Reducing Channel Interaction Through Cochlear Implant Programming May Improve Speech Perception

Subjects

Seven (eight ears) postlingually deafened adults along with and one peri-lingually deafened (S40) adult, all of whom wore the Advanced Bionics HiRes90K CI, participated. Their details are shown in Table 1. One postlingually deaf subject, who was bilaterally implanted, was tested in each ear and is identified as S23-L and S36-R. This subject’s two ears are treated separately, and therefore, for the purposes of analysis, nine subjects are considered. The Human Subjects Review Board at the University of Washington approved all procedures.

OPEN IN VIEWER

Table 1.

Subject Demographics.

Subject ID	Age (years)	Duration of deafness (years)	Duration of CI use (years)	Etiology	Average performance with everyday strategy (% correct)	Electrode array type
S22	75	18	9	Unknown (genetic)	72	Helix
S23 (L)	71	18	8	Unknown	84	HiFocus 1J
S28	76	43	7	Autoimmune	45	HiFocus 1J
S30	52	22	11	Genetic	97	HiFocus 1J
S36 (R)	71	17	6	Unknown	74	HiFocus 1J
S38	51	5	5	Otosclerosis	44	HiFocus 1J
S40	53	49	3	Enlarged vestibular aqueduct	31	HiFocus 1J
S42	65	33	14	Unknown	96	HiFocus 1J with positioner
S43	70	18	2	Unknown	56	Mid-scala

Stimuli for Detection Threshold

All threshold estimations were conducted with the pTP electrode configuration. Biphasic, charge-balanced, cathodic phase first pulse trains were used. Phase durations were 97 µs and the pulse rate was 997 pulses per second. Pulse trains were 200 ms in duration. These pulse settings are consistent with what have been used previously (e.g., Bierer, 2007; Bierer et al., 2015). Stimuli were presented in the pTP configuration with a return current fraction (σ) of 0.9, which allowed for focused stimulation while remaining within the voltage compliance limits of the device. All stimuli were presented and controlled using research hardware and software (“BEDCS”) provided by Advanced Bionics (version BEDCS 18, 1.18.315, Valencia, CA, USA). Programs were written using the Matlab programming environment, which controlled low-level BEDCS routines. Prior to subject testing, all stimuli were checked using a test implant and digital storage oscilloscope.

Signal detection thresholds were measured using an adaptive three-down, one-up, two-interval forced-choice procedure that converged on 79% correct (Levitt, 1971). Step size was 1 dB for the first two reversals and 0.25 dB thereafter. The mean of the last four of six turnpoints was used to estimate threshold. Subjects were asked, “Which interval contained the sound?” and responded using a computer mouse. Four or five repetitions were performed and averaged for each measurement. Thresholds were measured for all available channels (usually 2 through 15).

Channel Selection

Channel selection for deactivation was based on the behavioral thresholds obtained with the pTP, σ = 0.9 electrode configuration. The following procedure was followed in the order described. A contrast enhancement filter (space constant of four electrodes, and gain of four) was applied to the threshold data to enhance the differences between peaks and troughs. We first applied a symmetric high-pass filter to the threshold profiles (with normalized coefficients of −.14, −.18,1, −.18, −.14, modified to account for electrodes at the ends of the array) to accentuate differences in threshold between each electrode and its apical and basal neighbors. The mean and standard deviation were calculated and a criterion was established such that more channels were selected for deactivation if the mean and standard deviation were high. Channels were dropped progressively until the standard deviation of the enhanced threshold data for the remaining channels was less than a subject-specific criterion or such that the total number of the remaining channels was at least eight. For each subject, the criterion was set based on the average threshold. Specifically, the criterion equaled 3 dB if the average threshold was 42 or below, and 1.5 dB if the average level was 48 or above. The distribution of thresholds for all of the subjects tested with these stimuli had a mean of approximately 45 dB re 1 uA +/− 5 dB. One-half of the standard deviation was used above and below the mean to set these criteria. For subjects with average thresholds between 48 dB and 42 dB, the criterion was set as −(A−42)/(48−42) × 1.5 + 3. The subject-specific criterion was added so that more channels were dropped for subjects in whom the tripolar thresholds were higher.

Channels with high thresholds were selected in the following manner. The number of channels selected ranged from 1 to 6. If the algorithm selected two channels in a row, both channels were deactivated. If three channels in a row were identified, however, then the middle channel was not selected to maintain stimulation in that cochlear region. An equal number of relatively low-threshold channels were selected manually from the same region(s) (apical 2–6, middle 7–10, basal 11–15) as the high-threshold channels when possible, to serve as a control condition.

Experimental Mapping Procedure

Six different experimental programs were created for each subject using BEPS + software (version 1.10.19.27375, Advanced Bionics Corp., Valencia, CA, USA). All experimental programs were created on a Harmony research processor dedicated for use in the laboratory. The first two strategies were 14-channel programs including electrodes 2 through 15. Electrodes 1 and 16 were excluded because they could not serve as active electrodes with focused stimulation. Programs were created either in the pTP or MP configurations and are referred to as “pTP-all” and “MP-all,” respectively. The pTP-all program was created first because higher current levels are required for focused strategies and the software was set to automatically adjust the pulse duration and rate to reduce the power requirements (i.e., pulses that are longer in duration but smaller in amplitude have equal charge to short duration and larger amplitude but require less power). The goal was to create a pTP program with the highest degree of focusing possible for most subjects and channels; therefore, a σ = 0.875 was targeted for each electrode, for each subject. For some subjects and channels, the most comfortable listening level could not be achieved with that degree of focusing and sigma was reduced to either 0.75 or 0.625. Table 2 shows how often the sigma was changed for some channels in each subject. The final pulse duration and stimulation rate are also listed in Table 2. Once the pulse duration and stimulation rate was set for the pTP-all program, the same pulse duration was used for the remaining five programs. This relatively slow rate could have reduced the performance with all of the experimental programs but was kept consistent to eliminate rate as a confounding factor for performance.

OPEN IN VIEWER

Table 2.

Experimental Program Details.

Subject ID	Number of channels with a σ < 0.875	Pulse width (µsec/phase)	Pulse rate (pulses/ second)
S22	5; σ = 0.75	50.3	710
S23 (L)	5; 4 with σ = 0.7 and 1 σ = 0.625	192.2	200
S28	1; σ = 0.75	65.6	545
S30	0	141	253
S36 (R)	0	88.9	402
S38	7; σ = 0.75	88	406
S40	10; σ = 0.75	49.4	723
S42	0	36.8	970
S43	0	68.2	523

The all-channel programs were used as baseline strategies for the reduced-channel programs. Channels were selected for deactivation based on pTP threshold. The reduced-channel program using focused stimulation with the relatively high-threshold channels deactivated is referred to as “pTP high-off.” The other focused program with relatively low-threshold channels deactivated is referred to as “pTP low-off.” The “MP high-off” and “MP low-off” reduced-channel programs were created in the same manner using the MP-all as a baseline program. When channels were deactivated, both the overall pulse rate and per channel pulse durations were held constant to match the all-channel strategies across stimulation type. As a result, the inter-pulse interval was increased when the number of active channels was reduced.

To create each experimental program, threshold was estimated manually for each channel to the level in which the subject first heard the signal. Then, most comfortable level was estimated using a subjective loudness rating scale provided by Advanced Bionics that goes from 0 to 10. Stimulation level was increased until the subject indicated a rating of 7 described by Loud but comfortable and then the level was decreased until the subject indicated it was back to a 6 described as Most comfortable. Stimulus levels were loudness balanced at most comfortable level MCL in sets of four electrodes beginning with Electrode 2. The subject was instructed to inform the audiologist if the sounds were equally loud. Adjustments were made accordingly until those channels were perceived as not equally loud. The next set of electrodes was then loudness balanced with one electrode overlapping from the previous set until all 14 channels were balanced (2 through 5, 5 through 8, 8 through 11, and 11 through 14).

The following procedures were performed for each processor program. Once all channels were equally loud, the volume was reduced on the processor, and the microphone was activated. The volume was gradually increased until it was at 0 gain or 12 o’clock on the dial. The subject was then asked to describe the overall volume of speech and the M-levels were globally adjusted until the subject indicated it was comfortable. Fine-tuning was conducted to optimize the stimulation levels for speech sounds. For instance, if the subject indicated the sound of their own voice was loud then the M levels were decreased for Electrode 2. Likewise, if the program had too much high frequency sound the M-level was reduced for Electrode 15. The Ling sounds (“ah,” “ee,” “oo,” “mm,” “sh,” “ss”) were presented through a screen to ensure audibility across the speech frequencies. If the Ling sounds were confused, minor adjustments in M-levels were made with the goal of correct identification of the sounds with each program.

Prior to speech discrimination testing, subjects were given approximately 20 min of real-world listening experience with each program. This consisted of going outside for a walk or going to a nearby coffee shop with the audiologist. Each listener made six or seven visits to the laboratory, each visit lasting 3 to 4 h. The order of testing with the experimental programs was randomized for each listener.

Speech Discrimination

Two speech discrimination tests were administered; 16 medial consonants were presented in the “ah” context using a 16-choice closed-set task (/p/, “aPa”; /t/, “aTa”; /k/, “aKa”; /b/, “aBa”; /d/, “aDa”; /ɡ/, “aGa”; /f/, “aFa”; /θ/, “aTHa”; /s/, “aSa”; /ʃ/, “aSHa”; /v/, “aVa”; /z/, “aZa”; /dʒ/, “aJa”; /m/, “aMa”; /n/, “aNa”; /l/, “aLa”) (Shannon, Jensvold, Padilla, Robert, & Wang, 1999; Tyler, Preece, & Tye-Murray, 1986). Medial vowels were in the “hVd” context in a 10-choice closed-set task (/i/, “heed”; /ɪ/, “hid”; /eɪ/, “hayed”; /ɛ/, “head”; /æ/, “had”; /ɑ/, “hod”; /u/, “who’d”; /ʊ/, “hood”; /o/, “hoed”; /ʌ/, “hud”). Vowel stimuli were recorded for these experiments with one male and one female Pacific Northwest talker, as regional dialect has been found to influence recognition of vowel sounds (Wright & Souza, 2012). A head-mounted close talking microphone was used to record vowel sounds in a double-walled sound-treated booth. Recordings were digitized at 44.1 kHz using 16-bit quantization and were resampled to 22.5 kHz. Listeners were given one practice set, where each token was presented three times; listeners could repeat the token and were given feedback. Following the practice set, two more sets of three repetitions were completed. If the average score of the two sets differed by more than 10%, a third set was run and all three were averaged to determine the percent correct. If listeners performed better than 70% correct on vowel or consonant identification in quiet, testing was performed with four-talker babble noise (Auditek) at a + 10 signal-to-noise ratio. Speech scores were then converted to rational arcsine units (rau; Studebaker, 1985).

Stimuli were presented through an external A/D device (SIIG USB SoundWave 7.1), amplified by a Crown Amplifier (D75) and presented at 60 dB sound pressure level (SPL) in the sound field inside a double-walled sound attenuating both. The sound files were presented from a desktop PC using custom software (ListPlayer2 version 2.2.11.52, Advanced Bionics). The stimuli were calibrated to a 1 kHz tone with a sound level meter (Bruel and Kjaer, Hand-held Analyzer Type 2250 and ZC 0032 microphone) and presented through a loudspeaker (Bose 161) placed at ear level height, at 0° azimuth and 1 m from the subjects’ head.

Performance Questionnaire

Following the testing with each experimental strategy, subjects were asked to complete a questionnaire that involved rating the sound quality and clarity on a scale from 1 to 10 in comparison to their everyday listening strategy. A rating of “1” was considered worse than their everyday strategy, a rating of “10” was considered better, and “5” was considered equivalent. Subjects were blinded to the specific programs that were tested and did not see their scores following testing with each program.

Focused Stimulation: Perceptual Measures

Consistent with previous studies, focused thresholds were high and variable across the CI array (Bierer, 2007; Bierer & Faulkner, 2010; Long et al, 2014; Pfingst et al., 2004). Previous studies suggest that elevated thresholds are an indication of a poor electrode-neuron interface either because of a large distance between the stimulating electrodes and the target neurons or because of poor neural health (Bierer & Faulkner, 2010; DeVries et al., 2016; Goldwyn et al., 2010; Long et al., 2014). Focused thresholds are often correlated with distance between the electrodes and the inner wall of the cochlea; therefore, it is difficult to separate the contributions of neural health to focused thresholds (e.g., Long et al., 2014).

It is well established that CI listeners have significant channel interactions. Interactions have been quantified with broad tuning (Bierer & Faulkner, 2010; Chatterjee, Galvin, Fu, & Shannon, 2006; Nelson, Donaldson, & Kreft, 2008; Shannon, Zeng, Kamath, Wygonski, & Ekelid, 1995), indicating an inability of implant listeners to access independent spectral information as well as normal hearing listeners (for review, see Shannon, Fu, & Galvin, 2004). In addition, greater channel interaction and poor spectral resolution have also been correlated and vary across the array within individual listeners (Anderson, Nelson, Kreft, Nelson, & Oxenham, 2011; Jones et al., 2013).

Channel Deactivation

In Experiment 2, another method used to reduce channel interaction was to deactivate a subset of channels from listeners’ programs. This method of channel deactivation has shown some success for improved speech perception and spectral resolution in previous studies (Garadat et al., 2012; Noble et al., 2013, 2014; Saleh et al., 2013). However, Noble et al. (2013, 2014) did not provide either psychophysical or electrophysiological evidence to support that the channels selected for deactivation were those with high levels of channel interaction. Indeed, the results from the present study suggest that the choice of which channels should be deactivated may not be critical because both reduced-channel MP programs were beneficial for many listeners, especially those with poorer speech perception performance. The improvements observed by Garadat et al. (2012) may not have been observed in the present study because we did not require the selection of a single channel in each of the quadrants of the CI array. In the Garadat study, the channels were deactivated based on a temporal resolution measure and the relationship between modulation detection and focused thresholds is yet unknown.

In the second experiment of the present study, one hypothesis tested was if improvements beyond focusing could be achieved by channel deactivation in the tripolar strategies. In fact, rather than improved speech perception, the scores were generally unchanged or even reduced by channel deactivation. Because focused thresholds alone were used to select channels and the contributions of electrode position and neural health to the elevated thresholds have not been differentiated, it might be that deactivating channels with focused stimulation simply reduces the number of neurons activated. Consider the model shown in Figure 8. In the top two panels with electrical fields created in the MP configuration, even with a reduced number of active channels, the number of neurons activated is still complete, whereas in the lower two panels with fields created in the pTP configuration, there are neurons available that might not ever be stimulated. OPEN IN VIEWER

The small improvements observed in speech perception tasks for listeners using programs designed to reduce channel interaction largely did not reach statistical significance. Part of the reason for this could be that a small sample of subjects participated in the experiments and only the poorer performing subjects seemed to benefit. In previous studies, the poorer performing listeners have typically shown a higher degree of channel interaction (e.g., Anderson et al., 2011; Jones et al., 2013), elevated thresholds (Long et al., 2014), or small evoked potential peak amplitudes (DeVries et al., 2016). This may be the reason those listeners, in particular, benefit from focusing and channel-deactivation. In addition, previous studies allowed listeners to acclimate to the reduced-channel programs for one month (Noble et al., 2013, 2014). When channels are deactivated, the frequencies are reallocated to the available channels and this change in the channels which carry certain sound frequencies can take some time for listeners to adapt (e.g., Fu, Shannon, & Galvin, 2002). It is possible that in the present study, experience with each strategy might have enhanced the effects of reducing channel interaction and led to better performance. Future studies will examine the effects of listening experience with strategies that reduce channel interaction.