of electrical stimulation. This objective measure of interaction was compared to psychoacoustic EAS interaction. The magnitude of ECochG EAS interaction was lower than the psychoacoustic EAS interaction. In Krüger et al. (2020), the method introduced by Koka and Litvak (2017) was combined with cone beam computed tomography (CBCT) to compare ECochG EAS interaction and psychoacoustic EAS masking across the relative spatial locations at which electric and acoustic stimulation was provided inside the cochlea. The outcome of this study showed that psychoacoustic masking was most pronounced if electric and acoustic stimuli were presented around the same spatial location inside the cochlea and that the mean CM/DIF interaction obtained from ECochG recordings was significantly lower than the mean psychoacoustic EAS interaction.

In Koka and Litvak (2017) and in Krüger et al. (2020), the acoustic stimuli used in the psychoacoustic interaction experiments were presented at threshold level while the ECochG interaction was measured using acoustic stimuli at most comfortable level (MCL) to maximize the CM/DIF and ANN/SUM responses. The difference in stimulation level used in both experiments may be an explanation for the discrepancy between the observed electrophysiological and psychophysical interaction. Even if CM/DIF responses are not directly related to behavioral responses, the difference in stimulation level used in both experiments could contribute to the discrepancy between the observed electrophysiological and psychophysical interaction. For instance, studies investigating the compound action potential (CAP) in animals showed an effect of acoustic probe level on EAS interaction. Nourski et al. (2007) observed less amount of masking with increasing probe level. Stronks et al. (2010) and Stronks et al. (2011) also reported more suppression of acoustically evoked CAP responses in the presence of electric stimuli for low acoustic probe levels than for high acoustic probe levels. Therefore, these results indicate that to understand the peripheral mechanisms of EAS interaction, it is necessary to record electrophysiological EAS interaction through CM/DIF or ANN/SUM at threshold level such that it can be compared to EAS interaction based on psychoacoustic threshold detection.

This work characterizes CM/DIF and ANN/SUM amplitude growth functions (AGF) derived from intracochlear ECochG recordings. The AGFs are used to determine the CM/DIF and ANN/SUM thresholds as well as the response amplitudes at psychoacoustic threshold level and MCL. The AGFs are used in an electrophysiological experiment to objectively investigate the effect of electrical stimulation on CM/DIF and ANN/SUM potentials in response to acoustic stimuli presented at various levels from threshold to MCL. Moreover, the electrophysiological measures are supplemented by a psychoacoustic EAS interaction experiment to assess the relation between electrophysiological and psychoacoustic EAS interaction and to investigate the origin of EAS interaction.

Two experiments were designed to measure the ECochG amplitude growth and to compare psychoacoustic and electrophysiological interaction. The psychoacoustic experiment was designed to investigate the effect of electric stimulation on acoustic hearing thresholds. The electrophysiological experiment used intracochlear ECochG recordings to determine peripheral EAS interaction objectively. ECochG amplitude growth was recorded to estimate CM/DIF thresholds and to investigate the effect of the acoustic stimulus level to the amount of interaction.

Psychoacoustic thresholds, MCLs, ECochG response recordings, and CBCT image data were analyzed for ten CI users having residual hearing in the implanted ear. Figure 1 shows the unaided air conduction pure tone thresholds measured via headphones (Sennheiser electronic GmbH & Co. KG, Wedemark, Germany) and a clinical audiometer (Audio 4000, Homoth Medizinelektronik GmbH & Co. KG, Kaltenkirchen, Germany) at the study appointment. The thresholds are given in dB hearing level (HL) according to DIN ISO 389:-8:2004. All subjects were implanted with an Ultra HiFocus implant and a SlimJ electrode array (Advanced Bionics, Valencia, CA). Table I contains detailed demographic data for the ten CI users.

The last column of Table I shows the audiometric frequencies selected for acoustic stimulation in the psychoacoustic and electrophysiological experiments. For adequate testing time, only one frequency was selected for each subject. From previous findings in Krüger et al. (2017), Krüger et al. (2020), and Imsiecke et al. (2018), it was expected to observe EAS interaction for low acoustic frequencies corresponding with the tonotopic frequency associated with the electrode locations. In Krüger et al. (2017), the electric acoustic frequency difference (EAFD) was defined as the difference between the acoustic probe frequency and the tonotopic frequency corresponding to the electrode location in octaves. Psychoacoustic masking was observed up to EAFDs of two octaves. For this reason, the selection of acoustic probe frequencies for the current study was limited to 250, 500, and 750 Hz, which resulted in a mean EAFD of 0.8 ranging from −0.24 to 1.8 octaves. A detailed characterization of the tested subject's cochleae (see Fig. 12) including the implant's insertion angle and insertion depth is presented in the  Appendix (see Fig. 13). Moreover, from these three possible frequencies, we selected the one for which the subjects were able to perceive acoustic stimulation at MCL without reaching the amplification limit. Note that to record the AGF, it was desirable to reach MCL perception. Additionally, we visually checked which of these three frequencies presented the most promising signal-to-noise ratio (SNR) at MCL though a pilot measurement before starting the actual experiment.

Figure 2 shows the hardware instrumentation used for the psychoacoustic and the electrophysiological experiments. The setup was adapted from Koka and Litvak (2017). The subject's implant was connected to a PC via the clinical programming interface (CPI-2, Advanced Bionics, Valencia, CA, USA) and a portable speech processor (PSP, Advanced Bionics, Valencia, CA) to provide electrical stimulation. Acoustic stimulation was delivered using an audio amplifier (Sony PHA-2, Sony Corporation, New York, NY) and inserted earphones (ER2, Etymotic Research, Elk Grove Village, IL). The audio amplifier was connected to a digital-to-analog converter (NI DAQ 6216 BNC, National Instruments Corporation, 11 500 Mopac Expwy, Austin, TX) which used a USB connection to communicate with the PC. The trigger output of the CPI-2 was used to synchronize the electric and the acoustic stimulation. Electric and acoustic stimuli were generated using the Bionic Ear Data Collection System (BEDCS, Advanced Bionics, Valencia, CA). BEDCS was also used for telemetry data collection from the CI. Acoustic stimuli were calibrated in the subject's ear canal using a probe tube (ER3-14C, Etymotic Research, Inc, Elk Grove Village, IL 60 007 USA) connected to a microphone amplifier (ER-7C Series B Clinical Probe Microphone System, Etymotic Research, Inc, Elk Grove Village, IL 60 007 USA) and an analog-to-digital converter of the NI DAQ 6216.

Figure 3 shows the recording configuration of the HiRes Ultra implant and the SlimJ electrode array (Advanced Bionics, Valencia, CA). The second most apical electrode was used as recording electrode $elrec$⁠. The implant's ring electrode was used as reference electrode $elref$⁠. The internal buffer of the HiRes Ultra implant was configured to perform 54.4 ms recordings by setting the sampling rate of its analog-to-digital converter (ADC) to 9280 samples/s.

ECochG was recorded in response to acoustic stimulation (⁠$fstim$⁠). Therefore, sinusoidal tones with durations of 50 ms and Hanning window ramps at onset and offset were used. Each ECochG recording consisted of 240 trails with alternating polarities (120 rarefaction and 120 condensation). CM/DIF responses were derived by subtraction of the rarefaction and condensation responses as a representation of the CM. ANN/SUM responses were derived by the summation of rarefaction and condensation responses as a representation of the ANN. If electrical stimulation was present, unmodulated pulse trains with a pulse duration of 25 μs and a stimulation rate of 1019 pulses per second (pps) were delivered using the most apical electrode $(elstim)$⁠. This procedure allows the recording of ECochG responses even under the presence of an electrical stimulation artifact as demonstrated by Koka and Litvak (2017). CM/DIF and ANN/SUM responses were evaluated regarding the amplitudes of the first six harmonics. For this, the response recordings were transformed into the frequency domain through a discrete Fourier transform (DFT) of 404 samples. Zero-padding was used for adequate amplitude estimation.

The measurement of ECochG (CM or ANN) thresholds is limited by the recording system's inherent noise quality. For this reason, the ECochG threshold was defined as the minimum sound pressure level required to elicit a significant response above the noise floor of the CI's telemetry system, which was in the range between 0.1 and 0.3 μV (mean 0.19 μV, standard 0.03 μV) across frequencies. The noise floor was used to estimate ECochG responses at psychoacoustic thresholds and MCLs in dB above this noise floor.

The noise floor was estimated for each subject to evaluate the significance of the recorded CM/DIF and ANN/SUM responses. For this purpose, ECochG responses without acoustic or electric stimulation were recorded. The recorded time series were transformed into the frequency domain using a DFT. Next, the significance level (noise floor) was estimated using a bootstrap method with 1000 iterations to obtain the 99% confidence interval for each frequency DFT bin (Koka et al., 2017b).

ECochG amplitude growth was recorded in response to acoustic stimulation from psychoacoustic MCL to a level below the threshold using an individualized step size. Depending on the absolute CM/DIF amplitude at MCL, the step size was selected to obtain an AGF consisting of four to five measurement points and at least three points within the subject's dynamic range (DR), defined as the difference between psychoacoustic MCL and T-Level. The MCL was determined by an initial loudness scaling procedure in which the acoustic stimulus level was successively increased, from inaudible to MCL. On a ten-scale loudness table, ranging from “extremely soft” to “extremely loud,” the MCL was indicated as number six. The MCL was estimated as the mean of two repetitions of the loudness scaling procedure.

Psychoacoustic interaction was investigated comparing acoustic detection thresholds of sinusoidal tones in the presence (masked condition) and in the absence (unmasked condition) of electrical pulse trains presented at single electrodes using the method described in Krüger et al. (2017), Imsiecke et al. (2018), and Krüger et al. (2020). The amount of interaction was obtained as the difference between the unmasked and the masked condition and was defined as threshold elevation. A three-interval forced-choice (3-IFC) procedure with a two-down one-up rule was used to determine the detection thresholds. On the psychometric function (Levitt, 1971) this is defined as 71% correct.

In the unmasked condition, three intervals, indicated by tree numbered buttons (1, 2, and 3) shown on a PC screen were presented to the user. The buttons lit up sequentially for 500 ms with a pause of 500 ms in between. Randomly, centered in one of these intervals, the acoustic probe stimulus was presented for 200 ms including a 5 ms Hanning window ramp at onset and offset. In the masked condition, electrical stimulation was delivered for 500 ms in all three intervals simultaneously. Unmodulated electric pulse trains were delivered at the most apical electrode at MCL for 500 ms in all three intervals. In both conditions, the subject's task was to indicate the interval where the acoustic stimulus was presented.

The level of the acoustic probe stimulus was adapted to determine the acoustic detection threshold. The initial level of the acoustic stimulus was set to MCL to ensure it was audible. The probe level was decreased with a step-size of 8 dB. The step-size was halved after each reversal until the final step-size of 2 dB was reached. Eight additional reversals were measured using the final step-size and the detection threshold was estimated as the mean of the last four reversals.

The subjects were instructed and trained to the experiment performing one run of the 3-IFC in the unmasked condition. Finally, one run in the unmasked and masked condition was used to estimate the psychoacoustic threshold elevation for each acoustic frequency. The probe level converging from MCL to threshold level was visually checked. If the standard deviation of the detection thresholds of the last four reversals were larger than 4 dB, the measurements were discarded.

Electrophysiological interaction was investigated using CM/DIF and ANN/SUM response recordings to acoustic pure tones (A), electric pulse trains (E), and simultaneously presented acoustic and electric stimuli (E + A). The electrical artifact in the response E + A was minimized by subtracting the response E. The resulting response (A′) includes the response to an acoustic stimulus and the interaction with the electric stimulus (Koka and Litvak, 2017; Krüger et al., 2020). The time series of A and A′ were transformed into the frequency domain using a DFT to determine the EAS interaction. The difference in the magnitude of the amplitude spectrum at the acoustic stimulation frequency indicated the amount of EAS interaction. Figure 4 illustrates the procedure to estimate electrophysiological EAS interaction.

The AGFs were obtained from a sequence of CM/DIF and ANN/SUM recordings in response to acoustic tones presented at various stimulation levels. Figure 5 shows an example of the recoded CM/DIF response waveforms for stimulation levels from −5 to 75 dB sound pressure level (SPL).

The AGFs of the different harmonic components were also analyzed for ten subjects. Figure 6 shows the amplitude growth of the CM/DIF and ANN/SUM recordings in response to acoustic stimulation for each of the first four harmonics. Each CM/DIF and ANN/SUM was recorded in response to 120 repetitions of rarefaction and condensation for each stimulation level. The resulting standard error is given for each measurement. Psychoacoustic thresholds and MCLs determined by the psychoacoustic experiment are indicated by vertical dashed and dashed-dot lines for each subject.

As shown in Fig. 6, the recordings were dominated by the first harmonic for the CM/DIF or the second harmonic for the ANN/SUM of the stimulation frequency. Significant first harmonic CM/DIF responses at MCL could be recorded for nine of ten subjects. Second harmonic ANN/SUM responses at MCL were significantly above noise floor for eight of ten subjects. The mean CM/DIF response at MCL for the first harmonic was 8.76 μV and it was significantly larger than the mean ANN/SUM response of 0.98 μV (p < 0.01, paired-samples t-test 1-tailed) for the second harmonic.

CM/DIF amplitudes at the third harmonic were significantly above noise floor but low (mean 0.24 μV, all below 0.36 μV) at MCL for subjects 1, 2, 3, and 9. Subject 3 showed a significant fourth harmonic ANN/SUM response amplitude (0.39 μV) above noise floor at MCL. Except for subject 8, for all other analyzed harmonics (harmonics 1–6) and subjects, the recorded CM/DIF and ANN/SUM responses were below 0.2 μV at MCL and therefore below or just at noise floor.

In contrast to the other subjects, the hearing threshold of subject 8 at 250 Hz was in a normal hearing range (hearing loss of 5 dB HL). Subject 8 showed significant CM/DIF amplitudes at the third harmonic and ANN/SUM amplitudes at the fourth harmonic for stimulation levels above 24 dB HL. Due to the high hearing thresholds of most study participants, subject 8 was the only one that could be tested at 5 dB above MCL to limit the risk of exceeding the uncomfortable level. The recorded response was compressed or saturated for acoustic levels above MCL (Fig. 6, ID 8). The presence of amplitude responses at additional harmonics could be an indication for saturated or compressed CM/DIF or ANN/SUM responses.

As expected, the error in the recordings was additive to the response amplitudes in the linear scale. The error stayed approximately constant when the CM response amplitude increased. Consequently, in the logarithmic scale, the error increases with lower CM amplitudes. The relation between the error in dB (⁠$errdB)$⁠, the error in μV (⁠$errμV)$⁠, and CM/DIF amplitude in μV (⁠$CM/DIFampl.$⁠) is given in Eq. (1):

From this analysis, it seems that the use of an exponential model fit on a linear scale is more suitable, whereas a fit on the logarithmic scale would require further consideration of the amplitude dependent error.

Because the first harmonic in the CM/DIF response was more prominent than the ANN/SUM responses and the ANN/SUM responses were low or absent, the first harmonic of the CM/DIFs was used for further analyses. An exponential model fit was used to characterize the CM/DIF amplitude growth (⁠$CM/DIFampl.$⁠) as a function of the acoustic stimulation level (⁠$stimLevel$⁠) in dB:

where $a$ and $b$ are parameters to be fitted by the model to the data using the trust-region-reflective algorithm implemented in matlab 2019a (Mathworks, Natick, MA). Figure 7 presents the results of the exponential fit, the recorded CM/DIF (first harmonic) responses, the psychoacoustic T-levels and MCLs as well as the R-square values for each subject. Except for subjects 4 and 7, the R-square of the exponential fit was larger than 0.99. Subjects 4 and 7 showed a lower R-square of 0.82 and 0.21 which could be explained by their overall lower CM/DIF response amplitudes.

The mean CM/DIF amplitude in response to an acoustic stimulus presented at psychoacoustic threshold level was 1.37 μV. This value, estimated from the exponential model fits, was significantly above noise floor (p = 0.02, paired-samples t-test 1-tailed). The estimated CM/DIF threshold was 17.5 dB lower than the psychoacoustic threshold.

Eight of ten subjects showed significant CM/DIF amplitudes at psychoacoustic threshold level. These responses were all above the 99% confidence interval with respect to the noise floor. The recorded CM/DIF amplitude and the 99% confidence interval of the noise floor were 0.23 and 0.28 μV for subject 2 and 0.14 and 0.19 μV for subject 4. As described above, subject 4 did not show significant CM/DIF amplitudes for acoustic stimulation at MCL.

R-square values close to 1 demonstrated that exponential models were well suited to approximate the individual CM/DIF AGFs. However, for high stimulation levels, CM amplitudes seem to saturate and the model fits may become biased. Subjects 1, 2, 3, 8, and 9 showed significant third harmonics which may be caused by the saturation (non-linearity) effect of the first harmonics at higher stimulation levels.

Figure 8 shows the CM/DIF AGFs for all subjects. The acoustic probe levels were related to the subject's psychoacoustic threshold level (THL_p) and the recorded CM/DIF amplitudes were also related to the CM/DIF amplitude recorded at the subject's threshold level (THL_cm).

The CM/DIF amplitude growth in Fig. 8 shows a linear growth with increasing acoustic probe level for subjects 1, 2, 3, 5, 6, 8, 9, and 10. These subjects showed prominent absolute CM/DIF amplitudes in μV. Because the R-squares of the exponential fits, the CM/DIF AGFs of subjects 4 and 7 were lower in comparison to the other subjects, no linear relation between acoustic probe level and CM/DIF amplitude could be found in logarithmic scale. In Fig. 8, it can be observed that the first harmonic of the CM/DIF response saturated for stimulation levels above 20 dB.

Psychoacoustic masking was determined for seven subjects using the selected frequencies as acoustic probes (Table I) and the most apical electrode to present the electrical maskers. Table II shows the result of the psychoacoustic masking experiment including the threshold elevation in dB as well as the T-levels and the MCLs in dB SPL are given in Table II.

Figure 9 shows the CM/DIF amplitudes of the first harmonic in response to an acoustic tone with (masked) and without (unmasked) simultaneous electrical stimulation. The response to an acoustic tone in the presence of an electric stimulus was obtained from the E + A and the A conditions. T-Level and MCL were estimated from the psychoacoustic measurement.

The CM/DIF interaction was analyzed across all stimulation levels (CM/DIF mean), additionally acoustic stimulation levels were divided into three groups: acoustic stimulation levels between 0 and 33%DR (CM/DIF low), between 30 and 67%DR (CM/DIF mid) and between 67 and 100%DR (CM/DIF high). The number of measurements in each group was N = 7 (corresponding to the number of tested subjects). For each group, the mean electrophysiologic and psychoacoustic interaction across subjects was computed as shown in Fig. 10. Because ANN/SUM response amplitudes were only present at MCL, the ANN/SUM interaction is only given for this acoustic level (ANN/SUM MCL). Psychoacoustic interaction was only measured at acoustic threshold level and therefore it is only reported at threshold level.

For each condition (CM/DIF mean, CM/DIF low, CM/DIF mid, CM/DIF high, ANN/SUM MCL, and psychoacoustic) a one-sample Wilcoxon signed rank test was performed to assess if the obtained interaction was significantly different from 0 dB. Significant interaction was observed for CM/DIF mean (mean = 1.15 dB, p = 0.028, N = 7), CM/DIF low (mean = 1.52 dB, p = 0.028, N = 7), CM/DIF high (p = 0.018, mean = 1.04 dB, N = 7), and psychoacoustic interaction (mean = 2.38 dB, p = 0.18, N = 7). No significant interaction was observed for CM/DIF mid (mean = 0.89 dB, p = 0.091, N = 7) and ANN/SUM at MCL (mean = −0.01, p = 0.866, N = 7). The CM/DIF mid interaction of 0.89 dB was not significant, probably because of the inter subject variability at these levels. A related sample Wilcoxon signed rank test showed that the psychoacoustic interaction (threshold elevation) was significantly larger than the mean CM/DIF interaction (amplitude attenuation) across all stimulation levels (p = 0.043, N = 7). A related-samples Friedman's two-way analysis of variance (ANOVA) showed no significant difference between CM/DIF low, CM/DIF mid, and CM/DIF high (p = 0.565, N = 7, TS = 1.143, df = 2).

Because the impact of the additive noise is greater for lower stimulation levels in the logarithmic scale, a larger standard deviation at lower stimulation levels is expected. Indeed, the observed standard deviations were 1.43 dB for CM/DIF low, 1.16 dB for CM/DIF mid, and 0.63 dB for CM/DIF high. These results can be confirmed by applying a linear regression model to psychoacoustic and electrophysiological EAS interaction measures. Figures 11(A)–11(C) show the correlation between psychoacoustic and electrophysiological EAS interaction for the CM/DIF high, mid, and low conditions. For CM/DIF high the linear regression showed an R-square of 0.84 with p = 0.004. For CM/DIF mid, the linear regression showed an R-square of 0.02 with p = 0.76. For CM/DIF low, the linear regression showed an R-square of 0.51 with p = 0.07. In summary, this analysis shows that a significant correlation between psychoacoustic and electrophysiological interaction could only be observed for the CM/DIF high condition.

In the presented study, the amplitude growth of CM/DIF and ANN/SUM responses derived from intracochlear ECochG recordings in response to acoustic stimulation were analyzed in ten EAS users. Significant CM/DIF responses for acoustic stimulation at MCL could be recorded in nine out of ten subjects and in eight out of ten subjects for ANN/SUM responses. The mean CM/DIF response at MCL was 8.76 μV and the mean ANN/SUM response at MCL was 0.98 μV. Moreover, significant CM/DIF amplitudes at psychoacoustic threshold could be recorded in eight out of ten subjects. However, ANN/SUM responses could not be measured at psychoacoustic threshold level. A compression of CM/DIF responses for high acoustic stimulation levels was observed as saturation of the amplitude growth (Fig. 8) as well as by the increase of additional harmonics in the spectral analysis (Fig. 6).

The CM/DIF AGF was characterized through exponential models. These models were used to estimate the CM/DIF amplitude corresponding with the psychoacoustic threshold level. The mean CM/DIF threshold was 17.5 dB lower than the psychoacoustic detection threshold. No significant ANN/SUM amplitude responses could be observed at psychoacoustic detection thresholds.

For seven EAS users, the CM/DIF amplitude growth was used to objectively characterize the interaction between electric and acoustic stimulation for various acoustic stimulation levels. Additionally, electrophysiologic and psychoacoustic EAS interactions were compared to each other. EAS interaction could be shown using both psychoacoustic measurements and intracochlear ECochG recordings. A mean threshold elevation of 2.38 dB was observed from the psychoacoustic masking experiment. The mean CM/DIF interaction determined by ECochG recordings was 1.15 dB and thus less sensitive than the psychoacoustic measurement (related sample Wilcoxon signed rank test: p = 0.043). No significant difference between the CM/DIF low, CM/DIF mid, and CM/DIF high interaction could be observed, indicating that the level of the acoustic stimulus has only a minor effect on measurable CM/DIF interaction. Under this assumption, acoustic stimulation levels close to MCL appear best suited to evaluate psychoacoustic interaction using CM/DIF recordings because the resulting CM/DIF response amplitudes increased proportionally to the acoustic stimulation level, which leads to decreased impact of noise [Eq. (1)]. This could be confirmed analyzing the correlation between psychoacoustic interaction and CM/DIF interaction, which was largest for acoustic stimulation levels close to MCL (Fig. 11).

In previous studies, CM/DIF responses were used to estimate post-operative hearing threshold and to monitor cochlear functionality during electrode insertion (e.g., Koka et al., 2017b). In these previous studies, the CM/DIF threshold was typically defined as the stimulation level where the recorded amplitude falls below a certain value. This value could be estimated as the first visually detectable amplitude peak or it could be set to the estimated system noise floor. Using the proposed exponential model fit to the CM/DIF AGFs allows the prediction of thresholds below this noise floor. However, in Campbell et al. (2015), the measured CM/DIF thresholds were always below or at the same level as the psychoacoustic thresholds. Abbas et al. (2017) observed strong correlation between CM/DIF and ANN/SUM thresholds and behavioral thresholds. Some of the CM/DIF and ANN/SUM thresholds were below the behavioral thresholds. Similar observations were made by Fontenot et al. (2019), who found a significant correlation between ECochG thresholds and audiometric thresholds. However, they observed that many ECochG thresholds were lower than the audiometric thresholds. This supports the results presented in the current study; CM/DIF threshold measurements showed a mean CM/DIF threshold, which was 17.5 dB lower than the psychoacoustic threshold. Note that, in contrast to Abbas et al. (2017) and Fontenot et al. (2019), who compared physiological thresholds to behavioral thresholds determined from a typical clinical procedure, in the current study the psychoacoustic thresholds were measured using a 3-IFC procedure with the same stimulation system as the one used for ECochG recording. An explanation for the difference between these physiological and behavioral thresholds is that CM/DIF responses are probably only indirectly connected to audiometric thresholds since audiometric thresholds require intact connection of inner hair cells to the auditory nerve fibers and the CM/DIF responses are mainly generated by the outer hair cells. Fontenot et al. (2019) excluded a causal relationship between ECochG thresholds and audiometric thresholds after observing that ECochG thresholds lay below audiometric thresholds. Fontenot et al. (2019) argued that evoked potentials require the summed synchronous activity of numerous responding elements, while behavioral thresholds may require relatively weak activity of a small number of sensory neurons. This argument is supported by animal studies that also found differences between physiological thresholds obtained from CMs and thresholds obtained from CAP responses (e.g., Choudhury et al., 2011).

The method used to analyze the ECochG could affect the magnitude of the CM/DIF amplitude. In the current study, the first harmonic was obtained computing the difference response of two ECochG recordings to acoustic stimuli with alternating polarities and it was defined as the CM/DIF response. This is a common method used in ECochG analysis and is applied in the clinical routine of the Hannover Medical School for intraoperative monitoring. However, other methods have been proposed by Fitzpatrick et al. (2014) and McClellan et al. (2014) to estimate the CM/DIF response amplitudes based on the summation of the first harmonics, this method was termed total responses. As shown in Fig. 6, the CM/DIF responses were most present at the first harmonic and therefore, the use of total responses would not significantly change the CM/DIF amplitudes in the tested population of CI users.

For the tested subjects we observed a linear relation between acoustic stimulation levels and CM/DIF amplitudes in dB. For high stimulation levels, saturation in the AGF of the first harmonic of the CM/DIF responses was observed (Fig. 8). This saturation effect (non-linearity) is supported by the presence of second or third harmonics for subjects 1, 2, 3, 8, and 9 (Fig. 6). The linear relation between acoustic stimulation levels and CM/DIF amplitudes in dB holds the potential to estimate the CM/DIF threshold from a single recording by linearly extrapolating the threshold from any CM/DIF response amplitude. However, particularly subject 8 demonstrates that in case of good residual hearing, the saturation of CM/DIF amplitude responses can already occur at low stimulation levels (24 dB HL). In this case, the relation between harmonics could be used to compensate for the saturation to estimate CM/DIF threshold using a single acoustic stimulation level. In Fontenot et al. (2017) and Choi et al. (2004), second order Boltzmann functions were used to model the CM amplitude saturation. These models may be used to estimate the CM/DIF saturation from ECochG recordings at a single acoustic stimulation level. Alternatively, the recording of the AGF would be the best method to determine CM/DIF thresholds. However, the recording at multiple stimulation levels is time consuming and so far not applicable in the clinical routine for objective threshold estimation or for intraoperative response monitoring during surgery.

In the current study, we used the CM/DIF amplitude growth to determine electrophysiological EAS interaction at various acoustic stimulation levels. The feasibility of measuring EAS interaction via intracochlear ECochG using the CI was shown by Koka and Litvak (2017) and by Krüger et al. (2020). These works demonstrated electrophysiological EAS interaction if acoustic and electric stimuli were presented simultaneously at MCL. However, they observed less electrophysiological EAS interaction in comparison to psychoacoustic EAS interaction measured for acoustic stimuli at threshold level and electrical stimuli presented at MCL. Koka and Litvak (2017) and Krüger et al. (2020) suggested that the effect of electrical stimulation could be greater for acoustic stimuli presented at threshold level than at MCL. Findings from animal studies by Nourski et al. (2007), Stronks et al. (2010), and Stronks et al. (2011) showed that the effect of EAS interaction became lower with increased acoustic stimulation level in CAP responses. In the current study, the ECochG amplitude growth was measured in the presence and in the absence of electrical stimulation and the EAS interaction was analyzed at various acoustic stimulation levels. The results showed no significant difference between electrophysiological EAS interaction at CM/DIF low, CM/DIF mid, or CM/DIF high levels. A possible explanation for the differences between electrophysiological and psychoacoustic EAS interaction could be that psychoacoustic EAS interaction consists of peripheral and central interaction, whereas the ECochG is a measurement technique that captures only the peripheral interaction. These results may indicate that the origin of psychoacoustic EAS interaction is not only at the level of CMs. However, accurate measurements of ANN/SUM EAS interaction were not possible in this study. Therefore, it cannot be excluded that EAS interaction also occurs at the level of the auditory nerve through electro-neural interaction.

Koka and Litvak (2017) observed no significant correlation between psychoacoustic interaction and interaction estimated from intracochlear CM/DIF responses. In contrast, the present study shows a significant correlation between psychoacoustic and CM/DIF interaction at high stimulation levels. However, no significant correlation to CM/DIF low or CM/DIF mid was observed. The mean interaction was largest for low stimulation levels. Because the error of the responses increased for lower amplitude recordings in the logarithmic scale, it is possible that the precision of electrophysiological EAS interaction was higher at high acoustic stimulation levels while the sensitivity was higher at low acoustic stimulation levels.

In ten EAS users, the CIs were used to measure intracochlear ECochG in response to acoustic stimulation presented at various stimulation levels. Derived CM/DIF responses were characterized regarding their amplitude growth and their harmonic components. CM/DIF responses at psychoacoustic threshold level were estimated by exponential models fitted to the CM/DIF amplitude growth. Thereby, the presence of CM/DIF responses at and below psychoacoustic threshold level could be demonstrated. Saturation of the first harmonic was observed at high stimulation levels indicated by CM/DIF amplitude growth and the presence of additional harmonics.

The ECochG was used to measure electrophysiological EAS interaction at various acoustic stimulation levels. The results suggest a constant effect of electric stimulation to CM/DIF responses independent of the acoustic stimulus level. In comparison to psychoacoustic EAS interaction, the electrophysiological EAS interaction was less sensitive, which could indicate that the origin of psychoacoustic EAS interaction is not only occurring at the level of CMs. However, EAS interaction could be observed using both psychoacoustic and electrophysiological measurements.

This work was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy, EXC 2177/1, Project ID 390895286, and by the DFG, Project number: 396932747 (PI: W.N.). Without the subjects who participated in the experiments, this work would have not been possible; therefore, the authors would like to thank the subjects for their invaluable contribution.

Electrode insertion angle and cochlear coverage were determined from CBCT images. Insertion angles were determined using post-op CBCT scans. The insertion angle is given by the posterior margin of the round window, the modiolus, and the target electrode contact. If the electrode contact was visually detectable, the artifact center was used as electrode position. Otherwise, the most apical electrode contact was marked. Starting from that position, a spline was fitted through the electrodes array. With the known electrode spacing, further electrode angles were determined.

The cochlear length was determined using pre-op CBCT scans. The cochlear length was defined as the distance between the posterior margin of the round window and the apex along the lateral wall. The insertion angles were transferred to the pre-op CBCT scans and were converted to insertion depth in mm or cochlear coverage in %, respectively. Equation (A1) defines the cochlear coverage in % as a function of the electrode's insertion depth in mm and the cochlear length in mm:

Figure 12 shows the schematic used reference points.

Figure 13 shows the insertion angle, the insertion depth in mm, the cochlear coverage in % as a function of the electrode number, and the cochlear coverage in % as a function of insertion angle.