Skip to main content
NCBI home page
AIP Publishing Selective Deposit logoLink to AIP Publishing Selective Deposit
. 2022 Sep 16;2(9):094403. doi: 10.1121/10.0013428

Speech recognition as a function of the number of channels for pediatric cochlear implant recipientsa)

René H Gifford 1,b),✉, Linsey W Sunderhaus 1, Jourdan T Holder 1, Katelyn A Berg 1, Benoit M Dawant 2, Jack H Noble 2, Elizabeth Perkins 3, Stephen Camarata 1
PMCID: PMC9488908  PMID: 36182337

Abstract

This study investigated the number of channels required for asymptotic speech recognition for ten pediatric cochlear implant (CI) recipients with precurved electrode arrays. Programs with 4–22 active electrodes were used to assess word and sentence recognition in noise. Children demonstrated significant performance gains up to 12 electrodes for continuous interleaved sampling (CIS) and up to 22 channels with 16 maxima. These data are consistent with the latest adult CI studies demonstrating that modern CI recipients have access to more than 8 independent channels and that both adults and children exhibit performance gains up to 22 channels.

1. Introduction

Cochlear implant (CI) stimulation relies on electrical pulses propagated through highly conductive fluid resulting in spread of electrical excitation (i.e., channel interaction). Channel interaction coupled with an unknown underlying neural survival pattern (e.g., Won et al., 2014) results in poor spectral resolution (e.g., Jeon et al., 2015; Padilla and Landsberger, 2016) and relatedly poor speech recognition in quiet and noise (e.g., Goehring et al., 2021). Seminal CI channel studies demonstrated that CI users had access to 5–8 independent channels despite having up to 22 intracochlear electrodes (e.g., Fishman et al., 1997; Friesen et al., 2001; Shannon et al., 2011). Recently, Berg et al. (2019) replicated the Friesen et al. (2001) study with adult CI recipients implanted with current technology via minimally traumatic surgical approaches. We showed that adult CI recipients with perimodiolar or precurved electrode arrays located in scala tympani (ST) achieved significantly greater performance with 16 vs 8 channels for speech in quiet and that 16 channels in a 16 maxima (16-of-m) strategy yielded significantly higher outcomes than both 8- and 10-channel continuous interleaved sampling (CIS) for speech and auditory perceptual tasks as well as subjective sound quality (Berg et al., 2019). This finding was consistent with other reports. For example, Schvartz-Leyzac et al. (2017) reported a significant improvement for sentence recognition with 20 vs 8 active electrodes. Similarly, Croghan et al. (2017) demonstrated significant speech recognition benefits with 22 vs 12 active electrodes using an n-of-m strategy with 8 maxima for all tested electrode conditions.

In subsequent studies of adults with translocated precurved electrode arrays and straight electrode arrays, Berg et al. (2020, 2021) reported that speech recognition plateaued with 8–10 electrodes, consistent with the classic literature (Fishman et al., 1997; Friesen et al., 2001; Shannon et al., 2011); however, this was in contrast to our earlier study for which adults with precurved electrode arrays completely in ST showed significant improvement up to 16 channels (Berg et al., 2019). Our working hypothesis was that straight and translocated precurved electrode arrays likely yielded poorer spatial selectivity and channel independence (i.e., number of “independent” channels irrespective of the number of intracochlear electrodes) due to the lateral wall and/or scala vestibuli (SV) location, which places electrodes further from the neural targets in the modiolus (e.g., Goldwyn et al., 2010; Litvak et al., 2007), thereby requiring higher charge for auditory perception (Davis et al., 2016). This theory was supported by the finding that there was a significant effect of electrode-to-modiolus distance (in mm) on the number of channels required for speech recognition in noise (Berg et al., 2019, 2021) and vowel perception (Berg et al., 2021).

Several studies have investigated the number of channels required for performance asymptote on various measures of speech recognition for children with normal hearing (NH) (Dorman et al., 2000; Eisenberg et al., 2000; Warner-Czyz et al., 2014). Dorman et al. (2000) investigated sine-wave vocoded speech recognition for NH adults and children using 4, 6, 8, 10, 12, and 20 channels. They showed that children required 12 channels to reach asymptote as compared to 8 channels for adult word recognition using the lexically easy Lexical Neighborhood Test (LNT) (Kirk et al., 1995). For the lexically hard LNT, children showed continuous improvement up to 20 channels, whereas NH adult listeners demonstrated a clear performance plateau with just 10 channels (Dorman et al., 2000).

In a similar study, Eisenberg et al. (2000) investigated noise-band vocoded speech recognition for 4-, 6-, 8-, 16-, and 32-band conditions in two groups of children with NH (5–7 years and 10–12 years) as well as a group of NH adults. They used monosyllabic words, sentences, and a discrimination task of speech feature contrasts (e.g., /da/ vs /za/). They found that both groups of children and adults achieved performance plateau with 8 channels for sentence recognition and speech feature contrasts. For monosyllabic words, however, NH children required 16 channels for plateau.

Warner-Czyz et al. (2014) investigated behavioral vowel discrimination (/ta/ vs /ti) for noise-band vocoded stimuli with 16 and 32 channels for 52 infants (5–7 months) with NH. Using a visual habituation protocol, infants demonstrated vowel discrimination with 32 channels, but were unable to discriminate /ta/ vs /ti/ with 16 channels. Thus, their findings were consistent with the previous studies such that infants and children required greater spectral representation for speech recognition or discrimination than adult listeners.

Despite consistent reports of children requiring greater spectral detail for speech recognition with spectrally degraded stimuli, channel independence—referring to the number of independent spectral channels along the array irrespective of the number of electrodes—and its impact on speech recognition has yet to be studied in children with CIs. This investigation is warranted given that young children are known to rely more heavily on sensory processing for speech recognition as compared to adults (e.g., Nittrouer and Boothroyd, 1990). Thus, the purpose of the current study was to investigate the number of channels needed to achieve performance plateau for speech recognition in a group of congenitally deafened children with CochlearTM CI systems and precurved electrode arrays. Our hypothesis was that speech recognition would increase beyond 8 electrodes based on recent findings for adult CI users with precurved electrode arrays (Berg et al., 2019) as well as the literature investigating the number of channels required for spectrally degraded speech understanding for NH children.

2. Methods

2.1. Study participants

We recruited 10 prelingually deafened pediatric CI users (4 female) with precurved electrode arrays. Nine of the 10 participants had postoperative computed tomography (CT) scanning to determine scalar electrode location via image processing (Zhao et al., 2018). Of the 9 children with postoperative imaging, 4 had electrode arrays completely within ST, 1 had their array completely in SV, and 4 had transcalar displacement from ST to SV (n = 3) or SV to ST (n = 1). The parents of the child without postoperative CT declined scanning, but were still interested in participation. Inclusion criteria required at least 12 months of CI experience and ≥18 active electrodes in their clinical map. Table 1 displays demographic information for the 10 study participants.

Table 1.

Participant demographics including age at implantation, age at enrollment, electrode type, electrode scalar location, mean electrode-to-modiolus distance (mm), number of active electrodes in everyday map, number of maxima used in everyday n-of-m map, and etiology.

ID Age at CI (years) Age at study enrollment (years) Electrode type Electrode scalar location Mean electrode-to-modiolus distance (mm) Active electrodes in everyday map Maxima used in everyday map Etiology
1 1.3 9.7 CI512 ST 0.38 22 10 Unknown
2 1.7 9.6 CI512 SV 0.49 22 14 Meningitis
3 4.8 12.3 CI512 No postop CT N/Aa 22 8 Premature (26 weeks), NICU,b mechanical ventilation
4 1.1 8.5 CI512 SV-ST-SV 0.73 22 12 Unknown
5 5.8 13.3 CI532 ST 0.42 18 12 Unknown
6 2.2 11.1 CI512 ST-SV 0.54 20 12 Unknown
7 1.4 12.3 CI512 ST-SV 0.50 21 10 Unknown
8 1.3 10.0 CI24RE(CA) ST 0.45 21 12 Connexin 26 GJB2
9 2.0 14.5 CI24RE(CA) ST 0.49 22 10 Connexin 26 GJB2
10 1.7 15.3 CI24RE(CA) ST-SV 0.67 22 11 Premature (26 weeks), NICU, mechanical ventilation
Mean 2.3 11.7 — — 0.52 21.2 11.1 —
Open in a new tab
a

Not applicable (N/A).

b

Neonatal intensive care unit (NICU).

2.2. Materials and conditions

This study was completed in accordance with institutional review board (IRB) approved protocols at our institution. We tested 5 spatially selective channel programs using 4, 8, 12, 16, and “all on” active electrodes. These conditions were used to match the spatially selective programs described by Friesen et al. (2001) and the replication studies completed by Berg et al. (2019, 2020); however, in contrast to the previous studies, we opted to use 12 vs 10 channels to capture the range of intracochlear electrodes that are available for Food and Drug Administration (FDA) approved CI systems. Figure 1 displays the electrodes activated and corresponding frequency allocations for each of the spatially selective maps.

Fig. 1.

Fig. 1.

Open in a new tab

Active electrodes and associated frequency allocations for all tested conditions. For participants who had electrode(s) deactivated in their clinical map, the closest available electrode was activated to maintain the greatest spatial separation between activated electrodes. For example, if E1 had been deactivated in the participant's map, E2 would be activated instead for the 8, 16, and “all on” conditions.

Similar to past studies (Berg et al., 2019, 2020; Friesen et al., 2001), the input frequency range was fixed across the different conditions. The number of maxima matched the number of active electrodes to create CIS maps (Wilson et al., 1991). For the “all on” condition, we used 16-of-m as this was the upper maxima limit available in the clinical software. For the “all on” 16-of-m condition, participants had 18–22 active electrodes, consistent with their everyday map. All children in this study had long-term experience with a channel stimulation rate of 900 Hz, and 9 of 10 children were using a pulse duration of 25 μs—both of which were unchanged for experimental activities. Participant 10 was using a 37-μs pulse duration for his everyday map, which we decreased to 25 μs prior to testing. The number of maxima used in each child's everyday map ranged from 8 to 14 with total stimulation rates ranging from 7800 to 12 600 Hz (Table 1). All children in the current study had their upper stimulation level profile set using electrically evoked stapedial reflex thresholds; thus, to achieve adequate and equivalent perceived loudness across all tested channel conditions, we globally adjusted upper stimulation levels without adjusting individual electrodes. For n-of-m conditions, upper and lower stimulation levels were held constant, and we asked each child if there were perceived loudness differences across the maps. For children reporting loudness differences across maps, we globally adjusted the upper stimulation profile to achieve equal loudness across the n-of-m maps (8-of-m and 16-of-m). Lower stimulation levels were not adjusted from the participant's own map but were verified to be providing CI-aided detection thresholds in the range of 20–30 dB hearing level (HL) from 250 through 6000 Hz. Also, consistent with our past adult studies (Berg et al., 2019, 2020), front-end processing features were deactivated, except for autosensitivity control (ASC) and adaptive dynamic range optimization (ADRO), which all children used in their everyday programs.

Order of assessment for the different channel conditions and speech recognition measures was randomized using a Latin square design, and all testing was completed acutely following a brief period during which a few sentences or words were presented for practice purposes. Testing included use of a single loudspeaker placed at 0° azimuth at a distance of 1 m from the study participant in a single walled sound booth using the following measures: CNC monosyllabic words (Peterson and Lehiste, 1962) and pediatric AzBio (i.e. BabyBio) (Spahr et al., 2014) sentences at a +5 dB signal-to-noise ratio (SNR) in the presence of continuous 20-talker babble. A +5 dB SNR was chosen to be consistent with our previous studies (e.g., Berg et al., 2019, 2020, 2021) as well as the fact that +5 dB SNR is a commonly encountered SNR in realistic listening scenarios, lending to the ecological validity of this listening condition (e.g., Smeds et al., 2015). Speech stimuli were presented at a calibrated level of 60 dB sound pressure level (SPL).

3. Results

Data analysis involved a mixed effects model with number of channels as the independent variable and speech recognition as the dependent variable. Post hoc analyses were completed via all-pairwise, multiple comparisons applying Sídak corrections. The top row of Fig. 2 displays individual and mean speech recognition scores, in percent correct, for CNC monosyllabic words and BabyBio sentences in noise. All scores displayed in Fig. 2 were transformed to rationalized arcsine units (RAU) (Studebaker, 1985) prior to statistical analysis.

Fig. 2.

Fig. 2.

Open in a new tab

Individual and mean performance, in percent correct, for all channel conditions. Error bars represent ±1 standard deviation. Statistical significance for post hoc, all comparison analyses are displayed for RAU transformed data: *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001; ns, not significant (p > 0.05).

3.1. CNC word recognition

Mean CNC word recognition was 16.6% (16.3 RAU) for 4 channels, 56.8% (56.5 RAU) for 8 channels, 70.6% (70.2 RAU) for 12 channels, 74.8% (73.9 RAU) for 16 channels, and 84.8% (85.8 RAU) for the “all on” 16-of-m condition. There was a significant main effect of number of channels [F(2.1,18.4) = 77.41, p < 0.0001, ηp2 = 0.89]. Post hoc analyses revealed significant performance differences between 4 electrodes and all other electrode conditions (p < 0.0001 for all comparisons) as well as 8 channels and all other conditions (p < 0.011 for all comparisons). Additionally, there were significant differences between “all on” 16-of-m and 12-channel CIS (p = 0.004) as well as 16-channel CIS (p = 0.018). No other comparisons were found to be statistically significant.

3.2. BabyBio sentence recognition in noise

Mean scores for BabyBio sentence recognition at +5 dB SNR were 14.3% (8.4 RAU) for 4 channels, 73.0% (73.2 RAU) for 8 channels, 85.4% (87.1 RAU) for 12 channels, 86.5% (87.1 RAU) for 16 channels, and 90.7% (96.6 RAU) for the “all on” 16-of-m condition. There was a significant main effect of number of channels [F(1.7,15.4) = 112.8, p < 0.0001, ηp2 = 0.93]. Post hoc analyses revealed significant performance differences between 4 channels and all other conditions (p < 0.02 for all comparisons) as well as 8 channels and all other conditions (p < 0.004 for all comparisons). No other comparisons were found to be significant.

3.3. Number of spatially selective channels: Individual clinical significance and number of maxima

Figure 3 was constructed to visualize the effect of channels within the context of the 95% confidence interval for test-retest variability for CNC (Thornton and Raffin, 1978) and BabyBio (Spahr et al., 2014) represented by the solid lines in each panel. Considering the first 4 panels in each row, individual scores are displayed for 12, 16, 8-of-m, and 16-of-m plotted against 8-channel CIS. Data points located above the top line represent an individually significant increase in performance relative to the 8-channel condition. Data points falling within the two lines represent no individually significant change in performance relative to 8-channel CIS. Data points falling below the lower line represent individually significant decrement in performance relative to the 8-channel condition; however, none of the participants exhibited a decrement.

Fig. 3.

Fig. 3.

Open in a new tab

Individual CNC (top row) and BabyBio (bottom row) sentence scores, in percent correct. The first 4 panels in each row display individual performance for the 12, 16, 8-of-m, and 16-of-m conditions as a function of 8-channel CIS. The last panel in each row displays the effect of maxima plotting scores with 16-of-m vs 8-of-m. Lines in each panel represent the 95% confidence interval for test-retest variability for CNC (Thornton and Raffin, 1978) and BabyBio (Spahr et al., 2014).

For CNC word recognition (Fig. 3, top row), the number of participants achieving individually significant improvement relative to 8-channel CIS was as follows: 3 for 12-channel CIS, 4 for 16-channel CIS, 7 for 8-of-m, and 8 for 16-of-m. For BabyBio sentences at +5 dB (Fig. 3, bottom row), the number of participants achieving individually significant improvement relative to 8-channel CIS was as follows: 4 for 12-channel CIS, 5 for 16-channel CIS, 7 for 8-of-m, and 8 for 16-of-m.

The rightmost panel in each row of Fig. 3 displays scores for the “all on” 16-of-m condition plotted against scores obtained with “all on” 8-of-m. Mean CNC scores for the “all on” conditions were 81.4% (78.2 RAU) and 84.8%, (85.8 RAU) for 8 and 16 maxima, respectively. Mean BabyBio sentence in noise recognition scores for the “all on” conditions were 88.8% (92.6 RAU) and 90.7% (96.2 RAU) for 8 and 16 maxima, respectively. Paired t-tests were completed to investigate the effect of maxima for 16-of-m versus 8-of-m for the RAU transformed scores. For CNC word recognition, there was a significant effect of maxima [t(9) = 2.4, two-tailed p = 0.039, dCohen = 0.68], whereas there was no effect of maxima for BabyBio sentence recognition at +5 dB SNR [t(9) = 1.62, two-tailed p = 0.141, dCohen = 0.25].

4. Discussion

Consistent with our hypothesis, children with precurved electrode arrays demonstrated significantly higher outcomes with 12 and 16 channels as compared to 8 channels for both CNC word recognition and BabyBio sentence recognition at +5 dB SNR. When including n-of-m conditions, children demonstrated additional significant performance gains for the “all on” 16-of-m condition as compared to 8-, 12-, and 16-channel CIS for CNC word recognition and BabyBio sentences at +5 dB SNR. These results are not only relevant for understanding channel independence for pediatric CI recipients, but there is potential impact for scientific and clinical approaches aimed at improving spatial selectivity via selective electrode deactivation (e.g., Noble et al., 2016). That is, for children implanted with precurved electrode arrays, the current results suggest that 12 should be the minimum number of active electrodes with potential for additional increased performance with up to 16 in a CIS strategy and 18–22 electrodes with 16 maxima in an n-of-m strategy.

It is unknown whether the additional benefit conferred by the “all on” 16-of-m condition as compared to 8-, 12-, and/or 16-channel CIS is due to increasing the number of electrodes, the use of a familiar “n-of-m” strategy, and/or different input frequency ranges between the conditions (see Fig. 1). The “all on” 16-of-m condition was identical to the participants' everyday programs with respect to input frequency range for each electrode and the use of an n-of-m strategy to which the children were accustomed. Thus, one could argue that there was bias for the “all on” 16-of-m condition; however, consistent with reports in our adult channel studies (Berg et al., 2019, 2020), the “all on” 16-of-m program used 16 maxima, which increased overall stimulation rate per frame to 14 400 Hz as compared to the participants' everyday programs using 8–14 maxima with corresponding overall stimulation rates in the range of 7200–12 600 Hz (see Table 1). Future studies should investigate the impact of map familiarity related to overall stimulation rate, number of active electrodes, number of maxima, and frequency allocation for both adult and pediatric CI recipients.

4.1. Comparison of current dataset with adult CI channel studies

The current findings were similar to recent studies of adult CI users with precurved electrode arrays. For 11 adult CI users with precurved electrodes completely in ST, Berg et al. (2019) demonstrated continuous performance gains up to 16 channels as compared to 8 channels in a CIS strategy for CNC words with additional significant improvement in the “all on” 16-of-m condition. The current study, however, showed significant performance gains up to 12 channels but no additional improvement for 16-channel CIS. We also observed a statistically significant performance gain for the “all on” 16-of-m condition as compared to 16-channel CIS for CNC words, which was similar to the adult precurved CI study (Berg et al., 2019). While the number of electrodes firing on a given frame is identical for 16-of-m and 16-channel CIS, the 16-of-m condition incorporates place-based cues as the electrodes stimulated on each cycle can vary across the array. With CIS, however, the same electrodes and corresponding frequency ranges are constant across frames, and thus place-based cues are limited to a channel specific amplitude envelope.

For sentence recognition at +5 dB SNR, we observed significant performance differences between 4- and 8-channel CIS with no additional gains for other channel conditions, including the “all on” 16-of-m condition. For our adult CI study, we observed significant performance differences between 4-channel CIS and both 10- and 16-channel CIS as well as an additional significant improvement for the “all on” 16-of-m condition vs all other conditions (Berg et al. (2019)). In contrast, children in this study exhibited significant performance gains with 8 vs 4 channels, whereas the adult CI recipients with precured electrodes did not show significant improvement with 8 over 4 channels (Berg et al., 2019). These across-study differences for sentence recognition in noise could be due to several factors. First, we did not use a 10-channel condition in the current study, and thus an exact across-study comparison could not be completed; however, we did have both 8- and 16-channel CIS and did not demonstrate a significant difference between these conditions, though this was significant for adult CI users (Berg et al., 2019). Second, sentence in noise scores were considerably poorer for the adult CI recipients (Berg et al., 2019) as compared to the children in the current study. For example, children with CI in the current study scored 96 RAU for sentences at +5 dB SNR in the “all on” 16-of-m condition, whereas the adult CI users scored 46 RAU for sentences at +5 dB SNR in the same listening condition. Thus, it is possible that had we used a more difficult measure for the children in the current study, across-study comparisons may have been more similar. Third, only 4 of the children in the current study had all electrodes within ST, whereas all 11 CI users in our adult study had confirmed ST electrode placement. The classic CI literature (Fishman et al., 1997; Friesen et al., 2001; Shannon et al., 2011) and a recent study of adult CI users with translocated electrodes (Berg et al., 2020) showed performance plateau with 8 or fewer spatially selective channels. Thus, it is possible that inclusion of children with ST and ST-SV scalar location impacted the current findings. Further research is needed to investigate the number of spatially selective channels required for asymptotic speech recognition in noise for a measure that is not affected by ceiling effects and for children with precurved electrode arrays with confirmed ST placement.

Considering the effect of spectral maxima (n-of-m), both adult (Berg et al., 2019) and children with precurved electrode arrays demonstrated significant differences between 16 and 8 maxima in the “all on” condition for CNC word recognition. In contrast to the adult CI study, we did not observe a significant difference between 16 and 8 maxima for sentence recognition in noise in this study. Furthermore, just one child in the current study demonstrated an individually significant improvement with 16 vs 8 maxima (Fig. 3). There are several possible reasons that could account for across-study differences and the limited clinical significance observed despite statistical significance for CNC words. First, all but one child in the current study scored ≥82% correct (92.6 RAU) for BabyBio sentences at +5 dB in the 8-of-m condition, leaving little room for further improvement. Second, all but one child had long-term CI experience with 10–14 maxima, whereas all adult CI users in Berg et al. (2021) had long-term experience with just 8 maxima. Third, while the difference between 16 and 8 maxima represented a doubling of both spectral maxima and overall stimulation rate for the adult CI users (Berg et al., 2019), all but one of the children in the current study were using 10–14 maxima in their everyday CI programs, and thus increasing to 16 maxima represented a less extreme change for both maxima and overall stimulation rate. Additional research examining the effects of increasing maxima for children using the clinical software default of 8-of-m is warranted, particularly given that the children in this study exhibited performance gains up to 18–22 channels in a 16-of-m strategy as well as gains up to 12-channel CIS for monosyllabic words. Finally, the children in the current study had an average electrode-to-modiolus distance of 0.52 mm with a range of 0.38–0.73 mm. While the adults in the Berg et al. (2019) study had a similar mean (0.49 mm, range = 0.20–0.80 mm), they had a broader range of distances with a statistically significant inverse correlation between electrode-to-modiolus distance and benefit observed for 16 vs 8 maxima for sentence recognition in noise. Furthermore, none of the adult CI recipients with mean electrode-to-modiolus distances ≥0.55 demonstrated benefit with 16 vs 8 maxima (Berg et al., 2019). To aid this comparison, we completed Pearson correlation analyses for the current study comparing average electrode-to-modiolus distance and degree of benefit with 16 vs 8 maxima. This analysis revealed no statistically significant relationship for either CNC word recognition (r = −0.30, p = 0.43) or BabyBio sentence recognition in noise (r = −0.58, p = 0.10); however, the correlation coefficient for sentence recognition in noise (−0.58) was similar to that observed in the adult study (−0.52). Thus, it is likely that we were underpowered for this analysis as there were 23 CI users in the adult study (Berg et al., 2019) and just 9 CI recipients in the current study with confirmed device placement and electrode-to-modiolus distances. Thus, there is a need for future investigations on the default number of spectral maxima for children with CIs to achieve their highest auditory outcomes. This is particularly critical for our youngest CI recipients as Warner-Czyz et al. (2014) demonstrated that infants with NH achieved vowel discrimination with a 32-channel simulation but were unable to discriminate vowels with just 16 simulated channels.

4.2. Comparison of current pediatric CI dataset to CI simulation studies

Dorman et al. (2000) suggested that children with access to 12 channels could presumably achieve similar outcomes to NH adults and children listening to vocoded speech with 6–8 spectral channels. Our group of 10 pediatric CI users performed similarly to the NH children in Dorman et al. (2000) such that word recognition was significantly higher with 12 vs 8 channels in both studies. Also, similar to Dorman et al. (2000), there was additional significant improvement in CNC word recognition with the “all on” 16-of-m condition as compared to both 12- and 16-channel CIS. Across-study comparisons were also similar for the current data set and Eisenberg et al. (2000) with the exception that their children with NH exhibited significant performance gains with 16 vs 8 channels for monosyllabic word recognition, whereas this comparison was not found to be significant in the current study. This could have been due to a smaller sample size in the current study (n = 10) as compared to the 16 NH children in the Eisenberg et al. (2000) paper. Additionally, Eisenberg et al. (2000) used two different, unequal groups of children with one group (n = 10) tested with 4, 6, and 8 noise-band vocoder channels and the other group (n = 6) tested with 8, 16, and 32 channels.

Another way to compare the current findings to previous studies is to consider absolute word recognition across children with CI and children with NH listening to vocoded speech. The children in the current study scored 20-percentage points lower, on average, for monosyllabic word recognition with 12 spatially selective channels as compared to NH children with a 12-channel simulation (Dorman et al., 2000). Additionally, the children in the current study scored 10-percentage points lower, on average, for monosyllabic words in the “all on” 16-of-m condition as compared to NH children listening to a 20-channel simulation (Dorman et al., 2000). Comparing the current findings to NH vocoded data reported by Eisenberg et al. (2000), our participants scored 15-percentage points lower, on average, for monosyllabic word recognition with 16 spatially selective channels as compared to their group of NH children with a 16-channel simulation. These across-group comparisons (CI vs NH) yielded much smaller performance differences between the NH and CI groups as compared to the CI vs NH evaluation described by Dorman et al. (2000). In particular, Dorman et al. (2000) compared NH pediatric vocoder data to a group of 13 “early-implanted” children whose mean age at implantation was 3.3 years. The early-implanted children scored 51%, on average, for the multisyllabic lexical neighborhood test (MLNT) (Kirk et al., 1995), whereas NH children achieved MLNT scores of 34% and 79% in the 4- and 8-channel conditions, respectively. This narrowing of the performance gap between the NH children (Dorman et al., 2000) and modern-day pediatric CI recipients is consistent with auditory performance gains over time for children with CIs. Performance gains over time are thought to be due to universal newborn hearing screening affording earlier implantation and greater residual hearing prior to implantation (e.g., Teagle et al., 2019) as well as electrode array design improvements and various speech coding and pre-processing strategies currently available.

4.3. Limitations and future study considerations

The current study had a relatively small sample size and included children with different scalar electrode locations whose performance was approaching ceiling effects for speech recognition in noise. Thus, additional studies with larger samples, homogeneous scalar electrode location, and more difficult speech recognition measures would provide valuable information. Additionally, the current study increased overall stimulation rate with channels and included differences in electrode frequency allocation across conditions; future investigations controlling for overall stimulation rate via manipulation of interpulse interval as well as electrode-specific frequency mismatches would also be warranted. Addtionally, all conditions were tested acutely with limited training prior to data collection. Though the previously referenced studies had also assessed performance acutely, future investigations should investigate the effects of longer-term familiarity with the different channel conditions prior to experimentation. The current study also did not include tasks designed to evaluate speech and auditory processing that are highly dependent upon functional spectral resolution such as vowel recognition or tasks of across-channel spectral resolution such as spectral modulation detection or spectral ripple discrimination. Future studies should include both CIS and n-of-m strategies with the addition of across-channel measures of spectral resolution—the latter of which would be particularly informative in the comparison of 8-of-m vs 16-of-m. Note that the number of active CI channels as measured here reflects independent spectral information encoded in the peripheral auditory system and transmitted to the central auditory system for processing of the speech stimuli. In the current experimental design using behavioral measures of speech recognition, however, it is not possible to disentangle the number of spectral channels encoded in the periphery as compared to the number of spectral channels required for speech understanding at the level of the central auditory system. Finally, while the current study limited enrollment to children with precurved electrode arrays, future studies should also include lateral wall or straight arrays given the global prevalance of this electrode type.

5. Summary

Consistent with our hypothesis, children with precurved electrode arrays achieved performance plateau with 12 or more spatially selective electrodes for monosyllabic word recognition and sentence recognition in noise. Pediatric CI recipients also exhibited a statistically significant effect of 16 vs 8 maxima for monosyllabic word recognition in an n-of-m strategy. Thus, the current results are consistent with a recent study of adult CI recipients with precurved electrode arrays in ST (Berg et al., 2019), such that both children and adults derive speech recognition benefits from more than 5–8 channels—which is in contrast to the results presented in the classic CI literature. The current findings hold clinical implications for pediatric CI candidates and recipients, including electrode design, selective channel activation, and audiologic programming regarding selective electrode deactivation and default spectral maxima for n-of-m strategies. Specifically, clinical implementation of selective electrode deactivation for children with precurved electrode arrays should consider no fewer than 12 active electrodes spaced along the array with potential for additional gain up to 16 electrodes in a CIS strategy or up to 18–22 electrodes and 16 maxima in an n-of-m strategy.

Acknowledgments

The authors would like to thank Antonio “Drew” Schefano, Noah Whittenbarger, and Robert Dwyer, AuD, for their help with CT scans and image processing. The project is supported by Grant Nos. R01 DC017683 [co-principal investigators (PIs): R.H.G. and S.C.], R01 DC014037 (PI: J.H.N.), and R01 DC014462 (PI: B.M.D.) from the National Institutes of Health (NIH). The content is solely the responsibility of the authors and does not necessarily reflect the official views of the NIH.

a)

Portions of this dataset were presented at the 2019 Conference on Implantable Auditory Prostheses (CIAP), Tahoe City, CA, July 14–19, 2019.

References and links

  • 1. Berg, K. A. , Noble, J. H. , Dawant, B. M. , Dwyer, R. T. , Labadie, R. F. , and Gifford, R. H. (2019). “ Speech recognition as a function of the number of channels in perimodiolar electrode recipients,” J. Acoust. Soc. Am. 145(3) 1556–1564. 10.1121/1.5092350 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2. Berg, K. A. , Noble, J. H. , Dawant, B. M. , Dwyer, R. T. , Labadie, R. F. , and Gifford, R. H. (2020). “ Speech recognition with cochlear implants as a function of the number of channels: Effects of electrode placement,” J. Acoust. Soc. Am. 147(5), 3646–3656. 10.1121/10.0001316 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3. Berg, K. A. , Noble, J. H. , Dawant, B. M. , Dwyer, R. T. , Labadie, R. F. , and Gifford, R. H. (2021). “ Speech recognition as a function of the number of channels for an array with large inter-electrode distances,” J. Acoust. Soc. Am. 149(4), 2752–2763. 10.1121/10.0004244 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4. Croghan, N. B. H. , Duran, S. I. , and Smith, Z. M. (2017). “ Re-examining the relationship between number of cochlear implant channels and maximal speech intelligibility,” J. Acoust. Soc. Am. 142(6), EL537–EL543. 10.1121/1.5016044 [DOI] [PubMed] [Google Scholar]
  • 5. Davis, T. J. , Zhang, Y. , Gifford, R. H. , Dawant, M. , Labadie, F. , Yzjack, Ã. , and Noble, H. (2016). “ Relationship between electrode-to-modiolus distance and current levels for adults with cochlear implants,” Otol. Neurotol. 37(1), 31–37. 10.1097/MAO.0000000000000896 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6. Dorman, M. F. , Loizou, P. C. , Kemp, L. L. , and Kirk, K. I. (2000). “ Word recognition by children listening to speech processed into a small number of channels: Data from normal-hearing children and children with cochlear implants,” Ear Hear. 21(6), 590–596. 10.1097/00003446-200012000-00006 [DOI] [PubMed] [Google Scholar]
  • 7. Eisenberg, L. S. , Shannon, R. V. , Martinez, A. M. , and Boothroyd, A. (2000). “ Speech recognition with reduced spectral cues as a function of age,” J. Acoust. Soc. Am. 107(5), 2704–2710. 10.1121/1.428656 [DOI] [PubMed] [Google Scholar]
  • 8. Fishman, K. E. , Shannon, R. V. , and Slattery, W. H. (1997). “ Speech recognition as a function of the number of electrodes used in the SPEAK cochlear implant speech processor,” J. Speech Lang. Hear. Res. 40(5), 1201–1215. 10.1044/jslhr.4005.1201 [DOI] [PubMed] [Google Scholar]
  • 9. Friesen, L. M. , Shannon, R. V. , Baskent, D. , and Wang, X. (2001). “ Speech recognition in noise as a function of the number of spectral channels: Comparison of acoustic hearing and cochlear implants,” J. Acoust. Soc. Am. 110(2), 1150–1163. 10.1121/1.1381538 [DOI] [PubMed] [Google Scholar]
  • 10. Goehring, T. , Archer-Boyd, A. W. , Arenberg, J. G. , and Carlyon, R. P. (2021). “ The effect of increased channel interaction on speech perception with cochlear implants,” Sci. Rep. 11(1), 10383. 10.1038/s41598-021-89932-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Goldwyn, J. H. , Bierer, S. M. , and Bierer, J. A. (2010). “ Modeling the electrode-neuron interface of cochlear implants: Effects of neural survival, electrode placement, and the partial tripolar configuration,” Hear. Res. 268(1), 93–104. 10.1016/j.heares.2010.05.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12. Jeon, E. K. , Turner, C. W. , Karsten, S. A. , Henry, B. A. , and Gantz, B. J. (2015). “ Cochlear implant users' spectral ripple resolution,” J. Acoust. Soc. Am. 138(4), 2350–2358. 10.1121/1.4932020 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Kirk, K. I. , Pisoni, D. B. , and Osberger, M. J. (1995). “ Lexical effects on spoken word recognition by pediatric cochlear implant users,” Ear Hear. 16(5), 470–481. 10.1097/00003446-199510000-00004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. Litvak, L. M. , Spahr, A. J. , and Emadi, G. (2007). “ Loudness growth observed under partially tripolar stimulation: Model and data from cochlear implant listeners,” J. Acoust. Soc. Am. 122(2), 967–981. 10.1121/1.2749414 [DOI] [PubMed] [Google Scholar]
  • 15. Nittrouer, S. , and Boothroyd, A. (1990). “ Context effects in phoneme and word recognition by young children and older adults,” J. Acoust. Soc. Am. 87(6), 2705–2715. 10.1121/1.399061 [DOI] [PubMed] [Google Scholar]
  • 16. Noble, J. H. , Hedley-Williams, A. , Sunderhaus, L. , Dawant, B. , Labadie, R. F. , Camarata, S. , and Gifford, R. H. (2016). “ Initial results with image-guided cochlear implant programming in children,” Otol. Neurotol. 37(2), e63–e69. 10.1097/MAO.0000000000000909 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17. Padilla, M. , and Landsberger, D. M. (2016). “ Reduction in spread of excitation from current focusing at multiple cochlear locations in cochlear implant users,” Hear. Res. 333, 98–107. 10.1016/j.heares.2016.01.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18. Peterson, G. E. , and Lehiste, I. (1962). “ Revised CNC lists for auditory tests,” J. Speech Hear. Disord. 27, 62–72. 10.1044/jshd.2701.62 [DOI] [PubMed] [Google Scholar]
  • 19. Schvartz-Leyzac, K. C. , Zwolan, T. A. , and Pfingst, B. E. (2017). “ Effects of electrode deactivation on speech recognition in multichannel cochlear implant recipients,” Cochlear Implants Int. 18(6), 324–334. 10.1080/14670100.2017.1359457 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Shannon, R. V. , Cruz, R. J. , and Galvin, J. J. (2011). “ Effect of stimulation rate on cochlear implant users' phoneme, word and sentence recognition in quiet and in noise,” Audiol. Neurotol. 16(2), 113–123. 10.1159/000315115 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. Smeds, K. , Wolters, F. , and Rung, M. (2015). “ Estimation of signal-to-noise ratios in realistic sound scenarios,” J. Am. Acad. Audiol. 26(2), 183–196. 10.3766/jaaa.26.2.7 [DOI] [PubMed] [Google Scholar]
  • 22. Spahr, A. J. , Dorman, M. F. , Litvak, L. M. , Cook, S. J. , Loiselle, L. M. , Dejong, M. D. , Hedley-Williams, A. , Sunderhaus, L. S. , Hayes, C. A. , and Gifford, R. H. (2014). “ Development and validation of the pediatric AzBio sentence lists,” Ear Hear. 35(4), 418–422. 10.1097/AUD.0000000000000031 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Studebaker, G. A. (1985). “ A ‘rationalized’ arcsine transform,” J. Speech. Lang. Hear. Res. 28(3), 455–462. 10.1044/jshr.2803.455 [DOI] [PubMed] [Google Scholar]
  • 24. Teagle, H. F. B. , Park, L. R. , Brown, K. D. , Zdanski, C. , and Pillsbury, H. C. (2019). “ Pediatric cochlear implantation: A quarter century in review,” Cochlear Implants Intl. 20(6), 288–298. 10.1080/14670100.2019.1655868 [DOI] [PubMed] [Google Scholar]
  • 25. Thornton, A. R. , and Raffin, M. J. (1978). “ Speech-discrimination scores modeled as a binomial variable,” J. Speech Hear. Res. 21(3), 507–518. 10.1044/jshr.2103.507 [DOI] [PubMed] [Google Scholar]
  • 26. Warner-Czyz, A. D. , Houston, D. M. , and Hynan, L. S. (2014). “ Vowel discrimination by hearing infants as a function of number of spectral channels,” J. Acoust. Soc. Am. 135(5), 3017–3024. 10.1121/1.4870700 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27. Wilson, B. S. , Finley, C. , Lawson, D. T. , Wolford, R. D. , Eddington, D. K. , and Rabinowitz, W. M. (1991). “ Better speech recognition with cochlear implants,” Nature 352(6332), 236–238. 10.1038/352236a0 [DOI] [PubMed] [Google Scholar]
  • 28. Won, J. H. , Humphrey, E. L. , Yeager, K. R. , Martinez, A. A. , Robinson, C. H. , Mills, K. E. , Johnstone, P. M. , Moon, I. J. , and Woo, J. (2014). “ Relationship among the physiologic channel interactions, spectral-ripple discrimination, and vowel identification in cochlear implant users,” J. Acoust. Soc. Am. 136(5), 2714–2725. 10.1121/1.4895702 [DOI] [PubMed] [Google Scholar]
  • 29. Zhao, Y. , Dawant, B. M. , Labadie, R. F. , and Noble, J. H. (2018). “ Automatic localization of closely spaced cochlear implant electrode arrays in clinical CTs,” Med. Phys. 45(11), 5030–5040. 10.1002/mp.13185 [DOI] [PMC free article] [PubMed] [Google Scholar]

Articles from Jasa Express Letters are provided here courtesy of American Institute of Physics

RESOURCES