Effect of Stimulus Polarity on Detection Thresholds in Cochlear Implant Users: Relationships with Average Threshold, Gap Detection, and Rate Discrimination

Eight ears from seven post-lingually deafened adults wearing the Advanced Bionics HiRes90K CI were studied; listener details are shown in Table 1. One subject, who was bilaterally implanted, was tested in each ear and is listed as S30L and S39R in the table. This subject’s two ears were treated as completely separate, and therefore, for the purposes of analysis and for discussion in the remainder of this article, there were eight “subjects.” They had all participated in the study by Cosentino et al. (2016) and the same subject codes are used here. Five of the subjects were implanted and tested in Cambridge, UK, whereas the other three were implanted and tested in Seattle, WA, USA. All procedures were approved by the respective Human Subjects Review Boards.

Triphasic pulses were used, in which the amplitude of the central phase was twice that of each of the first and third phases. They are named here in terms of the central high-amplitude phase, which was anodic for stimulus TP-A and cathodic for stimulus TP-C. (Note that here we use the abbreviation “TP” to refer to the triphasic pulse shape rather than to tripolar stimulation mode as in some other articles.) Phase durations were 43 μs, the pulse rate was 99 pulses per second, and the duration of each pulse train was 400 ms. Each subject was tested on the same four or five electrodes as used for the rate discrimination and gap detection measures by Cosentino et al. (2016). All stimuli were presented in monopolar mode. Initially, MCLs were obtained for each combination of subject and electrode, so as to guide the starting level for the threshold measurements and to set a safety limit for those procedures. After each stimulus was presented, the subject indicated its loudness using the Advanced Bionics loudness rating scale for which a “6” is “Most Comfortable” and a “7” is “Loud but Comfortable.” The level was increased gradually until the rating was 7 and then reduced until the listener reported a 6 again. All stimuli were presented and controlled using research hardware and software (“BEDCS”) provided by the Advanced Bionics company. Programs were written using the MATLAB programming environment, which controlled low-level BEDCS routines. Stimuli were checked using a test implant and digital storage oscilloscope. The same software and type of hardware were used at both testing sites (Cambridge and Seattle).

Signal detection thresholds were measured using a three-down, one-up, two-interval forced choice adaptive procedure that converged on 79 % correct. Step size was 1 dB for the first two turnpoints and 0.25 dB thereafter. The mean of the last four of six turnpoints was used to estimate threshold. Four repetitions were performed for each measurement and then averaged. Subjects were asked “Which interval contained the sound?” and responded by selecting a button on a computer screen. Correct-answer feedback was provided at the end of each trial.

To assess between-electrode correlations, we performed univariate ANOVAs with one factor as the dependent variable, the other as a covariate, and with subject entered as a random effect. This revealed a significant positive correlation between the polarity effect and the average of the TP-A and TP-C thresholds (r = 0.49, F(1,24) = 7.59, p = 0.011). Our statistical approach is mathematically equivalent to subtracting each subject’s mean score from every data point, so as to obtain a normalized score, and then correlating these normalized scores. The resulting correlation is shown in Fig. 2. It is in the direction predicted by the hypothesis that lower/negative polarity effects and lower overall thresholds both reflect better survival of the peripheral processes. Note that, although the difference between two scores will a priori be correlated with each score alone, no such a priori relationship holds between the difference and the average of two scores, under the null hypothesis that the variances of each score are equal (Oldham 1962; Tu and Gilthorpe 2007). This is true even when the two scores correlate with each other. For example, if EMD and neural survival combine additively to affect thresholds, then the across-electrode variation in EMD will produce a correlation between TP-A and TP-C thresholds; however, this additive effect will not cause the mean and difference to correlate with each other.

Although there was a significant across-electrode correlation between the polarity effect and the average thresholds, there was no significant correlation between either of them and any of the supra-threshold measures obtained in our previous studies. Those measures were the log of the gap detection thresholds (GDTs) measured by Bierer et al. (2015), and two rate discrimination measures (“RDR100” and “RDR400”) obtained by Cosentino et al. (2016). One of those, RDR100, is the log of the ratio between two just-discriminable pulse rates, where the standard had a rate of 100 pps and the signal, whose rate was always above 100 pps, was adaptively varied to obtain threshold. The other, RDR400, was similar except that the standard had a rate of 400 pps and the signal always had a rate lower than 400 pps. The two measures reflect low-rate discrimination and the upper limit of temporal pitch, respectively. In both cases, low values represent better performance. The between-electrode correlations are shown in Table 2.

We also investigated the effect of the longitudinal electrode position in the array. Bierer (2007) reported a trend for thresholds to decrease from the base to the apex of the electrode array for symmetric biphasic pulses presented in partial tripolar mode, but not in bipolar or monopolar mode. They suggested that this could be due to the progression of spiral ganglion degeneration from base to apex in several pathologies (Zimmermann et al. 1995), and/or to a variation in the EMD along the array. Subsequently, Bierer et al. (2015) found that thresholds decreased from base to apex for symmetric biphasic pulses presented both in partial tripolar and monopolar mode. A similar finding was observed in the present data for the average of the TP-A and TP-C thresholds, as assessed by a univariate ANOVA with average threshold as dependent variable, subject as a random effect, and electrode number as covariate (F(1,24) = 11.41, p = 0.002). However, no such finding was observed for the polarity effect (F(1,24) = 2.76, p = 0.16). Hence, if the polarity effect reflects neural survival, the present results provide no evidence that, for this group of subjects, the effect of longitudinal electrode position on average thresholds was due to a variation in the pattern of neural survival.

Between-subject correlations were obtained by calculating the average score across electrodes for each subject and then performing product-moment correlations. The correlation between the polarity effect and the average threshold was in the same direction as for the between-electrode correlations, but did not reach statistical significance (r = 0.60, df = 6, p = 0.16). The correlations between each of these measures and the supra-threshold scores obtained previously (GDT, RDR100, and RDR400) are shown in Table 3. All of the correlations were positive, three reached statistical significance, and two did so after Bonferroni correction for multiple comparisons. One of these was between the average threshold and GDT (r = 0.87, df = 6, p = 0.0050, corrected p = 0.03). The other was between the polarity effect and RDR400 (r = 0.91, df = 6, p = 0.0017, corrected p = 0.01). As noted in the Introduction, it is unlikely that the second of these can be explained in terms of cognitive factors, such as the ability or willingness to concentrate on the task, because the polarity effect is a difference score.

A large number of psychophysical and electrophysiological studies have demonstrated CI users’ greater sensitivity to anodic than to cathodic current for moderately and comfortably loud stimuli (Macherey et al. 2006; Macherey et al. 2008; van Wieringen et al. 2008; Undurraga et al. 2010; Macherey et al. 2011; Carlyon et al. 2013). More recently, two studies have shown that, although detection thresholds do not differ overall between stimuli of opposite polarity, individual CI listeners show idiosyncratic but consistent polarity effects.

Macherey et al. (2017) measured thresholds for trains of symmetric biphasic pulses with either anodic or cathodic leading polarity (“SYM-A” and “SYM-C”), as well as for trains of so-called quadraphasic (“QP”) pulses. The QP pulses were constructed by abutting two SYM pulses of opposite leading polarity, such that the central portion consisted of two anodic (QP-A) or cathodic (QP-C) phases. A previous study had shown that, as for the TP-A and TP-C pulse trains used here and for pseudomonophasic pulse trains, the loudness of supra-threshold QP-A pulse trains was consistently greater than that of QP-C pulse trains at the same current (Carlyon et al. 2013). Macherey et al. (2017) found that, across subjects, the difference between QP-A and QP-C thresholds correlated strongly with that between SYM-A and SYM-C thresholds, demonstrating that the idiosyncratic differences observed across subjects were not simply due to measurement noise. In addition, they reported that, for some electrode-listener combinations, the loudness growth function was non-monotonic, and that this unusual pattern was significantly more likely to occur when thresholds were lower for QP-C than for QP-A pulse trains. They suggested, as we do here, that low QP-C thresholds may reflect good local survival and activation of peripheral processes. They further suggested that non-monotonic loudness growth might then be explained by “cathodal block” (Ranck 1975) impeding the propagation of action potentials as level was increased. Interestingly, they only observed non-monotonic growth for users of the Cochlear CI; when they tested five Advanced Bionics users on one electrode each, the loudness growth was always monotonic. The fact that we reported several instances of substantially negative polarity effects raises the possibility that, for those electrode-subject combinations, the loudness growth may also have been non-monotonic. A stronger test of this hypothesis would involve measuring loudness growth functions for the subject-electrode combinations showing the most negative polarity effects, such as C3 (electrode 15, − 2.2 dB), C5 (electrode 7, − 1.3 dB), and S22 (electrode 5, − 2.45 dB).

Mesnildrey et al. (2017) measured thresholds using a fast one-interval adjustment procedure for trains of TP-A and TP-C pulses presented in partial tripolar mode. Their measurements, obtained with multiple electrodes in 16 ears, revealed that thresholds were lower for TP-C than for TP-A pulse trains in 22 % (48/219) of electrodes. This percentage was very similar to the corresponding value of 27 % (9/33) in the present study. Another similarity is that they found a positive across-electrode correlation between the polarity effect and the threshold for a SYM-C pulse train, analogous to our correlation between the polarity effect and overall threshold. However, their correlation was quite weak, accounting only for about 5 % of the variance. More recently, Goehring et al. (2018) used a method similar to that of Mesnildrey et al. (2017) to measure TP-A and TP-C thresholds in eight subjects. They also reported a weak positive correlation which, in their case, was not significant. Clearly, then, the size and significance of the correlation depend on the particular group of subjects and/or electrodes studied. Indeed, inspection of Fig. 2 reveals that the correlation was largely driven by the results of subjects S22, C6, and (to a lesser extent) C3. In terms of our hypothesis, it may be that large correlations occur when the EMD is constant, and neural survival varies markedly, along the length of the cochlea. More generally, it is worth noting that although the across-electrode comparisons reported here and elsewhere involve a reasonably high number of degrees of freedom, they are obtained from only a modest number of subjects.

With the above caveats, of particular interest in the study by Mesnildrey et al. (2017) is their analysis of CT scans from nine ears, which allowed them to derive an estimate of the electrode-modiolar distance (EMD) for each subject. Although the EMD accounted for between 63 and 68 % of the between-subject variance in thresholds, it accounted for only a small proportion of the between-electrode variance. The authors suggested that this was due to EMD varying substantially between but not within ears. They also showed that the EMD did not correlate significantly with the polarity effect. This is important for the interpretation of the present results because it is possible that the site of excitation, and hence the polarity effect, was affected by the EMD. If such a correlation had been found, this would have provided an alternative to our interpretation that the correlation between the polarity effect and overall thresholds was mediated by neural survival.

Zhou and Pfingst (2016) also obtained thresholds for multiple electrodes and estimated EMDs from the data relating EMD to electrode number reported by Long et al. (2014). They did not report the correlation directly but did state that MPI, defined as the difference between thresholds for 80- and 640-pps pulse trains, correlated with the EMD. Their data showed generally greater across-electrode threshold variations at 80 pps than at 640 pps, and so it is likely that the variation in the MPI was driven mainly by the 80-pps thresholds. If so, then those thresholds would have correlated positively with the EMD. However, they do not state how much of the variation in 80-pps thresholds was accounted for by EMD, nor whether a substantial across-electrode (rather than across-listener) correlation was observed. In addition, it is possible, as suggested by Mesnildrey (2017), that the across-electrode variation in EMD for a given subject may be smaller for Advanced Bionics listeners (as tested here) than for the Cochlear CI24RE listeners studied by Zhou and Pfingst (2016) and by Long et al. (2014). Mesnildrey reported that the average range of EMDs, across listeners, was 0.75 mm in his study compared to 1.2 mm in Long et al.’s study; an intermediate value of 1.00 mm is revealed by an analysis of the 10 Advanced Bionics listeners studied by DeVries et al. (2016).

Table 3 shows the across-subject correlations between the two measures obtained here—average thresholds and the polarity effect—and the supra-threshold measures reported previously by our laboratories (Bierer et al. 2015; Cosentino et al. 2016). Two correlations were statistically significant after Bonferroni correction: average thresholds vs. log GDTs, and the polarity effect vs RDR400 (which is a measure of the upper limit of temporal pitch). This does not mean, though, that log GDTs are more strongly related to average thresholds than to polarity, or that RDR400 is more strongly correlated with the polarity effect than with average thresholds. Such a conclusion would require that these two correlations were significantly larger than, respectively, those between polarity vs log GDT and average threshold vs. log400R. This was not the case (Williams test, df = 5, two-tailed p = 0.07 and 0.14 for GDT and RDR400 respectively). The safest conclusion is that both low average thresholds and a small or negative polarity effect correlate with good performance on some supra-threshold tasks, including gap detection and a measure (RDR400) of the upper limit of temporal pitch. There is also evidence from Mesnildrey et al. (2017) that, across subjects, small or negative polarity effects correspond to low thresholds on a spectro-temporal ripple test (Aronoff and Landsberger 2013). In addition, Zhou and Pfingst (2016) showed that high thresholds for 80-pps pulse trains—similar to the rate used here—correlated significantly, both across subjects and electrodes, with poor spatial selectivity as measured using forward masking. A caveat is that they measured spatial selectivity by measuring the slope relating masker position to the amount of masking of a fixed probe. This measures the spread of excitation produced by the maskers rather than by the signal.

Finally, it is worth noting that Zhou (2017) has recently observed substantial and significant improvements, both in speech perception and on a spectro-temporal ripple test, by deactivating electrodes with high low-rate thresholds. As they have pointed out, such high thresholds may reflect a combination of large EMD and poor local neural survival. If, as we have argued, the polarity effect is more affected by local neural survival (specifically that of the peripheral processes), then this may have two implications for the potential improvement of patient outcomes by re-programming the clinical map. First, in devices such as the one studied here, where EMD variation across the array may be less than in the Cochlear CI24RE device, variations in spatial selectivity may be more strongly driven by neural survival, and it may be beneficial to have a measure that is specific to that parameter. Second, computational models (e.g., Goldwyn et al. 2010) indicate that, in devices that allow focused stimulation methods such as the tripolar mode, the appropriate intervention in cases of locally high thresholds depends on the cause. Specifically, Bierer and Litvak (2016) have argued that electrodes with a large EMD should be stimulated in tripolar mode, whereas those in neural “dead regions” should be deactivated. Again, a measure that differentiates between these two situations could provide significant clinical advantages.