Signal processing & audio processors

Abstract

Signal processing algorithms are the hidden components in the audio processor that converts the received acoustic signal into electrical impulses while maintaining as much relevant information as possible. Signal processing algorithms should be smart enough to mimic the functionality of external, middle and the inner-ear to provide the cochlear implant (CI) user with a hearing experience as natural as possible. Modern sound processing strategies are based on the continuous interleaved sampling (CIS) strategy proposed by B. Wilson in 1991, which provided envelope information over several intracochlear electrodes. The CIS strategy brought significant gains in speech perception. Translational research activities of MED-EL resulted in further improvements in speech understanding in noisy environments as well as enjoyment of music by not only coding CIS-based envelope information, but by also representing temporal fine structure information in the stimulation patterns of the apical channels. Further developments include “complete cochlear coverage” made possible by deep insertion of the intracochlear electrode, elaborate front end processing, anatomy based fitting (ABF), triphasic pulse stimulation instrumental in the suppression of facial nerve stimulation, and bimodal delay compensation allowing unilateral CI users to experience hearing with hearing aids on the contralateral ear. The large number of hardware developments might be exemplified by the RONDO, the world's first single unit audio processor in 2013. This article covers the milestones of translational research around the signal processing and audio processor topic that took place in association with MED-EL.

Graphical Abstract

Chinese abstract

信号处理算法是音频处理器中的隐密组件。它将接收到的声音信号转换为电脉冲, 而在转换过程中不会丢失任何信息。信号处理算法应高明到可以模仿外耳、中耳和内耳的功能, 从而为CI植入者提供自然的听觉体验。信号处理方面的突破发生在1991年, 当时美国的Wilson教授提出了连续交错采样（CIS）策略。该策略使得电极通道的非同时刺激成为可能。精细结构处理（FSP）策略是对CIS策略的改进, 它在MED-EL的信号处理策略中起着关键作用。该信号处理策略通过以相同的频率并与低频中的声频同步施加电脉冲来模仿自然。本章涵盖了MED-EL的转换研究活动, 评估了FSP的功效、前端处理功能以及音频处理器的开发。根据解剖特征的安装（ABF）是一个新概念, 即根据个人的频率图和电极插入深度进行安装。三相脉冲刺激是另一个有助于抑制面神经刺激的新概念。双峰延迟补偿这一概念允许单侧CI使用者利用对侧耳朵的助听器来体验相配的听力。

Keywords:

• Continuous interleaved sampling
• fine structure processing
• single unit processor
• anatomy-based fitting
• triphasic pulse stimulation
• bimodal delay compensation

5.1. Introduction

In the cochlear implant (CI) systems, the sound signal is captured by the microphone in the externally worn audio processor. The audio processor converts the sound signal into detailed digital signals using signal-processing algorithms and transmits those to the implantable electronics via an inductive link. The implant electronics then convert these digital signals to electric impulses and transfer them to the inner ear through the intracochlear electrode array. The auditory nerve then transfers the electric impulses received from the electrode array to the brain to interpret the sound signals. While the proper placement of the electrode array inside the cochlea – which covers the entire frequency range – without causing any structural damage is essential, it is equally crucial for the audio processors to process the sound signals without losing any of its key elements.

Before detailing how the audio processor in a CI system processes sounds, it is of importance to canvass through the same process in a normal hearing ear first. In a healthy hearing ear, the sound signal is processed at three levels – in the outer, middle and inner ear, as shown in Figure 1. Pinna and the external ear canal constitute the outer ear, and it is mainly the pinna that gives some form of directionality to the listener [1]. The pinna allows the normal-hearing listener to hear better with the acoustic signal coming from the front than from the rear side of the head. The outer and the middle ear filter the sound signals and provide some pre-emphasis, especially to frequencies around 1,000Hz, which carries speech information necessary for typical conversation [2]. The inner ear processes the sound signal in a frequency-specific manner, called tonotopicity, i.e. high frequencies (HF) are processed along the basilar membrane (BM) in the basal turn of the cochlea which is closer to the outer and middle ear, and the low frequencies (LF) are processed in the apical end of the cochlea which is further away from the outer and middle ear. Tonotopicity is the first important element in the frequency coding of a normal hearing ear. The other aspect of inner ear sound processing is the travelling wave latency/LF delay, i.e. HF sound signals are processed with the neural responses reaching the auditory cortex relatively faster than the LF sound signals. The LF delay comes from the delayed mechanical vibrational response of the BM that takes longer for LF than for HFs [3]. In other words, LF signals have longer latencies than HF signals, i.e. latency increases with decreasing frequencies and vice versa. Both tonotopicity and travelling wave latency are specific to certain locations along the BM. The BM in a normal acoustic ear acts as a gain-controlled amplifier, i.e. it amplifies low-level sound and compresses (compression) high-level sound – allowing the listener to hear even very soft sounds [4]. Around the apex of the cochlea where the LF signals are processed, the neural responses are produced in synchrony with sound frequency. In other words, the neural response rate (neural responses/second) is equivalent to the sound frequency in the apical location of the BM that is responsible for processing LFs, and this is called phase-locking – a phenomenon predominantly happening in the LF region [5].

Figure 1. left-hand side picture depicts the functionality of the external, middle and the inner ear in a normal acoustic ear. The right-hand side picture shows the tonotopic arrangement of frequency in the inner ear. Image courtesy of MED-EL.

The neural responses are phase-locked to the acoustic stimulus in the LF and are clearly clustered, whereas, in the HFs, those are not clustered but rather smeared. Phase locking is the second important element in the frequency coding of a normal hearing cochlea (Figure 2).

Figure 2. Low-frequency sound signals in the cochlear apex produce clusters of neural responses which the brain can identify separately, whereas the high-frequency sound signals in the cochlear base produce neural responses that are close to each other so that the brain cannot identify them separately. Image courtesy of MED-EL.

To sum-up the sound processing in a normal acoustic ear, the pinna offers directionality to the listener, the middle ear emphasises the frequencies around 1,000Hz, enhancing the speech perception. The tonotopic inner ear works as a frequency analyser and adds latency to the sound signal over the entire frequency range, whereas the phase-locking is specific to the LFs. The signal processing in the audio processor of the CI shall model or mimic all the functionalities of the normal acoustic ear, providing near-natural hearing experience to the hard of hearing patients. The signal processing chain can be divided into two major blocks, namely the front-end processing and sound coding. The front-end processing block tries to model the functionality of external and middle ear while the sound coding block aims at modelling the functionality of the inner ear.

This article starts with the brief introduction to the continuous interleaved sampling (CIS) strategy which is the basis of multi-channel sound coding strategies in modern CI systems among brands, with some modifications matching their inbuilt electronics. The article aims to cover the developments in the signal processing and audio processors over time within the MED-EL CI system. Following the developments in chronological order, sound coding will be discussed first along with the scientific evidence that demonstrated its benefits in CI users, followed by the front-end processing. Individualisation within sound coding will be briefly covered in the last section as well, highlighting MED-EL’s latest unique advancements in sound coding strategies.

5.2. Sound coding

One of the blocks of signal processing is the sound coding strategy that aims to model the inner ear functionality, including tonotopicity, temporal processing via phase-locking and the travelling wave delays. This section covers all the sound coding strategies that were implemented in MED-EL’s audio processors from 1990 until 2021, along with the scientific pieces of evidence that evaluated these strategies in MED-EL CI users.

5.2.1. Continuous interleaved sampling strategy

In 1990, when MED-EL hired its first team members, the audio processor of its CI system was body-worn, as shown in Figure 3, and was further developed to a behind-the-ear (BTE) audio processor in 1991, making it the world’s first of its kind. The Continuous Interleaved Sampling (CIS) strategy was not a part of these audio processors at the time.

Figure 3. MED-EL’s body-worn audio processor before 1991 and the world’s first BTE audio processor in 1991 (image courtesy of MED-EL).

Before the invention of the CIS strategy, various stimulation strategies, ranging from low rate pulsatile stimulation using some kind of feature extraction, to multichannel analogue stimulation, were investigated. Simultaneous multichannel analogue stimulation raised concerns because significant interactions among channels reduced extractable information [6]. Bipolar stimulation to reduce these interactions resulted in considerable power consumption. Nevertheless, MED-EL’s frequency adjusted, elaborate dynamic range compressed analogue stimulation signals, presented via just a single stimulation site (selected out of four possible sites), provided fortunate users with a monosyllabic word understanding of up to 40% and allowed them enjoyable music perception [7]. Its low power consumption made the world’s first BTE audio processor possible [8].

In 1991, Prof. Wilson and his colleagues from the Research Triangle Institute in the USA proposed the CIS strategy to be set as a base in the CI – which result is of existential importance in the nowadays CI filed (Figure 4).

Figure 4. Prof. Blake Wilson proposed the CIS strategy, currently affiliated to Duke University, USA.

The CIS strategy that was newly developed at the time addressed the problem of simultaneous channel interaction using interleaved non-simultaneous stimuli which heavily reduced the power consumption.

In the classic CIS sound coding strategy, the microphone signal is first processed through a pre-emphasis filter that attenuates strong components in the speech above 1.2 kHz and emphasises signals that are below 1.2 kHz, as the speech information that is needed for normal conversation is around that frequency (stage 1). The output of the pre-emphasis is further passed through multiple channels of processing that includes bandpass filters (BPF) (stage 2) for splitting the broadband signal into different frequency bands, rectification, as well as lowpass filtering for envelope extraction (stage 3). The envelope signals are compressed into the narrow dynamic range of electrically evoked hearing (stage 4). Trains of charge-balanced biphasic pulses are sequentially interleaved in time across electrodes to eliminate any overlap across channels, as shown in Figure 5 by the red dotted vertical lines. The pulse amplitudes derive from the envelopes of the bandpass filter outputs and are directed to intracochlear electrodes (EL-1 to EL-12) (stage 5).

Figure 5. Block diagram of the CIS strategy. The pre-emphasis filter (Pre-emp)/automatic gain control attenuates strong components in the speech above 1.2 kHz. This filter is followed by multiple channels of processing, with each channel including stages of bandpass filtering (BPF), envelope detection, compression and modulation. The envelope detectors generally use a full-wave or half-wave rectifier (Rect.), followed by a lowpass filter (LPF). Carrier waveforms for two of the modulators are shown immediately below the two corresponding multiplier blocks (circle with an x mark). The outputs of the multipliers are directed to intracochlear electrodes (EL-1 to EL-12). The inset shows an x-ray image of the implanted electrode (in a cochlear model) to which the outputs of the speech processor are directed. Scheme created from Wilson et al. [9].

All commercially available MED-EL CI systems feature the CIS strategy in their sound coding portfolio, with some modifications matching to their inbuilt electronics, resulting in further developed variants of the CIS strategy (CIS+, HDCIS). Prof. Wilson was awarded the Lasker-DeBakey Clinical Medical Research Award in the year 2013 for his contributions to the CI field, along with Dr Ingeborg Hochmair from MED-EL in Austria and Prof. Clarke from the University of Melbourne in Australia. Since the early '90s, MED-EL and Prof. Wilson had a close scientific collaboration that helped MED-EL to implement the CIS strategy in its CI system.

In January 1994, MED-EL introduced the COMBI 40 implant system, which was the world’s first multichannel high-rate CI. It was an eight-channel system, designed to implement Prof. Wilson’s CIS sound coding strategy faithfully. The system featured a maximum overall stimulation rate of 12,120 non-overlapping biphasic pulses/second (pps), allowing the implementation of a high-rate CIS strategy on eight channels (1,515pps/channel for eight active electrode channels). It featured a 31.5 mm long flexible electrode array for coverage of the entire cochlear length. The COMBI 40 implant electronics included individual safety capacitors, serially added to all its eight stimulating channels to prevent any direct current (DC) component from being delivered inside the cochlea.

Prof. Helms from the University of Würzburg in Germany was the primary investigator in the study that aimed at evaluating the hearing performance with the COMBI 40 CI system [10]. Dr Ingeborg Hochmair had drafted the study protocol and it had been refined and agreed upon in first of the COMBI 40 workshops in Alpbach, in Tyrol Austria in fall of 1993. These workshops have taken place regularly since then and are a welcome platform now for the presentations and exchange of new research outcomes and discussions. They also offer further educational credits.

The first sixty adult patients who received a COMBI-40 device at 19 prominent ENT-clinics in 7 different countries in Europe, took part in a multicentric clinical study. The mean age of the participants was 47.5 years with a mean duration of deafness of 5.3 years. The patients were evaluated with different speech tests, involving two-digit number test, sixteen consonant tests, the eight-vowel test, sentence test and monosyllabic word test without lip reading. Figure 6 shows the sentence and monosyllabic word score results that were collected at different time points, starting from the preoperative testing interval until twelfth-month post-fitting.

Figure 6. The sentence and monosyllabic word test conducted in patients implanted with MED-EL’s COMBI 40 CI system at different time points, starting from preoperative to twelfth-month post fitting [10]. Reproduced by permission of Karger AG, Basel.

The score was zero prior to the CI surgery, which increased to 34% during the first month’s test. This further increased to 48% at the sixth month, and to 54% at the first-year’s test. The maximum value achieved after six months was 90%. The results were published in the year 1997.

Overall, improvements in speech understanding with the fast CIS strategy occurred soon after switch-on, and a very rapid learning curve with the implant was observed in most patients without lip reading. At the time, the CIS strategy was seen as a real tipping point, and the CI system COMBI 40 that implemented the CIS strategy was reported to be safe and effective in adults. Users who achieved 50% and more in monosyllable word understanding could typically use the telephone.

In 1996, MED-EL further upgraded the body-worn processor with CIS PRO + that helped to reduce the number of batteries from 4xAA to 2xAA, and this was the time when MED-EL also upgraded its implant system from COMBI 40 to COMBI 40+ that featured twelve stimulating channels. The COMBI 40+ system had a maximum overall stimulation rate of 18,181.8pps across the twelve channels, which was the fastest system at the time.

In 1999, MED-EL launched the world’s first BTE high-rate audio processor, TEMPO + BTE (Figure 7), based on a patent by Prof. Zierhofer from the University of Innsbruck in Austria (US patent number: 5983139). Prof. Zierhofer and Mr Stöbich, who is currently employed at MED-EL and who was a PhD student of Prof. Hochmair at the University of Innsbruck at the time, were highly instrumental in the development of TEMPO + BTE audio processor.

Figure 7. Prof. Clemens Zierhofer from the University of Innsbruck and Dr Bernhard Stöbich (from MED-EL and a PhD student at the University of Innsbruck at the time) were instrumental in the development of TEMPO + BTE audio processor. Image courtesy of MED-EL.

The TEMPO + processor is capable of using high-rate stimulation (up to more than 18,000pps) and uses Hilbert transform instead of rectification and low-pass filtering for envelope detection. The Hilbert transform allows a more accurate determination of the signal envelope containing loudness-over-time and pitch information. The analysed frequency range was extended to 200–10,000Hz in the TEMPO + processor, compared to only 300–5,500Hz in the CIS PRO + body-worn processor. The TEMPO + obsoleted the body-worn processor in the year 1999.

In 2001, a multicentric study led by Prof. Helms from the University of Würzburg in Germany reported on the comparison of the TEMPO + BTE, and CIS PRO + body-worn processor in adult MED-EL CI experienced users [11]. The study comprised of forty-six post-lingually deaf adults who were native German-speaking and experienced users of MED-EL COMBI 40/40+ CI system. All participants partook in two test sessions, the first one immediately after receiving and fitting of the TEMPO + processor, and the second one approximately four weeks later. In both sessions, speech understanding with both processors with the same signal-to-noise ratio (SNR) was assessed in a free-field test using monosyllabic words which were widely used in German-speaking areas. Group scores for the monosyllabic word test are displayed in Figure 8. The grey column on the left shows the mean CIS PRO + score for the first test session, while the grey column on the right represents the respective mean score for the second session. TEMPO + results are shown in red in the same configuration as CIS PRO + results. Group mean scores are well within the target score range (30% to 70%). In addition, the group mean values for the CIS PRO + in the first test session are close to 50%.

Figure 8. Mean correct scores for monosyllabic words (n = 46). Grey columns: mean group results for the CIS PRO + in the first (left) and the second (right) test session; red columns: mean group results for the TEMPO + in the first (left) and the second (right) test session. Histogram created from the data given in Helms et al. [11].

In general, TEMPO + scores were higher than the CIS PRO + scores. Relating to the increase in mean scores of the first session from 44.6% to 46.7%, through the second session from 46.1% to 48.1%, towards the target score range, gives a relative increase of approximately 5%. The results obtained in the study indicated the superiority of the TEMPO + over the CIS PRO+. With the miniaturisation of the device, speech understanding has not been compromised. On the contrary, the TEMPO + provided higher levels of speech understanding than the CIS PRO+.

5.2.2. Limitations of CIS strategy

The CIS strategy has undoubtedly overcome the limitations of CA strategy by eliminating the significant interactions among channels by interleaved non-simultaneous stimuli. The CIS strategy presents the channel-specific envelope information of the sound signal derived from the bandpass filter outputs via rectification and low-pass filtering (stage 2 of CIS strategy, as shown in Figure 5) or via Hilbert Transform (stage 3 of CIS strategy, as shown in Figure 5). However, the envelope extraction largely discards the temporal fine structure (TFS) information present in the bandpass outputs [12].

In a normal acoustic ear, the envelope information is represented most prominently in neurons tuned to high frequencies, and temporal fine structure (TFS) is represented via phase-locking most prominently in neurons tuned to low frequencies. It is reasonably well-established that human sensitivity to such phase-locking to the TFS of stimuli is limited to frequencies below 1,500 Hz [13]. All CI systems that use CIS type strategies convey mainly the envelope information in different frequency bands, whereas TFS is largely missing, at least apart from the envelope modulations that mainly code the fundamental frequency of the sound only. Figure 9 shows the envelope and fine structure components of a sound wave.

Figure 9. Simple illustration of the envelope and fine structure components of a sound signal (image courtesy of MED-EL).

5.2.3. Fine structure processing (FSP) strategy

In the normal acoustic ear, phase locking is an important phenomenon, predominantly occurring in the LF region, i.e. around the apex where the LF signals are processed and where the neural responses are generated in synchrony with the sound frequency. Every aspect of MED-EL’s CI system is inspired by nature, and this concept is also applied to sound coding strategies. In 1997, Prof. von Ilberg from the Goethe University Frankfurt in Germany came up with the concept of electric-acoustic stimulation (EAS) to treat partially deaf patients with electric stimulation in the HF region and acoustic amplification in the LF region. The EAS concept gives users access to fine structure information through acoustic amplification of the LF region (Figure 10). Research has shown that EAS users hear better in comparison to the regular CI users, and especially so with regards to speech perception in noise and music appreciation. As a consequence of the results with EAS, MED-EL has developed the FineHearing™ technology to better model the natural acoustic hearing by providing sound-rate stimulation to apical electrode channels while retaining the constant-rate stimulation to the basal electrode channels, as depicted in Figure 10.

Figure 10. Illustration of EAS and FineHearing™ concept. EAS provides acoustic amplification of the functional LF region using hearing aids, and electric stimulation of the HF deaf region using a CI electrode array. FineHearing™ concept provides sound-rate stimulation (electric stimulation in synchrony with the sound rate) to the apical electrode channels that are physically placed well beyond the basal turn of the cochlea to have a place match and constant-rate stimulation to the basal electrode channels (image courtesy of MED-EL).

In 1999, the FineHearing™ technology (FSP) was initiated based on a patent by Prof. Zierhofer (US patent number: 6594525) from the University of Innsbruck in Austria (academic-industrial partnership program with MED-EL until 2005). In this patent, Prof. Zierhofer proposed the concept of channel-specific sampling sequences (CSSS) which allow the representation of the temporal fine structure of a sound signal through a CI in addition to the envelope, and this was implemented in the OPUS BTE audio processor.

In 2006, the development of the FineHearing™, the fine structure processing (FSP) coding strategy by Dr Nopp und Dr Schleich from MED-EL was finished. FSP was the first coding strategy approved for clinical use to overcome the limitations of envelope based coding strategies, which do not use the timing of stimulation pulses as a carrier of information. The fine-structure coding strategy uses CSSS that monitors the bandpass filter output (stage 2 in the CIS strategy, as shown in Figure 5) for zero-crossings in band-pass signals, as shown in Figure 11. At times of positive zero-crossings, stimulation pulses are triggered in synchrony with the instantaneous frequency of the bandpass signal. The fine structure is derived up to frequencies of approximately 350 Hz and results in different electric stimulation patterns for apical and basal electrodes. Fine structure stimulation in the apical region of the cochlea results in neural responses in synchrony with the instantaneous sound frequency – in other words, the apical electrode channels will apply electric pulses at the same rate and in synchrony with the sound frequency. This brings the phase-locking functionality of the normal acoustic hearing to MED-EL’s CI system. In contrast to the FSP strategy, the CIS strategy stimulates at a fixed rate. In addition, the lower cutoff frequency is decreased to 70 Hz for FSP, compared to 250 Hz for the CIS strategy. For FineHearing™ to work optimally, the apical electrode channels should reside in the LF region, i.e. covering the entire cochlea with an electrode array, which is called Complete Cochlear Coverage (CCC). All in all, the MED-EL CI system uses a flexible, atraumatic electrode array that is long enough to achieve CCC, and fine structure stimulation aims to closely mimic the functions of the normal hearing ear in deaf and hard of hearing recipients.

Figure 11. Differences between CIS and fine structure stimulation. Band-pass filter represents the original filtered signal, and the blue line represents the envelope function (A). Fine structure stimulation applies pulses at zero crossings of the band-pass output with an amplitude according to the envelope function (B). CIS stimulates at a fixed stimulation rate according to the envelope function (C). Image courtesy of MED-EL.

FSP was implemented in OPUS 2 BTE processors (US patent numbers: 8639359 and 9566434). The OPUS 2 processor was the first processor to feature a remote control – called FineTuner™. The latter enabled the user-friendly processor setting changes without a need for removing it from behind the ears. Dr Stöbich (Project Leader) and his colleagues from MED-EL developed the OPUS 2 processor.

In 2007, Prof. Arnoldner and his colleagues from the Medical University of Vienna in Austria evaluated speech perception, music perception and the general acceptance of the new FSP strategy implemented in the OPUS processor, compared to the CIS strategy implemented in the TEMPO + processor [14] (Figure 12). Fourteen postlingually deaf patients implanted with MED-EL CI devices participated in the study. The speech perception tests consisted of the two-digit number test, Freiburger monosyllabic word test and the HSM sentence test. Tests were presented in a soundproof room at 60- and 80-dB HL in quiet and in noise. Tests were performed at the baseline visit with TEMPO+ (CIS) speech processor, immediately after fitting of the new OPUS (FSP) processor, and consecutively at fourth, eighth and twelfth week after the first fitting.

Figure 12. Clinicians from the Medical University of Vienna, Austria, compared the hearing performance with TEMPO+ (CIS) and OPUS (FSP) processors in adult CI users, implanted with MED-EL CI system (image courtesy of MED-EL).

Figure 13 shows the results of the speech recognition tests in quiet and in noise. Mean results improved for all patients from baseline visit (CIS) to visit 4 (FSP).

Figure 13. Speech test results for all patients at the baseline visit with TEMPO+ (CIS) immediately after fitting with OPUS (FSP) processors (visit 1) and four, eight and twelve weeks after that (visits 2–4). Bars indicate 95% confidence interval and the continuous improvement of mean values overtime for all tests is seen. Asterisks indicate statistical significance compared with the baseline visit [14]. Statistical analysis: Paired samples two-sided t-test (p ≤ .05). Reproduced by permission of Taylor and Francis Group.

For the number test, scores rose from 78.9% (TEMPO+) to 85% (OPUS, visit 4), for the monosyllable test from 45.12% to 48.49%, and for the HSM test from 57.97% to 69.25%. In the noise condition, scores improved even more evidently for the HSM test – from 45.89% to 57.48% at 15 dB SNR, from 22.51% to 45.00% at 10 dB SNR, and from 8.83% to 21.63% at 5 dB SNR. These improvements were statistically significant for the numbers and the HSM tests in all conditions, with and without noise. The data presented in this study were the very first results of the new FSP coding strategy. The excellent outcomes with significant improvements in the speech tests encouraged MED-EL to implement and further fine-tune the coding strategy in its sound processors.

In 2010, Prof. Van de Heyning and his colleagues from Antwerp Medical University in Belgium investigated the effects of the new FSP strategy on speech perception in noise and quality of life through a prospective study, followed up to twelve months [15] (Figure 14).

Figure 14. Clinicians from Antwerp Medical University, Belgium, studied the long-term effects of the FSP strategy in experienced CI users, implanted with MED-EL CI system.

Thirty-two patients were implanted with 31.5 mm long electrode array, as shown in Figure 10 (FineHearing™ segment) and were fitted with the TEMPO+ (CIS + strategy) processor. After an average of seventeen months of experience with the TEMPO + processor, participants switched over to the OPUS 2 processor. Twenty-two participants switched to the FSP sound coding strategy, and in remaining ten, the processor upgrade resulted in high definition continuous interleaved sampling (HDCIS) maps with no fine-structure channels assigned because of longer pulse durations in their maps. Thus, the latter participants were not able to benefit from improved fine-structure coding but were only able to benefit from the extended sound frequency range in FSP. The MAESTRO fitting software automatically did this conversion to HDCIS. Speech perception tests, including speech recognition in noise, were tested using the Leuven Intelligibility Sentence Test (LIST) that consists of thirty-five lists of ten sentences that are representative of daily communication. Participants were tested with the TEMPO + processor just before the switchover, and then with the OPUS processor at switchover, and after one, three, six and twelve months of OPUS use. At the twelfth-month interval, they were also tested in an acute manner with the TEMPO + processor that was fitted with the map they had been using just before switchover. Speech reception threshold (SRT) in noise for the FSP and HDCIS groups are shown in Figure 15.

Figure 15. Results of speech reception threshold (SRT) in noise for the FSP group (A) and HDCIS group (B)—lower the SRT value, better the result is. Statistical analysis: Parametric student’s test with significance calculated for p < .01. (Histograms created from the data given in Vermeire et al. [15]).

Before switchover, i.e. with the TEMPO+ (CIS+) processor, the mean SRT was 16.2 dB, and it deteriorated to 19.5 dB at the acute switchover to OPUS processor using the unfamiliar FSP strategy, with no significant difference. During the first twelve months of FSP use, the SRT gradually improved to 9.7 dB, reaching statistical significance at the twelfth-month interval. When the participants switched back to the TEMPO + processor at the twelfth-month interval, acute testing showed a mean SRT of 10.6 dB, which was not significantly different to the result at twelve-months of OPUS processor use (Figure 15(A)). For the group of participants using HDCIS, mean values for speech perception in noise at different time intervals are shown in Figure 15(B). In this group, the mean SRT changed from 17.5 dB before the switchover to 17.7 dB at the acute switchover stage and 12.5 dB after twelve months of use. After twelve months of HDCIS use, when the participants switched back to TEMPO+ (CIS+) processor and were acutely tested with the CIS strategy, results did not show any significant difference in SRT compared to twelve months of use with HDCIS.

In 2011, a follow-up study of the abovementioned study was published by the clinicians from Antwerp Medical University, evaluating the long-term effects in the range of twenty-four months with a focus on the improvement in speech perception with FSP coding strategy [16]. Figure 16 shows the mean SRTs in noise for the FSP and HDCIS group at each test interval that extends to twenty-four months. After twenty-four months of FSP experience, the SRT decreased significantly from 9.7 dB SNR at twelve months to 3.0 dB SNR, as shown in Figure 16 within the FSP group with an asterisk. Whereas with the HDCIS group, the SRT in noise resulted in 10.9 dB SNR, which was not significantly different from the twelfth month's value of 12.5 dB SNR.

Figure 16. Mean SRT in noise for the FSP and HDCIS group up to twenty-four months after upgrade to the OPUS 2 audio processor. The white bar indicates CIS, red bars indicate FSP, and grey bars indicate HDCIS. Statistical analysis: Post hoc pairwise signed-rank tests to assess SRTs in noise change over time (p < .05). Histograms created from the data given in Kleine Punte et al. [16].

The results presented in these two studies show that by focusing on fine structure coding in the LFs, speech perception in noise can be enhanced. An important learning effect can be seen, indicating that it can take patients some time to be able to benefit from the FSP strategy.

In 2010, Prof. Skarzynski and his colleagues from the Institute of Physiology and Pathology of Hearing in Poland published about the benefit of CIS+, HDCIS and FSP strategies, both qualitatively and quantitatively, in sixty children implanted with a MED-EL CI system comprising with a long electrode array length of 31 mm [17] (Figure 17).

Figure 17. Clinicians from the Institute of Physiology and Pathology of Hearing, Poland, compared the benefits of CIS, HDCIS and FSP sound coding strategies, both quantitatively and qualitatively in children implanted with MED-EL CI system.

CI surgery had been performed in all children at an average age of 3.8 years, the average time of device use was 6.3 years, and the average age at upgrade was ten years. Adaptive Auditory Speech Test (AAST) is a closed-set procedure with the presented stimuli as trisyllabic words where the child chooses an answer from the six pictures shown. In the adaptive procedure, the speech level varies to obtain the SNR for a 50% correct score (speech reception threshold). The AAST was conducted for the HDCIS strategy at the interval I, and for all three strategies at intervals II and III. Visual Analogue Scales (VAS) were completed by the children to reflect subjective judgement with each coding strategy for music stimuli, as well as to make comparisons between coding strategies. The VAS scale for satisfaction required the child to mark whether the strategy was bad, average or good with smiley faces on a 20 cm scale to assist children in decision making.

Figure 18(A) shows the AAST test in noise with no significant interaction effect between strategy and interval and no overall significant effect for the interval. However, an overall significant effect was reached for strategies with FSP better than CIS + by 0.7 dB, and HDCIS better than CIS + by 0.8 dB HL. No statistically significant difference was found for HDCIS and FSP. Figure 18(B) shows the VAS satisfaction rating for music stimuli. VAS results for music stimuli at interval II revealed an overall positive effect for the strategy with FSP better than CIS + by 27.1% and HDCIS better than CIS + by 31.5%. However, no significant difference was seen in music stimuli between FSP and HDCIS. Results for music stimuli at interval III were also significant for strategy showing FSP better than CIS + by 32.4% and HDCIS better than CIS + by 22.3%. No differences were found for FSP versus HDCIS.

Figure 18. Results for the AAST test in noise as a function of interval (A) and VAS satisfaction scaling using the music stimuli (B) [17]. Statistical analysis: Two-way repeated measures ANOVA test. Reproduced by permission of Elsevier Ireland Ltd.

Overall, the FSP strategy offered better SRTs in noise and music acceptance compared to HDCIS and CIS+, and the importance of the study increased with the fact that the tested patients were children who did not need to undergo reimplantation in order to benefit from the new developments in the CI technology. This applies to other patients as well, as they do not require reimplantation to acquire the newest technological upgrades and hearing benefits.

In 2012, a clinical trial evaluating the effectiveness of the FSP strategy in experienced MED-EL CI users was reported by clinicians from various CI centres in Germany and was led by Prof. Müller [18] (Figure 19).

Figure 19. Clinicians from different German clinics and engineers from MED-EL, involved in the clinical trial results evaluating FSP strategy in experienced MED-EL CI implant users. ¹University of Würzburg, ²Technical University of München, ³Goethe University Frankfurt, ⁴Carl Gustav Carus University Hospital Dresden, ⁵University of Innsbruck, Austria, and ⁶MED-EL, Innsbruck, Austria.

Forty-six postlingually deaf adults with a minimum of six months MED-EL CI experience participated in this study. Their mean age at implantation was fifty-four years, and the mean age at testing was fifty-six years. They all had at least two years of experience with a TEMPO + processor (CIS + strategy) prior to switchover to an OPUS processor (HDCIS/FSP). The study aimed to compare CIS+, HDCIS and FSP strategies, mainly in terms of speech perception test results in noise using a vowel test, the Freiburger monosyllable word test and the German OLSA sentence test in noise at three months post-switchover. For the OLSA test, the speech level was constant at 70 dB SPL, and the noise level varied in order to determine the SNR that resulted in a 50% correct for each individual. Data of speech perception tests are shown in Figure 20.

Figure 20. Speech perception scores at third-month post-switchover to OPUS processor (HDCIS/FSP) from TEMPO + processor (CIS). Participants were tested with vowels, Freiburg monosyllable and OLSA using the FSP, HDCIS and CIS coding strategies. Statistical analysis: Paired sample t-tests (p < .05). Box plot created from data given in Müller et al. [18].

At the test intervals, vowel scores were similar for FSP (64.4 ± 10.9%) and HDCIS (65.4 ± 12.5%). Those for FSP were significantly higher than those for CIS+ (59.6 ± 11.2%). HDCIS vowel scores were significantly higher than CIS + scores. Monosyllable scores showed the same behaviour for FSP (44.8 ± 19.03%) and HDCIS (42.3 ± 18.8%) however, FSP and HDCIS showed significantly higher scores in comparison to CIS+ (38.9 ± 17.8%). With the OLSA test, SRTs were slightly lower for FSP (3.0 ± 6.7 dB) and HDCIS (2.9 ± 7.0 dB) than for CIS+ (3.4 ± 7.7 dB), with no significant differences among them. Pitch scaling was another important test that was carried out in this study and which showed LFs sounding low-pitched with FSP strategy compared to CIS+, which reflects the benefits of the more natural low-frequency coding via sound-rate stimulation with FSP. The results from this clinical study demonstrated that users of FSP or HDCIS, as implemented in the OPUS processor, performed equal or better when tested with CIS+, as implemented in the TEMPO+.

In 2013, the interest in FSP coding strategy moved to China to evaluate the benefits of FSP strategy in Mandarin-speaking CI users. Mandarin is a tonal language in which, the pitch is used to distinguish different words. The study was conducted by Prof Han and his colleagues from Beijing Tongren Hospital, involving ten MED-EL CI users (with OPUS 2 processor), aged eighteen years or older [19] (Figure 21). The mean age at implantation was 31.1 years, and the speech performance was assessed before and after cochlear implantation using monosyllables in quiet and sentences in quiet test, called Mandarin Speech Test Materials (MSTM), Mandarin Hearing in Noise Test (MHINT) and Mandarin tone perception test. The Mandarin tone perception test was designed by Dr Krenmayr, an engineer from MED-EL [20].

Figure 21. Prof. Demin Han and Dr Xueqing Chen from the Beijing Tongren Hospital, China, conducted the study along with her colleagues. Dr Andreas Krenmayr is an employee at MED-EL who designed the Mandarin Perception Test.

Figure 22 shows the monosyllables in quiet, sentences in quiet, MHINT, and tone perception results in percentage correct. All the audiological tests showed no statistical significance at first fitting compared to preoperative scores. However, at three months, there was a significant improvement compared with preoperative scores with all speech tests. Monosyllables in quiet and sentences in quite improved significantly at six months, compared to preoperative scores. There was a significant improvement in speech perception in all speech tests at three months compared with the first fitting, and at six months compared with the first fitting. Tone perception did not improve significantly at first fitting compared to preoperative results, nor at three months – but there was a significant improvement at six months. This was the first study that evaluated the FSP strategy in the Mandarin language, and overall, it showed significant improvement in Mandarin speech and tone perception of adult CI users who had no prior CI experience.

Figure 22. Results are shown in percentages correct for monosyllables in quiet (monosyll), sentences in quiet (Sentences), MHINT, and the tone perception test (Tone perc) over time. Horizontal lines on the box plot represent median values; black squares represent mean values. Ff: first fitting; 3-m: 3 months; 6-m: 6 months [19]. Statistical analysis: Repeated measurements ANOVA test (p < .05). Reproduced by permission of Taylor and Francis Group.

The six studies presented in this section reported on the benefits of FSP strategy implemented in OPUS audio processor over the CIS + strategy implemented in TEMPO + audio processor. By keeping the fine structure information in the sound signal by applying phase-locking low-frequency pulses to the LF apical channels in the FSP strategy, along with electric stimulation covering the entire frequency range, helps MED-EL CI users, including Mandarin speakers, to experience near-normal hearing and music acceptance, compared to CIS strategy.

5.2.4. Advancements in FSP strategy

With scientific evidence showing better hearing experience for patients with the FSP strategy, it shall be noted that the benefits of FSP over CIS came by adding fine structure information in the LFs between 70–350Hz. In general, FSP provided fine structure information on up to two apical channels, depending on the individual map parameters – in other words, with the FSP strategy, up to two apical channels are stimulated in synchrony with the sound frequency. FSP monitors for zero-crossings of the bandpass filter output and triggers stimulation pulse packages (CSSS) in synchrony with the sound frequency, as previously described in Figure 11.

In 2010, MED-EL got further improved its FSP strategy by providing fine structure information to four apical channels instead to only two, and hence the name FS4 strategy (Figure 23(B)). Within the FS4 strategy, if two channels are identified with the zero-crossing at the same time, as shown in Figure 23(A), then the system picks the channel that has higher amplitude for providing the fine structure information (Figure 23(B)). An additional variant was added to the FS4 strategy, called FS4-p. The FS4-p strategy presents fine structure information for more than one channel if zero-crossings appear at the same time, or in parallel (Figure 23(C)). However, two channels receiving zero-crossing at the same time is unlikely, but still possible.

Figure 23. Zero-crossing of the bandpass output (A). FS4 strategy with single-pulse CSSS. In the event of zero-crossings coinciding on two or more FS channels, FS4 picks the channel with the highest instantaneous pulse amplitude for stimulation (B). The FS4-p strategy provides simultaneous stimulation pulses on two channels with coinciding zero crossings (C). Image courtesy of MED-EL.

This was mainly proposed by Dr Nopp, Dr Schleich, Dr Meister and Dr Schatzer (at the time with the University of Innsbruck) within the Sound Coding research group of MED-EL. The FS4-p strategy provides the possibility to have parallel stimulation in more than one apical channel at a time and needs an additional algorithm to compensate for the effects of simultaneous stimulation (Figure 24).

Figure 24. Signal Processing engineers from MED-EL who proposed the initial concept of FS4 and FS4-p. They were also instrumental in writing algorithms to handle the stimulation frame, sequence of channels, channel groups to be stimulated and in testing parameter variations.

This algorithm is referred to as the channel interaction compensation (CIC) algorithm, patented by Prof. Zierhofer. CIC compensates for the effects of simultaneous channel interaction by computing reduced amplitudes such that after direct electric field summation with simultaneous stimulation, the field distribution resulting from sequential stimulation is approximated. It should also be noted that FS4-p strategies can only be applied to implant systems that allow parallel stimulation and are not applicable to implant systems like COMBI 40 and COMBI 40+.

In 2014, Prof. Rajan and Dr Távora-Vieira from the University of Western Australia published data on subjective preferences and speech perception of unilaterally deaf CI users with FS4 and FS4-p [21] (Figure 25). Thirteen users who had received a CI from MED-EL with the FLEXSOFT™ electrode array were fitted with OPUS 2 processor, and all patients had at least three months experience of using FSP strategy.

Figure 25. Clinicians from the University of Western Australia subjectively assessed the FS4 and FS4-p strategies in unilaterally deaf CI users implanted with MED-EL CI device.

The patients were provided with two maps – FS4 and FS4-p – in a blinded manner for assessing their subjective preference towards these different coding strategies. They were asked to rate the two maps on five qualitative attributes daily, as indicated in Figure 26. While speech perception scores were not significantly different among FS4 and FS4-p, all patients showed a subjective preference towards the FS4-p strategy. Providing the fine structure information to the four apical channels and in addition, providing such information simultaneously in more than one channel offered subjectively a more natural hearing experience to the MED-EL CI users.

Figure 26. The group results for each of the five questions. FS4 is shown in grey boxes, and FS4-p is shown in red boxes. Mean values are depicted as black squares, medians as horizontal lines and asterisks are the outliers—statistical analysis: Wilcoxon signed-rank tests (p < .05). Box plot adapted from Távora-Vieira et al. [21].

In the same year, Dr Riss and his colleagues from the Medical University of Vienna in Austria compared FS4 and FS4-p with FSP strategy in thirty-three postlingually deaf adults. FSP was used as the default strategy [22], but each participant was fitted with these three different strategies for four months in a randomized and blinded order. After each run, an adaptive sentence test in noise (Oldenburger Sentence Test (OLSA)) and a monosyllable test in quiet were performed. Scores of the OLSA did not reveal any significant differences among the three strategies, but the monosyllable word test showed a statistically significant effect (p = .03) with slightly worse scores for FS4 (49.7%) compared with FSP (54.3%). Performance of FS4-p (51.8%) was comparable with other strategies. The results of this crossover study showed great variability between CI users, and overall, the average results were similar among all three tested sound coding strategies with regards to speech perception in noise. Nevertheless, the majority of the participants subjectively preferred one of the strategies of FS4 or FS4-p over FSP. The study showed that each user might have an individual preference for FS coding strategy.

Today, MED-EL has all these sound coding strategies (HDCIS, FSP, FS4 and FS4-p) in its product portfolio giving the choices to the audiologists to try the ones that would provide the best hearing experience to the individual MED-EL CI users.

5.3. Front-end processing

Front-end processing aims to model the functionality of external and middle ear covering the directionality and filtering processes, respectively. This section will cover the innovations in front-end processing that were implemented in MED-EL’s audio processors and the scientific studies that assessed the benefits of those in MED-EL CI users.

5.3.1. Automatic sound management 1.0

Automatic Sound Management (ASM) is a term created to bring together a set of front-end features that were implemented in the audio processors at various time points at MED-EL. Automatic Gain Control (AGC) is one of the features within ASM that attenuates high-level signal and enhances low-level signal, enabling the CI user to hear even a very soft sound signal. AGC recreates or models the sound level compression function of the basilar membrane and compresses the range of sound levels by mapping a dynamic input range of 75 dB to a narrower output dynamic range. This feature is available in all MED-EL audio processors, including in off-the-ear processors, existent since 2013. AGC is the first-ever and the only front-end feature that was implemented in MED-EL’s COMBI 40 body-worn audio processor and is still available along with other advanced features in the latest SONNET2 BTE audio processor. The modern AGC in CI audio processor carries a dual time constant compression system (slow and fast detector). The slow detector is generally in control of the system gain and mainly determines the dynamic properties of the AGC. The exceptions are sudden intense transient sounds (like door slamming) when the AGC gain is determined by the fast detector, which immediately reduces the system gain.

In 1999, Dr Stöbich (at the time a PhD student), Prof. Zierhofer and Prof. Hochmair from the University of Innsbruck published on the evaluation of the AGC with dual time constant compression system under six different settings in MED-EL CI users fitted with COMBI 40+ audio processor [23] (Figure 27).

Figure 27. Group of scientists from the University of Innsbruck evaluated the AGC with dual time constant compression system under six different settings in MED-EL CI users fitted with COMBI 40+ audio processors.

In linear mode, the AGC operates as a linear amplifier with a fixed gain of +20dB. The setting standard is the standard AGC (compression limiter) of the MED-EL COMBI 40 body-worn processor that has only one peak detector. The remaining four configurations, i.e. 3:1 rapid, 6:1 rapid, 3:1 slow, 6:1 slow, are slow-acting dual time constant structures. The Göttingen German language sentence test was used to test the CI users hearing performance under the abovementioned six different AGC settings, and the results are given in Figure 28. The results showed that CI users performed significantly better with all four dual front-end configurations than with the standard AGC in situations where intense transient sounds were present.

Figure 28. Mean correct score of six users fitted with MED-EL’s COMBI 40 body-worn audio processor, tested with six different AGC settings. Histogram created from the data given in Stöbich et al. [23].

Overall, the results indicated that slow-acting front-end AGC could be used effectively in speech processors for CIs to expand the range of input levels that can be heard by the CI users compared with a linear amplifier without any need to adjust a processor control. This was an encouraging result that made MED-EL implement it in its audio processor. AGC was the only front-end feature that was a part of ASM 1.0 in all the audio processors, including the COMBI 40+ (CIS PRO+) BTEs including OPUS and OPUS2, and off-the-ear single unit, including RONDO and RONDO 2 which are described below.

In 2010, an important report was published by Dr Haumann, Prof. Lenarz and Prof. Büchner from Hannover Medical School in Germany in which they evaluated CI patients fitted with audio processors of various CI brands in more realistic listening situations [24] (Figure 29).

Figure 29. Clinicians from Hannover Medical School, Germany, who evaluated audio processors of various brands under more realistic listening situations.

Groups of eleven participants each matched for performance in quiet with five different CI systems, with a total of fifty-five participants (similar age group), were tested with an adaptive test regime where the presentation level of the speech signal roved by ±10 or ±15 dB. The HSM sentences were presented at a roving level, and the noise was adapted to obtain the SNR for a 50% correct score, commonly referred to as the Speech Reception Threshold (SRT). Within each test sentence list, the presentation level of each sentence was randomly roved by either 0, +10 or −10dB in the ±10dB roving condition, and by either 0, +15 or −15dB in the ±15dB roving condition. The base (0 dB roving) presentation level was 65 dB SPL for both tests, hence ranges of 55–75 dB (±10dB roving) and 50–80 dB SPL (±15dB roving) were explored. Speech-shaped noise (CCITT noise) was used as the competing signal and it started 0.5 s before the sentence and finished 0.5 s after the sentence.

Although not significant, a clear trend (p = .083) was found for SRT values being higher (i.e. worse) for the ±15dB roving condition than for the ±10dB roving condition. Figure 30 shows the comparison between results for the individual CI brands. Results are widely scattered for all devices, although the scatter seems to be more pronounced in users of the Espirit 3 G processor. The users of OPUS 2 processor from MED-EL showed significantly lower SRT values (p = .045) for the ±15dB roving condition than for the ±10dB roving condition. For all other groups, these differences were not significant. Within ±10dB roving condition, post-hoc testing found significantly smaller SRT values for the OPUS 2 and Harmony group, than for the Freedom group. Within ±15dB roving condition, post-hoc testing revealed that all other groups apart from the Freedom group showed significantly smaller SRT values than the Espirit 3 G group. Similarly, the Harmony group and the OPUS 2 group showed significantly smaller SRT values than the Freedom group. Finally, the OPUS 2 group showed significantly smaller SRT values than the Auria group.

Figure 30. Individual results (small crosses) as well as mean values (crosses on the left and right) and standard deviation values (error bars) as a function of speech processor and test condition [24]. (±15dB is more difficult listening condition than ±10dB). Reproduced by the permission of Karger, Basel.

The investigators of this study thought that the technical parameters most challenged by a roving-level test are the input dynamic range (IDR) and automatic gain control (AGC) of the speech processors. The Auria and Harmony speech processors by Advanced Bionics are reported to have an IDR of 80 dB and for the OPUS 2 speech processor by MED-EL, an IDR of 75 dB is given. In contrast, for the Espirit 3G and Freedom speech processors by Cochlear Corporation, an IDR of only 30–45 dB is recommended. Furthermore, the speech processors by MED-EL and Advanced Bionics are reported to feature dual-loop AGCs, whereas for those by Cochlear™, a single loop AGC is reported. They concluded that their results show that speech processors featuring a wider IDR and a dual-loop AGC are advantageous when tested under more realistic test conditions like the roving-level test used in their study.

In 2013, MED-EL, as the first CI manufacturer, developed a single unit audio processor that combined the processing unit, battery pack and the head-piece all in one unit, named RONDO. Figure 31 compares RONDO with OPUS 2 processor.

Figure 31. Comparison of BTE (OPUS 2) and single unit (RONDO) processor (image courtesy of MED-EL).

This was a significant change in the audio processor design, led by Dr Stöbich and his colleagues from MED-EL. It featured advantages over the BTE processor in terms of cosmetic look, as it could be hidden under hair and give comfort to people wearing glasses. One of the questions that arose was how the position of the single-unit audio processor, which is positioned away from the pinna, would affect the hearing performance of the CI users.

In 2014, Prof. Mertens and her colleagues from Antwerp University Hospital published on the assessment of the SSD patients who had received a MED-EL CI with a BTE audio processor and were offered the single unit RONDO audio processor, to study if there was any difference in hearing performances with the two different sound processor designs [25] (Figure 32).

Figure 32. Clinicians from Antwerp Medical University, Belgium, compared the effectiveness of a BTE processor with the single-unit audio processor by evaluating the hearing thresholds between them.

Ten SSD patients with severe tinnitus with an average age of fifty-six years were included in the study. All of them had an average of eight years of CI experience. The hearing performance assessment was first made with their BTE processor, followed by the application of the RONDO processor for twenty-eight successive days. Outcome measures included unaided and aided hearing thresholds, speech perception in noise and sound localisation. Subjective assessments included sound quality assessment, hearing (dis)ability using Speech, Spatial and Qualities (SSQ), spatial hearing abilities, tinnitus loudness and user feedback questionnaire, and microphone position. None of these tests showed any significant difference in hearing performance between these two audio processors.

The positive aspects of the single-unit processor (RONDO) were no skin pressure, no skin irritation and the comfort to wear glasses as observed from the feedback questionnaire. The study concluded that long-term BTE audio processor of SSD users could be upgraded to a single-unit audio processor without compromising their speech performance, aided hearing thresholds, sound localisation, objective speech quality, hearing abilities and tinnitus reduction.

In 2016, group of Clinicians from Germany led by Prof. Mlynski published their findings on the effect of RONDO audio processor on speech perception of experienced CI users compared to OPUS2 processor [26] (Figure 33).

Figure 33. Clinicians from different clinics in Germany who were involved in the assessment of RONDO audio processor. ¹Ruhr-University Bochum, ²Deutsches HörZentrum Hannover, ³Ludwig-Maximilians University, Munich, ⁴Tübingen University Hospital and ⁵Rostock Medical University Center.

Fifty subjects were enrolled in the study with a mean age of 56.1 years and mean duration of hearing loss of 20.2 years. The subjects had at least 3.2 years of OPUS 2 experience before upgrade to RONDO processor. Freiburg Monosyllable word test showed little changes between OPUS 2 (range 62.4–63.4% correct) and RONDO (range 60.3–61.9% correct). The German OLSA in noise showed again little changes between OPUS 2 (range 2.2–4.1 dB SNR) and RONDO (range 1.9–4.6 dB SNR) audio processors. The study concluded that RONDO provides comparable speech perception to the OPUS 2 and it is a suitable and safe alternative to traditional BTE audio processors.

In 2017, MED-EL launched the RONDO 2 single-unit processor, which is an advanced design of RONDO. RONDO2 came up with wireless charging, making it easy to power up the device without the need of removing the battery from the processor. Because of its smaller size, it gets easily hidden under hair and gives comfort to people wearing glasses. It comes with WaterWear, which is a reusable waterproof cover that is easily attached and may be used in any type of water (Figure 34). Within MED-EL, it was Dr Duftner who had the project leadership role in product development.

Figure 34. RONDO-2 single-unit processor. Dr Alexander Duftner from MED-EL assumed the project leader role for this product development.

5.3.2. Automatic sound management 2.0

Directionality is an important function of the external ear. With the help of the pinna, the ear can better collect sound signals from the front than from the rear side of the head. Using dual-microphone technology, the directionality feature in MED-EL’s front-end processing offers three possibilities to its users – omnidirectionality, natural/fixed directionality, and the adaptive directionality. Omnidirectional functionality (the front microphone is enabled while the rear is disabled) treats any sound signal coming from all directions equally, whereas the natural/fixed directionality mimics the pinna that is focused to the sound signal coming from the front rather than the rear end. Similar to the human ear pinna, natural directionality is omnidirectional in the low frequencies and increasingly directional towards the front with increasing frequency. The adaptive directionality adapts the directionality patterns in a frequency-depending manner, depending on the acoustic scenario in the back hemisphere of the user. Finally, auto-adaptive directionality switches between omnidirectional for lower signal levels and adaptive for medium and high signals. Wind noise reduction (WNR) is another function that was added to the front-end processing of the SONNET processor. The signals from the microphones are used to monitor any wind noise, and in case of any detection of such, it applies the wind noise suppression network. Wind noise is mainly in the LFs, and the mild mode of WNR acts without reducing the target sound signal. The strong mode provides heavier wind noise reduction but also affects the target sound signal to some extent.

Directionality and WNR functions were made available in the SONNET and SONNET EAS audio processors. SONNET and SONNET EAS are available since 2014 with these two additional front-end processing features under the term Automatic Sound Management 2.0 (ASM 2.0). The SONNET processors are complemented by the AudioLink universal connectivity device that allows to stream from mobile phones, tablets, TVs and much more, directly to the SONNET audio processor. At MED-EL, it was Dr Aschbacher and his colleagues from the Signal Processing research group who were responsible for implementing dual microphones (Figure 35).

Figure 35. Dr Ernst Aschbacher (Team Leader- Front-end processing) and his colleagues from Signal Processing research group at MED-EL was responsible for implementing dual-microphone in SONNET audio processor. SONNET audio processor showing dual microphone that gives the directionality function and WNR to the audio processor. AudioLink is a universal connectivity device that connects the audio processor with media players and mobile phones using Bluetooth connectivity. Image courtesy of MED-EL.

In 2018, Prof. Baumann and his colleagues from Goethe University Frankfurt in Germany compared the speech perception in quiet and in noise between two EAS audio processors (DUET and SONNET EAS) to assess the impact of front-end processing, including microphone directionality (MD) and WNR [27] (Figure 36).

Figure 36. Clinicians from Goethe University Frankfurt, Germany, compared the speech perception in quiet and noise between two audio processors to assess the impact of front-end processing, including microphone directionality and wind noise reduction function.

DUET has fixed omnidirectional microphone directionality, whereas SONNET EAS processor offers three modes of microphone directionality (MD), as mentioned above. Ten EAS patients, implanted with MED-EL EAS system and with at least one year of DUET processor use prior to the switchover to SONNET EAS processor, were enrolled in this study. Speech perception in quiet was assessed with Freiburg Monosyllables test for both processors, and mainly this test served as reference and additional verification of proper fitting of the SONNET EAS processor. Speech perception in noise was assessed with Oldenburg sentence test with the noise level fixed at 65 dB SPL and speech level was set adaptively according to the number of words perceived correctly to measure the SRT.

The results of speech perception in quiet are shown in Figure 37(A), and the scores ranged between 75.8 ± 10.7% (SONNET with mild WNR) and 80 ± 12.8% (DUET EAS) with no statistical significance. SRT with DUET EAS was −1.7 ± 2 dB SNR, and with the SONNET EAS using the omnidirectional microphone and WNR off was −2.3 ± 1.9 dB SNR, with no statistically significant difference. Compared with DUET EAS, the SRT with fixed MD natural (SONNET EAS default setting) was 2.2 dB better, and with adaptive MD 3.5 dB better (Figure 37(B)). The results obtained from these experiments showed that the SONNET EAS processor with the front-end features like directionality and WNR provide experienced EAS users with significantly better speech perception, particularly in noisy conditions.

Figure 37. Boxplots of monosyllable scores obtained with the DUET EAS and SONNET EAS processors in three different WNR settings. (A). Boxplots of SRTs with audio processors DUET EAS and SONNET EAS with MD omnidirectional, natural and adaptive directional microphones (B). Grey circles indicate the mean value; open circles indicate outliers [27]. Statistical analysis: RM-ANOVA and Bonferroni-Holm correction method (p < .05).

In 2018, Dr Dorman and his colleagues from Arizona State University in the US published data on the effectiveness of dual-microphone technology in the SONNET audio processor in bilateral CI adult users (n = 10) [28]. Sentence understanding scores in terms of percentage of words correct were tested under one CI and in two CIs in quiet and in noise, simulating real-life test environment. In a restaurant simulating type test environment, the listeners were seated in the centre of eight loudspeakers arrayed in a 360° arc. Sentences from AzBio sentence lists were presented from the loudspeaker at 0° azimuth, and directionally appropriate restaurant noise was presented from all eight loudspeakers, including the speaker from which the target sentences were delivered. For the single CI omni-in-quiet conditions, the mean scores were as follows: omni-in-quiet was 83% correct, omni-in-noise was 28% correct, natural-in-noise was 44% correct, adaptive-in-noise was 51% correct. In bilateral CI test conditions, the mean scores were as follows: omni-in-noise was 40% correct, natural-in-noise was 59% correct, adaptive-in-noise was 71% correct. The results (Figure 38) show that in both single and bilateral CI conditions, the natural and adaptive settings allowed significantly higher scores than the omni- setting. Comparing the single and bilateral CI test conditions, the bilateral CI condition scores were significantly better than the corresponding scores in the single CI test conditions. The data show that both, the natural and adaptive microphone settings significantly improved speech understanding in the noisy environment under both, single and bilateral CI condition, and that they do not impair sound source localisation, with retaining low-frequency sensitivity to signals from the rear. However, the bilateral CI scores were higher than the single CI scores.

Figure 38. Percentage correct word recognition in quiet and in restaurant noise with one CI and with bilateral CIs as a function of microphone setting (Histogram created from the data given in Dorman et al. [28]).

In 2020, Prof. Caversaccio and his colleagues from the Bern University Hospital in Switzerland published their work comparing the sound-source localisation of bilateral CI users with omnidirectional (OMNI) and pinna-imitating (PI) – which is the dual microphone in the SONNET audio processor [29] (Figure 39).

Figure 39. Researcher and clinicians from the University of Bern, Switzerland. ¹Hearing Research Laboratory and ²Bern University Hospital.

Twelve experienced bilateral CI users who wore the SONNET audio processor for at least four weeks were included in the study, and they scored 70% or better with monosyllabic words at 60 dB SPL. The default audio processor modes consisted of FS4 strategy with activated PI directionality mode and WNR disabled. Participants were seated in the centre of a horizontal circular loudspeaker setup with a radius of 1.1 m in the acoustic chamber. Mainly, the static sound source localisation was evaluated in the study of participants with the OMNI and PI microphone directionality modes. Within the sound source localisation test, participants indicated the estimated position using a 1° angle resolution dial-on touchpad. For each test step, two pink noise stimuli with a duration of 200 ms at a sound pressure level of 65 dB were used, separated by a one-second intra-stimulus interval. The perceived stimulus shift was indicated by the participants using a touchpad.

Figure 40 shows the absolute localisation accuracy for each stimulus direction for the OMNI and PI modes. The localisation performance in OMNI mode was the worst in the dorsal azimuth at 150°, 180° and 210° angle with RMSE values of 42 ± 18° angle, 41 ± 27° and 44 ± 27° angle, respectively. The best localisation performance was observed at a 120° angle, 240° angle with 17 ± 9° angle, and 15 ± 9° angle, respectively. In PI model, the localisation errors at the dorsal azimuths (150°, 180° and 210° angles) were reduced, leading to a similar performance compared to the frontal azimuths (330°, 0° and 30° angles). In simple words, for the static sound localisation, the greatest benefit was a reduction in the number of front-back confusions (FBCs). The FBC score was reduced from 27% with OMNI mode to 18% with PI mode. Also, the ability to discriminate sound sources at the sides was only possible with the PI mode.

Figure 40. Averaged RMSE for the omnidirectional (OMNI-circles) and pinna-imitating (PI; red crosses) microphone modes in the static sound source localisation test [29]. Statistical analysis: Two-sided Wilcoxon signed-rank tests (p < .05). Reproduced by permission of Wolters Kluwer Health, Inc.

In 2020, Prof. Hagen and his colleagues from the University Hospital of Würzburg in Germany and engineers from MED-EL published a comparison of the speech understanding in noise and hearing in the real-life situation of MED-EL CI users when fitted with OPUS 2 and SONNET audio processors [30]. Thirty-one participants were assessed for speech understanding in two types of acoustic noise and wind noise. A four-speaker setup was used, and speech was presented from 0° and noise from 90°, 180°, and 270°. Wind noise was simulated with a fan. Oldenburg Sentence Test (OLSA) was used to assess the 50% speech recognition threshold (SRT in dB) in a noisy setting (SRT is a measure of the level difference at which 50% of speech can be correctly identified in the presence of simultaneous masking noise). Figure 41(A) shows speech perception results tested under speech shaped noise (S0N0). Under the test condition, Nat + Mild and OMNI + Off, SONNET was not significantly (statistical and clinical) better than OPUS 2. The same trend was seen with SONNET in Adp + Strong compared to OPUS 2 processor. Within the SONNET processor, none of the modes showed any significance both statistically and clinically. Figure 41(B) shows speech perception results tested under babble noise (S₀N_III), and the SONNET in both Nat + Mild and Adp + Mild were significantly (statistical and clinical) better than with the OPUS 2, but not SONNET in OMNI + Off or OMNI + Mild with OPUS 2. Within SONNET, Nat + Mild was significantly (statistical and clinical) better than with OMNI + Mild. Figure 41(C) shows speech perception results tested under speech shaped noise (S₀B_III) and the SONNET in Nat + Mild and Adp + Mild was significantly (statistical and clinical) better than with the OPUS 2, but not SONNET in OMNI + Off or OMNI + Mild with OPUS 2. Within SONNET, Nat + Mild and Adp + Mild better (statistical and clinical) than with OMNI + Mild. Figure 41(D) shows the speech perception results tested under wind noise (S₀W₄₅), where OPUS2 was significantly (statistical and clinical) better than with the SONNET in all the MD modes. Within SONNET, participants in SRT were better when WNR was activated.

Figure 41. Scores of OLSA test in (A) speech shaped noise (S₀N₀ setup), (B) babble noise (S₀N₁₁₁ setup), (C) speech shaped noise (S₀B₁₁₁ setup), and (D) in the wind without additional noise (S₀W₄₅ setup). Black squares represent mean values; horizontal lines are the median. The black circles depict outliers. The black crosses depict extreme outliers [30]. Statistical analysis: the asterisks depict significance differences (p ≤ .05). Reproduced by permission of Taylor and Francis Group.

The study concluded that SONNET provides the same or significantly improved speech understanding when compared to OPUS 2 in noise. While OPUS 2 was superior in the wind condition when compared to the SONNET in some settings, SONNET was superior in real-life listening situations (SONNET with WNR acts better than OPUS2).

Overall, the studies listed in this section demonstrate the benefit of having dual-microphone in the audio processor, which helps the users to regain natural directionality mimicking the pinna function.

5.3.3. Automatic sound management 3.0

In 2019, MED-EL further advanced the front-end processing features by including Ambient Noise Reduction (ANR), Transient Noise Reduction (TNR) and the Adaptive Intelligence (AI) into the ASM portfolio. These three features, in addition to all other above-listed features, were altogether brought under the term ASM 3.0 and were made available in SONNET 2 audio processor, which is the most advanced BTE audio processor in 2021. ANR monitors for ambient/stationary noise level and reduces the stimulation level on each channel based on the signal-to-noise ratio. TNR reduces transient noises by controlling stimulation levels on the high-frequency channels only. AI classifies the sound signal into one of the five classes – Quiet, Speech, Speech in Noise, Noise, and Music – and controls the ASM 3.0 features (directionality, ANR, TNR, WNR) accordingly to maximise the benefits for the CI user. It was Dr Aschbacher and his colleagues from MED-EL who added these front-end features to SONNET 2 audio processor. Apart from the technical advancements, this processor has the AudioLink universal connectivity device to connect to any media devices, and WaterWear may be used for water resistance.

In August 2020, MED-EL received FDA approval for its RONDO 3 single-unit processor. The ASM 3.0, wireless charging, wireless connectivity and smaller size made it the most advanced single-unit audio processor. Within MED-EL, it was Mr Philipp Schmidt, MSc, who had assumed the project leadership role in developing the RONDO 3 processor as a product.

5.4. Individualisation in sound coding strategy

Literature reveals that the size, shape, anatomy and the frequency map of human cochleae vary individually [31]. This is the valid reason for MED-EL to offer electrode arrays in different lengths to achieve an electrode-place match inside the cochlea. A perfect electrode-place match inside the cochlea would enhance FSP coding strategy to work at its best in helping CI users to hear naturally. MED-EL came up with a unique concept called Anatomy-Based Fitting (ABF) that would assign centre frequencies to individual electrode channels based on patient-specific Greenwood’s frequency map. The other situation where the individualisation in sound coding is needed is when the patient is wearing HAs on one side and CI on the other side. It is known from the literature that HAs have higher latencies compared to the CI. To address this issue, MED-EL implemented artificial time delays to its CI system and named it Bimodal Delay Compensation. Undesirable facial nerve stimulation is experienced by some CI patients due to their special inner-ear anatomical condition. To address this patient group, MED-EL modified the shape of the biphasic stimulation pulses to triphasic stimulation pulses with the aim of minimising the undesirable stimulation of the facial nerve. This section details all these three different individualisation concepts in MED-EL’s sound coding strategies.

5.4.1. Anatomy based fitting

In 2015, Dr Landsberger from New York University School of Medicine in the US, Prof. Van de Heyning from Antwerp University Hospital in Belgium, and their colleagues jointly reported that reliable low-frequency pitch perception in CI requires apical electrodes and a rate-place match [32]. They re-analysed pitch-matching data in SSD MED-EL CI recipients presented earlier by Schatzer et al. [33] with concluding that for a perceptually accurate encoding of sound frequency via temporal rate of stimulation, as in MED-EL’s fine-structure coding strategies, fine-structure rate stimulation has to be presented on cochlear locations not shallower than 430°. The ratio of the change in acoustic frequency (in dB) and the corresponding change in rate required for pitch-match (also in dB) has to be 1.0 as shown in Figure 42 (horizontal black dotted line). For cochlear locations deeper than 430°, the ratio is not significantly different from 1 (red symbols in Figure 42), whereas it is significantly different from 1 at shallower cochlear locations (yellow symbols in Figure 42). In addition, Landsberger et al. [32] showed that low-rate stimuli are only perceived as clean, not noisy, and not annoying when presented on electrode channels in the second cochlear turn. When presented more basally, most of the tested MED-EL CI recipients perceived them as not clean, noisy, and annoying.

Figure 42. Data plot from table 3 from Schatzer et al. (2014) [33]. The ratio of the change in perceived acoustic frequency (in dB) to the change in stimulation rate (in dB) is plotted as a function of the mean insertion angle for each of the cochlear regions. For a rate of stimulation of properly encoded pitch, the relationship between the rate of stimulation and the frequency corresponding to the perceived pitch must equal one, as shown by the black dotted line. The data from a similar study by Blamey et al. [34]. (1996) in Cochlear™ CI22 users are consistent with the data in Schatzer et al. (crossed symbol). Reproduced by permission of Wolters Kluwer Health, Inc.

In 2016, Prof. Baumann and his colleagues from the Johann Goethe University Frankfurt in Germany demonstrated that in SSD patients (n = 7) implanted with FLEXSOFT™ (array length = 31.5 mm) and FLEX28™ (array length = 28mm), the FSP strategy enabled CI users to have matched low pitch perception in the implanted ear (ipsilateral), compared to the normal acoustic hearing ear (contralateral) [35]. FLEXSOFT™ and FLEX28™ electrode arrays reach an angular insertion depth of close to 700° and 600°, respectively, which is closer in place to LFs <300Hz, as shown in Figure 43.

Figure 43. Angular insertion depth of FLEXSOFT™ and FLEX28™ electrode arrays in an average-sized cochlear model (image courtesy of MED-EL).

When apical channels of these electrodes are electrically stimulated at a defined rate (pulses/second) representing the corresponding acoustic frequency, then these SSD patients can subjectively match with their normal hearing on the contralateral ear and say whether the electric hearing matches with their natural acoustic hearing, as given in Figure 44.

Figure 44. Individual frequency-place functions for electric stimulation obtained with place independent electric stimulation rate. When the apical channels are supplied with fixed-rate stimulation providing only place cues, then their LF pitch perception was highly variable across participants and generally not matching with the Greenwood’s frequency function in the LFs (A). When the apical channels are supplied with place-dependent stimulation rates, then the LF pitch perception was closely matching with the Greenwood’s frequency function in the LFs (B) [35]. Reproduced by permission of Elsevier B.V.

Per Greenwood’s frequency function, at an intracochlear insertion depth of 630°, the neural fibres are responsible for processing frequencies closer to 150 Hz. To match the 150 Hz perception with CI, the electrode array shall be physically placed at that insertion depth, and the electric stimulation should be provided at a rate of 150pps. If the apical channels were placed at an insertion depth of 630° and provided with fixed stimulation rates (1,500pps), it sounded more like above 300 Hz (Figure 44(A)). Whereas, if the apical channels were provided with place-dependent stimulation rates, the SSD users felt like it sounded more natural as they were able to match the CI pitch percept with their normal-hearing ear (Figure 44(B)). Similar pitch matching results in SSD patients with MED-EL CI were reported earlier by Vermeire et al. in 2010 [15] and Schatzer et al. in 2014 [33].

Like with any other features in the MED-EL CI system that are inspired by nature, it is the wish of MED-EL to follow this principle by allocating centre frequencies to the patient-specific electrode contact positions based on Greenwood’s frequency-place map. In order to achieve a reliable LF pitch perception with a CI, it requires a good match between electrode place and stimulation rate. The fact that the cochlear size varies a lot among the human population [30], selecting a proper electrode array length which matches the cochlear size, plays a key role in achieving a good match between electrode place and stimulation rate. Measuring the basal turn diameter of the cochlea, commonly called as A-value, could be used in the estimation of cochlear duct length (CDL) by applying dedicated mathematical equations [36–38]. Based on the predicted CDL, applying Greenwood’s frequency map would provide the patient-specific frequency map. By combining the CDL, frequency map and the audiogram of the patient, choosing an appropriate electrode array length would be the concept proposed towards patient-specific CI electrode array selection (Figure 45). This was a concept MED-EL proposed in the year 2011 and developed the research-based CDL software in 2014 (link to download), which was even clinically used, as reported by Dr Stefanescu et al. in 2018 [39].

Figure 45. Illustration of a single measurement of the cochlea (A-value) from the preoperative images, applying dedicated mathematical equations, the patient-specific CDL is estimated. With the estimated CDL, applying Greenwood’s frequency function would provide the patient-specific frequency map. Image courtesy of MED-EL. Preoperative audiogram of the patient shows if there is any functional LF residual hearing. By applying these parameters, an optimal electrode array length may be chosen. The patient-specific electrode array length selection concept was proposed by Dr Dhanasingh and Dr Jolly (US patent number: 9037253) from MED-EL. Dr Assadi is acknowledged for translating CDL research software to OTOPLAN^®.

With technological advancements over time, the CDL research software was further finetuned, and today it is available for clinical use under the name OTOPLAN^® which includes other features, including three-dimensional segmentation of the key temporal bone anatomical structures, identify the individual electrode channel insertion depths from the post-operative image. OTOPLAN^® is a tablet-based otological planning software tool that was developed in collaboration with CAScination AG, a Swiss company, and Dr Assadi from MED-EL was taking care of the project logistically.

One of the key features in the OTOPLAN^® software is that by loading the postoperative image of the electrode inside the cochlea, the software offers the possibility to identify the individual electrode array channels and its corresponding angular insertion depths, as shown in Figure 46. By combining this information with the patient-specific Greenwood’s frequency map, it is possible to assign the channel frequency bands to the individual electrode contacts, based on the tonotopic frequency of that electrode contact.

Figure 46. A screenshot from OTOPLAN® software that shows the identification of individual electrode channels, its corresponding angular insertion depth along with the centre frequency based on Greenwood’s function. Image courtesy of MED-EL.

In 2020, Dr Nopp, Dr Kals, and Dr Penninger from MED-EL took the overall concept and added additional algorithms to the information received from the OTOPLAN^® tool to bring it inside the MED-EL’s CI clinical system software MAESTRO 9.0 (Figure 47).

Figure 47. Engineers from the Signal Processing research team who combined the patient-specific electrode array length selection tool, OTOPLAN^®, and added additional algorithms to come up with the concept of anatomy-based fitting (ABF).

From the MAESTRO system software, it is then possible to assign the patient/anatomy-specific centre frequencies to each of the electrode array channels. The algorithms developed within the anatomy-based fitting concept mainly offer electrode place-rate match to the mid-frequencies (800–3,000 Hz) where the speech information is mainly coded. This is done on an individual basis taking the cochlear size variation and the electrode insertion depth seen from the post-operative imaging into consideration.

In 2019–20, clinicians from the University of North Carolina at Chapel Hill in the US applied OTOPLAN^® to their clinical practice to identify the angular insertion depths (AID) associated with MED-EL’s various FLEX electrode array variants [40] (Figure 48).

Figure 48. Clinicians from the University of North Carolina at Chapel Hill who applied OTOPLAN^® in the clinical practice in investigating the AID of MED-EL’s FLEX electrode array variants and analysed the frequency-to-electrode place mismatch for the CI-alone and EAS™ users.

The study aimed at investigating a patient population implanted with MED-EL electrode arrays of various lengths to establish if the variations in angular insertion depths in different cochlear sizes result in frequency-to-place mismatch [40]. Forty-seven patients were implanted with FLEXSOFT™/STANDARD array, forty-eight with FLEX28™, and eleven with FLEX24™. From the postoperative CT scans (n = 106), OTOPLAN^® estimated that the CDL ranged between 29.4 mm and 39.5 mm. The CDL was found to be negatively correlated with the electrode angular insertion depths (Figure 49(A)). Every cochlea is unique in its size and shape and has its own frequency map and if the chosen electrode array is short (e.g. FLEX24™), then there will be a higher mismatch between the frequency allocation and the electrode place (Figure 49(B)) for maps with frequency allocations covering at least the speech frequency range.

Figure 49. Correlation between CDL and angular insertion depth of the apical electrode contacts for complete insertions of FLEX24™, FLEX28™ and FLEXSOFT™/STANDARD electrode arrays (A). Relationship between absolute frequency-to-electrode place mismatch at 1,500Hz and electrode array type for CI-alone users with complete insertion (B) [40]. Statistical analysis: Pearson correlation used in evaluating the relationship between AID and CDL and multiple linear regression was used in assessing the relationship between the degree of frequency mismatch and angular separation between electrode contacts. Reproduced by permission of Wolters Kluwer Health, Inc.

It makes more sense to choose an electrode array length which matches the cochlear size, and therefore the optimal AID, and minimised frequency-to-electrode place mismatch can be achieved. These two preliminary studies are encouraging results for the clinical application of OTOPLAN^®, which can be confidently used in the ABF. ABF is an emerging concept and a key component of individualized CI fitting that aims to make the fitting process simple and efficient, saving time for audiologists.

5.4.2. Bimodal delay compensation

Travelling wave latency is a function of the inner ear which the MED-EL CI system models in the filtering process of the sound coding strategy (stage 2, as shown in Figure 5). With natural acoustic stimulation, there is a certain time needed for the acoustic wave to travel from the external ear canal to reach the auditory cortex. All the steps in between result in certain latency/time delay, which can be measured from the wave V of electrically evoked auditory brainstem response (eABR). In contrast, with electric stimulation, all delays in the transmission of a sound wave in the external ear canal are missing. A further complication is that interaural stimulation timing in bilateral CI, SSD, or bimodal stimulation varies with the frequency/pitch of the sound signal. If this interaural stimulation timing is not adjusted during the sound processing stage, then this could create an imbalance or mismatch in the interaural stimulation timing in the SSD patients with CI on their deaf ear in bimodal setting, thus resulting in a much-degraded hearing on the deaf ear and compromising spatial hearing [41].

MED-EL’s CI system implements group delays through its sound coding strategies across all frequencies resulting in stimulation pulses with some delays, thus mimicking natural hearing. Latencies measured through eABR for various frequencies in MED-EL CI users (Figure 50, red curve) fitted with OPUS audio processor, represent closer match to the latencies measured in a normal acoustic ear (Figure 50, green curve), as reported by Zirn et al. from the Freiburg Medical University in Germany in the year 2016 [42].

Figure 50. ABR waves V latencies across four different frequencies in normal hearing participants wearing a hearing aid (black curve), in normal hearing ears of SSD CI recipients without a hearing aid (green curve: nature), and the implanted ears [42] of those CI recipients (red curve: MED-EL implantees). Reproduced by permission of Elsevier B.V.

In general, the latencies increase with decreasing frequencies which is reflected in the MED-EL CI system, although it is lower by 1 ms above 1 kHz in the MED-EL CI system compared to normal hearing latencies. The black curve in Figure 50 represents the latencies caused by the hearing aid, which take longer to process the sound signal before it is amplified and released to the ear canal.

In 2019, group of clinicians from Medical University of Innsbruck in Austria, led by Prof. Stephan investigated the effects of adding additional delays (0.5-, 1.0-, 2.0- and 4.0-ms) to frequencies above 1 kHz [43]. The test was conducted in twelve SSD adults fitted with MED-EL’s OPUS 2 processor. The effects of adding additional delays were evaluated in terms of sound source localisation score and SRT. The overall performance in sound source localisation and SRT measured as SNR (SNR for 50% speech intelligibility was used) is given in Figure 51(A,B), respectively. The study participants achieved their best performance, i.e. the maximum percentage of correct answers and smallest angular errors, at a tested signal delay of 1 ms. For the larger signal delays of 2- and 4-ms, performance in sound localisation progressively decreased. In terms of speech performance, the SRTs observed in this group of SSD CI users were between −4 and −5dB SNR for all tested signal delays, which is close to the performance of a normal-hearing person with SRT usually between −7dB and −8dB SNR.

Figure 51. Scores for correct sound localisation in per cent (A) and SRT expressed in SNR for signal delays of 0 ms (standard use), 0.5 ms, 1 ms, 2 ms and 4 ms (B) [43]. Statistical analysis: Repeated measures ANOVA (p < .05). Reproduced by permission of Elsevier B.V.

The results show that the signal delay in the pre-processing of a CI audio processor affects the binaural hearing performance of CI users with SSD to a certain degree. In particular, in sound localisation, an improvement was seen at 1 ms signal delay. The effects of signal delay on speech intelligibility in noise was that performance deteriorated with larger signal delays with no improvement at any particular signal delays.

RONDO 3, which is available since 2020, is compatible with any HA if a HA is used on the contralateral ear. HAs result in relatively longer time delays across the frequencies compared to a CI, as shown in Figure 50 (black curve), since the sound processing delay of the HA is in the order of 3–10ms, added on top of the travelling wave and neural delays in the acoustically stimulated ear. If the CI is used on one side and HA is used on the other side, then there will be a mismatch in the interaural time difference (ITD), experienced by a bimodal listener. To avoid this mismatch, the bimodal delay compensation feature in the RONDO 3 and SONNET 2 processors can adjust for the higher time delays in the HA.

5.4.3. Triphasic pulse stimulation

Facial nerve stimulation (FNS) is characterised by facial muscle movement or facial tickling sensation, which can be a side effect of intracochlear electric stimulation with the CI in some cases, regardless of a CI brand. In severe cases, it may lead to patient intolerance to the extent of preferring not to use the CI. The undesired FNS after CI surgery is reported with cochlear conditions like osteoporosis, otosclerosis, bony dehiscence between the facial nerve (FN) and basal turn of the cochlea, and inner ear malformations. The reason for the FNS in such cochlear conditions is mainly due to unusual current leakage from the cochlea, and as a result, the threshold and maximum comfort levels (MCL) of the auditory nerve need to be increased to get the desired hearing/loudness sensation. Due to the increased stimulation levels, the FN which runs close to the basal turn of the cochlea can also get stimulated. It is known from the literature that FNs have a higher sensitivity to electric pulse shapes than the hearing nerve. To control the FN costimulation and to reduce the stimulation levels, certain CI fitting procedures are established, such as raising the pulse width or increasing the interphase gaps. If reprogramming is not enough, then deactivation of single electrode contacts may be necessary, but this comes at the price of compromising speech comprehension [44]. Although the pre-curved modiolar hugging electrodes are generally believed to reduce the FNS due to its closer proximity to the central modiolar trunk and sufficient distance from the FN, clinical data has shown that electrode designs and modiolar proximity do not influence the prevalence of FNS [45].

The generally applied biphasic pulses (Figure 52(A)) in CI are charge-balanced and consist of two opposing polarities of similar phase durations (T) in which the negative charge cancels the positive charge and keeps the neuronal elements in a state of equilibrium. In contrast to biphasic pulses, triphasic pulses (Figure 52(B)) consist of two negative phases of the same duration (T/2) and one positive phase of double of that duration (2 × T/2), all with the same amplitude, thus resulting in an overall charge-balanced pulse. Because of two-phase reversals in triphasic pulses (2 × T/2), or in other words, splitting the negative phase (cathodic phase), it becomes less effective for extracochlear activation of the facial nerve. These two properties of the triphasic pulse stimulation are favourable for keeping the inadvertent FNS under control or at a minimum level which patients do not detect. Nevertheless, with a reduction in stimulation effect, the loudness sensation and neural responses evoked by triphasic pulses would be lower than the loudness sensation evoked by biphasic pulses with the same current level. To keep the loudness sensation to the desired level in patients fitted with triphasic pulses, the MCL may be raised without eliciting FNS, as shown in Figure 52(C).

Figure 52. Balanced biphasic pulse stimulation (A) and triphasic pulse stimulation (B) showing two negative phases of duration (T/2) and one positive phase duration (T) (image courtesy of MED-EL). Model of expected benefit with triphasic pulse stimulation on FNS (C) (image recreated from Bahmer et al. [44]).

In 2017, MED-EL received CE marking for the triphasic pulse stimulation fitting option which was made available for clinical use through its fitting software MAESTRO 7.0.

In 2017, Prof. Löwenheim and his team along with engineers from MED-EL in Austria evaluated the effectiveness of triphasic pulse stimulation in suppressing undesired FNS in a group of patients (n = 15) who underwent CI surgery between 2014 and 2017 [46] (Figure 53).

Figure 53. Clinicians from Tübingen University Hospital, Germany, and engineers from MED-EL applied triphasic pulse stimulation in CI patients to eliminate/minimise inadvertent FNS. Mr Werner Sürth and Dr Reinhold Schatzer came up with the triphasic pulse stimulation concept (US patent number: 9265944).

The aetiology of HL in these patients were hypoplastic cochlear nerve, temporal bone fracture, EVA, otosclerosis, sudden HL and unknown reasons. Before evaluating the reduction of FNS and hearing ability, the patients experienced a triphasic map for a mean period of at least twenty-five months. Out of fifteen patients, ten had a complete suppression of undesired FNS, and three had partial suppression with triphasic settings. Two patients, however, did not show any suppression of FNS with triphasic settings and had EVA as well as unknown reasons as the HL aetiology. The MCL was evaluated for biphasic versus triphasic pulse setting in these fifteen patients. The median MCL level in the triphasic setting (51.12qu) was found substantially higher than in biphasic setting (36.37qu), and the mean MCL levels could be raised by 150% in triphasic, compared with the biphasic setting, without triggering the FNS (Figure 54(A)). The hearing results, as measured with Freiburger monosyllable test at 65 dB and 80 dB SPL, showed patients achieving median speech scores of 28% and 45%, respectively, with the biphasic setting. The results increased to the median speech scores of 40% at 65 dB SPL and 63% at 80 dB SPL with the triphasic setting (Figure 54(B)). The difference in speech scores between these two fittings showed a significant improvement for the triphasic setting, especially at 65 dB and not at 80 dB SPL, which patients appreciated in everyday usage.

Figure 54. Average MCL values in the biphasic and triphasic pulse stimulation modes (A). Freiburger monosyllables at 65- and 80-dB SPL (B) [46]. Statistical test: bivariate analyses were performed using independent sample t-tests and Mann-Whitney U tests. Reproduced by permission of Wolters Kluwer Health, Inc.

In 2020, the triphasic pulse stimulation mode reached the Middle East and was successfully applied by Prof. Alzharani and his colleagues in eleven CI recipients (sixteen ears) who had unintended FNS with the activation of the audio processor [47,48] (Figure 55).

Figure 55. Clinicians from ¹ King Abdullah Ear Specialist Centre, Saudi Arabia, ² Menoufia University Hospital, Egypt, and engineer from MED-EL applied triphasic pulse stimulation in CI patients to eliminate/minimise inadvertent FNS.

The average number of electrode channels responsible for FNS were six for biphasic stimulation – with the average audio processor usage of 23.6 months – with no success of controlling the FNS. The fitting program was changed to triphasic stimulation which showed successful resolution of the FNS in thirteen ears, and the remaining three ears were resolved by deactivating one channel. The patients used their audio processor with the triphasic stimulation fitting map for an average of 17.5 months. The speech discrimination score level at 65 dB HL was better with triphasic stimulation (average of 75.25% ± 26.13) compared to biphasic stimulation (average 58.25% ± 26.13) and the improvement in the speech discrimination score was seen with triphasic stimulation (Figure 56).

Figure 56. Comparison of the measured speech discrimination scores at 65 dB for the biphasic and triphasic pulse pattern groups. Statistical test: Parametric paired t-test to test the significance between the group data. Histogram created from the raw data provided by Alhabib et al. [47].

With positive results of triphasic stimulation pulses in resolving FNS, as well as in improved speech discrimination scores and MCL, as shown by these two scientific pieces of evidence, the triphasic mode may be recommended to all patients with FNS following CI surgery with MED-EL CI devices.

5.5. Conclusion

Signal processing is a highly technical topic that is often perceived as complex for people untrained in the field to grasp it in depth. In this article, MED-EL’s signal processing was approached in an easy language and compared it with the functionalities of the normal acoustic ear. The overall aim of signal processing in a CI system is to capture essential information hidden in any meaningful sound signal and provide it to the inner ear in the form of electric pulses. Dual microphone, AGC compression function, compensation of artificial time delays and phase-locking the rate of LF stimulation pulses with the sound frequency are some of the features that support MED-EL’s signal processing that aims in modelling the normal acoustic hearing. The audio processor design at MED-EL, starting from the body-worn CIS PRO type in the early 90 s until the latest RONDO 3 version of the single-unit processor in 2020, is an achievement by itself as every version of the audio processor included improved features adding more benefits and comfort to the users.

The signal processing algorithms implemented in MED-EL audio processors, starting from the CIS PRO and TEMPO + to the RONDO 3, were evaluated in close collaboration with clinicians around the world to demonstrate the safety and efficacy of the audio processors. It involved numerous hours of efforts from the signal processing research team and as well from clinicians. As a result, all the strategies mentioned above were successfully implemented in the audio processors and are being used by MED-EL CI patients successfully. Signal processing and the audio processor is yet another topic within MED-EL that followed the translational science path in successfully bringing the concept from the laboratory setting to patients.

Acknowledgments

The authors would gratefully like to acknowledge the key contributors to the development of the subject matter. Their contributions are outlined in this article. The authors further acknowledge Peter Nopp and Reinhold Schatzer from MED-EL for their valuable input and comments during several rounds of review meetings that contributed to the final version of this article.

Disclosure statement

This article is sponsored by MED-EL and has not undergone the regular peer-review process of Acta Oto-Laryngologica. Both the authors are affiliated with MED-EL.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.