Comparison of Speech Recognition with an Organ of Corti versus Spiral Ganglion Frequency-to-Place Function in Place-Based Mapping of Cochlear Implant and Electric-Acoustic Stimulation Devices

Abstract

Objective:

To compare acute speech recognition with a cochlear implant (CI) alone or electric-acoustic stimulation (EAS) device for place-based maps calculated with an organ of Corti (OC) versus a spiral ganglion (SG) frequency-to-place function.

Patients:

Eleven adult CI recipients of a lateral wall electrode array.

Intervention:

Postoperative imaging was used to derive place-based maps calculated with an OC versus SG function.

Main Outcome Measure:

Phoneme recognition was evaluated with consonant-nucleus-consonant (CNC) words presented using an OC versus a SG place-based map.

Results:

For the 9 CI-alone users, there was a non-significant trend for better acute phoneme recognition with the SG map (mean 18 rau) than the OC map (mean 9 rau; p=0.071, 95% CI [≤−1.2]). When including the 2 EAS users in the analysis, performance was significantly better with the SG map (mean 21 rau) than the OC map (mean 7 rau; p=0.019, 95% CI [≤−6.2]).

Conclusions:

Better phoneme recognition with the SG frequency-to-place function could indicate more natural tonotopic alignment of information compared to the OC place-based map. A prospective, randomized investigation is currently underway to assess longitudinal outcomes with place-based mapping in CI-alone and EAS devices using the SG frequency-to-place function.

Introduction

Significant variability in angular insertion depth (AID) is present both across and within cochlear implant (CI) electrode arrays.^1–2 While current mapping procedures take advantage of cochlear tonotopicity by assigning low- to high-frequency information across apical to basal electrodes, they do not individualize frequency allocation using the precise location of electrodes. This can result in discrepancies between the electrically-represented frequency information and the natural place of acoustic transduction, known as frequency-to-place mismatch.^1–2 For CI-alone users, recipients of shorter arrays experience large mismatches; for example, a 24 mm lateral wall array introduces a 1–2 octave mismatch.^1–2 Among electric-acoustic stimulation (EAS) users, the majority experience mismatches greater than ½ octave.² This is clinically relevant as an increasing body of literature supports the notion that closer alignment between the default frequency filters and characteristic frequencies of the stimulated neural populations confers speech recognition benefit.^2–4 While some CI recipients can acclimate to mismatch,^5–7 others may experience incomplete acclimatization even with months to years of listening experience.^5–6,8

One approach to minimize mismatch and mitigate the need for acclimatization is place-based mapping. In place-based mapping, a frequency-to-place function estimates the place frequency corresponding to each electrode, and these values are used to set the frequency filters. Early investigations demonstrated a benefit of reducing mismatch using the estimated linear insertion depths. (9–14) Recently, algorithms using postoperative computed tomography (CT) have allowed for precise determination of intracochlear location of electrodes. (15–17)

An accurate frequency-to-place function is critical to the success of place-based mapping. For CI recipients, the precise function depends upon the site of neural excitation, which could occur on radial nerve fibers near the organ of Corti (OC) or more proximally at the level of the central axon of spiral ganglion (SG) cells.(18) Figure 1.A plots the place frequency estimates across AID using Greenwood’s OC function¹⁹ and an average SG function(20), with resultant mismatch from default frequency filters plotted for reference. Figure 1.B depicts the deviation between functions across AID, which differ by ~5 semitones in the first 1.5 turns. A SG function may support a more accurate place-based map than the OC function because SG cell survival has been demonstrated following long durations of hearing loss, even when peripheral dendrites are largely absent.(21) Alternatively, an OC function may support a more accurate place-based map for CI recipients with intact dendritic processes; this may be more likely in patients with residual hearing in the CI ear. In practice, there is no evidence that the differences between an OC and SG place-based map are large enough to impact speech recognition. The present investigation compared acute speech recognition of CI-alone and EAS users at device activation when listening to an OC versus SG place-based map.

Figure 1:

A) Estimates of cochlear place frequency by angular insertion depth with the organ of Corti (OC) frequency-to-place function and the average spiral ganglion (SG) frequency-to-place function, as described in the text and Supplement 1. For reference, a comparison of the angular insertion depth and frequency-to-place relationship with the default frequency filters for recipients of the FlexSOFT (31 mm), Flex28 (28 mm), and Flex24 (24 mm) arrays are provided. The grey shading indicates the 95% confidence interval around the mean of the CI recipient cohort reviewed by Canfarotta et al (citation #²). B) Difference in the cochlear place frequency between the OC and SG frequency-to-place functions by angular insertion depth.

The present investigation compared acute speech recognition of CI-alone and EAS users at device activation when listening to an OC versus SG place-based map.

Material and Methods

Procedures were approved by the Institutional Review Board, and subjects provided consent prior to participation. Adult CI recipients (≥18 yrs) of MED-EL (Innsbruck, Austria) arrays were recruited at initial CI-alone or EAS activation. Exclusion criteria included prelingual deafness and CI revision. Subjects were fit with the FS4 coding strategy using the place-based mapping procedure.

Place-Based Mapping Procedure

The AID for individual electrodes was determined with postoperative CT using OTOPLAN (MED-EL, Innsbruck, Austria),(15) and used to derive the place frequency of each electrode. For the OC function, this entailed: 1) calculating the cochlear duct length (CDL) and electrode linear insertion depth at the OC using the elliptic-circular approximation method,(22) and 2) using the proportional distance from the apex to determine frequency based on Greenwood’s function.(19) For the SG function, place frequencies were derived from the average SG map.(20) These computations are described in Supplement 1. A custom MATLAB (MathWorks) script aligned the frequency filters of individual channels to the place frequency for each function to at least 3000 Hz. The 3000-Hz cutoff was selected due to the importance of mid-frequency information for speech recognition.(23) High frequencies were logarithmically distributed across the remaining intracochlear electrodes.

For recipients with shallow insertions, place-based mapping limits the electric low-frequency information. For CI-alone users, this may result in a loss of available low-frequency cues. For EAS users, this may result in a gap in frequency information when the low edge of the lowest frequency filter is above the upper cutoff of acoustic hearing, which may result in poorer performance.²⁴

For EAS users with deep insertions, place-based mapping can introduce an overlap in frequency information when an electrode(s) resides in the acoustic hearing region; when this occurs, a place-based map could duplicate the acoustic frequency information. In these cases, reducing the electric stimulation below the listener’s detection for the electrode(s) within the acoustic hearing region could avoid peripheral masking.²⁵

Speech Recognition

Acute speech recognition was assessed in a soundproof booth with the subject seated one meter from the speaker at 0° azimuth. The recorded consonant-nucleus-consonant (CNC) word test²⁶ was presented at 60 dB SPL. The contralateral ear was plugged or masked with an insert phone. The tester and subject were blinded to the randomized map configuration.

Data Analysis

Phoneme recognition was assessed due to floor effects for word recognition. Percent correct scores were converted to rationalized arcsine units (RAUs) to normalize error variance.²⁷ For normally-distributed data, paired, one-tailed t-tests were used to compare performance between the OC and SG maps for the CI-alone listeners and the combined cohort (CI-alone & EAS) using Prism (version 8.4.1). A significance criterion of ∝ = 0.05 was adopted.

Results

The demographic information for the 11 subjects is listed in Table 1. The mean age at implantation was 61 years (SD: 11 years). The AID of the most apical electrode ranged from 311° to 700° (mean: 601°, SD: 113°). Nine subjects listened with a CI-alone and two with EAS. Subject EAS1 did not have a CT scan, therefore the intraoperative x-ray was used to estimate the AID with a method shown to be accurate within 10° of CT estimates.²⁸ The stimulation levels of the two most apical electrodes were reduced for subject EAS2 to avoid peripheral masking.

Table 1:

Demographic information for the study subjects and associated map information. The initial letters of the subject identifier indicate whether the subject listened with a cochlear implant alone (“CI#”) or an electric-acoustic stimulation (“EAS#”) device. Age (years) indicates the subject’s age at implantation. The angular insertion depth (AID) of the most apical electrode (E1) is reported in degrees. Subjects are sorted by E1 AID for each device (CI-alone and EAS). Cochlear place frequency for two functions (OC and SG) is indicated for each electrode (E1-E12). Grey cells indicate electrodes that were not matched to the cochlear place frequency. Dashed cells indicate that the electrode’s stimulation level was reduced below detection to limit peripheral masking of acoustic and electric stimulation. The acoustic crossover (Hz) for the EAS devices were obtained from the clinical programming software.

Subject	Age	Array	E1 AID	Cochlear Place Frequency (Hz)													If EAS, Acoustic Crossover
				Fn	E1	E2	E3	E4	E5	E6	E7	E8	E9	E10	E11	E12
CI1	50	Flex24	311°	OC	1513	2055	2918	4201	5712	7647	10856	15054	18312	out	out	out	NA
				SG	1139	1500	2095	3053	4213	5695	8074	11042	13327	out	out	out
CI2	70	FlexSOFT	500°	OC	491	758	1001	1307	1756	2580	3779	5262	7055	9739	13345	17465	NA
				SG	404	603	782	999	1304	1859	2737	3870	5248	7261	9852	12726
CI3	65	FlexSOFT	552°	OC	357	555	781	1045	1522	2238	3224	4712	6418	9025	13491	18095	NA
				SG	306	449	616	812	1146	1621	2318	3429	4755	6724	9902	13172
CI4	48	FlexSOFT	613°	OC	235	355	522	923	1444	2012	2939	4169	5903	8001	11651	17467	NA
				SG	215	303	426	724	1087	1464	2110	3010	4349	5951	8648	12639
CI5	75	FlexSOFT	629°	OC	206	341	555	917	1396	2039	3022	4074	5663	8147	12139	17672	NA
				SG	193	293	450	718	1058	1482	2166	2949	4170	6051	8998	12740
CI6	70	Flex28	638°	OC	194	309	645	1016	1327	1642	2305	3158	4266	6367	9028	14245	NA
				SG	183	271	516	788	1006	1224	1664	2254	3097	4700	6705	10381
CI7	55	FlexSOFT	666°	OC	152	261	525	892	1292	1835	2697	3840	5618	8149	12659	17883	NA
				SG	148	235	427	698	983	1347	1926	2767	4120	6073	9347	12990
CI8	51	Flex28	688°	OC	126	240	397	728	1216	1702	2461	3547	5106	7318	11080	16111	NA
				SG	121	218	334	575	932	1258	1764	2553	3724	5422	8177	11660
CI9	79	FlexSOFT	700°	OC	112	232	461	795	1147	1608	2321	3441	5267	7786	11533	16293	NA
				SG	107	213	381	629	885	1203	1679	2480	3870	5800	8563	11904
EAS1	49	FlexSOFT	649°	OC	176	303	478	726	1161	1630	2371	3392	4569	6258	8533	13937	150
				SG	169	266	393	578	896	1218	1713	2443	3335	4635	6362	10264
EAS2	54	FlexSOFT	669°	OC			461	832	1260	1711	2475	3441	4902	7185	11416	17263	250
				SG			380	657	962	1266	1773	2480	3587	5339	8430	12484

Figure 2 plots phoneme recognition of each subject when listening with the OC and SG maps. Symbol shapes identify individual subjects, as defined in the legend. For the CI-alone users, there was a trend for better performance with the SG map as compared to the OC map (t(8)=−1.63, p=0.071, 95% CI [≤−1.2]). Two (CI8 and CI9) of the three subjects who experienced better performance with the OC map had the deepest AID (i.e., 700° and 688°). With the addition of the EAS users, performance was significantly better with the SG map compared to the OC map (t(10)=-2.40, p=0.019, 95% CI [≤−6.2]).

Figure 2:

Acute phoneme recognition at initial CI or EAS activation with an organ of Corti (OC) versus a spiral ganglion (SG) place-based map. Symbol shape indicates the results for individual subjects, as defined in Table 1. Subjects are sorted by the angular insertion depth of the most apical electrode for the CI-alone and EAS device users. Results are reported in rationalized arcsine units (RAUs), with higher scores indicating better performance.

Discussion

There was a trend for better acute phoneme recognition for CI-alone users with a place-based map using the SG function as compared to the OC function. This indicates that the choice of frequency-to-place function used in place-based mapping likely influences speech recognition – at least initially. The most plausible explanation for the observed performance benefit with the SG map is that SG cells are the predominant site of electrical stimulation in CI-alone and EAS users.

Three subjects demonstrated better performance with the OC map (CI5, CI8, and CI9). It is possible that these subjects had remaining peripheral dendritic processes that supported speech recognition benefit with an OC map. This interpretation, however, is undermined by the fact that the EAS listeners experienced better speech recognition with the SG map, despite evidence of peripheral dendrite survival via their residual hearing.

It is also possible that differences in low-frequency bandwidth between the OC and SG maps are responsible for performance differences. For CI3, for example, the place frequency of E1 was 1513 Hz for the OC map and 1129 Hz for the SG map; in this subject, greater access to low-frequency information could be responsible for the better performance obtained with the SG map. Another consideration when comparing maps concerns spectral representation of important speech information (~300–3000 Hz); this region of the spectrum is distributed over more electrodes with the SG map, which could improve spectral resolution. Lastly, place-based mapping with FS4 reduces mismatches in both spectral and temporal properties of stimulation. Place of stimulation is known to affect the perception of low-frequency rate pitch,^29–30 but it is unclear what role fine-structure cues play in the benefit observed with the SG map in the present dataset. While difficult to disentangle the mechanistic contributions related to bandwidth, spectral resolution, and electric encoding of fine-structure information, results are nonetheless suggestive regarding an approach for maximizing outcomes.

Potential limitations of the present study include the use of an average SG function, whereas the OC function was based on individual estimates of CDL. Patient-specific SG maps may provide a more accurate frequency-to-place function. As more accurate representations of the cochlear apex and neural elements in this region become available,^31–33 and these findings are modeled such that cochlear place estimates can be generated from clinically available in-vivo imaging studies, comparisons between functions will need to be reassessed. Nevertheless, the findings herein remain compelling in that subtle differences (e.g., <6 semitones) in place-frequency can reduce speech recognition acutely.