During middle ear surgery the cochlea is vulnerable to acoustic trauma and pressure transients (Tos et al 1984, Mair and Laukli 1986, Bauchet St. Martin et al 2008, Martin and Stecker 2008). Despite this vulnerability, middle ear intra-operative monitoring (MEIOM) using auditory evoked potentials has not been employed routinely to determine which surgical manipulations are problematic (as discussed later; Wazen 1994, Legatt 1995, Martin and Stecker 2008, Møller 2011a, Ren et al 2016). Rather, surgeons have had to rely on longitudinal outcomes studies to assess any cochlear trauma, including pre- and post-operative air- and bone-conduction audiometry, but this provides no specific information on which surgical manipulation caused a problem, and when it occurred. High-frequency information is particularly hard to obtain because of the poor high-frequency efficiency of commercially available bone conductors (BCs). Unfortunately the same BCs are also needed for MEIOM: air-conducted sound is not feasible during middle ear surgery because there is no room in the surgical field, and the ear canal’s acoustical and electrical properties can change with surgery, blood and fluid. As a result, bone-conducted stimulation is required, in combination with contra-lateral acoustic masking to ensure that the evoked responses are from the operated ear only. Apart from their poor high-frequency response, BCs have yet another problem when used in MEIOM: available transducers produce large electrical/magnetic stimulus artefacts that contaminate the small auditory evoked potentials. The artefacts are acceptable if BC thresholds are good and the electrical drive to the BC is low, but if BC thresholds are poor and the required BC electrical drive is high, the large stimulus artefact contaminates the evoked responses, reducing the dynamic range of MEIOM testing.

Unfortunately MEIOM has other problems too. The technique most commonly used to monitor auditory sensitivity in IOM is waveform averaging of the auditory brainstem response (ABR), resulting in waveforms with standard Jewett peaks (Jewett and Williston 1971; Kartush et al 1991; Neu et al 1999; Zamel et al 2009), but this has its own limitations. As discussed later, waveform averaging is too slow to allow the surgeon to identify and avoid any problematic manipulations in a timely way. In addition, if the stimulus level is increased in an attempt to speed testing (by improving the ABR’s signal-to-noise ratio; SNR), the ABR loses its sensitivity to cochlear damage: although the neural responses in normal and damaged cochleas are clearly different at low sound levels, they are very similar at high sound levels, due to the nonlinearity of neural recruitment. As a result, the ABR at high sound levels is relatively insensitive to moderate cochlear damage, especially if the damage is only in the high-frequency basal cochlear turn (where BCs have poor efficiency). In the end, while IOM has been performed in some otological procedures for over 40 years (see Discussion), it is almost never used during middle ear surgery (Møller and Jannetta 1981, Grundy et al 1982, Møller et al 1982, Raudzens 1982, Levine et al 1984, Nuwer 1986, 1994, Martin and Stecker 2008, Møller 2011b).

Given the vulnerability of the cochlea, the limitations of ABR waveform averaging, and the technological limitations of available equipment, we have developed novel equipment and procedures to enable rapid and sensitive monitoring of cochlear sensitivity throughout middle ear surgery. We had hypothesised that it should be possible to: (i) increase the frequency response of BCs to allow stimulation of the vulnerable basal cochlear turn; (ii) reduce the electrical artefact produced by BCs, to increase the dynamic range of MEIOM; (iii) develop a compact and cost-effective bio-amplifier suitable for the cluttered and noisy OR environment; (iv) develop an efficient method to place and remove the electrodes, BC and masker to streamline MEIOM; (v) develop a process to continuously and rapidly monitor cochlear sensitivity at low sound levels in a fully anaesthetised patient; and ultimately (vi) provide surgeons with an indication of cochlear changes that is fast enough to be useful in avoiding damage, but does not significantly increase their workload or the cost of surgery.

In this preliminary report we describe the equipment and methods we have developed to present bone-conducted chirps while monitoring the auditory steady state response (ASSR; Rance 2008, Rampp et al 2017), as an indirect measure of cochlear sensitivity. Under general anaesthesia and at high stimulus rates the ASSR is dominated by generators in the auditory brainstem (McNeer et al 2009; Haghighi et al 2018) that are not themselves vulnerable during middle ear surgery, so that we assume that any ASSR changes are due to cochlear changes alone (see later discussion). To automate the ASSR testing we have developed software to continuously adjust the level of the chirp stimuli to evoke an ASSR with a pre-set amplitude (‘the ASSR target’). The chirp level required to evoke this ASSR target we define as the subject’s ‘ASSR threshold’, and use it as a running estimate of cochlear sensitivity. Once the hardware and software are set up at the start of surgery (taking a few minutes), the system automatically tracks ASSR threshold throughout the full procedure without intervention, providing the surgeon with a simple real-time display of cochlear sensitivity.

2.1. Overview of system development

The development of our system occurred iteratively over five years, in three stages. First, the development and testing of early equipment and software was performed in a laboratory environment using the researchers as subjects. Once prototypes of the software and hardware were available, they were trialled in the more demanding hospital environment on anaesthetised middle ear patients. This later clinical work occurred in two stages. The first group of patients allowed us to identify the sources of electrical interference in the operating environment, and taught us how to cope with other issues (including the preparation of patients for MEIOM, the layout of the equipment around the operating table and staff, and the avoidance of any disruptions to the workflow of the surgeon, anaesthetist and other staff). In the first group of patients we collected standard ABR averaged waveforms as a measure of the quality of our electrodes and electrode contact, and of the bio-amplifier, BC and procedures (such as the securing of the BC, electrodes and contra-lateral masker). From recordings in this first group it became clear that ABR waveform averaging was too slow and cumbersome to be useful to the surgeon in identifying problems rapidly, certainly without a senior audiologist or neurophysiologist in attendance. Moreover, any post-surgery analysis of the large number of collected ABR waveforms took as much time as the surgery itself. As a result, in the second group of patients we switched from ABR waveform averaging to automatic threshold tracking using the ASSR, which provided a simpler and faster output for the surgeon (a simple chart of cochlear sensitivity over the full duration of a procedure). In this final phase of development we also improved the system’s sensitivity, speed and usability, to the point where the main limitation seems to be inadvertent acoustic masking within the OR (see later).

2.2. Patient recruitment and inclusion

Software and hardware development was largely done in Perth, Australia. All clinical work between 2019 and 2024 was performed at the Christchurch Hospital or St. George’s Hospital in Christchurch, New Zealand (ethics clearance HDEC13/STH/49). All procedures were consistent with the declaration of Helsinki, and all patients provided informed written consent to take part in the study. Only middle ear patients were recruited, with full pre-op air and bone audiometry indicating that monaural recordings from the operated side were possible. That is, cochlear sensitivity on the operated side and air-conduction on the non-operated side had to allow effective contra-lateral noise masking, without over-masking the response from the operated side. We did not discriminate between male and female patients, but accepted cases as they were scheduled for surgery, and no data have been excluded from our considerations if responses were measurable. Because each patient’s pre-operative audiometry and their surgical procedures were unique, data have not been averaged across patients: during the development stage of our system more insight is gained by presenting each patient’s results in detail.

In the first phase of our work, four normally hearing subjects were used to test the equipment and software: two of the researchers (ages 65yo and 37yo) and two younger subjects (15yo and 17yo). In the second phase of the study where ABR waveforms were collected in the OR, 5 patients were tested to understand and eradicate the interference and stimulus artefact with our early prototype gear. Another 25 patients were tested successfully to obtain ABR waveforms, now without interference or stimulus artefact. ABR waveforms evoked by BC stimulation were available in all 25, but air-conduction results were available in only 1 (in middle ear surgery, air-conducted stimulation is normally unavailable). Only representative ABR waveforms from 4 patients are presented in this report (figure 3), including one anomalous case where a significant hearing loss was produced (figure 3(d)). Our goal was not to study the ABR per se, but to use it as a test of improvements in the BC, bio-amplifier, electrodes and procedures, and to understand fully the limitations of ABR in MEIOM. In the third phase of the study in which ASSR threshold tracking was investigated, 17 patients were tested, with ASSR tracking results available in 14 patients. In only 1 of these 14 was there a problem with poor BC contact, and that was before the introduction of the latex retention band. Tracking results were not available in 2 early subjects because their hair was impenetrable, in 2 patients ABR/ASSR were available but the BC had moved (before latex band use), and 1 patient had no BC audiometric thresholds better than 70 dB nHL, so that they acted as a control to demonstrate the absence of stimulus artefact (figure 5(b)). In the 47 patients tested in the OR, the middle ear procedures were a mix of cholesteatoma (ABR = 5, ASSR = 3), revision cholesteatoma (ABR = 4, ASSR = 1), blind sac closure (ABR = 1), ossiculoplasty (ABR = 2, ASSR = 1), stapedotomy (ABR = 8, ASSR = 3), tympanoplasty (ABR = 1, ASSR = 6), myringotomy (ASSR = 3), exostoses removal (ABR = 2, ASSR = 2). Although an exostoses procedure is not strictly middle ear surgery, the cases were valuable as tests of equipment performance, and because these patients’ often had excellent air- and bone-conduction thresholds, they served as useful tests of contra-lateral masking and over-masking).

2.3. Insights from early failures

Testing in the laboratory environment was crucial, but it could not determine whether our electrodes, bio-amplifier and software had sufficient SNR in the operating room to allow the reliable and rapid detection of brainstem responses in anaesthetised patients without interference or electrical artefacts. In our first patient we identified the main sources of electrical interference in the recordings: the cautery, the drill, the mains chargers for the operating table’s battery power supply and the EEG wakefulness monitor (BIS). While the cautery and drill interference could not be avoided, mains interference from the operating table and BIS monitor were abolished by running the equipment on battery power during MEIOM. In our second patient the vertex electrode and the BC were secured with taping (Coban 3 M self adherent wrap). This was inconvenient, time-consuming, disruptive for the anaesthetist and surgeon, and was too close to the surgical field. Although excellent ABR responses were obtained, it also became clear that the large increase in slow EEG components (delta, alpha and spindle waves) with anaesthesia could overload the bio-amplifier, given its original filtering. We subsequently increased the bio-amplifier’s lower corner frequency to 60 Hz, which reduced the intrusion of EEG and ECG, leaving the brainstem responses intact.

In two later patients the intrusive taping was replaced by an over-the-head elastic strap that held the BC and electrodes in place. It was secured at the back by a strap around the chest, and at the front with a mouth hook, itself designed to minimise interference with the laryngeal mask airway (LMA). While this worked, it became clear that the size of the mouth hook had to be different for each patient, and was inconvenient for the anaesthetist. In subsequent patients the over-the-head elastic strap was replaced with a simple non-adhesive latex ‘exercise band’, taped to the back of the neck and to the throat at the front. The band was simply split at the front to avoid the nose and LMA (see figure 1).

Figure 1. (a) Our IOM system included a battery-powered bio-amplifier optically coupled to a nearby laptop via a plastic light guide, a high-frequency bone conductor on the forehead, an earbud speaker for contra-lateral masking, a stainless steel active electrode on the hairline, and a stainless steel indifferent electrode on the nape of the neck. (b), (c) The electrodes and bone conductor were secured by a non-adhesive latex band, itself secured with adhesive tape to the nape of the neck and to the throat. The split in the band at the front avoided the nose and laryngeal mask airway (LMA). The masking earbud was inserted into the non-operated ear and taped in position. The bone conductor was placed centrally on the forehead under the band just above the brow line. The bio-amplifier (not visible) was placed behind the head on the non-operated side.

Download figure:

Standard image High-resolution image

Two early failures occurred because the patients had such thick hair that it was impossible to obtain a stable, low-impedance vertex contact, even when the vertex electrode was secured with the over-the-head elastic strap. From then on we switched from a vertex electrode to a forehead electrode to avoid hair and dry scalp, and changed from adhesive Ag/AgCl electrodes to larger stainless steel electrodes that improved the electrode SNR. We finally settled on a large indifferent electrode at the nape of the neck to avoid the surgical field, and a narrow active electrode along the hair line to avoid the anaesthetist’s BIS electrodes.

After all of these changes our final system was as shown in figure 1(a). It consisted of a small battery-powered bio-amplifier (optically coupled to a nearby laptop via a 1 mm diameter plastic light guide), a small high-frequency BC centrally on the forehead just above the brow, an earbud speaker for contra-lateral masking (Sony MDR-EX15 LP), a narrow stainless-steel active electrode (100 mm × 10 mm) close to the hairline, and a stainless steel indifferent electrode on the nape of the neck (50 mm × 30 mm). Both electrodes were curved to sit comfortably on the skin along their entire length. Custom software written in National Instruments LabVIEW and running on a battery-powered laptop alongside the operating table provided the electrical drive to the BC and masker via two external USB soundcards. The laptop software also collected the ABR and ASSR, and had an algorithm to keep the ASSR at a pre-set amplitude (‘the ASSR target’), so that the required chirp level represented what we call the ‘ASSR threshold’. The ASSR target was typically set to twice the level of the biological noise floor at 80 Hz (the chirp repetition rate), and the ASSR threshold was typically within 10 dB of each patient’s psychoacoustic threshold obtained in the ante room.

2.4. Preparation in the ante-room and note-taking during surgery

One problem in MEIOM can be the delay in moving patients from the ante room to the operating table (due to skin preparation, electrode attachment, and any re-check of BC thresholds). At the operating table, workspace clutter and disruption of the anaesthetist’s workflow can also cause problems. To avoid these issues, in the ante room the skin at the nape of the neck and hairline was cleaned for the electrodes with alcohol and light abrasion. The skin of the throat was also wiped lightly with alcohol to improve tape adhesion. A single non-adhesive latex exercise band (figures 1(b) and (c)), was then taped to the back of the neck and left dangling (to be attached at the throat later when the patient was on the operating table). The psycho-acoustic threshold of the patient was then estimated in the ante room using the same equipment to be used in surgery, and only then was the patient moved to the operating table. After anaesthetic induction the latex band was passed over the head, where it split either side of the nose to avoid the LMA, before being pulled taut and taped to the throat (figure 1(b)). The stainless steel scalp and neck electrodes were then smeared with the electrode gel (Spectra 360, Parker, New Jersey U.S.A) and slipped with back padding under the latex band. The bio-amplifier was then placed under the band at the back of the neck and connected to the electrodes. The BC was then placed under the latex band, centrally on the forehead and just above the brow line, so that it firmly contacted the patient’s skin.

Notes detailing the timing and nature of surgical events were taken by the experimenters in consultation with the surgeons throughout all procedures. The bio-amplifier’s raw output and the bone conductor’s drive signal were also recorded on the laptop as a stereo wav file as a backup and for later post-operative analysis if required. Annotation of the traces below is based on the detailed written notes. Although it would be possible to use either audio- or video-logging to capture these events, this would require revisiting the recordings for the full duration of the procedures to itemise the events. This analysis was avoided by taking notes in real time.

2.5. The custom-made wideband bone conductor

As mentioned earlier, MEIOM requires bone-conduction stimulation, but available BCs (e.g. B71) are limited in their frequency response (Mauldin and Jerger 1979, Cornacchia et al 1983) and dynamic range, and produce large electrostatic and electromagnetic artefacts (Mauldin and Jerger 1979, Foxe and Stapells 1993, Campbell et al 2004, Small and Stapells 2004, Hatton et al 2012). As a result, available BCs cannot stimulate the basal cochlea (most at risk in middle ear surgery) without producing excessive artefacts in ABR and ASSR recordings (Mauldin and Jerger 1979, Hatton et al 2012). To overcome these problems, we have developed our own compact high-output high-frequency BC that produces minimal artefact (see Results). After much experimentation, we constructed our BC with an inexpensive 1” piezo disc, driven by a reverse-connected audio transformer (Tamura, MET-23), glued immediately behind the piezo disc and its black plastic shell, with all but the front face of the piezo potted in epoxy. The left and right channels of a relatively powerful external USB sound card (Buddy 6 G) drove the secondary of the transformer (acting as its primary) in balanced push-pull mode. This doubled the electrical drive (+6 dB) and helped cancel electrical artefacts. The piezo was connected directly across the transformer’s primary winding (acting as a secondary) with extremely short wires to reduce stimulus artefact. The small transformer had a 1.6 kOhm primary with a centre tap (unused) and a 3.2 Ohm secondary winding, which provided a ×22.4 voltage step-up ratio when connected in reverse. That is, a 3Vpp push-pull drive from the sound card gave 67Vpp across the piezo (+27 dB compared with the drive directly from the sound card). To make an efficient acoustic contact between the vibrating piezo disc and the patient’s skull, a ‘sound post’ was created by pressing a small amount of two-component epoxy putty into the hole in the piezo’s plastic shell. The potted transformer and piezo were then electrostatically shielded with a continuous shroud of copper foil. The 1.5 m low-voltage drive wires from the soundcard were a light-weight twisted pair to the transformer secondary (functionally its primary), shielded with an outer light-weight braided shield. This shield was itself connected at the BC end to the copper foil around the piezo and transformer, and to the laptop ‘ground’ point at other end (because the laptop was battery-powered there was no true ground connection). Finally, another thin insulating layer of plastic was positioned between the bone conductor’s copper foil shield and the patient’s skin, adding further electrical isolation. The combination of the low-level balanced voltages on the long drive wires, the extremely compact construction of the higher-voltage transformer/piezo combination, the balanced push-pull drive on both the primary and secondary sides of the transformer, and the electrostatic shielding of the foil and braided shield all combined to reduce the electrical artefact from the BC so that it was undetectable in the ABR and ASSR recordings (see Results). This artefact reduction was also helped by the lack of frequency overlap of the stimulus and response (the stimuli had a frequency content above 2 kHz, while the ABR and ASSR had a frequency content below 2 kHz).

To test our BC’s frequency response we collected high-resolution Bekesy audiograms from four normally hearing and unanaesthetised subjects using custom-written software in LabVIEW. In Bekesy audiometry a subject presses a button while they can hear a sound stimulus, causing the software to progressively decrement the stimulus intensity, and releases the button while they cannot hear the stimulus, which automatically increments the stimulus intensity. This process produces a zigzag chart of stimulus intensity versus time, oscillating around the subject’s detection threshold. In our testing, tone bursts lasting 100 ms were presented at 300 ms intervals. With the button pressed the tone-bursts decreased in intensity by 0.5 dB with each presentation, and increased in intensity by 0.5 dB otherwise. The frequency of the tone-bursts advanced logarithmically from 500 Hz at a rate of 1 octave per 200s. Each Bekesy audiogram took about 15 min to collect, and four replicate audiograms were collected in each subject. These repeat trials provided information on the frequency response of the BC, and reliability of the acoustic contact between the BC and the subject’s skull. Between each trial the BC was completely removed and re-placed, with a 5 min rest in between.

2.6. The custom-made optically-coupled bio-amplifier, interference and electrical noise reduction

Our system uses a custom-built battery-powered bio-amplifier, optically coupled to the analysing laptop via a 1 mm diameter plastic light guide. To reduce electrical artefact the amplified biological signal was encoded onto the light signal as an amplitude modulated (AM) signal with a 9 kHz carrier frequency. The bio-amplifier is inherently safe, is electrically ‘floating’ (better at rejecting common-mode interference), and needs no external reference connection (so one less electrode). It also avoids problems inherent with mains-powered bio-amplifiers: larger size, increased likelihood of mains interference, and high cost (patient safety requires extreme design). The bio-amplifier had a fixed gain, a second-order passband of 60 Hz–500 Hz (rejecting ECG and the delta and theta waves of EEG), with an unfiltered and un-averaged noise floor of ∼2 μV,

Because a doubling of the noise/interference quadruples the averaging required for the same SNR, it was worth reducing interference as much as possible by (i) turning off unnecessary equipment; (ii) distancing or shielding radiating cables; (iii) using twisted-pair drive wires to the BC and masking ear-bud; (iv) placing the small bio-amplifier behind the head as shielding; and (v) using short, light-weight leads to reduce interference and stimulus artefact.

Although we started with Ag/AgCl ECG electrodes on the scalp and mastoid, we later improved our SNR and success rate by using custom-made stainless steel plate electrodes, curved to the neck and forehead. Stainless-steel is inexpensive, readily available as sheet or foil, has relatively low noise and drift, can be cut and shaped, and is easily cleaned. Although stainless-steel is electrically noisier than Ag/AgCl electrodes per unit area, the larger surface area of our stainless steel electrodes meant that they were quieter than ECG electrodes overall, with electrode noise well below the physiological and bio-amplifier noise. Although stainless-steel electrodes are often more polarised than Ag/AgCl (producing a small DC offset), the bio-amplifier high-pass filtered the electrode signal at 60 Hz, and patient movement and therefore movement artefacts were minimal. In addition, the large stainless-steel electrode at the hairline avoided the problem we had in some patients who had coarse hair at the vertex that hampered electrode contact. The neck electrode also avoided using a mastoid electrode that was too near the surgical field. Both electrodes were centred under the latex band on the midline, although their exact placement was not crucial, because ABR waveform morphology was unimportant for ASSR monitoring. The leads from the electrodes to the bio-amplifier were light-weight, and as short as possible to minimise interference, and the bio-amplifier was tucked under the patient’s neck, so that their head and body acted as an electrical shield (this only works if the operating table is not radiating interference).

Our software also included an efficient mains-harmonic nulling process, common in digital signal processing: the incoming biological signal is delayed by a full cycle of the mains fundamental frequency, and subtracted from the un-delayed signal, producing a digital comb filter with nulls at every harmonic of the mains frequency. Frequency components between the nulls of the comb filter are modified (by a known amount), including noise components that can increase or decrease depending on their frequency relative to the comb filter nulls. Nevertheless, we found that this mains suppression algorithm worked well to suppress mains interference, but in our case it was rarely needed because of the effectiveness of the battery power supplies of the laptop and bio-amplifier, the electrically-floating bio-amplifier design, and the effectiveness of our electrostatic shielding. Most often no mains interference was evident in the raw bio-amplifier output, even without the mains suppression algorithm.

2.7. The chirp stimulus

Brief clicks are often used as simple broadband stimuli in evoked response averaging (Martin and Stecker 2008), but they do not ensure simultaneous nerve firing (because of the acoustic travel time along the cochlea), and their high peak amplitude can distort through the sound system. A logarithmic up-chirp, however, has a similar frequency content, and partly compensates for cochlear travel time by staggering the delivery of its frequency components (Shore and Nuttall 1985, Elberling et al 2007). Its lower peak amplitude for the same energy/Hz (and so the same neural stimulation) also avoids distortion. For these reasons, we have used logarithmic up-chirps (5 ms 2–16 kHz phase-alternating sweeps at 80 s−1).

2.8. ASSR detection, system setup and automatic threshold tracking

During all ABR and ASSR recordings the raw bio-amplifier output was displayed clearly on the laptop. The software used synchronous detection to determine the amplitude and phase of any specific frequency component in the bio-amplifier output, including that of the ASSR’s first, second and third harmonics, and also the amplitude of the background electrical noise at 70 Hz (a frequency adjacent to the ASSR’s own first harmonic of 80 Hz). The magnitudes of these signals were charted continuously (see Results). To improve the SNR of recordings, the synchronous detection was performed using a concatenated input waveform, constructed from 32 contiguous input epochs (500 ms). This concatenation (and the phase coherence it maintained) reduced the noise floor almost six-fold. With this process the ASSR detection noise floor was <50nV, which was much lower than the ASSR itself. Overall we had a constant indication of the SNR of our recordings, based on the real-time display of the raw input, the displayed 70 Hz noise floor, and the magnitude of the ASSR harmonics, and we could easily distinguish physiological changes from changes due to electrical interference.

To ensure that the ASSR originated from the operated ear alone, the contra-lateral/non-operated ear was simultaneously masked with band-pass noise (2–16 kHz), played into the contra-lateral ear through an ear-bud taped in position, and adjusted manually at the start of surgery to produce effective contra-lateral masking (estimated from pre-operative audiograms).

To estimate cochlear sensitivity indirectly from the brainstem response, our custom-written software automatically adjusted the chirp level every 250 ms, in a process similar to the Bekesy audiometry described earlier. The chirps were presented continuously at 80 s−1 via the BC, and their level was adjusted automatically by the software to produce an ASSR (first harmonic) near to a pre-set ‘ASSR target’ amplitude (conservatively set to two times the system’s noise floor at 80 Hz). The stimulus level was controlled using a continuous closed-loop feedback system acting on the error between the measured ASSR amplitude and the predefined target value. The controller adjusted stimulus intensity in real time to minimise this error, incorporating proportional and integral components to achieve rapid convergence while avoiding steady-state offset. Controller gain was empirically tuned in real-time to balance responsiveness against stability, recognising that excessive gain produced oscillatory behaviour in the estimated threshold, due to the inherent latency and variability of the biological response. At higher controller gains, the system exhibited oscillatory behaviour consistent with an underdamped feedback loop (as illustrated in figure 4), confirming the expected trade-off between convergence speed and stability.

As for system setup, at the start of recording the noise floor was estimated with the stimulus and masking off. The level of the unmasked (binaural) bone conduction stimulus was increased manually using the software, to produce an ASSR that was twice the measurement noise floor (if a binaural response cannot be detected, then a monaural response from the operated ear will also be unobtainable). The software’s ASSR target was then set to this value, and the software was switched to automatic mode to maintain this ASSR target value by constantly adjusting the chirp level. The contra-lateral masking was then switched on, and a slight rise in ASSR threshold was observed (if the contra-lateral ear in a particular patient was contributing to the ASSR at that chirp level). This whole setup process took about two minutes, after which the system was left to track ASSR threshold automatically. The setup procedure was repeated (rarely) if there was a change in the measurement noise floor or there was some other major disruption. In some cases, this standard tracking was interrupted occasionally to check the measurement noise floor (marked by N), or to demonstrate some aspect of the system performance (e.g. an ASSR input-output (IO) function, or changes in ASSR target setting). These checks were not necessary but were done to demonstrate the behaviour of the system.

3.1. Performance of the custom-made bone conductor

The frequency response of our custom BC was tested in four normally hearing subjects using Bekesy audiometry. The four repeat audiograms from each subject are overlaid in figure 2(a) (light grey; stimulus level shown in dB re 1 Vrms drive). Each set of audiograms from each subject are successively shifted by 20 dB for clarity. The means of these four replicates are shown in black. Within each subject the raw Bekesy audiograms (light grey) showed only a small variance SD (<3.5 dB in every subject; BC removed and replaced between trials), but there was a larger variation between subjects (see figure 2(b)). This suggests that the variability in BC efficiency between subjects was individual to each subject, and probably due to skull structure (Gorga et al 1993). The high-frequency thresholds were more elevated in the two older subjects, presumably by presbycusis (65yo and 37yo, figure 2(b)), but in the two younger subjects (15yo and 17yo) the BC audiograms extended to well above 16 kHz. Overall, the audiograms were not flat from 2–16 kHz, but dropped at about 6 dB/octave (figure 2(b)): with a fixed electrical drive across frequency, the chirps stimulated the 2–4 kHz cochlear region about 15 dB more than the 8–16 kHz region. Although the 2–16 kHz chirps could have been pre-emphasised at 6 dB oct−1 to compensate, the idiosyncrasies of each patient’s skull would still have remained. We could have compensated for each patient’s pre-operative BC audiogram, but this would have required much more time, would bias the responses toward areas of pre-existing hearing loss, and any improvement may not justify the effort. As it stands, our BC allows automatic tracking of broadband hearing threshold over a 70 dB dynamic range above normal threshold, without a detectable electrical artefact (see later).

Figure 2. (a) Pure tone Bekesy audiograms for four representative subjects with nominally normal hearing (aged 65, 37, 17 and 15 yo) using our custom-made high-frequency bone conductor (vertical axis is dB relative to the maximal electrical drive to the bone conductor of 1V rms). For clarity, traces in figure 2(a) are successively shifted downward by 20 dB. Four repeat trials for each subject are overlaid (raw Bekesy traces in light grey), with the bone conductor removed and reapplied between trials. The average audiogram for each subject is shown in black. (b) Mean Bekesy audiograms from the four subjects of (a), overlaid without the vertical shifting. Note that over the range of our standard chirp (2–16 kHz), the bone conduction efficiency dropped at about 6 dB oct−1 (dashed line fitted by eye).

Download figure:

Standard image High-resolution image 3.2. ABR waveforms as evidence of the quality of the recording system

Figure 3 presents sample ABR waveforms to illustrate the quality of the electrode contacts, the reproducibility of responses, the bio-amplifier’s low-noise, and the insignificant stimulus artefact from the BC. Figure 3(a) shows a montage of unmasked (binaural) responses from 30 dB to 60 dB nHL in one patient, after anaesthesia but before surgery. Note the reproducibility and lack of stimulus artefact. Figure 3(b) shows the drop in ABR amplitude with contra-lateral masking, demonstrating the effectiveness of the contra-lateral earbud noise masker. Note also that the collection of a single set of four ABR waveforms required 200 s (four waveforms with n = 1000 at 20 s−1) as discussed later. Figure 3(c) shows masked (monaural) responses from an exostoses patient at the start (grey) and end (black) of surgery, showing little change in ABR waveforms after prolonged drilling in the ear canal. Conversely, figure 3(d) shows an abrupt change in the masked/monaural ABR waveforms at first lifting the tympanomeatal flap in one cholesteatoma patient, which presumably over-stimulated the cochlea and produced a (temporary) threshold shift. In this case, MEIOM did indicate a problem with that surgical manipulation on this occasion. We only observed such a change in this single early case, when using the ABR waveform averaging, because we had not yet developed our ASSR threshold tracking process. We were careful in later cases to avoid similar changes.

Figure 3. Examples of ABR waveforms to demonstrate the quality of the electrodes, bio-amplifier and software. (a) A montage of 25 ms automatically averaged ABR waveforms (n= 1000) at four interleaved chirp levels (30, 40, 50, 60 dB nHL) in a deeply anaesthetised patient early in cholesteatoma surgery (16 s−1 5 ms logarithmic 2–20 kHz bone-conducted up-chirps with alternating polarity; vertical scale bar 200nV, horizontal scale bar 5 ms, same scales for a-d). Four replicates are shown at each chirp level (light grey), with a grand average at each level (n = 4000; black). A response is clearly visible at all levels. Note the high reproducibility and the absence of stimulus artefact. The upward peak near 10 ms is the Jewett wave V. (b) As for (a), showing averaged ABR waveforms (n = 1000) for unmasked/binaural stimulation (two replicates; grey) and masked/monaural stimulation (two replicates; black). (c) As for (a), showing masked/monaural average waveforms (n = 1000) before (black) and two hours later (grey) at the end of exostoses surgery. Despite 40 min of intensive drilling, there was little change in the waveforms, even though the patient had near normal thresholds at the start of surgery, and was therefore at risk of noise damage. (d) In this cholesteatoma patient a change in ABR waveform was observed on first lifting the tympanomeatal flap early in surgery (grey, just before lift; black just after).

Download figure:

Standard image High-resolution image 3.3. ASSR threshold tracking

To illustrate the process of ASSR threshold tracking in detail, figure 4 shows ASSR data from the end of an exostoses procedure (additional data from the same patient is presented in figure 6(a)). Five system variables available to the surgeon in real time are shown: detection noise level at 70 Hz (light grey; close to the horizontal axis), ASSR amplitude (dark grey; 80 Hz first harmonic with a score of 20 equivalent to 1 μV rms), and chirp sound level (solid black). First, between 0 and 1.5 min the noise floor was estimated (indicated as N), then the ASSR target was set to twice the noise floor, before the system automatically tracked threshold reaching ASSR target (near 3 min). Quiet periods in this example and all subsequent figures are marked Q. Note the slow oscillation in chirp level between 1.5 and 3.5 min (and again between 10 and 12 min), because the negative feedback gain was set slightly too high (it can be adjusted in real time but was most often set once at the start of surgery). Although not a standard part of ASSR tracking, figure 4 includes an example of a manual IO function, estimated between 4 and 8 min (plotted IO function not shown), before the system again tracked threshold. At 12.5 min the chirp was turned off to check measurement noise (N), but in this case the feedback loop was left on to demonstrate how the software rapidly increased the chirp level when no response was present. This also illustrates how fast the system is in signaling a rapid elevation of threshold. Again, although not a standard part of ASSR threshold tracking, from 18 to 24 min the ASSR target was changed in a stepwise fashion to demonstrate the system’s tracking speed and the ASSR threshold’s sensitivity to the ASSR target setting.

Figure 4. (a) Typical ASSR threshold tracking data from the end of an exostoses procedure, showing measurement noise level at 70 Hz; (lower dotted trace), ASSR amplitude (first harmonic at 80 Hz; solid grey; right axis) and chirp level (2–16 kHz up-chirps at 80 s−1; solid black; left axis). The measurement noise floor was estimated (0–1.5 min; indicated by N) before the system automatically tracked threshold in quiet, reaching ASSR target level near 3.5 min (note the slow threshold oscillation between 1.5 and 3.5 min, and 10 and 13 min, due to excessive feedback gain). A manual input-output (IO) function was estimated between 4 and 8 min, before the system again tracked threshold in quiet (Q). At 12.5 min there was a brief re-check of the noise floor with the negative feedback of the software intact, chirp level rose when the stimulus was switched off and no response was present, then from 18 min to the end the ASSR target was changed to demonstrate the system’s tracking speed, and the sensitivity of estimated chirp threshold to the ASSR target setting. (b) Inset shows ABR waveforms over two stimulus cycles (n = 1000; 10 dB to 60 dB nHL bottom to top; three replicates at each chirp level; phase fixed but arbitrary). Waveforms were recorded during the manual IO function from 4 to 8 min in (a).

Download figure:

Standard image High-resolution image

Figure 5. (a) ASSR threshold tracking in an awake normally hearing subject with a binaural noise masker ON and OFF (illustrating tracking speed, stability and resolution). (b) ASSR threshold tracking in an anaesthetised patient with no 2 kHz to 16 kHz hearing thresholds better than 70 dBHL in pre-operative audiometry. When the system was set to automatic tracking, the chirp level rose without evoking an ASSR response, so the system could not converge to an ASSR threshold (showing system dynamic range and absence of artefacts). (c) ASSR threshold tracking during cholesteatoma revision surgery demonstrating a temporary threshold shift. The measurement noise floor was estimated repeatedly throughout the procedure (N), and with the chirp stimulus off and the software feedback disabled the chirp level dropped temporarily to zero. After initial setup, the ASSR threshold converged to 33 dBHL, consistent with the pre-operative audiometry. At about 33 min, blunt dissection commenced and the ASSR threshold rose slowly until the dissection stopped at 80 min, after which the ASSR threshold slowly recovered. Near 125 min and 140 min during quiet (Q) the ASSR threshold was very near its initial value. This slow rise and fall in ASSR threshold was not due to slow tracking speed, which was much faster (see a and b). The rise in ASSR threshold near the end of the procedure was due to post-operative masking noise from staff (150 min onwards).

Download figure:

Standard image High-resolution image

Figure 6. (a) ASSR tracking during an exostoses procedure (same patient as figure 4). Shown are the typical 70 Hz measurement noise floor (dotted black near horizontal axis), the ASSR magnitude (80 Hz first harmonic, grey; right vertical axis), and the chirp sound level (black, left vertical axis). The ASSR target (dotted grey line) and the system’s feedback gain (not shown) were adjusted to give fast and reliable threshold tracking, without responding to noise when the target was too low, or producing slow oscillations when feedback gain was too high, as in figure 4(a). The patient had near-normal pre-operative BC audiometry (10 dB nHL), consistent with the bone-conduction threshold measured in the OR ante room and the ASSR threshold established automatically in quiet (Q; 5–10 min). Two more measurement noise estimates were made at 10 min and 23 min, this time with the chirp off but the software feedback still enabled, producing a transient rise in nominal chirp level, demonstrating how rapidly the system responds to a drop in the ASSR. At 24 min drilling commenced, and ASSR threshold was markedly elevated (presumably by masking noise). During brief quiet periods (Q), the ASSR threshold dropped towards its initial value in quiet, and when given sufficient time in quiet at 38 min the ASSR threshold was very near its initial value of 22 dBHL. At the very end of the procedure, the ASSR threshold was very close (18 dB HL, see figure 4) to its initial value (22 dB). (b) As for (a), in this stapedotomy, after checking the measurement noise floor (N) and setting the ASSR target, the system tracked binaural ASSR threshold in quiet (Q) near 27 dB nHL (without contra-lateral masking). Near 7 min, contra-lateral masking was turned on, which elevated the estimated threshold to near 40 dB nHL, partly because of the smaller monaural ASSR, and partly due to the higher audiometric threshold in the operated ear. Between 33 and 35 min the measurement noise floor was determined with the stimulus and feedback off (N), and immediately after (35–38 min) a manual ASSR input-output function (IO) was determined. Near the end of the procedure (49 min), contra-lateral masking was turned off, and the estimated threshold returned to near its original binaural value. (c) In this relatively quiet tympanoplasty, after a check of measurement noise floor (N) and setting the ASSR target, contra-lateral masking noise was turned on, and the target was re-adjusted. After another re-check of noise floor (N), the system tracked threshold reliably in quiet (Q). When room masking noise was present, the estimated threshold rose, but during brief quiet (Q) threshold was near its initial value. Binaural psychophysical threshold was 0dB nHL in the ante-room. (d) In this noisy revision tympanoplasty, initial threshold was acquired at 23 dB nHL (dashed line), but threshold was elevated through much of the procedure by the acoustic noise of the hissing aspirator from 30 min on. During brief quiet periods (Q), threshold dropped close to its initial value (23 dB nHL; psychophysical threshold just before surgery was 10 dB nHL).

Download figure:

Standard image High-resolution image

Apart from tracking ASSR thre