Single trial detection of human vestibular evoked myogenic potentials is determined by signal-to-noise ratio

J Appl Physiol 109: 53 59, 2010. First published May 6, 2010; doi:10.1152/japplphysiol.01139.2009. Single trial detection of human vestibular evoked myogenic potentials is determined by signal-to-noise ratio Neil P. McAngus Todd, 1 Sally M. Rosengren, 2 Sendhil Govender, 2 and James G. Colebatch 2 1 Faculty of Life Science, University of Manchester, Manchester, United Kingdom; and 2 Prince of Wales Clinical School and Medical Research Institute, University of New South Wales, Randwick, Sydney, New South Wales, Australia Submitted 6 October 2009; accepted in final form 5 May 2010 Todd NP, Rosengren SM, Govender S, Colebatch JG. Single trial detection of human vestibular evoked myogenic potentials is determined by signal-to-noise ratio. J Appl Physiol 109: 53 59, 2010. First published May 6, 2010; doi:10.1152/japplphysiol.01139.2009. Vestibular reflexes in humans can be assessed by means of acoustically evoked responses of myogenic origin. For the vestibular-collic pathway this is termed the vestibular evoked myogenic potential (or VEMP) and for the vestibular-ocular pathway the ocular VEMP (or OVEMP). Usually VEMPs require an averaging process to obtain a clear response against the background myogenic activity, but depending on the combination of target reflex and stimulus mode, in some cases clear responses can be observed in single trials without averaging. We aimed to test whether this difference in detectability was simply related to signal-to-noise ratio (SNR), or a manifestation of some other difference in the reflex pathways. In four healthy subjects we recorded VEMPs and OVEMPs in response to 2-ms, 500-Hz sound pips and 10-ms, 100-Hz transmastoid vibrations at four intensity levels, and also determined thresholds. A plot of probability of detection P vs. SNR for all subjects and conditions fell onto a single sigmoid curve. When fitted by a logistic function after linearization a regression yielded an R 2 of 0.89 (n 64, p 0.001), with parameter estimates of 2.9 and 2.0. Three patients with superior canal dehiscence, characterized by significantly lowered thresholds for sound-activated responses, exhibited a similar detection curve. We conclude that single trial detection of evoked myogenic potentials is a property mainly determined by SNR. Thus vestibular reflexes, differing in both their response magnitude and in their levels of myogenic activity by more than an order of magnitude, can be described by a single relationship when their magnitude is expressed relative to background activity, demonstrating the fundamental importance of the SNR. vestibular reflexes; signal detection; vestibular evoked myogenic potential IN HUMANS, the integrity of the vestibular reflex systems may be investigated by means of evoked myogenic potentials activated by sound or vibration. For the vestibulo-collic system the preferred target muscle is the sternocleidomastoid (SCM) and the potentials are recorded using a belly-tendon montage with a background muscle contraction. The response is termed the vestibular evoked myogenic potential (or VEMP) and characterized by a short-latency biphasic positive-negative wave of latency 13 and 23 ms (4). The VEMP is interpreted as being a manifestation of an inhibitory trisynaptic neuronal arc passing through the ipsilateral medial vestibulospinal tract (23) and may be activated by sound or vibration (4, 16, 26). The optimal stimulus frequency for sound lies between about 400 and 800 Hz and often short tone bursts or pips of 500 Hz are used to obtain a response (17). VEMPs may also be recorded in Address for reprint requests and other correspondence: N. P. McAngus Todd, Faculty of Life Science, Stopford Bldg., Univ. of Manchester, Manchester M13 9PL, UK (e-mail: neil.todd@manchester.ac.uk). response to impulsive head acceleration, but in this case the responses are bilateral, consistent with vestibular apparatus activation on both sides of the head (13). More recently a response analogous to the VEMP has been described for the vestibulo-ocular system, where the targets are the extraocular muscles (EOMs) and the response is termed the ocular VEMP (or OVEMP) (18). The OVEMP is also recorded using a differential montage, with surface electrode pairs positioned inferior to the eyes on the face. The OVEMP is characterized by both excitatory and inhibitory potentials, which for sound are largest contralaterally with an initial negativity at 10 ms (2, 18). OVEMPs are also easily produced by head vibration or impulsive head acceleration and show a bilateral negative response (19). For head vibration it has been found that the OVEMP shows a well-defined resonance at 100 Hz (20). In a series of recent studies investigating the sensitivity and tuning of the VEMP and OVEMP produced by sound and vibration, we found considerable variation in their response properties, which we have suggested may be attributable to biomechanical properties of the end organs, electrical properties of vestibular hair cells, or features of the vestibulo-ocular pathway (20 22). In the course of these investigations we observed an effect whereby large responses could be commonly observed on single trials without averaging, whereas typically 100 200 repetitions are usually required for the detection of VEMPs or sound-evoked OVEMPs. On first inspection this remarkable effect appeared to be specific to vibration-activated OVEMPs. A plausible alternative hypothesis is that the signal-to-noise ratio (SNR) is higher for vibration-activated OVEMPs, thus improving their detectability. Our aim was to test this hypothesis by determining the empirical probability that an observer would judge a response to be present in single trials of VEMPs and OVEMPs produced by short sound and vibration stimuli. This question is of both theoretical and practical importance since we are comparing two systems where the muscles, the SCM and EOM, have differing physiological characteristics, and the modes of stimulation, sound and vibration, have differing actions on the vestibular receptors. We measured the noise level for each condition from the prestimulus interval and the response amplitude from the average in order to estimate the SNR for each condition. In addition we estimated the effective intensity of the stimuli by measuring the thresholds for each of the conditions in the normal subjects. For sound stimulation we compared the normal subjects to patients with superior canal dehiscence, a condition that increases the amplitude of sound-evoked vestibular reflexes due to their low thresholds (25). http://www.jap.org 8750-7587/10 Copyright 2010 the American Physiological Society 53

54 SINGLE TRIAL DETECTION OF HUMAN VESTIBULAR EVOKED MYOGENIC POTENTIALS METHODS Subjects. Four normal subjects, with no history of auditory or vestibular disease, participated (1 woman, 3 men; age: 23 54 yr old). The subjects vestibular thresholds were in the normal range. Data from three patients with superior canal dehiscence (SCD) (2 men, 1 woman; age: 31 59 yr old) were also included. The diagnosis of SCD was made on the basis of clinical features, pathologically low VEMP threshold and evidence of dehiscence shown by high-resolution CT imaging (9). All participants gave informed written consent according to the Declaration of Helsinki, and the study was approved by the local ethics committee (Human Research Ethics Committee, Northern Network, South Eastern Sydney and Illawarra Area Health Service, NSW, Australia). Stimuli. Subjects were stimulated with 100-Hz, 10-ms sine waves delivered to the mastoid (i.e. directly posterior to the external auditory meatus) by a handheld minishaker (model 4810, Brüel and Kjaer P/L, Denmark). The minishaker was fitted with a custom cylindrical Perspex rod (diameter 2.5 cm, length 9.2 cm) and held normal to the head by the experimenter with approximately 1 2 kg force. The stimuli were delivered at four intensities, produced by input voltages of 20, 10, 5, and 2.5 V peak to peak (pp). Head acceleration in the interaural (y) direction was measured in all normal subjects with two linear accelerometers (model 751 100, Endevco). The accelerometers were placed normal to the skull on the temporal bone above the external auditory meatus and held in place by tight elastic bandages. The input voltages produced an initial mean peak head acceleration at 2.7 ms, with amplitudes of 0.39, 0.19, 0.09, and 0.05 g at each intensity level, respectively (averaged over the left and right sides of all normal subjects). With increasing intensity a second acceleration peak occurred at 13.7 ms, which had an amplitude of 0.57 g at maximum intensity. The normal subjects were also stimulated with 500-Hz, 2-ms air-conducted (AC) tone bursts delivered at a single intensity of 136 db peak sound pressure level (SPL) (5 V pp) with calibrated headphones (TDH 49, Telephonics). The patients with SCD were stimulated with AC sound only as we were primarily interested in observing high-amplitude responses evoked by sound. They were additionally stimulated at lower sound intensities in steps of 6 db (i.e., 136 100 db peak SPL). Both stimuli were generated by means of customized software, using a laboratory interface (1401plus, Cambridge Electronic Design, Cambridge, UK) and custom amplifier. A total of 100 stimuli were delivered at a rate of 5 Hz. VEMPs. Subjects were reclined to 30 above the horizontal, and surface potentials were recorded using pairs of Ag/AgCl electrodes. The recording electrode was placed over the middle of the SCM muscle belly and referred to an electrode over the medial clavicle. An earth was placed on the sternum. Rectified EMG activity was monitored online and recorded, and care was taken to ensure constant background contraction of the SCM between trials. EMG was sampled at 5 khz from 20 ms before to 100 ms after stimulus onset, amplified, filtered (8 Hz to 1.6 khz) and sampled with a second CED power1401 and SIGNAL software (version 2.15, Cambridge Electronic Design, Cambridge, UK). Negative potentials at the active electrodes were displayed as upward deflections. All individual trials were recorded. For the vibration stimulus, responses in both SCM muscles were recorded. The vibration VEMP signal was defined as the initial positive/negative wave in the ipsilateral SCM and the initial positive wave in the contralateral SCM. These components of the response have been shown to be vestibular-dependent using a stimulus waveform similar to 100 Hz (13). Later waves are likely to include stretch reflexes and were therefore not included in the analysis. The signal amplitude was measured peak to peak for the ipsilateral response and base to peak for the contralateral response. The amplitude was measured from the average of 100 trials. In the raw trials, responses were included as hits when both peaks were present on the ipsilateral side and when one peak was present on the contralateral side. Some normal subjects have an early negativity preceding the positivity on the contralateral side (13), and if this occurred, the contralateral negativity/positivity was treated as a whole. The peak-to-peak amplitude was measured from the average trace and hits were registered when both peaks were present. The morphology and latencies of the peaks in the average were used as a guide for each subject. The noise was calculated as the mean rectified EMG recorded over the 20-ms prestimulus period. For the sound stimulus, only the ipsilateral response was recorded, and the signal was defined as the initial positive/negative peak, i.e., p13 n23 (4). OVEMPs. Subjects sat upright and directed their gaze to a target located 1 m away at an elevation of 20. Surface potentials were recorded using pairs of 9 mm or self-adhesive Ag/AgCl electrodes (Cleartrace 1700 030, Conmed). The active electrode was placed on the orbital margin below the eye and referred to an electrode 15 mm below it on the cheek. The earth was placed on the sternum. EMG was amplified and filtered (8 Hz 1.6 khz), and sampled at 10 khz for 70 ms, from 10 ms before to 60 ms following stimulus onset. Electrode impedance was maintained below 5 k. Online automatic artefact rejection was used to reject voluntary blinks. Negative potentials at the active electrodes were displayed as upward deflections. For the vibration stimulus, OVEMPs were recorded from both eyes. As the entire OVEMP waveform appears to be vestibular dependent (20, 21), the OVEMP signal was defined as the largest biphasic wave, beginning with either polarity, and was measured peak to peak from the average. In the raw trials, responses were included as hits when any biphasic wave present in the average trace could be identified at the appropriate latency. The noise was calculated as the mean rectified EMG recorded over the 10-ms prestimulus period. For the sound stimulus, only the contralateral response was recorded (2), and the signal was defined as the largest biphasic wave. To investigate the range of EMG at the extraocular electrodes, we recorded rectified EMG from all subjects without stimulation at four levels of gaze: neutral, 10 and 20 upward, and maximal up-gaze. VEMP/OVEMP thresholds. Thresholds were measured for each subject by successively reducing stimulus intensity in 6-dB steps until the target waves could no longer be observed in the average waveform. The intensity was then increased in steps of 3 db until a response could be reliably observed in repeated averages. The procedure was carried out separately for each response, stimulus modality, and side. The result, labeled V T, was expressed in db relative to (db re) the maximum output of the stimulator. For sound this was 5Vpp (136 db peak SPL) and for the minishaker 20 V pp. Data analysis. A total of 100 trials were measured for each stimulus and recording site and the probability P of detecting a response defined as the number of hits divided by 100. The SNR was defined as the ratio of peak-to-peak amplitude to the mean rectified EMG from the prestimulus period. Univariate ANOVAs were carried out separately for the sound and vibration conditions for each of the four dependent variables, i.e., signal, noise, SNR, and P. For the vibration conditions the ANOVA was three-way, with factors of pathway (VEMP or OVEMP), laterality (ipsilateral or contralateral), and intensity (0, 6, 12, and 18 db re 20 V). For the sound conditions the ANOVA was one-way, with the single factor of pathway. The test levels of significance are denoted by p (lowercase) to distinguish them from detection probability P (uppercase). The form of the relationship between the probability of detecting a signal and the SNR r is typically sigmoidal. This is commonly modeled using a logistic function, such that if P(r) is the probability of detection as a function of r the signal to noise ratio, then P(r) 1 exp r s 1 where and s are parameters which may be related as the mean and standard deviation of the associated logistic probability density func-

SINGLE TRIAL DETECTION OF HUMAN VESTIBULAR EVOKED MYOGENIC POTENTIALS tion. The standard deviation is given by s/公3. The equation may be linearized by taking logs, i.e. Y ln(1 P(r) 1) r s s so that a linear regression can be done. The parameter corresponds to the SNR at which P 0.5, i.e. the SNR at which a response is 55 detected for half the single trials while the parameter gives a measure of the steepness of the sigmoid function. RESULTS VEMPs and OVEMPs were present in each normal subject following stimulation with both vibration and sound, with the exception of the ipsilateral VEMP to vibration, which was absent in some subjects and intensities (these cells were not included in the analyses). Vibration stimuli activate the vestibular organs bilaterally whereas sound exclusively activates the ipsilateral vestibular apparatus. Figure 1 shows the averaged responses compared with the first 10 raw responses for a single normal subject and an SCD patient. At maximum stimulus intensity, OVEMPs evoked by transmastoid vibration were present in each raw trace, but sound-evoked OVEMPs and VEMPs to both stimuli were detected much less frequently. Properties of the signal and the noise. The absolute signal amplitudes were much larger for the VEMP than the OVEMP, as expected (vibration: 63.5 vs. 10.7 V over all intensities and sides, F1,42 42.0, p 0.001; sound: 128.6 vs. 3.5 V at 136 db, F1,6 30.5, p 0.001) (Table 1). For the vibrationevoked responses, the signal tended to be larger contralaterally for the VEMP and ipsilaterally for the OVEMP (pathway and laterality interaction, F1,42 4.0, p 0.052). There was also a large difference in the background noise for the VEMP and OVEMP (vibration: 63.1 vs. 2.6 V over all intensities and sides, F1,42 711.9, p 0.001; sound: 68.4 vs. 2.3 V at 136 db, F1,6 240.9, p 0.001) (Table 1). There was a trend for the signal to increase with intensity, and there was some variation in noise between the stimulus intensities for each pathway for vibration, but none of these effects reached significance in the three-way ANOVAs. When we measured the effect of vertical gaze on OVEMP background EMG activity, we found no significant change in activity, despite large increases in angle of gaze (F3,9 1.8, p 0.212), although for each subject the background EMG level was slightly larger with maximum up-gaze than with neutral gaze. Properties of signal-to-noise ratio and response detection probability. The largest SNR and probability of detection P overall were seen for vibration-evoked OVEMPs. In response to vibration, the SNR and P were both much larger for OVEMPs than VEMPs, despite the fact that the VEMP signal was at least an order of magnitude larger than the OVEMP (main effect of pathway, SNR: 4.7 vs. 1.0, F1,42 53.0, p 0.001; P: 0.75 vs. 0.12, F1,42 338.7, p 0.001). SNR and P increased with increasing stimulus intensity for both the VEMP Fig. 1. Illustration of average and raw traces in a single normal subject (A) and a patient with superior canal dehiscence (B). For each response [vestibular evoked myogenic potential (VEMP) or ocular VEMP (OVEMP)] and stimulus type (sound or vibration) the average of 100 raw trials as well as the first 10 raw trials are shown. Vibration-evoked responses were recorded bilaterally in the normal subject. Sound-evoked responses were measured from the ipsilateral (Ipsi) sternocleidomastoid (SCM) for VEMPs and the contralateral (Contra) eye for OVEMPs in both the normal subjects and patients. In the normal subjects the most readily detected response was the vibration-evoked OVEMP. In patients with superior canal dehiscence (SCD) sound-evoked OVEMPs and VEMPs were large compared with the sound-evoked responses in normal subjects. Note the different gain bars for VEMPs and OVEMPs.

56 SINGLE TRIAL DETECTION OF HUMAN VESTIBULAR EVOKED MYOGENIC POTENTIALS Table 1. Mean values of signal, noise, signal-to-noise ratio (SNR), and probability of response detection (P) OVEMP VEMP Vibration Sound Vibration Sound Intensity* Ipsi Contra Contra Ipsi Contra Ipsi Signal 0 18.6 11.0 3.5 65.0 92.6 128.6 6 13.6 10.6 50.1 81.9 12 10.3 8.6 42.8 84.0 18 6.7 5.9 37.6 54.3 Noise 0 2.3 2.8 2.3 62.7 57.5 68.4 6 2.3 3.2 58.5 61.5 12 2.2 2.9 63.9 63.2 18 2.4 3.0 71.5 66.6 SNR 0 8.5 4.5 1.4 1.0 1.5 1.9 6 6.1 4.6 0.9 1.2 12 4.8 4.0 0.7 1.3 18 2.9 2.5 0.5 0.8 Intensity (re threshold) 0 46 42 12 22 36 19 6 40 36 16 30 12 34 30 10 24 18 28 24 3 18 P 0 0.97 0.78 0.29 0.09 0.20 0.23 6 0.93 0.77 0.09 0.17 12 0.79 0.76 0.06 0.18 18 0.54 0.47 0.07 0.13 *Intensity is given in db relative to maximum (for sound 5 V pp; for vibration 20 V pp). Signal and noise values are given in V. Noise is the mean rectified level of tonic EMG. Intensity (re threshold) is given in mean db level above threshold, which was measured for each stimulus and response separately (n 4). Ipsi, ipsilateral; Contra, contralateral; VEMP, vestibularevoked myogenic potential; OVEMP, ocular VEMP. and OVEMP (main effect of intensity, SNR: F 3,42 3.2, p 0.05; P: F 3,42 7.5, p 0.001). Similar to the pattern seen for the signal, SNR and P were larger ipsilaterally for the OVEMP and contralaterally for the VEMP (laterality and pathway interaction, SNR: F 1,42 4.2, p 0.05; P: F 1,42 9.0, p 0.01). At the highest intensity, the SNR for the largest OVEMP (in the ipsilateral eye) was about five times larger than that for the largest VEMP (in the contralateral SCM) and eight times larger than the ipsilateral VEMP (Table 1). Similarly, the probability of detection at the highest intensity was almost 1 (0.97) for the ipsilateral OVEMP, larger than the contralateral OVEMP (0.78) and much larger than the VEMP on both sides (ipsilateral 0.09, contralateral 0.20). Indeed for probability of detection, there was a ceiling effect for the vibration-evoked OVEMP at the highest intensities (intensity and pathway interaction, F 3,42 4.7, p 0.01). In contrast to vibration, for sound stimulation in the normal subjects, there were no significant differences in SNR or probability for the OVEMP and VEMP, despite large differences in the magnitude of the OVEMP and VEMP signal, i.e., peak-to-peak amplitudes (main effect of pathway, SNR: 1.4 vs. 1.9, F 1,6 0.768, p 0.414; P: 0.29 vs. 0.23, F 1,6 0.278, p 0.617). Relationship between response probability and signal-tonoise ratio. The form of the relationship as illustrated by the scatterplot of P vs. SNR showed the expected sigmoid form (Fig. 2A). Although there was a large disparity in the values of P and SNR between the different conditions, all conditions fell within a similar sigmoid distribution. A regression carried out on these data yielded parameter estimates of 2.9 and 2.0 with R 2 0.89 (Fig. 2B). Thus the single logistic function accounted for 90% of the variance in the P vs. SNR plot across all four conditions (n 64). Thresholds and effective intensities. The lowest threshold was found for the vibration-evoked OVEMP on the side ipsilateral to stimulation, which had a mean V T of 46 db re 20 V pp. The highest threshold condition was the soundevoked OVEMP, which had a mean V T of 12 db re 5 V pp. On average, the V T for vibration-activated OVEMPs was lower than for vibration-activated VEMPs ( 44 vs. 29 db, respectively), but this relationship was reversed for sound-activated OVEMP vs. VEMP responses ( 12 vs. 19 db, respectively). The mean values of the stimulus intensities above threshold are given in Table 1. The highest SNRs were present for the condition for which the effective intensities were highest, i.e., for the vibration-activated OVEMP on the side ipsilateral to stimulation. In this condition the mean SNR reached 8.6 at a mean effective intensity of 46 db. At the other end of the scale the lowest mean SNR, obtained for the vibration-activated VEMP on the side ipsilateral to stimulation, was 0.5 and corresponded to a mean effective intensity of 3 db. The relationship between SNR and effective intensity (expressed in db re threshold and determined for each individual subject) is shown in Fig. 3 as a scatterplot, which indicates a linear relationship between SNR and effective intensity. There was an overall positive correlation, r 0.72, p 0.001, n 66, which corresponded to a linear regression with slope 0.47 and intercept -6.7. The correlations for sound- and vibration-evoked responses separately were r 0.76 (p 0.05, n 8) and r 0.73 (p 0.001, n 58). The slopes for sound and vibration were similar (0.53 and 0.50). Superior canal dehiscence. When plotted on the same axes as the normal subjects (Fig. 2, C and D), the SCD data fell onto a similar distribution, suggesting that the same general logistic function relating P and SNR held. The noise levels for the VEMP and OVEMP were comparable to the normal subjects (for VEMPs: mean SCD 56.5 V, normal 62.9 V; for OVEMPs: SCD 4.3 V, normal 2.6 V). However, for AC sound stimulation, the signal values, SNR, and P in the patients were much greater than in the normal subjects for both pathways, when compared at the same high intensity, i.e., at 6dB (VEMPs: mean SCD 216 V, 3.4, 0.57, normal 129 V, 1.9, 0.23 for signal, SNR and P, respectively; OVEMPs: SCD 54.0 V, 16.7, 0.96, normal 3.5 V, 1.4, 0.29). The SNR in the patients was particularly high for the sound-evoked OVEMP, which were as high as those seen in normals with vibration, and produced a ceiling effect for probability of detection at the highest intensities (Fig. 2C). Regression analyses were carried out on individual subjects and SCD patients and compared (Table 2). These showed that the SCD patients had similar values to the normal subjects for both the and parameters ( : SCD 2.0, normal 2.0; : SCD 3.0, normal 2.9).

SINGLE TRIAL DETECTION OF HUMAN VESTIBULAR EVOKED MYOGENIC POTENTIALS 57 Fig. 2. Scatterplots of probability of detection P vs. signal-to-noise ratio (SNR) r (A, C) and regressions of linearized data points of P vs. r (B, D). Data from all normal subjects are shown in A and B, and all SCD patients in C and D. For C, four data points from 1 SCD patient had r values 16 (20.5, 25.8, 26.6, 33.2) and P 1, which are not illustrated. For the scatterplots (A and C), data from both the normal subjects and SCD patients fell within a similar sigmoid distribution, despite the fact that there was a large difference in the values of P and r between the different stimulus [sound (SND) vs. vibration (VIB)] and reflex (VEMP vs. OVEMP) conditions. When linearized the logistic functions (B and D) accounted for 80 90% of the variance in the P vs. r plot across all stimulus and reflex conditions. DISCUSSION Our experiments demonstrated that single trial detection of vestibular-dependent myogenic responses is characterized by a sigmoid relationship where the primary factor in determining the presence of the response is the SNR. Noise has been quantified using the mean rectified level of activity, which is closely related to standard deviation and root mean squared (RMS) levels, two other commonly used measures of noise (3). The mean rectified measurement includes both physical and biological sources of interference. We have assumed that good SNR ( Fig. 3. Scatterplot of signal-to-noise ratio rvs. effective stimulus intensity. To compare reflexes evoked by different modes of stimulation (sound vs. vibration) in normal subjects we expressed stimulus intensity in terms of the level in db above the threshold for each stimulus and response. Overall there was a positive linear correlation between r and effective intensity (0.72, p 0.001, n 66). Similar positive correlations were obtained for sound- and vibrationevoked responses separately (r 0.76, p 0.05, n 8 for sound vs. r 0.73, p 0.001, n 58 for vibration). practice and electronic design will ensure low levels of physical interference (e.g., interference related to the electrical supply frequency). There is considerable variability in the level of the SNR depending on the mode of activation (sound or vibration) and the pathway through which the response is measured (to the neck or to the eyes), but the response characteristic appears to be similar in all cases. Both VEMPs and OVEMPs are normally recorded using averaging techniques, by which the increase in signal with every stimulus repetition is greater than that of the noise, resulting in an increased SNR. Analysis of individual trials, on the other hand, allows measurement of physiological properties not accessible using the standard average. These include temporal and amplitude jitter of the signal, which relate to the processes that generate the signal and affect the appearance of both the rectified and unrectified averages (3). Our findings for the normal subjects are remarkable given the 50-fold difference in the magnitude of the background activity in the vestibular-ocular vs. vestibular-collic systems. For the SCD patients the SNRs for the sound-evoked responses were larger, by more than an order of magnitude in the OVEMP case; however, the parameters characterizing single trial detection were indistinguishable from those for the normal subjects. The sigmoid relationship for both normal and SCD subjects represents a combination of both physiological properties of the reflexes, including the variability in both the background muscle activity and the signal, plus psychophysical characteristics of the observer, including the likely presence of false positives at low SNRs. While the responses from both the normal subjects and SCD patients fell on the same sigmoid distribution, there were clear differences in SNR and probability of detection between conditions. This effect can at least partly be explained by differences in stimulus intensity relative to threshold. It is not mean-

58 SINGLE TRIAL DETECTION OF HUMAN VESTIBULAR EVOKED MYOGENIC POTENTIALS Table 2. Parameter estimates for the single trial detection function for individual subjects R 2 n Normal subjects Individual analyses mean (range) 2.0 (1.6 2.4) 2.9 (2.6 3.3) 0.9 (0.83 0.96) (12 16) Combined analysis (all normal subjects*) 2.2 2.9 0.89 64 SCD patients Individual analyses mean (range) (AC only) 2.0 (1.8 2.4) 3.0 (2.2 3.6) 0.93 (0.86 0.98) (8 13) Combined analysis (all SCD patients) (AC only) 1.9 2.8 0.83 34 *Data from all normal subjects are illustrated in Fig. 2, A and B. Illustrated in Fig. 2, C and D. SCD, superior canal dehiscence. AC only indicates that only air-conducted sound was used. ingful to compare evoked responses when using modality-specific measures of intensity, particularly as each of our stimuli were designed to fall within the preferred frequency range for each modality (i.e., 500-Hz air-conducted tone bursts and 100-Hz vibration; 21). For this reason we employed a measure of effective intensity, i.e., the intensity level (in db) above the threshold for each stimulus and response, when making crossmodality comparisons. We found that the vibration-activated OVEMP had the lowest threshold and hence the highest effective intensity at the maximum input voltage, consistent with the high SNR and probability of detection for this response. In contrast, thresholds to sound stimuli in the normal subjects were higher, consistent with the lower SNR and probability values reported. The nature of the 100-Hz stimulus is also responsible for the apparent reversal of laterality of the main response in both the eyes and neck (i.e., larger responses contralaterally for the VEMP and ipsilaterally for the OVEMP). Similarly, for the SCD patients the form of the sigmoid detection function was indistinguishable from that of the normal subjects, but there was an upward extension in the SNR and probability of detection following stimulation with sound. This is due to the well-established reduction of vestibular threshold to sound in patients with SCD (1, 25). The soundevoked OVEMP was particularly large in the patients, and this has previously been suggested to be a consequence of the excitatory nature of the OVEMP compared to the inhibition underlying the VEMP, the size of the latter being intrinsically limited by the level of tonic activity as a consequence of its origin as an inhibition (12). Despite this, the probability of detection for the VEMP reached 0.92, suggesting that, although the OVEMP was easier to detect overall, with even higher intensities the VEMP might become detectable in all raw traces. The findings for the SCD patients indicate that similar levels of SNR can be obtained using sound as for vibration, if the stimulus is sufficiently intense compared with threshold. However, it is unlikely that the afferents stimulated using such high relative intensities in the patient are the same as those for normal subjects, as there may be spread to other vestibular receptors (12). There is some evidence that the different modes of stimulation produce different patterns of activity in the vestibular end organs, and the otolith organs in particular. Whereas airconducted sound may selectively activate the sacculus (6, 8, 11, 27), low-frequency vibration at 100 Hz in the transmastoid plane may be selective for utricular afferents (5, 19, 21). These differences are likely to have contributed to the variation we found in threshold values. For example, we have suggested that the sensitivity to 100-Hz vibration is produced by a combination of the biomechanical properties of the utricle and amplification in utriculo-ocular pathways (22), possibly related to ocular microtremor (15). Type I hair cells, which are characterized by irregular spontaneous rates and high gain, are concentrated around the striola and are especially sensitive to vibration stimuli (5). There are larger absolute numbers of afferent fibers arising from the utricle than the saccule (14). Therefore, sensitivity of the OVEMP to vibration may be due to larger number of type I afferent fibers projecting from the utriculus to ocular muscles than to neck muscles (10). The relative strength of the utricular-ocular vs. saccular-ocular projections is also supported by intracellular recordings in response to electrical stimulation of the end-organs (7, 24). These differences may reflect the relative importance of the utriculo-ocular reflexes. Some caution is required, however, in interpreting the differences between VEMPs and OVEMPs, as different methods were used for estimating a response for each. For the VEMP, only the initial vestibular-dependent peaks were measured, while for the OVEMP, the entire waveform was measured. The present investigation raises issues of potential practical significance. Based on the current threshold and signal data, vibration appears to be the more effective stimulus in normal subjects, as safe stimulation is possible at levels above threshold much higher than those possible for sound stimulation. However, the shape and direction of application of the vibration will affect the morphology, polarity, and latency of both the VEMP and OVEMP (13, 19), which may change the detectability of the responses. Vibration- and sound-evoked OVEMPs are also likely to depend on different vestibular receptors and so cannot be considered to be equivalent. Given the very high effective intensities that are achievable with vibration, spread to other receptor species with very intense stimulation is also a potential consideration. Our findings for the OVEMP depended on cooperative subjects who were able to minimize facial muscle activity during the recordings. Although we found that the background activity (i.e., noise) for the OVEMP was relatively much smaller than the VEMP, it did not increase significantly with increasing angle of vertical gaze. In both cases, noise means more than just electrical interference and is mainly of biological origin. For the VEMP, the noise originates mainly as tonic activity in SCM muscle and is a prerequisite to recording the response (as the reflex is inhibitory). For the OVEMP, a large proportion of noise probably primarily originates from muscles unrelated to the reflex. In fact, poor relaxation of facial muscles produces excessive interference in many patients, which can particularly affect sound-evoked OVEMPs (Rosengren, unpublished observations). Our data suggest that for

OVEMPs, it may be useful to simultaneously record averages of both rectified and unrectified EMG, similar to the VEMP, to monitor and minimize noise during the recording. SNR can be used as a measure both of the quality of individual recordings as well as, more broadly, a measure of the strength of neural connections underlying reflex responses, assuming differences in relative stimulus intensities are allowed for. GRANTS This work was supported by the National Health and Medical Research Council of Australia and the Garnett Passe and Rodney Williams Memorial Foundation. DISCLOSURES No conflicts of interest, financial or otherwise, are declared by the author(s). REFERENCES SINGLE TRIAL DETECTION OF HUMAN VESTIBULAR EVOKED MYOGENIC POTENTIALS 1. Brantberg K, Bergenics J, Tribukait A. Vestibular-evoked myogenic potentials in patients with dehiscence of the superior semicircular canal. Acta Otolaryngol 119: 633 640, 1999. 2. Chihara Y, Iwasaki S, Ushio M, Murofushi T. Vestibular-evoked extraocular potentials by air-conducted sound: Another clinical test for vestibular function. Clin Neurophysiol 118: 2745 2751, 2007. 3. Colebatch JG. Properties of rectified averaging of an evoked-type signal: theory and application to the vestibular-evoked myogenic potential. Exp Brain Res 199: 167 176, 2009. 4. Colebatch JG, Halmagyi GM, Skuse NF. Myogenic potentials generated by a click-evoked vestibulocollic reflex. J Neurol Neurosurg Psychiatry 57: 190 197, 1994. 5. Curthoys IS, Kim J, McPhedran SK, Camp AJ. Bone conducted vibration selectively activates irregular primary otolithic vestibular neurons in the guinea pig. Exp Brain Res 175: 256 67, 2006. 6. Didier A, Cazals Y. Acoustic responses recorded from the saccular bundle on the eighth nerve of the guinea pig. Hear Res 37: 123 128, 1989. 7. Isu N, Graf W, Sato H, Kushiro K, Zakir M, Imagawa M, Uchino Y. Sacculo-ocular reflex connectivity in cats. Exp Brain Res 131: 262 268, 2000. 8. McCue MP, Guinan JJ. Acoustically responsive fibers in the vestibular nerve of the cat. J Neurosci 14: 6058 6070, 1994. 9. Minor LB. Clinical manifestations of superior semicircular canal dehiscence. Laryngoscope 115: 1717 1727, 2005. 10. Minor LB, McCrea RA, Goldberg JM. Dual projections of secondary vestibular axons in the medial longitudinal fasciculus to extraocular motor nuclei and the spinal cord of the squirrel monkey. Exp Brain Res 83: 9 21, 1990. 11. Murofushi T, Curthoys IS. Physiological and anatomical study of clicksensitive primary vestibular afferents in the guinea pig. Acta Otolaryngol 117: 66 72, 1997. 12. Rosengren SM, Aw ST, Halmagyi GM, Todd NPM, Colebatch JG. Ocular vestibular evoked myogenic potentials (OVEMPs) in superior canal dehiscence. J Neurol Neurosurg Psychiatry 79: 559 568, 2008. 13. Rosengren SM, Todd NPM, Colebatch JG. Vestibular evoked myogenic potentials evoked by brief interaural head acceleration: properties and possible origin. J Appl Physiol 107: 841 852, 2009. 14. Rosenhall U. Vestibular macular mapping in man. Ann Otol Rhinol Laryngol 81: 339 351, 1972. 15. Sheahan NF, Coakley D, Malone JF. Neuro-muscular model of ocular microtremor. Eng Med Biol Soc Proc Ann Int Conf IEEE 4: 1619 1620, 1992. 16. Sheykholeslami K, Murofushi T, Kermany MH, Kaga K. Bone-conducted evoked myogenic potentials from the sternocleidomastoid muscle. Acta Otolaryngol 120: 731 734, 2000. 17. Todd NPM, Cody FWJ, Banks JR. A saccular origin of frequency tuning in myogenic vestibular evoked potentials? Implications for human responses to loud sounds. Hear Res 141: 180 188, 2000. 18. Todd NPM, Rosengren SM, Aw ST, Colebatch JG. Ocular vestibular evoked myogenic potentials (OVEMPs) produced by air- and boneconducted sound. Clin Neurophysiol 118: 381 390, 2007. 19. Todd NPM, Rosengren SM, Colebatch JG. Ocular vestibular evoked myogenic potentials (OVEMPs) produced by impulsive transmastoid acceleration. Clin Neurophysiol 119: 1638 1651, 2008a. 20. Todd NPM, Rosengren SM, Colebatch JG. Tuning and sensitivity of the human vestibular system to low-frequency vibration. Neurosci Lett 444: 36 41, 2008b. 21. Todd NPM, Rosengren SM, Colebatch JG. A utricular origin of frequency tuning to low-frequency vibration in the human vestibular system? Neurosci Lett 451: 175 180, 2009. 22. Todd NPM, Rosengren SM, Govender S, Colebatch JG. Low frequency tuning in the human vestibular-ocular system is determined by both peripheral and central mechanisms. Neurosci Lett 458: 43 47, 2009. 23. Uchino Y. Connections between otolith receptors and neck motoneurons. Acta Otolaryngol Suppl 528: 49 51, 1997. 24. Uchino Y, Sasaki M, Sato H, Imagawa M, Suwa H, Isu N. Utriculoocular reflex arc of the cat. J Neurophysiol 76: 1896 1903, 1996. 25. Watson SRD, Halmagyi GM, Colebatch JG. Vestibular hypersensitivity to sound (Tullio phenomenon): structural and functional assessment. Neurology 54: 722 728, 2000. 26. Welgampola MS, Rosengren SM, Halmagyi GM, Colebatch JG. Vestibular activation by bone-conducted sound. J Neurol Neurosurg Psychiatry 74: 771 778, 2003. 27. Young ED, Fernandez C, Goldberg JM. Responses of squirrel monkey vestibular neurons to audio-frequency sound and head vibration. Acta Otolaryngol 84: 352 360, 1977. 59