The impact of auditory nerve functional states on the correlations between human and computer decisions for electrically evoked compound action potential threshold.

Objectives: The electrically evoked compound action potential (ECAP) is widely used in clinical to reflect the functional states of the auditory nerve in cochlear implant (CI) recipients, especially in pediatric CI users. Currently, the software can automatically provide the ECAP threshold, which is convenient and not affected by the subjective judgement of the clinicians. However, it remains unclear whether the correlations between human and computer decisions for ECAP threshold can be affected by auditory nerve functional states, which is also the main purpose of our present study.

Methods: Intracochlear electrical stimulation, which can decrease the excitability of the auditory nerve, was used to change the auditory nerve functional states of guinea pigs. Ten normal-hearing guinea pigs were implanted with CIs unilaterally. ECAPs were recorded both before and after the electrical stimulation, representing different functional states of the auditory nerve. Forward masking (FwdMsk) and alternating polarity (AltPol), two most commonly-used artifact-reduction methods, were applied to the measurements. All measurements recorded by the software were saved for computer and human analysis with linear regression and visual detection methods.

Results: The correlations between human and computer performance in the peak-picking process were not affected by auditory nerve states and artifact-reduction methods. However, complicated findings were observed for ECAP threshold. With FwdMsk utilized, weaker correlations between human and computer performance were observed in abnormal state compared to those in normal state. Regardless of the functional states of the auditory nerve, the results revealed stronger correlations in AltPol than those in FwdMsk. Furthermore, when compared with human decision, computer linear-regression threshold (C-LRT) was always less accurate than computer visual-detection threshold (C-VDT), which was not affected by auditory nerve states.

Conclusions: (1) the functional states of the auditory nerve can definitely affect the correlations between human and computer decisions for ECAP threshold, but the impact is limited to the FwdMsk method; (2) AltPol can produce stronger correlations compared with FwdMsk, which is not affected by auditory nerve states; and (3) regardless of the auditory nerve states, C-VDT can always show higher consistency with human decision, while C-LRT reveals more variability.