Electrocochleography-based Tonotopic Map: II. Frequency-to-Place Mismatch Impacts Speech Perception Outcomes in Cochlear Implant Recipients

Abstract

Objective:

Modern cochlear implants (CIs) use varying length electrode arrays inserted at varying insertion angles within variably sized cochleae. Thus, there exists an opportunity to enhance CI performance, particularly in post-linguistic adults, by optimizing the frequency-to-place allocation for electrical stimulation, thereby minimizing the need for central adaptation and plasticity. There has been interest in applying Greenwood or Stakhovskaya et al function (describing the tonotopic map) to postoperative imaging of electrodes to improve frequency allocation and place coding. Acoustically-evoked electrocochleography (ECochG) allows for electrophysiologic best-frequency (BF) determination of CI electrodes and the potential for creating a personalized frequency allocation function. The objective of this study was to investigate the correlation between early speech perception performance and frequency-to-place mismatch.

Design:

This retrospective study included fifty patients who received a slim perimodiolar electrode array. Post-implantation, five acoustic pure-tone stimuli ranging from 0.25–2 kHz were presented, and electrophysiological measurements were collected across all 22 electrode contacts. Cochlear microphonic tuning curves were subsequently generated for each stimulus frequency to ascertain the BF electrode or the location corresponding to the maximum response amplitude. Subsequently, we calculated the difference between the stimulus frequency and the patient’s CI map’s actual frequency allocation at each BF electrode, reflecting the frequency-to-place mismatch. Best-frequency electrocochleography-total response (BF-ECochG-TR), a measure of cochlear health, was also evaluated for each subject to control for the known impact of this measure on performance.

Results:

Our findings showed a moderate correlation (r = 0.51; 95% CI: 0.23 to 0.76) between the cumulative frequency-to-place mismatch, as determined using the ECochG-derived BF map (utilizing 500, 1000, and 2000 Hz), and 3-month performance on CNC words (N = 38). Larger positive mismatches, shifted basal from the BF map, led to enhanced speech perception. Incorporating BF-ECochG-TR, total mismatch, and their interaction in a multivariate model explained 62% of the variance in CNC word scores at 3-months. BF-ECochG-TR as a standalone predictor tended to overestimate performance for subjects with larger negative total mismatches and underestimated the performance for those with larger positive total mismatches. Neither cochlear diameter, number of cochlear turns, nor apical insertion angle accounted for the variability in total mismatch.

Conclusions:

Comparison of ECochG-BF derived tonotopic electrode maps to the frequency allocation tables reveals substantial mismatch, explaining 26.0% of the variability in CI performance in quiet. Closer examination of the mismatch shows that basally shifted maps at high frequencies demonstrate superior performance at 3-months compared to those with apically shifted maps (towards Greenwood and Stakhovskaya et al). The implications of these results suggest that electrophysiological-based frequency reallocation might lead to enhanced speech perception performance, especially when compared to conventional manufacturer maps or anatomic-based mapping strategies. Future research, exploring the prospective use of ECochG-based mapping techniques for frequency allocation is underway.

Introduction

Cochlear implants (CIs) are the standard of care for individuals with severe-to-profound hearing loss, who gain little benefit from conventional hearing aids (Buchman et al. 2020; Eshraghi et al. 2012; Lenarz 2017; Zwolan et al. 2020). These devices transform auditory signals into precise electrical pulses, delivered directly to the cochlea’s spiral ganglion and related dendrites (O. Adunka et al. 2005). The encoding of these signals is largely dependent on the cochlea’s tonotopic organization, assigning low-frequency sounds to the apical end and high-frequency sounds to the base of the electrode array.

Manufacturers of cochlear implants distribute frequencies across a fixed-length electrode array in a logarithmic fashion within their Frequency Allocation Tables (FATs). This approach, dictated by electrode constraints, optimizes frequency representation but does not precisely mimic the cochlea’s natural tonotopic organization. This system assigns each electrode contact a predetermined frequency band, encoding the received sounds (Peters et al. 2016). While Greenwood’s function describes the tonotopic organization of the cochlea based on the organ of Corti (Greenwood 1990), several researchers have pointed out the CI electrode array predominantly stimulates the spiral ganglion cells (Boëx et al. 2006; Dorman et al. 2007). Hence, Stakhovskaya et al. proposed a potentially more suitable frequency-place calculation for CIs, based on a histological analysis of the cochlea’s spiral ganglion cell map (Stakhovskaya et al. 2007). While Greenwood’s function and Stakhovskaya et al. application are based on normal hearing representation of frequency at threshold-level stimulation, this may not fully represent the typical CI recipient with acquired hearing loss. These recipients often experience auditory system reorganization, characterized by broader frequency tuning and higher levels needed for auditory perception due to severe to profound hearing loss (Dubno et al. 1992).

Despite the methods used to assign frequency band limits, the main source of bias in pre-set FATs stems from significant inter-individual variations in cochlear size (Erixon et al. 2013; Ketterer et al. 2018; van der Marel et al. 2014) and shape (Escudé et al. 2006; Meng et al. 2016). Variations also occur due to differences in electrode arrays in terms of length, caliber, stiffness, tip, and pre-curved shape, leading to highly variable insertion depths and positions inside the scala tympani, even when inserted by the same surgeon in the same manner. Prior research has highlighted a mismatch between default FATs for individual electrode contacts and the corresponding spiral ganglion frequency estimate across numerous arrays (Landsberger et al. 2015; Venail et al. 2015; Wess et al. 2017).

To precisely define the frequency-to-place mismatch and enhance CI outcomes, high-resolution computed tomography (HRCT) has been utilized (Kennedy et al. 2016; Piergallini et al. 2018). HRCT allows for the detection of sub-millimetric deviations in array insertion, enabling the localization and assignment of each electrode according to the Greenwood or Stakhovskaya frequency allocation predictions. The impact of Greenwood or Stakhovskaya based approaches to frequency allocation remains to be determined. Multiple studies suggest that a significant mismatch between the Greenwood or Stakhovskaya et al frequency map and the CI FAT may affect initial hearing outcomes in post-lingual CI recipients (Canfarotta et al. 2020; Fu et al. 2002; Venail et al. 2015). Despite these findings, prospective clinical trials attempting to fit patients to these anatomic-based maps have shown mixed results as it relates to CI performance (Creff et al. 2024; Di Maro et al. 2022; Dillon, Canfarotta, et al. 2023; Dillon, Helpard, et al. 2023; Fan et al. 2023; Gifford et al. 2022; Kurz et al. 2023; Kurz et al. 2022; Lambriks et al. 2023).

The frequency-position function, imposed by CIs, typically diverges from the user’s inherent cochlear tonotopy, necessitating adaptation to an altered auditory map. Research indicates variability in this adaptation process across individuals, with some subjects struggling to align their perception of electrode pitches with the CIs imposed mapping (Svirsky et al. 2015; Tan et al. 2017). This variability may contribute to a ceiling effect, where significant mismatches between the CI mapping and cochlear tonotopy limit the maximal achievable speech recognition improvements, highlighting the importance of minimizing such mismatches. This disparity underscores the need for CI programming to closely align with each user’s unique auditory characteristics, tailoring the frequency-position function to better match their residual auditory capabilities.

This disparity in perception and the inherent challenges in aligning CI programming with individual auditory characteristics highlight the necessity for more personalized approaches in CI technology. Recently, our team explored intracochlear electrocochleography (ECochG) as a potential tool to elucidate the in vivo tonotopic map in humans (Figure 2 in Walia et al. 2024). By utilizing ECochG, we measured acoustically-evoked responses following electrode array implantation, identifying the electrode that displays the largest response amplitude to a specific frequency, indicative of its ‘best frequency’ (BF). Importantly, both the Greenwood and Stakhovskaya et al frequency-position functions were formulated based on tonotopic measures derived at threshold levels of stimulation (Greenwood 1990; Stakhovskaya et al. 2007). Our ECochG research highlighted a significant basal or frequency-down shifted divergence between the tonotopic map operating at everyday conversational levels (~70 dB HL) and the conventional Greenwood map—derived at near-threshold levels (Figure 5 in Walia et al. 2024). This pivotal finding suggests that the ECochG-derived tonotopic map may be a more relevant target than Greenwood/Stakhovskaya et al for exploring frequency-to-place mismatch, offering strategies to enhance CI outcomes. No prior study has evaluated the mismatch between an ECochG-derived tonotopic map and CI performance.

In this study, our primary objective was to establish the extent of frequency-to-place mismatch using ECochG in patients implanted with perimodiolar electrode arrays and to assess its impact on CI speech performance. Our method involved measuring the anatomic position of all electrode contacts (expressed in degrees from the round window) using postoperative HRCT and 3-D reconstruction. ECochG derived BF electrodes for (250 to 2000 Hz stimuli) were localized and compared to the FAT according to the manufacturer’s software as mapped by the treating audiologist. Ultimately, we correlated the frequency-to-place mismatch between the ECochG-derived tonotopic map and the FAT with speech performance three months post-implantation. Additionally, as a secondary objective, we aim to compare the mismatches identified through ECochG-derived maps with those calculated based on the Greenwood/Stakhovskaya et al frequency-position functions.

Materials and Methods

Participants

This retrospective study, approved by the Institutional Review Board at the study institution (IRB #202007087), involved adult candidates for cochlear implantation exhibiting low-frequency hearing prior to surgery, confirmed by a low-frequency pure-tone average (LFPTA) of 125, 250, and 500 Hz ≤ 60 dB HL. Exclusions included candidates with middle ear pathology, those undergoing revision surgery, individuals with prelingual deafness, those lacking a patent external auditory canal (due to the air conduction delivery of the acoustic stimulus), non-English speakers, and individuals unable to provide informed consent.

Each participant was implanted with a slim pre-curved, perimodiolar electrode array (CI632) and provided with the N7 processor from Cochlear Ltd, Sydney, Australia. Following the surgery, participants underwent standard clinical protocols for cochlear implant activation and device settings optimization under the guidance of their clinical audiologists. While 19 subjects with preserved residual hearing (LFPTA ≤ 80 dB HL) were candidates for electroacoustic stimulation (EAS), only 2 used hybrid maps (O. F. Adunka et al. 2018). The remaining subjects used full frequency maps. This study primarily aimed to analyze the frequency allocations along the electrode, thus only the Frequency Allocation Tables (FAT) were extracted from the final programmed settings. The default setting (i.e., default FAT) according to the manufacturer’s software was maintained for the FAT in 48% of the participants, whereas in 52% of the cases (N = 26), specific electrodes, either in the apex or the base, were deactivated to enhance sound quality by the audiologist. Any subsequent adjustments to the frequency bandwidths were automatically managed by the manufacturer’s software, without manual intervention by the audiologist.

Surgical Technique

Three surgeons implanted all 50 electrode arrays using a standard transmastoid, transfacial recess (posterior tympanotomy) approach to access the cochlea. This process involved partial removal of the round window niche overhang. Depending on the orientation of the round window membrane, the array was inserted either via a round window incision or through a marginal cochlear opening created specifically for this purpose. The electrode was then advanced through the sheath following the manufacturer’s recommendations, ensuring a complete insertion as determined by the insertional (proximal) stop point on the electrode lead. Once fully inserted, the sheath was disengaged, and the cochleostomy was sealed using a small piece of previously harvested tissue. To confirm the functionality of the device, intraoperative impedances and electrically evoked compound action potentials were checked and found to be within normal ranges for all subjects. An intraoperative X-ray was used to check for tip rollover or kinks. When tip rollover was identified, the electrode array was removed, reloaded into the sheath, and re-inserted.

Default Frequency Allocation

CI programming involves a default frequency allocation process, which distributes frequencies logarithmically across a set-length electrode array, an approach that optimizes frequency representation within the constraints of the implant rather than emulating the cochlea’s natural tonotopic organization. These tables vary across different CI device manufacturers. For instance, the Cochlear (Sydney, Australia) electrode array strategy specifies center frequencies for electrodes 22 to 1, respectively, as follows: 250, 375, 500, 625, 750, 875, 1000, 1125, 1250, 1438, 1688, 1938, 2188, 2500, 2875, 3313, 3813, 4375, 5000, 5688, 6500, 7438 Hz. If electrodes are deactivated, then the frequencies are reallocated to the remaining electrodes by the manufacturer’s software.

Electrocochleography-based Derivation of Frequency Allocation for Electrodes

Our previous work has shown how the in vivo tonotopic map was derived for our human subjects using the CI electrode array and acoustically-evoked electrophysiologic responses (Figure 1 in Walia et al. 2024). Briefly, we used an ER3–14A insert earphone (Etymotic, Elk Grove Village, IL, United States) inserted in the external auditory canal, prior to sterilizing the surgical site. Following implantation of the electrode array into the cochlea, we ensured proper alignment of a telemetry coil with the CI antennae for direct device connectivity. ECochG potentials were measured using the Cochlear Research Platform Ver 1.2. Each of the 22 electrodes in the array were subsequently conditioned. Next, we administered tone burst stimuli across frequencies ranging from 250 Hz to 2000 Hz in both rarefaction and condensation phases. Electrophysiological responses were recorded from all electrodes, with stimulus intensities - set within the speaker’s maximum output capacity - ranging from 98 to 108 dB SPL for the different frequencies. This stimulus results in responses similar those obtained at louder conversational levels (~70 dB HL) (Figure 5 in Walia et al. 2024). Data analysis was performed offline utilizing MATLAB R2020a (MathWorks Corp., Natick, MA, United States). From the recorded responses, we derived a difference curve, and the ongoing portion of this curve was selected for Fast Fourier Transformation. This procedure facilitated the determination of response amplitudes to the varying stimulus frequencies and the generation of cochlear microphonic tuning curves. Significant responses were classified as those with magnitudes exceeding the noise floor by three standard deviations. We constructed a tonotopic map for all subjects and the BF electrode was being defined as the electrode that had the maximum amplitude response to a given frequency stimulus. The 250 Hz BF electrode could not be determined since the maximum response for this frequency was always electrode 22. The consistent maximum response for the 250 Hz stimulus at the apical-most electrode indicates that the pre-curved electrode array does not reach the BF location for this frequency.

Assessment of Frequency-to-Place Mismatch

The magnitude of the frequency-to-place mismatch was quantified using the semitone. For each patient, we compared the ECochG-determined best frequency (BF) values (500 Hz, 1000 Hz, and 2000 Hz) with the corresponding frequencies allocated by the manufacturer’s Frequency Allocation Table (FAT) (default or adjusted if electrodes deactivated) for each BF electrode. The difference between these frequencies was calculated using a log2 scale and then converted into semitones, facilitating a precise quantification of mismatch. In cases where electrodes were deactivated or hybrid maps utilized, adjusted center frequencies were employed. We calculated the mismatch (difference between BF electrode frequency and FAT frequency in semitones) individually for each BF electrode and also summed the deviations across the three stimuli to gauge overall mismatch. For our post-hoc analysis, we excluded subjects (N = 12) if CNC scores were not available at 3 months post-implantation and only used those ECochG-BF electrodes where responses were available (Supplementary Table 1).

Anatomy-based Derivation of Frequency Allocation for Electrodes

All patients underwent HRCT imaging after implantation. In our previous study (Figure 2 in Walia et al. 2024), the technique of angular insertion depth (AID) derivation for each electrode contact was detailed using reconstructed postoperative CT. To overcome the “bloom” effect, which complicates the interpretation of images from platinum iridium contacts, we employed a validated technique for accurate localization of the implanted electrodes within the cochlea (Holden et al. 2013; Skinner et al. 2007; Teymouri et al. 2011). This involved co-registering pre- and post-implant CT images to segment the electrodes and create a composite image. To visualize the scalar position, we aligned the composite CT volume with a high-resolution micro-CT cochlear atlas, which helped infer the location of soft tissue structures. The position of each electrode was determined by viewing the composite CT volume along the mid-modiolar axis, with the round window as the 0° start point. The angular position of each electrode and the number of turns was calculated based on rotation around the axis. This manual segmentation took approximately 30 minutes per subject.

For the purposes of making comparisons, a modified function, derived from synchrotron radiation phase-contrast imaging of cadaveric specimens (Helpard et al. 2021; Li et al. 2021), provided an anatomical reference for extrapolating cochlear frequency mapping in living subjects. We utilized this reference to estimate the Greenwood frequency location at the location of the BF electrode thereby enabling a direct comparison of mismatch (Greenwood frequency minus FAT frequency in semitones). The absolute semitone difference was taken as the total sum mismatch across 500, 1000, and 2000 Hz.

Best Frequency Electrocochleography-Total Response (BF-ECochG-TR)

Electrocochleography-total response (TR) measured at the round window, on the promontory, and recently at the BF electrode have all been shown to exhibit strong correlation with CNC performance post-activation (Fitzpatrick et al. 2014; Fontenot et al. 2019; Walia, Shew, Kallogjeri, et al. 2022; Walia, Shew, Lee, et al. 2022). Using the BF electrode (250, 500, 1000 and 2000 Hz) as described above, the cumulative sum of harmonics was calculated for each frequency. The sum of these measures is totaled across all frequencies (250 Hz, 500 Hz, 1000 Hz, and 2000 Hz) to determine the aggregate best-frequency electrocochleography-total response (BF-ECochG-TR). More details regarding validation of BF-ECochG-TR are available in (Walia et al. 2023).

Cochlear Implant Activation and Device Settings

All subjects received CI programming and auditory rehabilitation services from skilled audiologists at our study institution, following established clinical protocols for adult CI recipients. These include adjusting the minimum and maximum electrical stimulation levels for each electrode, and fine-tuning parameters such as gain, frequency allocation, input dynamic range, and the activation of electrodes. The primary goal of this was to ensure that soft sounds were audible, conversational speech was clear and at comfortable volume, and loud sounds remained within tolerable limits for each individual. All patients used the same processing strategy (ACE), and rate was not adjusted in any of the subjects. T levels were measured at a minimum of five electrodes, with intervening values interpolated, and C levels were determined to establish the maximum comfortable levels and then applying interpolation across the entire electrode array. Sound-field threshold levels, especially for frequencies ranging from 250 to 6000 Hz, were carefully monitored, targeting thresholds below 30 dB HL to guarantee audibility of soft speech and environmental sounds. Additionally, auditory training exercises and communication strategies for challenging listening situations, including telephone use, were integral components of the rehabilitation process, tailored to facilitate optimal device use in daily life.

Postoperative Speech Perception Testing

Three months post-activation, speech perception of the CI-alone ear was measured using Consonant-Nucleus-Consonant (CNC) word test at 60 dB SPL. This helped assess how the frequency-to-place mismatch might influence initial speech perception. Testing was conducted in a soundproof booth at 1 meter from the sound source. The better ear was plugged and muffed. The percentage of words correctly repeated from a 50-word list was calculated.

Statistical Analysis

The D’Agostino-Pearson omnibus test was used to check the normality of continuous variables. Paired t-test was used to compare the frequencies across the three different maps (default FAT, SG, and OC). The relationship between the AID of the most apical electrode and cochlear diameter was evaluated using Pearson correlation. Linear regression was used to assess the association between speech perception and the degree of mismatch. Multiple linear regression analyses were used to assess how the combination of variables could be used to predict performance in quiet. The independent explanatory variables selected for the multivariate regression had p-values < 0.05 in the simple regression between CI performance and the respective independent variables. The final multivariate model was constructed using a forward selection stepwise regression method, with each parameter in the final model significantly contributing to the increase in the r² value (p < 0.05). The statistical analyses were performed using SPSS version 27 for Windows (IBM Corp, Armonk, New York).

Results:

Participant Demographics, Surgical Techniques, and Postoperative Imaging Analysis

The cohort consisted of 50 participants (19 males and 31 females) with a mean age of 69.6 ± 18.3 years (Table 1). All participants were implanted with the CI632 electrode array. Postoperative CT scans with 3-D reconstructions confirmed correct placement of the array in the scala tympani for all subjects. The most apical electrode demonstrated a broad range of angular insertion depths (AIDs), from 345 to 475 degrees, averaging at 404.3 ± 35.3 degrees. Cochlear diameter averaged at 9.2 ± 0.4 mm. A negative correlation between cochlear diameter and the apical insertion angle (r = −0.45, 95% CI: −0.70 to −0.12; p=0.01) indicated that smaller cochleae tended to have deeper insertions (Figure 1A). Additionally, the most basal electrode showed wide variability in its angular insertion depth, ranging from 3 to 49 degrees (22.9 ± 14.4 degrees) (Figure 1B), implying both surgical and anatomic variability.

Table 1.

Demographic and audiologic information of subjects reviewed for electrocochleography-derived mismatch.

	Mean ± STD or N (%)	Poor Performers (CNC at 3-months <30%; N=12)	Excellent Performers (CNC at 3-months >60%; N=12)
Age (yrs)	69.6 ± 18.3	63.6 ± 20.9	72.3 ± 14.5
Gender
Female	31 (62.0)	5 (41.7)	6 (50.0)
Male	19 (38.0)	7 (58.3)	6 (50.0)
Duration of hearing loss (yrs)	29.0 ± 18.6 (range, 0.5 – 74)	24.0 ± 16.3	27.0 ± 8.5
Duration of severe-to-profound hearing loss (yrs)	8.6 ± 10.4 (range, 0.5 – 38)	6.4 ± 11.3	9.3 ± 8.4
Etiology of hearing loss
Unknown	43 (86.0)	9	11
Sudden sensorineural hearing loss	2 (4.0)	2	0
Congenital	5 (10.0)	1	1
Preoperative Low-frequency Pure Tone Average (LFPTA; 125, 250, 500 Hz; dB HL)	56.3 ± 25.5	53.4 ± 9.3	52.1 ± 6.8
Preoperative Pure Tone Average (PTA; 500, 1000, 2000, 4000 Hz; dB HL)	56.1 ± 20.8	62.3 ± 12.4	65.8 ± 9.5

Figure 1. Assessment of Cochlear Size and its Impact on Insertion Angle.

A, Illustrates a moderate linear correlation between apical insertion angle and cochlear diameter for the perimodiolar electrode, with deeper insertions observed in smaller cochleae. B, Displays no significant correlation between cochlear size and insertion angle of the basal-most electrode.

Frequency-to-Place Mismatch: Default FAT & Greenwood/Stakhovskaya et al Functions

Figure 2A illustrates the relationship between the mean AID of the 22 electrode contacts for the CI632 and the default FAT maps, defined by the manufacturer, for all 50 patients. The actual FAT (unadjusted/default and adjusted for electrode deactivation) for each patient is shown as a function of AID in Figure 2B. The solid blue line corresponds to the SG cochlear place frequency, while the Greenwood OC maps, illustrating varying cochlear sizes, are depicted by gray dotted (2.9 turns) and solid lines (2.1 turns). As mentioned above, the 250 Hz BF electrode could not be determined since the maximum response for this frequency was always electrode 22. The consistent maximum response for the 250 Hz stimulus at the apical-most electrode likely indicates that the pre-curved electrode array does not reach the BF location for this frequency.

Figure 2. Default (A) and Actual (B) Frequency Allocation Tables Compared to Established Frequency-Position Functions.

A, This panel represents the modified frequency-position functions of the organ of Corti (based on the Greenwood function), and spiral ganglion (according to Stakhovkskaya et al and Helpard et al., 2021). Manufacturer default frequency allocation data is derived from 50 subjects who underwent postoperative computed tomography imaging for the localization of each electrode after implantation with the CI632 device. The mean insertion angle across all subjects with the corresponding default central frequency is shown (i.e., default map). There is an approximate octave difference between the default frequency allocation table (FAT) and the established frequency-position functions, which lessens at higher frequencies. B, The grey lines represent each subject’s actual FAT, as set by the audiologist, with the black dot illustrating the mean AID of the best frequency electrodes derived from electrocochleography (ECochG BF), and the whiskers indicate two standard deviations. Significant variability is observed between each subject’s FAT and the ECochG BF map, indicating a substantial mismatch between these two maps. Although the ECochG BF for 500 Hz to 4 kHz is shown, only 500, 1000, and 2000 Hz were used for the mismatch analysis as only a proportion of patients (N=18) had acoustically-evoked ECochG responses at 3–4 kHz.

In all subjects and across all electrode contacts, the central frequency allocation was consistently lower than what would be expected based on the SG and OC maps. This consistent lower frequency allocation, referred to as ‘frequency downward shift’ or a ‘basal shift’, indicates that the CI electrodes were mapped to lower frequencies compared to the cochlear frequency-place mapping as outlined by the SG and OC maps. On average, the FAT frequency for the CI632 electrodes were approximately 2/3 of an octave (8 semitones) lower than the SG-predicted frequency for AIDs within the range of 0° to 183°; an octave equates to 12 semitones in music theory. Starting at ~20°, the mismatch between the FAT and the SG-predicted frequencies begins modestly at around 3.5 semitones, but this gap steadily enlarges, reaching 14.5 semitones by 183°. For electrode contacts with AIDs exceeding 183°, the disparities relative to the tonotopic cochlear place plateaued. Notably, the average deviation between the FAT and the SG map for the most apical electrode was 15.43 semitones. A paired t-test confirmed that the FAT map was statistically different than the SG (Paired samples t test: t = 10.64; df = 46; p < 0.001) and OC map (Paired samples t test: t = 13.45; df = 46; p < 0.001).

Frequency-to-Place Mismatch: FAT & ECochG Map Evaluation

There was significant variability in both apical and basal directions between the individual FAT maps and the average (+/− 2SD) patient-specific ECochG BF maps (Figure 2B). The individual subject ECochG BF data is presented in Supplementary Table 1. For each patient, the difference between ECochG BF and the FAT assigned frequency at the same location was calculated and shown in Figure 3 as a function of CNC word score at 3-months post-activation. The correlation analysis of the ECochG BF-FAT mismatch at 500 Hz (Figure 3A) was not significant (r = 0.14; 95% confidence interval: −0.20 to 0.45; p-value = 0.41). However, when comparing mismatch at 1000 Hz (Figure 3B) and 2000 Hz (Figure 3C), significant correlations were observed (Supplementary Figure 2). Specifically, larger positive or basally shifted mismatches (relative to the ECochG BF electrode map) were associated with greater speech-perception performance (1000 Hz - r = 0.43; 95% confidence interval: 0.11 to 0.66; p-value = 0.01; 2000 Hz - r = 0.44; 95% confidence interval: 0.13 to 0.67; p-value = 0.01). Figure 4 illustrate the relationship of mismatch direction relative to the ECochG BF electrodes on excellent (CNC >60%) and poor (CNC <30%) performance at 3 months post-activation. The Mann-Whitney U-test showed no significant difference in the duration of deafness (severe-to-profound hearing loss) between the two performance cohorts (p-value = 0.73). Among the excellent performers (CNC >60%), 41.7% (5 out of 12) had 1–3 basal electrodes deactivated by their mapping audiologist. In contrast, among the cohort of poor performers (CNC <30%), 75.0% (9 out of 12) had 1–3 basal electrodes deactivated during routine clinical mapping (usually based on clinical sound quality). There was no significant correlation between AID of the most active basal electrode contact and CNC at 3 months.

Figure 3: CNC Word Scores vs Electrocochleography Frequency-to-Place Mismatch. (ECochG BF minus FAT).

This figure illustrates the relationship of frequency-to-place mismatch to CNC word scores at 3-months post-activation at different frequencies: A, Mismatch at 500 Hz. B, Mismatch at 1000 Hz. C, Mismatch at 2000 Hz. A moderate linear correlation was observed for mismatches at 1000 Hz and 2000 Hz, with a positive or basally shifted (frequency downward) mismatch corresponding to improved performance. No correlation was found between the mismatch at 500 Hz and CNC scores at 3 months post-activation. Only subjects with responses to all three frequencies are shown in this figure.

Figure 4. Improved Performance Corresponds to Basally Shifted Frequency-Position Maps.

This figure depicts the individual frequency allocation tables (FAT) of subjects, represented by grey lines. The black dots denote the mean of the best frequency (BF) derived from electrocochleography (ECochG), with whiskers representing two standard deviations. A, The red lines illustrate the FAT of subjects with a CNC score less than 30% at 3 months post-activation. These maps are apically shifted (or frequency upwards) in relation to the average default frequency allocation table. B, The green lines depict the frequency allocation maps of subjects who achieved a CNC score greater than 60% at 3 months post-activation. These maps demonstrate a basal shift, particularly at high frequencies, compared to the average default frequency allocation table. Thus, this indicates that better performance is associated with basally shifted or frequency downward frequency-position maps.

The correlation between the total mismatch at 500, 1000, and 2000 Hz (sum of mismatches across all three frequencies) and factors such as cochlear size or insertion depth were also sought. There was no correlation between cochlear diameter, the number of turns, the apical insertion angle, and the total mismatch (Supplementary Figure 1). We then explored, in the same subjects, whether the frequency-to-place mismatch using Greenwood’s frequency-position function as a target (cumulative score at 500, 1000, and 2000 Hz), and no correlation with speech-perception testing at 3-months was observed (Figure 5). To ensure robustness of our findings, a sensitivity analysis was performed to assess the impact of outliers identified using the residual plot. Removing these outliers did not significantly alter the correlation coefficient. This analysis was also applied to the frequency-to-place mismatch using Greenwood’s frequency-position function for only 1000 and 2000 Hz, where no outliers were found, further substantiating the reliability of our conclusions regarding the lack of correlation (Supplementary Figure 3).

Figure 5. Correlation Analysis of CNC Word Scores and Greenwood Frequency-to-Place Mismatch.

This figure illustrates the relationship between CNC word scores and the frequency-to-place mismatch, calculated as the difference between Greenwood’s model and the frequency allocation table (FAT) provided by the cochlear implant manufacturer, as adjusted by the audiologist. We analyzed the absolute mismatch for 500, 1000, and 2000 Hz within the same cohort (used in the ECochG mismatch analysis). Our analysis revealed no significant correlation between CNC word scores and the Greenwood frequency-to-place mismatch. These findings underscore the importance of utilizing ECochG data for more accurate cochlear implant mapping, challenging the conventional reliance on standard FAT or Greenwood map.

Performance Prediction: Impact of Cochlear Health and Mismatch on Performance

Prior studies (Fitzpatrick et al. 2014; Walia, Shew, Kallogjeri, et al. 2022; Walia, Shew, Lee, et al. 2022) have shown a strong correlation between post-operative CNC word scores and ECochG-TR measurements taken at the promontory, round window and most recently, the BF electrodes (Walia et al. 2023). This relationship, specific to the current cohort, is depicted in Figure 6A. The BF-ECochG-TR accounted for 49% of the variance in CNC word scores at 3-months post-activation (p < 0.001). Figure 6B illustrates the correlation between CNC at 3-months and the total mismatch (ECochG BF minus FAT for 500, 1000, and 2000 Hz; N = 38). A moderate correlation was observed between these variables (r = 0.51; 95% confidence interval: 0.23 to 0.76; p-value = 0.002). The relationship can be quantitatively described by the linear regression equation: CNC score at 3-months = 14.19 + (1.06*total mismatch). Figure 6C visualizes the BF-ECochG-TR prediction error (residuals) for CNC word scores from Figure 6A, plotted as a function of the total mismatch. BF-ECochG-TR tended to overestimate performance for subjects with large negative total mismatches, while underestimating performance for those with large positive total mismatches. When BF-ECochG-TR (β = 26.81; 95% CI: 2.22 to 51.44) and mismatch (β = −2.84; 95% CI: −4.24 to −0.85), along with their interaction (β = 0.05; 95% CI: 0.03 to 0.13), were incorporated into a multiple regression model, they accounted for 62% of the variance in CNC word scores (p < 0.001, F = 13.52, Figure 6D). Given the lack of correlation between the 500 Hz mismatch and CNC scores at 3-months, we further analyzed the cumulative mismatch at 1000 and 2000 Hz, omitting the 500 Hz data. This revised analysis showed a similar correlation with CNC scores at 3-months (R² = 0.27) and revealed minimal difference in the multivariate predictive model combining ECochG-TR with the mismatch, maintaining an R² of 0.58 for CNC scores at 3 months.

Figure 6. Predicting 3-Month Post-Activation CNC Word Scores using BF-ECochG-TR and Total Mismatch.

This figure presents the correlation between CNC word scores at 3-months post-activation and two variables: best frequency electrocochleography-total response (BF-ECochG-TR) and the total mismatch between the best frequency (BF) identified by intracochlear electrocochleography (ECochG) and the frequency assigned by the subject’s frequency allocation table (FAT). A, Demonstrates the relationship between CNC word scores and BF-ECochG-TR. B, Illustrates CNC word scores as influenced by the total mismatch. C, Depicts the prediction error of CNC word scores based on BF-ECochG-TR, as determined by linear regression, and its relationship with total mismatch. Positive values indicate underprediction of CNC word scores by BF-ECochG-TR, while negative values signify overestimated predictions. It is observed that participants with larger positive mismatches outperform BF-ECochG-TR predictions, while those with larger negative mismatches underperform relative to the predictions. D, Shows observed CNC word scores plotted against predicted CNC word scores based on a model incorporating BF-ECochG-TR, total mismatch, and their interaction term. CNC refers to consonant-nucleus-consonant; TR represents total response, an aggregated ECochG measure.

Discussion:

Performance using a CI is dependent on the peripheral cochlear-neural substrate (i.e., cochlear health), and central auditory system integrity. In addition, device-related factors such as sensing and stimulating functions, electrode location, and electrode-frequency assignment are critical (Fitzpatrick et al. 2014; Holden et al. 2013; Walia, Shew, Kallogjeri, et al. 2022; Walia, Shew, Lee, et al. 2022). CI users commonly experience variability in frequency-to-place mismatch, which may explain, in part, the heterogeneity in speech perception outcomes previously reported among these population cohorts (Blamey et al. 1996; Blamey et al. 2013; Dillon et al. 2014; Gantz et al. 1993; Gifford et al. 2017; Green et al. 2007; Lazard et al. 2012; Polak et al. 2010). The degree of this mismatch is contingent on multiple factors, encompassing array length, the applied frequency filters, the surgeon’s insertion technique, individual variances in cochlear dimensions relative to the target frequency place. In contrast to previous studies that leveraged Greenwood and Stakhovksaya’s et al functions derived from psychoacoustic experiments near threshold and histological studies to model cochlear tonotopy, this study uses in vivo intracochlear electrophysiological measurements from CI recipients with hearing loss, offering a different assessment of frequency-to-place mismatch. This study’s primary objective was to characterize the frequency-to-place mismatch between an ECochG-derived BF map and the FAT among CI users implanted with a single, perimodiolar array. Our results demonstrate recipient variability in the extent of mismatch despite receiving the identical array, implanted according to the manufacturer’s specification. This finding reinforces the hypothesis that addressing this variability should provide an opportunity to improve speech perception performance. Optimization of frequency-to-place mismatch would be possible through alterations in the frequency allocation algorithms available to clinicians in the fitting software, assuming the right target for cochlear frequency place is known.

In our previous work, we successfully demonstrated the feasibility of detailing tonotopicity using in vivo electrophysiologic measures in humans during implantation (Walia et al. 2024). This offers a unique mapping target that diverges from the traditional frequency-position functions, namely organ of Corti and spiral ganglion maps that were derived at threshold levels of stimulation by Greenwood and modified by Helpard et al (Greenwood 1990; Helpard et al. 2021; Li et al. 2021; Stakhovskaya et al. 2007). Our work demonstrated that conversational or higher levels of stimulus presentation shift the operating tonotopic map in a basal (or frequency downward) direction relative to the organ of Corti and spiral ganglion maps. This effect is similar to that seen in both electrophysiological studies in animals and psychoacoustic experiments in normal hearing subjects (Liberman 1982; Liberman et al. 1984; Liberman et al. 1978; Moore et al. 2002). Optimizing CI stimulation to a frequency map derived at conversational levels may be a more relevant target, thus creating a personalized frequency-position function tailored to each patient’s cochlea. Here, we report no correlation between the mismatch of organ of Corti frequency-position function and FAT with postoperative speech perception, suggesting that the mismatch between the organ of Corti map and the FAT may not influence speech perception. This premise is further supported by behavioral studies using pitch matching approaches in single-sided deafness CI patients which suggest that the operating map is lower in frequency than predicted by the spiral ganglion/organ of Corti map (McDermott et al. 2009; Peters et al. 2016; Peters et al. 2019; Vermeire et al. 2015). For these pitch matching studies, it is important to consider the dynamic nature of pitch perception in CI patients, as studies have shown that pitch perception can shift over time post-implantation (Reiss et al. 2015; Reiss et al. 2007). Such shifts could influence the interpretation of frequency-to-place mapping and emphasizes the importance of postoperative neuroplasticity to modify the patient’s adaption to the CI-induced auditory map. In contrast to our previous findings, Canfarotta et al. (2020) reported a weak linear correlation with CI performance and the 1500 Hz mismatch (spiral ganglion frequency map minus default FAT). However, this disparity could stem from the sample size (N = 48) of the prior study or the use of lateral wall electrodes. Furthermore, their linear regression was influenced by a few subjects with a mismatch of >7 semitones, and no sensitivity analysis was conducted to determine the impact of those subjects on the regression. Importantly, recent clinical trials (Creff et al. 2024; Di Maro et al. 2022; Dillon, Helpard, et al. 2023; Fan et al. 2023; Kurz et al. 2023; Kurz et al. 2022; Lambriks et al. 2023) have shown limited benefit in improving speech-perception outcomes when using anatomical mapping (to Greenwood/Stakhovskaya et al) in most cases. Anatomical mapping may not optimally allocate frequencies due to its reliance on generalized models rather than individual cochlear physiology. Further research, considering device specifics and cochlear health, is crucial to understand variations in patient outcomes.

Variability in ECochG Map: Evaluating Cochlear Morphology & Surgical Characteristics

In the present study, we explored whether surgical/anatomic factors could account for the observed variability in ECochG maps; we found no such relationship. Measurements like cochlear size, number of turns, and both apical and basal insertion angles did not correlate with the observed mismatch (Supplementary Figure 1). There is remarkable variation in cochlear size and shape (Avci et al. 2014; Erixon et al. 2009; Hardy 1938; Helpard et al. 2021; Koch et al. 2017). That significant variability in the basal-most electrode’s insertion angle exists suggests that the surgeon’s technique, in part, accounts for some of the observed variability in electrode insertion depth (Figure 1B). That both surgical and anatomic variation exist implies that these variables likely cannot be captured well with a single, generalized function. Thus, electrophysiologic measures emerge as a precise tool for capturing an individual’s unique frequency-position function.

Frequency-to-Place Mismatch: ECochG Map and FAT

In our investigation of the frequency-to-place mismatch between FAT and ECochG map (Figure 2B), we discovered significant variability across subjects. The stratification of mismatch values at 500 Hz (Figure 3A), as opposed to the wider variability at higher frequencies (Figure 3B–C), may reflect the cochlear apex’s broader frequency mapping, where achieving precise alignment is inherently more challenging due to the dense spatial representation of frequencies at the basal end. Despite no notable correlation between the 500 Hz mismatch and early speech-perception performance, a moderate linear correlation was apparent between the mismatches at both 1 kHz and 2 kHz, and the 3-month CNC scores. This aligns with previous work, emphasizing the role of spectral alignment, especially within the 1 to 2 kHz range, in improving speech perception (Başkent et al. 2007). Moreover, recent work by Burwood et al. have shown that low-frequency pitch perception may rely more on temporal than spatial cues, as low-frequency hearing distributes sound-evoked responses across a broad cochlear region without a defined place code (Burwood et al. 2022). This could explain the lesser impact of electrode placement on pitch perception at 500 Hz, thus providing a potential rationale for the lack of correlation with CI performance at this frequency. Another consideration is that our analysis primarily involves the CM component, indicative of outer hair cell function, whereas CIs likely stimulate the SG. The spatial mismatch between the responding hair cells and spiral ganglion cells for 500 Hz stimuli could be more pronounced than for higher frequencies, possibly contributing to the observed variability in the correlation with speech-perception performance at this lower frequency.

Our results also demonstrate a clear association between the direction of the frequency-specific mismatch at 1 kHz and 2 kHz (Figure 3B–C) and the total mismatch (sum of 500, 1000, and 2000 Hz; Figure 6B), and early speech-perception performance. However, when juxtaposed with Greenwood’s frequency-position function, no such correlation emerged within the same patient group (Figure 5). These results suggest that our ECochG map provides a more accurate target of the actual tonotopic map than Greenwood’s function, likely due to Greenwood’s focus on threshold responses rather than loud conversational speech levels.

Notably, performance improved with positive mismatches, where frequencies deviated downwards from the Greenwood/Stakhovskaya et al function and even the ECochG BF map. Top performers (CNC >60%) were more likely to have FATs shifted basally, correlating to higher CNC performance compared to those with apically shifted maps (CNC <30%) (Figure 4). This superior performance could be attributed to a few interrelated factors. Basally shifted maps demonstrate improved alignment with high-frequency speech perception information (1–4 kHz), which includes formant transitions, fricatives, and plosives (Zeng et al. 2002). By contrast, apically shifted maps, with poorer performance, could result from basal electrode deactivation and thus, this was explored (Supplementary Figure 4). We found that limited basal electrode deactivation was present amongst both groups of performers, and when maps from patients with and without electrode deactivation were compared, a similar relationship between mismatch and performance was observed. The question that remains: were the basal electrode(s) deactivated because of initial poor performance and sound quality, or did basal deactivation actually result in poor performance in some patients? Further exploration of electrode reactivation is worthy of consideration in this group of patients. Clearly, the mechanistic underpinnings of mismatch direction, relative to ECochG determined BF electrodes remains to be determined.

Multivariate Modeling: BF-ECochG-TR and Frequency-to-Place Mismatch

In our previous work, we have documented a robust linear correlation between a measure of cochlear health (BF-ECochG-TR) and speech-perception performance (Walia et al. 2023). This measure accounted for approximately 56% of the variability in CI performance in a large cohort (n = 109), significantly surpassing any demographic, surgical, or audiologic factors (Walia et al. 2023). Therefore, incorporating this cochlear health indicator was vital to comprehend the influence of mismatch on CI performance variability. In our study, the total mismatch between the ECochG BF map and the FAT independently explained 27% of the variability in CNC scores at three months, while BF-ECochG-TR accounted for 49% of this variability. As depicted in Figure 6C, the prediction error for a model using BF-ECochG-TR as the only independent variable varied with mismatch. Notably, after adjusting for cochlear health using BF-ECochG-TR, performances exhibited a higher deviation than predicted with larger mismatches, especially in the negative direction (indicating an apically shifted frequency-position map or towards Greenwood). By constructing a multivariate model with BF-ECochG-TR, mismatch, and their interaction, we could explain 62% of the variability in CNC scores at three months. This observation underscores the significance of including cochlear health measures like BF-ECochG-TR when evaluating the impact of various strategies, such as mapping strategies and electrode design, on CI outcomes.

Limitations and Future Work

Our research proposes a novel tonotopic target for CI mapping, contrasting with anatomy-based strategies explored in recent literature (Dillon, Helpard, et al. 2023; Lambriks et al. 2023). While the disparity between the ECochG map and the FAT demonstrates a moderate correlation with early speech-perception performance in quiet and adds to the prediction model, our study has some limitations. We assessed only one type of recording system and electrode (i.e., perimodiolar electrode), thus leaving the viability of recording an intracochlear ECochG map using other systems and a lateral wall electrode unexplored. While both organ of Corti and spiral ganglion maps are substantially more apical than those described by our ECochG measures using a perimodiolar array, work by Peters et al (2019) suggests maps for lateral wall arrays are even more basal (frequency downward) shifted than perimodiolar arrays.

Although patients using hybrid/EAS configurations were included in our study cohort (N=2), this study did not consider the potential implications of this approach on the mapping strategy. Basal (downward frequency) shifts are inherent in electroacoustic (i.e. hybrid) stimulation maps, although the impact of this approach remains unknown. Moreover, our prior research with the CI632, which includes participants from the current study, has demonstrated that hearing preservation does not significantly impact speech perception outcomes (Shew et al. 2021). Nevertheless, there remains a concern that the standard mapping strategy—both as recommended by the manufacturer and as applied by audiologists—may not fully account for the acoustic-electric interaction at low frequencies, which could be crucial for optimizing success with hybrid stimulation. This study, while not designed to evaluate all predictors of speech-perception performance, concentrated on the effects of BF-ECochG-TR and mismatch. Ongoing larger studies aim to include a broader range of predictors, such as data logging, duration of deafness, electrode-to-modiolus distance and residual hearing, to more comprehensively predict speech-perception outcomes including performance in noise. Additionally, this study did not capture the specific timing of basal electrode contact deactivations, which may play a crucial role in postoperative adaptation and performance. Typically, in our clinical practice, electrode deactivation happens in the first month. However, this study’s retrospective nature limits our ability to precisely correlate the timing of these changes with speech-perception outcomes. The potential impact of such adjustments on performance, particularly when they occur outside the initial stabilization period, is an important consideration for future research.

Furthermore, our assessment of speech-perception performance was limited to quiet conditions at three months. It is plausible that central adaptation (Svirsky et al. 2015; Tan et al. 2017) over time might diminish the correlation between mismatch and speech-perception relative to early outcomes (Mertens et al. 2022; Svirsky et al. 2004). While most ECochG maps predominantly showed basal or apical shifts across all frequencies for a specific subject, two subjects presented apical shifts at one frequency and basal shifts at another. Although this resulted in near-zero total mismatch, a clear mismatch between the FAT and ECochG map existed. Another issue with the ECochG map is that it is dependent on measurable, residual cochlear function. While we have previously demonstrated ECochG responses in nearly all CI patients (>90%), the utility of this approach for mapping of individuals with limited physiological responses remains to be determined. We are currently investigating whether the entire map can be predicted once the location for one frequency has been identified. We are also presently conducting prospective studies to compare ECochG-based and default mapping procedures.

Conclusions:

Our study revealed that all patients using perimodiolar electrode arrays experienced a degree of mismatch between their FATs and the BF maps derived from ECochG. Notably, smaller magnitude or slightly positive mismatches correlated with improved early speech perception, accounting for approximately 26% of the variability in CI performance in quiet settings. In particular, participants whose maps were basally shifted at high frequencies demonstrated superior 3-month performance as compared to those with apically shifted maps. When combining mismatch with BF-ECochG-TR, a measure of cochlear health, these factors together could explain as much as 62% of the variability of CNC word scores in quiet. This leads us to believe that employing electrophysiological-based frequency reallocation could potentially enhance speech discrimination, as opposed to the standard default or anatomic-based mapping strategies. Future research efforts should explore the utility of ECochG-based BF mapping as a target for frequency allocation using an assortment of electrode arrays.