Christofer Bester

Aim: There are a range of approaches to intraoperative electrocochleography (ECochG) monitoring across different implant systems, and these show inconsistent results. Here, we synthesize by meta-analysis the evidence for real-time intracochlear ECochG recorded during cochlear implantation predicting residual hearing. Methods: Inclusion criteria included original studies in which real-time ECochG was undertaken on a commercial cochlear implant (CI) and residual hearing outcomes at least 4 weeks after implantation were reported. Thirteen studies comprising 313 individual patient data sets met inclusion criteria. Full data sets were sought from the authors. Studies were rated for quality and bias. Details about study design (observational or interventional) and electrode characteristics (lateral wall or perimodiolar) were extracted. Relative hearing loss across the low frequencies (250–1,000 Hz) was the main outcome, with decibel loss and functional hearing as secondary outcomes. Effects were explored by forest plot, and then linear mixed modelling on the full data sets. Sensitivity analyses included recalculation of hearing preservation using thresholds across different frequencies. Results: Thirteen studies, including 313 full patient data sets, were included, and meta-analysis was performed on lateral wall electrodes. A drop of ECochG amplitude of >30%, even if transient, predicted residual hearing across all metrics (decibel, relative, and functional), and data-synthesis methods (forest plots and linear modelling) for all systems where real-time measurements were available. A drop of amplitude of at least 60% across insertion predicted relative but not decibel hearing loss in linear mixed modelling. The “pattern” of ECochG drop, as described by Harris in 2017, did not predict residual hearing by any method. Discussion: The only truly “real-time” measurement that predicted residual hearing was an instantaneous drop in ECochG amplitude, with or without recovery. Large drops over the insertion did too, but these can only be calculated after insertion, and after the cochlear trauma has been done. All CI systems now offer real-time monitoring – these results provide guidance on how best to use it to optimize clinical outcomes.

Hearing loss is an impairment that affects a large population worldwide. Cochlear implant (CI) surgery is an effective solution, especially in patients with residual, natural hearing. However, the risk of intracochlear damage during CI surgery, which impairs residual hearing, remains as a concern. It has been shown that intra-operative monitoring (using electrocochleography, specifically the cochlear microphonic) during CI surgery can detect drops that are associated with such trauma, resulting in its potential prevention. Researchers have attempted to automate the process of drop detection by implementing supervised models. However, these have so far failed to achieve good reliability and generalizability. To address these issues, we develop an unsupervised model based on the Bayesian method with a scoring function derived from the logistic function. The proposed model, as a data-driven method, directly learns from data without the need for any supervision which enhances its adaptability to diverse patient populations. Empirical results demonstrate the proposed model’s high accuracy with a low false alarm rate (18 extra detection against 146 in previous work), as well as its robustness when processing new data. This indicates its potential for real-world application to improve safety during CI surgery.

Objectives: Real-time monitoring of cochlear function to predict the loss of residual hearing after cochlear implantation is now possible. Current approaches monitor the cochlear microphonic (CM) during implantation from the electrode at the tip of the implant. A drop in CM response of >30% is associated with poorer hearing outcomes. However, there is prior evidence that CM amplitude can fluctuate in a manner unrelated to hearing trauma, leading to false positives. By monitoring another cochlear response, the auditory nerve neurophonic (ANN), a differentiation between CM drops that result in reduced cochlear output from false positives may be possible. The hypothesis tested in the present work was that ANN/CM ratios measured during a CM drop will increase during drops not associated with postoperative hearing loss. Design: Twenty-eight adult participants with known CM drops during implantation were taken from a larger data set. This contains adult cochlear implant candidates scheduled to receive a Cochlear Nucleus cochlear implant with either the slim-straight or slim-modiolar electrode array with preoperative audiometric low-frequency average thresholds of ≤80 dB HL at 500, 750, and 1000 Hz in the ear to be implanted. Patients were recruited from eight international implant sites. Pure-tone audiometry was measured postoperatively and 4 to 6 weeks after implantation. Electrocochleography was measured during and immediately after the implantation of the array in response to a 500-Hz, 6-msec pure-tone pip at 110 dB HL. Results: The ANN/CM ratio rose during CM drops in 19 of these patients and decreased in 9. At the follow-up timepoint, patients with a decreasing ANN/CM ratio had a median hearing loss of 29.0 dB, significantly worse than the group with increasing ratio at 13.3 dB ( p = 0.004). Considering only the change in ANN amplitude during a CM drop led to smaller groups (ANN drop during CM drop N = 17, ANN increasing during CM drop N = 6) due to 5 patients having undetectable ANN during the CM drop. Using the ANN alone also led to as poorer prediction of hearing preservation, with median hearing preservation in the ANN increasing group of 12.9 dB, significantly better than the ANN decreasing group of 25 dB ( p = 0.02). The group with a decreasing ANN/CM ratio had maximum CM amplitude immediately after insertion lower than the maximum amplitude reached during insertion (mean maximum postinsertion amplitude of 98% of during-insertion amplitude). In comparison, the ANN/CM ratio increasing group tended to have a larger CM amplitude immediately after insertion (mean maximum CM amplitude postinsertion of 164% of the maximum during-insertion amplitude). Conclusions: These data show that the ANN/CM ratio is a measure that can differentiate between patients with CM drops that lead to a loss of residual hearing and those that do not. The ANN/CM ratio is easily measured and responds rapidly during a CM drop, showing clinical promise for improving current and developing approaches to intraoperative monitoring.

Finding effective representations for time series data is a useful but challenging task. Several works utilize self-supervised or unsupervised learning methods to address this. However, there still remains the open question of how to leverage available label information for better representations. To answer this question, we exploit pre-existing techniques in time series and representation learning domains and develop a simple, yet novel fusion model, called: \textbf{S}upervised \textbf{CO}ntrastive \textbf{T}emporal \textbf{T}ransformer (SCOTT). We first investigate suitable augmentation methods for various types of time series data to assist with learning change-invariant representations. Secondly, we combine Transformer and Temporal Convolutional Networks in a simple way to efficiently learn both global and local features. Finally, we simplify Supervised Contrastive Loss for representation learning of labelled time series data. We preliminarily evaluate SCOTT on a downstream task, Time Series Classification, using 45 datasets from the UCR archive. The results show that with the representations learnt by SCOTT, even a weak classifier can perform similar to or better than existing state-of-the-art models (best performance on 23/45 datasets and highest rank against 9 baseline models). Afterwards, we investigate SCOTT's ability to address a real-world task, online Change Point Detection (CPD), on two datasets: a human activity dataset and a surgical patient dataset. We show that the model performs with high reliability and efficiency on the online CPD problem ($\sim$98\% and $\sim$97\% area under precision-recall curve respectively). Furthermore, we demonstrate the model's potential in tackling early detection and show it performs best compared to other candidates.

Objective: Cochlear Implant (CI) programming based on subjective psychophysical fine-tuning of loudness scaling involves active participation and cognitive skills and thus may not be appropriate for difficult-to-condition populations. The electrically evoked stapedial reflex threshold (eSRT) is an objective measure that is suggested to provide clinical benefit to CI programming. This study aimed to compare speech reception outcomes between subjective and eSRT objectively determined CI maps for adult MED-EL recipients. The effect of cognitive skills on these skills was further assessed. Methods: Twenty-seven post lingually hearing-impaired MED-EL CI recipients were recruited, 6 with mild cognitive impairment (MCI– 4 male, 79 years ± 5), 21 with normal cognitive function (5 male, 63 years ± 12). Two MAPs were generated: a subjective MAP and an objective MAP in which eSRTs determined maximum comfortable levels (M-Levels). Participants were randomly divided into two groups. Group A trialled the objective MAP for two weeks before returning for outcome assessment. During the following two weeks, Group A trialled the subjective MAP before returning for outcome assessment. Group B trialled MAPs in reverse. Outcome measures included the Hearing Implant Sound Quality Index (HISQUI), Consonant-Nucleus-Consonant (CNC) word test, and Bamford-Kowal-Bench Speech-in-Noise (BKB-SIN) test. Results: eSRT based MAPS were obtained in 23 of the participants. A strong relationship was demonstrated between global charge between eSRT-based and psychophysical-based M-Levels (r = 0.89, p < .001). The Montreal Cognitive Assessment for the Hearing Impaired (MoCA-HI) testing identified 6 CI recipients with MCI (MoCA-HI total score ≤23). The MCI group was older (63, 79 years), but were not otherwise different in sex, duration of hearing loss or duration of CI use. For all patients, no significant differences were found for sound quality or speech in quiet scores between eSRT-based and psychophysical-based MAPs. However, psychophysically determined MAPs showed significantly better speech-in-noise reception (6.74 vs 8.20-dB SNR, p = .34). MoCA-HI scores showed a significant, moderate negative correlation with BKB SIN for both MAP approaches (Kendall’s Tau B, p = .015 and p = .008), with no effect on the difference between MAP approaches. Conclusion: Results indicate eSRT-based methods provide poorer outcomes than psychophysical-based method. While speech-in-noise reception is correlated with MoCA-HI score, this affected both behaviourally and objectively determined MAPs. The results suggest fair confidence in the eSRT-based method as a guide for setting M-Level for difficult-to-condition CI populations in simple listening conditions.

Objectives: Different patterns of electrocochleographic responses along the electrode array after insertion of the cochlear implant electrode array have been described. However, the implications of these patterns remain unclear. Therefore, the aim of the study was to correlate different peri- and postoperative electrocochleographic patterns with four-point impedance measurements and preservation of residual hearing. Design: Thirty-nine subjects with residual low-frequency hearing which were implanted with a slim-straight electrode array could prospectively be included. Intracochlear electrocochleographic recordings and four-point impedance measurements along the 22 electrodes of the array (EL, most apical EL22) were conducted immediately after complete insertion and 3 months after surgery. Hearing preservation was assessed after 3 months. Results: In perioperative electrocochleographic recordings, 22 subjects (56%) showed the largest amplitude around the tip of the electrode array (apical-peak, AP, EL20 or EL22), whereas 17 subjects (44%) exhibited a maximum amplitude in more basal regions (mid-peak, MP, EL18 or lower). At 3 months, in six subjects with an AP pattern perioperatively, the location of the largest electrocochleographic response had shifted basally (apical-to-mid-peak, AP-MP). Latency was analyzed along the electrode array when this could be discerned. This was the case in 68 peri- and postoperative recordings (87% of all recordings, n = 78). The latency increased with increasing insertion depth in AP recordings (n = 38, median of EL with maximum latency shift = EL21). In MP recordings (n = 30), the maximum latency shift was detectable more basally (median EL12, p < 0.001). Four-point impedance measurements were available at both time points in 90% (n = 35) of all subjects. At the 3-month time point, recordings revealed lower impedances in the AP group (n = 15, mean = 222 Ω, SD = 63) than in the MP (n = 14, mean = 295 Ω, SD= 7 6) and AP-MP groups (n = 6, mean = 234 Ω, SD = 129; AP versus MP p = 0.026, AP versus AP-MP p = 0.023, MP versus AP-MP p > 0.999). The amplitudes of perioperative AP recordings showed a correlation with preoperative hearing thresholds (r ²=0.351, p = 0.004). No such correlation was detectable in MP recordings (r ² = 0.033, p = 0.484). Audiograms were available at both time points in 97% (n = 38) of all subjects. The mean postoperative hearing loss in the AP group was 13 dB (n = 16, SD = 9). A significantly larger hearing loss was detectable in the MP and AP-MP groups with 28 (n = 17, SD = 10) and 35 dB (n = 6, SD = 13), respectively (AP versus MP p = 0.002, AP versus AP-MP p = 0.002, MP versus AP-MP p = 0.926). Conclusion: MP and AP-MP response patterns of the electrocochleographic responses along the electrode array after cochlear implantation are correlated with higher four-point impedances and poorer postoperative hearing compared to AP response patterns. The higher impedances suggest that MP and AP-MP patterns are associated with increased intracochlear fibrosis.

Objectives: Electrocochleography (ECochG) is emerging as a tool for monitoring cochlear function during cochlear implant (CI) surgery. ECochG may be recorded directly from electrodes on the implant array intraoperatively. For low-frequency stimulation, its amplitude tends to rise or may plateau as the electrode is inserted. The aim of this study was to explore whether compromise of the ECochG signal, defined as a fall in its amplitude of 30% or more during insertion, whether transient or permanent, is associated with poorer postoperative acoustic hearing, and to examine how preoperative hearing levels may influence the ability to record ECochG. The specific hypotheses tested were threefold: (a) deterioration in the pure-tone average of low-frequency hearing at the first postoperative follow-up interval (follow-up visit 1 [FUV1], 4 to 6 weeks) will be associated with compromise of the cochlear microphonic (CM) amplitude during electrode insertion (primary hypothesis); (b) an association is observed at the second postoperative follow-up interval (FUV2, 3 months) (secondary hypothesis 1); and (c) the CM response will be recorded earlier during electrode array insertion when the preoperative high-frequency hearing is better (secondary hypothesis 2). Design: International, multi-site prospective, observational, between groups design, targeting 41 adult participants in each of two groups, (compromised CM versus preserved CM). Adult CI candidates who were scheduled to receive a Cochlear Nucleus CI with a Slim Straight or a Slim Modiolar electrode array and had a preoperative audiometric low-frequency average thresholds of ≤80 dB HL at 500, 750, and 1000 Hz in the ear to be implanted, were recruited from eight international implant sites. Pure tone audiometry was measured preoperatively and at postoperative visits (FUV1 and follow-up visit 2 [FUV2]). ECochG was measured during and immediately after the implantation of the array. Results: From a total of 78 enrolled individuals (80 ears), 77 participants (79 ears) underwent surgery. Due to protocol deviations, 18 ears (23%) were excluded. Of the 61 ears with ECochG responses, amplitudes were <1 µV throughout implantation for 18 ears (23%) and deemed “unclear” for classification. EcochG responses >1 µV in 43 ears (55%) were stable throughout implantation for 8 ears and compromised in 35 ears. For the primary endpoint at FUV1, 7/41 ears (17%) with preserved CM had a median hearing loss of 12.6 dB versus 34/41 ears (83%) with compromised CM and a median hearing loss of 26.9 dB (p < 0.014). In assessing the practicalities of measuring intraoperative ECochG, the presence of a measurable CM (>1 µV) during implantation was dependent on preoperative, low-frequency thresholds, particularly at the stimulus frequency (0.5 kHz). High-frequency, preoperative thresholds were also associated with a measurable CM > 1 µV during surgery. Conclusions: Our data shows that CM drops occurring during electrode insertion were correlated with significantly poorer hearing preservation postoperatively compared to CMs that remained stable throughout the electrode insertion. The practicality of measuring ECochG in a large cohort is discussed, regarding the suggested optimal preoperative low-frequency hearing levels (<80 dB HL) considered necessary to obtain a CM signal >1 µV.