The relationship between time and place coding with cochlear implants with long electrode arrays

Abstract

The auditory system can theoretically encode frequencies by either the rate or place of stimulation within the cochlea. Previous work with cochlear implants has demonstrated that both changes in timing and place can be described as pitch changes but are perceptually orthogonal. Using multidimensional scaling, the present experiment extends the previous findings that timing and place changes are perceptually orthogonal into the cochlear apex using long 31-mm electrode arrays. However, temporal cues seem to be more reliable across subjects at the apex while place cues seem to be more reliable at the middle of the cochlea.

1. Introduction

Previous work with cochlear implants has demonstrated that at a fixed rate of stimulation on a single electrode, changing the electrode providing the stimulation is described as changing the pitch of the sound (e.g., Eddington et al., 1978). Electrodes which are placed more apically in the cochlea are reported to provide a lower pitch than electrodes which are placed more basally. Similarly, higher stimulation rates on a single electrode are reported as having a higher pitch than lower rates on the same electrode (e.g., Tong et al., 1983). When changes in rate and place are combined in the same direction (i.e., rate increases while stimulation shifts basally or rate decreases while stimulation shifts apically), the corresponding pitch change is increased relative to changing only rate or place in isolation. Furthermore, when a change in rate and place are combined in opposite directions (i.e., rate increases while stimulation shifts apically or rate decreases while stimulation shifts basally), the pitch shift is smaller than when changing only the rate or place of stimulation (Luo et al., 2012). This suggests that the two different physical manipulations each influence pitch. This idea is reinforced by Landsberger et al. (2016) who provided a pitch scaling map of rates between 100 and 1500 pulses per second (pps) at most electrode locations which shows the trade-off between rate and place on pitch-scaling.

With one subject, Tong et al. (1983) used a multidimensional scaling (MDS) technique to demonstrate that although rate and place have an influence on pitch, they represent orthogonal perceptual dimensions. This suggests that although rate and place coding may be used to encode pitch, rate and place are not perceptually interchangeable. Macherey et al. (2011) conducted a similar experiment yielding similar conclusions, although the place dimension was represented using an atypical stimulation mode, pseudo-monophasic bipolar stimulation. McKay et al. (2000) provided difference limen measurements that suggest rate change and place change cues were used independently on the task. It is interesting to contrast the McKay et al. (2000) results in which listeners independently use rate and place change cues with the Luo et al. (2012) finding that listeners are willing and able to combine rate and place change cues when making pitch judgements.

Experiments examining the relationship between rate and place changes with cochlear implants have previously been conducted with users of electrode arrays that are designed to be placed approximately only into the first (basal-most) cochlear turn. Therefore, it remains unknown if the relationship between rate coding and place coding at locations beyond the first cochlear turn is similar to the relationship at locations in the first cochlear turn. Research with either long electrode arrays inserted well into the second cochlear turn or with current shaping techniques pushing current deeper into the cochlea than the location of the most apical electrode of the array suggest that temporal and spectral coding may be different at different parts of the cochlea. For example, previous data have suggested that stimulation in the second cochlear turn provides better temporal discrimination (Macherey et al., 2011; Stahl et al., 2016) than stimulation in the first cochlear turn. Similarly, the low rates of stimulation which may encode temporal information have been described as sounding better in the second cochlear turn than the first cochlear turn (Landsberger et al., 2016). On the other hand, the pitch (e.g., Kenway et al., 2015) and sound quality (Landsberger et al., 2014) of adjacent electrodes in the second cochlear turn are often quite similar.

In the present experiment, a MDS technique is used to measure the perceptual spaces defined by a change in rate and place coding using contacts typically placed in either the first (MED-EL 31 mm array electrodes 6–8) or second (MED-EL 31 mm array electrodes 1–3) cochlear turns (e.g., Landsberger et al., 2015). The purpose of the experiment is to determine if there is a difference between perceptual spaces defined by rate and place coding in the two cochlear regions. The protocol used in this experiment is based on one used in Tong et al. (1983) with one subject. Therefore, a secondary purpose of the experiment is to replicate the findings of Tong et al. (1983) with more than the one subject to verify that the results generalize to a greater population of implant users.

2. Methods

2.1. Subjects

Six subjects with MED-EL cochlear implants and 31-mm electrode arrays (either Standard or FLEX^SOFT) participated in this experiment. All subjects provided informed consent in accordance with the IRB regulations for the University Hospital of Antwerp in Antwerp, Belgium. Specific demographics are presented in Table 1.

Table 1.

Subject demographics. For subjects UZA-M8 and UZA-M9, the onset of deafness is unknown so the onset of hearing loss is presented.

ID	Gender	Age	Onset deafness	Implant	Electrode	Strategy	Etiology
			**Onset hearing loss
UZA-M1	m	68	Unknown	Combi 40+	Standard	FS4	unknown
UZA-M3	m	35	12/27/2010	Sonata	Standard	FS4	bact. meningitis
UZA-M4	f	58	1993	Combi 40+	Standard	FSP	Meniere's disease
UZA-M7	m	70	Unknown	Pulsar	FlexSOFT	FS4	unknown
UZA-M8	f	60	1983**	Sonata	FlexSOFT	FSP	unknown
UZA-M9	m	61	1964**	Pulsar	Standard	FS4	unknown

2.2. Stimuli

Stimuli consisted of 600 ms biphasic cathodic-first single electrode pulse trains in monopolar mode. Phase durations were 50 μs without an inter-phase gap. All stimuli were presented at 5000 pps to ensure an adequate carrier rate for all modulation rates tested in this experiment. Amplitude-modulated stimuli had a 75% modulation depth. Stimuli were presented directly to the implant (bypassing the speech processor) using custom software on a windows computer and the RIB2 (Research Interface Box 2, University of Innsbruck) research interface.

Six electrodes were used representing apical (electrodes 1–3) and middle (electrodes 6–8) regions of the electrode array. Post-operative x-rays were not available to verify the insertion angles for each contact for each subject. However, the average insertion for apical electrodes 1, 2, and 3 are 636°, 552°, and 482° and the average insertion for middle electrodes 6, 7, and 8 are 295°, 247°, and 200° for patients implanted with the same electrode arrays at the same hospital by the same surgeon (e.g., Landsberger et al., 2016). Presumably these estimates of electrode position are similar to the average electrode position of the subjects in the present study.

2.3. Procedure

Before running the experiment, a rough dynamic range estimate was made for 5000 pps unmodulated pulse trains on each of the tested electrodes (1, 2, 3, 6, 7, and 8). Stimuli were presented sub-threshold and gradually raised in 5 μa steps until the maximum acceptable loudness was reported. As the amplitude was raised, subjects reported the loudness of the corresponding sound using a loudness scale provided by Advanced Bionics. Amplitudes corresponding to “Barely Audible,” “Soft,” “Most Comfortable,” “Loud But Comfortable,” and “Maximal Comfort” were recorded. After estimating the dynamic range, stimulation on electrodes 1, 2, 3, and 6 were presented at amplitude halfway between “Most Comfortable” and “Loud but Comfortable” and the subjects were asked to report any loudness differences between the stimuli. If all of the stimuli were not equally loud, the amplitudes were tweaked. The process was then repeated for stimulation on electrodes 3, 6, 7, and 8 until stimulation on all electrodes provided equally loud stimulation. The equally-loud amplitude for each electrode was used as the peak amplitude for the amplitude modulation (AM) stimuli used in the MDS technique.

A typical MDS protocol (e.g., Tong et al., 1983; Macherey et al., 2011; Landsberger et al., 2014) was used for two sets of stimuli to measure the perceptual dissimilarity between each of the stimuli in the set. There was a total of 9 stimuli in each set consisting of 3 electrodes providing stimulation at 3 AM rates (75, 150, or 300 Hz). One set of stimuli (the “apical” set) used electrodes 1–3 while the other set of stimuli (the “middle” set) used electrodes 6–8. The dissimilarities for one set were measured in its entirety before evaluating the dissimilarities in the other set. However, the order of sets evaluated was randomized across subjects.

Within a trial, two randomly-selected stimuli within a set were presented with a 300-ms inter-stimulus interval. Subjects were asked to scale how different the two sounds were by clicking on a line on the computer screen which represented a continuum from “Most Similar” (at the extreme left of the line) to “Most Different” (at the extreme right of the line). After each response, the line was moved to a new position on the screen to prevent a subject from clicking on the same location for each response. In a block of trials, all nine stimuli were compared to each other stimulus twice. In the two trials where a given set of two stimuli are compared, the stimuli were presented in opposite order. Each block consisted of a total 81 trials. Five blocks of trials were measured for each of the two stimuli sets.

3. Results

The perceptual differences between each stimulus were analyzed using a two-dimensional alternating least squares scaling (ALSCAL) algorithm (Young and Lewyckyj, 1979) for each subject and for each set (i.e., the apical and middle electrode sets). Two-dimensional ALSCAL fits provide r² values between 0.532 and 0.959 suggesting that the analyses provide good fits to the data. A paired t-test fails to detect any significant differences between r² values for the apical and middle electrode sets [t(5) = −1.167, p = 0.296]. Plots of the two-dimensional ALSCAL analysis for each subject are presented in Fig. 1. Each column of plots represents an individual subject. The top and bottom rows represent the data for the apical and middle electrode sets, respectively. Individual subject patterns typically demonstrate a dimension of electrode (organized by number) and a dimension of rate (organized by color and symbol). Notable exceptions include UZA-M7 and UZA-M8 whose data are organized by place (number) and not rate (symbol) for the middle electrode set but is organized by rate (symbol) and not place (number) for the apical set. The data for these two subjects are consistent with a horse-shoe shape corresponding to a one-dimensional space as described in many other publications including Kendall (1971) and Landsberger et al. (2014). Additionally, for the apical electrode set, it is worth noting that, consistent with Landsberger et al. (2014) and Kenway et al. (2015), the electrodes are often positioned out of order (e.g., UZA-M7, UZA-M9R).

Fig. 1.

(Color online) Two-dimensional multiple dimensional scaling plots for the apical electrode. Each column of plots represents individual subject data. The top plots represent results from the apical end of the array (electrodes 1, 2, and 3) and the bottom plots represent results from the middle of the array (electrodes 6, 7, and 8). In each of these plots, the perceptual location of each electrode (as indicated by the corresponding number) at each rate (as indicated by the symbol) are shown. Specifically, diamonds represent 75 Hz, triangles represent 150 Hz, and squares represent 300 Hz. The bottom-right corner of each panel indicates the r² fit.

To get an estimate of the perceptual space across subjects, the data were re-analyzed using a bootstrap MDS analysis (Jacoby and Armstrong, 2014). This technique not only allows representation of the dissimilarity ratings in a multidimensional space but also allows an estimation of the confidence region of the position of each stimulus within that space. First for each subject, the dissimilarity matrices for each block were averaged and normalized into an individual symmetrical matrix. Then six of these individual matrices were randomly picked with replacement (the same individual matrix could be picked many times) and averaged into a single mean matrix. This matrix was used to create a two-dimensional Euclidian space using the SMACOF algorithm (Borg and Groenen, 1997). As this solution is rotationally undetermined, the solution was rotated with a procrustean procedure toward a reference space created based on the difference in electrode places (abscissa) and in modulation frequencies (ordinate). This procedure was repeated 300 times in order to create the same number of different MDS solutions. Therefore, the position of each stimulus can be represented by a cloud of 300 points. The dimensions of each cloud were dependent on the variability of the ratings between subjects. A large cloud for a specific stimulus will indicate an unreliable position of this stimulus on the MDS solution. On the other hand, if the six individual matrices were very similar, the clouds of points would be very small. Figure 2 shows the bootstrap MDS solutions for the set of data with the apical and the mid-array electrodes (left and right panel, respectively). Because of how the solutions were rotated, the positions of the stimuli are ordered by place of stimulation (number) along the x axis and by AM rate (color) along the y-axis for both the apical and middle electrode sets, as in Fig. 1. Each cloud of points is represented by an ellipse, in which the orientation of the axes is determined by principal component analysis and their sizes represent the 95% confident interval of the distribution of points along each axes of the ellipse. The center positions of each ellipse for the bootstrap MDS solutions correspond to the position of each stimulus in a classical MDS.

Fig. 2.

(Color online) Two-dimensional multiple dimensional scaling plots averaged across all six subjects for the apical (left panel) and middle (right panel) portions of the array. In each of these plots, the perceptual location of each electrode (as indicated by the corresponding number) at each rate (as indicated by the symbol) are shown. Specifically, diamonds represent 75 Hz, triangles represent 150 Hz, and squares represent 300 Hz. The ellipses correspond to the 95% confidence interval of the location of each point in the two-dimensional space.

4. Discussion

The results of the present experiment expand on Tong et al. (1983) in that a change in rate and a change in place represent independent perceptual dimensions. While the data from Tong et al. (1983) were collected with only one subject, their results have now been replicated with six additional subjects. Furthermore, the results of these experiments confirm that independent perceptual dimensions related to rate and place exist across the cochlea. It is worth noting that the most apical electrode used in Tong et al. (1983) was less than 270° into the cochlea (Clark et al., 1979). Therefore, the placement of electrodes in Tong et al. (1983) were similarly or slightly more basally placed than the middle electrode sets used in the present experiment.

While the results of the current experiment suggest that changes in rate and changes in place are perceived as two orthogonal perceptual dimensions across the cochlea, there are perceptual differences between the apical and middle cochlear locations represented by a long electrode array. For the apical-stimulus set, the ellipses representing the 95% confidence interval are stretched along the horizontal axis, which represents a change in place of stimulation. This suggests greater variability among listeners along this perceptual dimension. This cue should then be used with care while designing a sound processing strategy and can be considered as less reliable than the temporal cue. Conversely, for the middle-electrode set, the effect of the temporal cue will differ among listeners and can be considered as less reliable on average than the place cue. It is worth noting that although many of the ellipses are greatly overlapping (e.g., electrodes 1 and 2 in the apex at 75 or 300 Hz or 150 and 300 Hz in the middle of array), this does not mean that they are very similar perceptually, but instead means that the MDS solution across subjects cannot distinguish them. It is likely that if more than six subjects were tested, the overlaps of the ellipses would be reduced (Jacoby and Armstrong, 2014). Nevertheless, the perceptual similarities between electrodes 1 and 2 with long 31-mm electrode arrays were also observed in other studies including Landsberger et al. (2014) and Kenway et al. (2015). It is plausible that the difference in reliability found between the groups of electrodes along the temporal dimension was influenced by the fact that each of the participants used a fine-structure strategy (e.g., FSP or FS4) in which variable lower rates of stimulation were clinically provided to the most apical electrodes effected the results in the present experiment.

In summary, we have confirmed that rate and place coding are not perceptually interchangeable but are instead perceptually orthogonal. Temporal and place coding are perceived independently in both the cochlear apex and first cochlear turn. However, temporal cues seem to be relatively more reliable at the apex while place cues seem to be relatively more reliable at the middle of the electrode array.