6th Annual International IEEE EMBS Conference on Neural Engineering San Diego, California, 6 - 8 November, 2013

Investigating OCULAR MOVEMENTS AND VE S T I B U L A R E V O K E D P O T E N T I A L S for a Vestibular Neuroprosthesis: Response to Pulse TRAINS AND

BASELINE STIMULATION

T. A. Khoa Nguyen, Student Member, IEEE, Wangsong Gong, Wigand Poppendieck, Jack DiGiovanna and Silvestro Micera, Senior Member, IEEE

ABSTRACT—No adequate treatment currently exists for bilateral vestibulopathy, which can result in significant decreases of social and physical functioning. To improve patients’ quality of life, vestibular neuroprostheses are being developed. Efficacy of current prototypes is evaluated by recording reflexive eye movements (vestibular ocular reflex, VOR). Vestibular Evoked Potentials (VEPs) provide real-time feedback about peripheral efficacy that could be used to adapt a closed-loop neuroprosthesis to improve performance (e.g., eye movement magnitude and direction). A key building block is the prediction of VOR with VEP. In earlier work, we correlated both in response to single stimulation pulses. While impulse responses are interesting, they do not reflect a typical operating mode. To learn more about VEP at expected modulations, we studied the impact of pulse trains and baseline stimulation on VEP here. At 250 pulses per second, VEP did neither change significantly for pulse trains nor over the course of 30-minute baseline stimulation. VOR, on the other hand, changed with the number of pulses, and was also influenced by baseline stimulation.

I . INTRODUCTION

Bilateral vestibular loss can be an incapacitating condition: Patients suffer from a significant deterioration of gaze stability, postural control and spatial orientation. For instance, oscillopsia caused by vestibular disease can lead to mild blurring or rapid jumping of objects in the visual field. Rehabilitation and medication are often insufficient treatment options.

Over the last 15-20 years, a number of research groups have been developing vestibular implants to restore functionality in such severe cases. Devices were successfully evaluated in various animal models (for a literature review, see [1]). More recently, a small number of human subjects

Manuscript received June 5, 2013. This work was supported by the European Commission within the 7th Framework Program (FET Open Project, project no. 225929).

T .A .K.N, J . D G and S .M are with the Translational Neural Engineering Lab, Center for Neuroprosthetics, B M 3115 - Station 17, Interfaculty Institute of Bioengineering, E P F Lausanne, 1015 Lausanne, Switzerland ([thuyanhkhoa.nguyen], [jack.digiovanna], [silvestro.micera]@epfl.ch). T .A.K.N is also with the Neuroprosthesis Control Group, Automatic Control Lab, E T H Zurich, Physikstrasse 3, 8092 Zurich, Switzerland. S .M is also with the BioRobotics Institute, Scuola Superiore Sant'Anna, P.za Martiri della Liberta', 33 - 56127 Pisa, Italy.

W . G is with the Jenks Vestibular Physiology Lab, Mass. Eye and Ear Infirmary, Room 421, 243 Charles Street, Boston, M A 02114 U S A , and also with the Department of Otology and Laryngology, Harvard Medical School, Boston M A 02114 U S A (wangsong_gong@meei.harvard.edu).

W . P is with Fraunhofer Institute for Biomedical Engineering, Department Medical Engineering and Neuroprosthetics, Ensheimer Strasse 48, 66386 St. Ingbert, Germany (wigand.poppendieck@ibmt.fraunhofer.de).

were instrumented with a hybrid cochlear-vestibular implant, proving its clinical viability [2].

Current prototypes use a motion sensor fixed to the subject to detect head rotation. A controller then processes that information and applies motion-modulated, pulsatile stimulation through electrodes implanted in the inner ear. The modulation gain is open-loop, i.e. not automatically adapted during operation. Finally, stimulation efficacy is assessed post-hoc by recording evoked eye movements called vestibular-ocular reflexes (VORs).

Further advances could be achieved by introducing a cascaded closed-loop control with vestibular evoked potentials (VEPs) as feedback signal. Stimulation efficacy could be assessed in real-time and modulation gains adjusted accordingly. We reported preliminary characterization results of V E P in response to single pulses and found a correlation with V O R [3]. These studies provided first insights on V E P and refined the stimulation and V E P recording methods.

To accomplish a closed-loop vestibular neuroprosthesis with V E P , additional characterization is required. In one experiment we recorded V E P induced by short bursts (pulse trains). In another experiment, V E P was measured in response to continuous baseline stimulation.

Short bursts and baseline stimulation are more representative of a typical scenario for a vestibular prosthesis. With a unilateral prosthesis, subjects are adapted to a (superphysiological) baseline to facilitate modulation for both inhibitory and excitatory responses [2]. Bursts have been applied to evoke stronger eye movement than with a single pulse alone. Results of these studies will help identify the V E P - V O R correlation, which is essential in the design of the closed-loop controller.

I I . MATERIALS AND METHODS

A. Animal preparation The institutional animal care and use committee approved

all experiments. A male guinea pig was prepared with three surgeries [3]. First, a container for connectors, wires and optional stimulation circuitry (‘headcap’) as well as a fiberglass-composite structure (‘headbolt’) were fitted to the animal’s skull. Second, a 3-turn stainless search eye coil was inserted into the left eye. Thirdly, a double-sided, ‘sandwich’ electrode array with eight stimulation sites [4] was implanted in the left lateral canal. During surgery, the array’s position was checked with a portable stimulator and monitoring eye movement. A wire electrode was inserted into the neck muscle as remote return electrode.

978-1-4673-1969-0/13/$31.00 ©2013 IEEE 855

L VEP

ms _ i

1 [ n 5x'

10x, 20x

I L

VEP

VEP

VOR

250 pps baseline

VEP I I I I I I I I I

0 mins Recordings every 3 mins

Figure 1. (A) Stimulation and recording for pulse trains. VEP was recorded after pulse trains of 1, 2, 5, 10, or 20 pulses at 250 pulses per second (pps). Not concurrently, VOR was recorded with the onset of the train. Phase widths were 25 us, phase gap 2.1 us, and current amplitude 230 uA. (B) VEP was recorded to 30 minutes baseline stimulation in 3-minute intervals (vertical black lines). Baseline pulse rate was 250 pps and baseline current amplitude 160 uA, phase widths 25 us. Experiments were performed on different days. (C) Example of a Vestibular Evoked Potential VEP after artifact removal through masker probe recording. VEP should feature N and P waves. Figure is not drawn to scale.

B. General experimental setup A MED-EL Research Interface Box RIBII (Innsbruck,

Austria) was used for stimulation and recording. It was connected to the implanted electrode array through a transmission coil and a PULSAR cochlear implant. Scripts with stimulation sequences and recording parameters were programmed in Matlab (MathWorks, Natick, MA, USA). After inspecting all sites on the electrode array, one site and the remote return were chosen for stimulation (typically the one with the strongest VOR response). For VEP measurements, two other sites on the array had to be used. The two sites, whose recordings in preliminary tests exhibited both negative and positive waves (Fig. 1C), were selected.

VEP was recorded in a 1.7 ms window starting 110 (j.s after pulse onset due to hardware blanking. The sampling rate was 1.2 MHz. Stimulation artifact was reduced with the masker-probe technique [5]. For our studies, masker and probe were 300 u.s apart. The masker amplitude was 20 uA larger than the probe amplitude and VEP responses were averaged across 30 iterations in Matlab.

For VOR measurements of pulse trains, the implanted search coil signal was low-pass filtered using a first-order filter with a cut-off frequency of 3000 Hz. This signal was sampled using a National Instruments DAQ card (Austin, TX, USA) with a sampling rate of 300 kHz. A 4th order low-pass, 500 Hz cut-off frequency Butterworth filter was applied in software. Eye movement velocity was differentiated from position using the Matlab 'filter' command. VOR responses were averaged across 60 iterations in Matlab.

C. Specific stimulation and recording paradigms Fig. 1 illustrates the two experiments. VOR and VEP

were both measured for the pulse train experiment. For the baseline stimulation only VEP was measured.

1) Pulse trains During this experiment, the animal was head-fixed and

restrained in a custom-chair. Recordings were performed with the subject alert in a small dark room. Prior to the application of the pulse trains, the subject was not adapted to any baseline stimulation.

VEP and VOR were not recorded concurrently, as the VEP signal would have been tainted by the magnetic search coil system used for VOR recording. First, VOR was recorded to pulse trains of 1, 2, 5, 10 or 20 pulses at a pulse rate of 250 pulses per second (pps). These trains consisted only of probe pulses with a constant amplitude of 230 uA. The phase width of the cathodic-first, symmetric pulses was 25 us, and the interphase gap was 2.1 |j,s. Afterwards the coil system was turned off.

Second, VEP was recorded to the same pulse trains. The set of pulse trains was repeated three times to record responses to probe, masker-probe and masker pulses.

2) Baseline stimulation The second experiment was performed on a different day,

and the animal was freely moving in a small cage. VOR was not recorded and the search coil system was off. Baseline stimulation was turned on for four 30-minute periods. Pulse rate was 250 pps, current amplitude 160 uA. Phase widths were again 25 |as, and interphase gap 2.1 |as.

VEP was recorded in intervals of three minutes. A complete recording cycle of 30 iterations for probe, masker-probe and masker took 765 ms, i.e. no baseline stimulation was applied during that period.

III. RESULTS

A. VOR and VEP to pulse trains Electrical stimulation of the subject's left lateral canal

evoked mainly horizontal eye movement (as physiologically expected). The ratio of vertical to horizontal eye movement was approximately 1 to 3. For the remainder of this report we decided to consider only horizontal eye movement.

Eye movement responses to pulse trains are depicted in Fig. 2. Peak positions were 0.01, 0.03, 0.07, 0.13, 0.16° for 1, 2, 5, 10, and 20 pulses, respectively. That is, peak eye position increased with a gain less than 1 when compared to the 2-pulse response. Eye movement peaks occurred approx. 8 ms after the end of the pulse train (e.g., at 28.6 ms for 5 pulses; train length was 20 ms). Afterwards, eye position returned to zero for 1, 2, and 5 pulses, but did not completely recover for 10 and 20 pulses.

Peak eye velocity also increased with less than unity gain up to 10 pulses (4.57s). Eye velocities peaked at 10.5 and 12.5 ms for 1 and 2 pulses, respectively, i.e. after the end of the pulse train. For the other cases, velocity peaked before the end of the train (e.g., at 15 ms for 10 pulses). After the peak, velocity returned to zero with a small negative overshoot for 1 and 2 pulses. For 5 pulses, velocity decreased rapidly with the end of the train and had a negative peak of-2.97s before reaching zero. For 10 and 20 pulses, velocity declined slowly first and then rapidly with the end of the train (inflection points at 40.8 and 81.8 ms, respectively). The negative overshoot was more pronounced at about -3.57s.

856

0.15

0.05

0

I I

! /f\ i /X i \

^A^L-

2 5 10 20

1*1 I - - -

2-KS-test, p > 0.47

& 1000

Figure 2. VOR and VEP responses to pulse trains. (A) Averaged horizontal eye position increased with number of pulses. Positive direction was leftward. The increase had a gain less than 1 when compared to the 2-pulse response. Positions peaked about 8 ms after the pulse trains ended (denoted by colored dashed lines). Position recovered completely for 1-5 pulses within 200 ms, but a residue remained for 10 and 20 pulses. (B) Averaged horizontal eye velocity also increased with pulse number, but only up to 10 pulses. Peaks occurred after the end of the pulse train for 1 and 2 pulses, but considerably before for 5-20 pulses. For 10 and 20 pulses, velocity decreased slowly after the peak, then plummeted with the end of the pulse train. (C) N-P voltage was highest for 1 pulse, and then stabilized from 5 pulses at 825 uV (2-sample KS-test, p > 0.47). (D) Also the latencies for the N and P waves were stable. The N wave had a latency of ca. 150 us, the P wave of ca. 250 us for all pulses. (Error bars smaller than markers).

N wave - P wave

Time [mins] Time [mins] Figure 3. VEP characteristics to 30-minute 250 pps baseline

stimulation. (A) N-P voltage had an increasing trend until the 24th

minute, but no significant difference (2-sample KS-test). The measurement at minute 27, marked with an asterisk, was affected by saturation and was excluded. (B) N wave latency remained constant during the 30-minute test (122.5 us). P wave latency varied between 220 and 240 us. Values for the 27th minute were excluded (red asterisk).

VEP characteristics are also shown in Fig. 2. N-P voltage was highest after 1 pulse (995 uV). It then decreased to 825 uV for 5 pulses and remained relatively unchanged for more pulses. The differences in N-P voltage were not significant (2-sample Kolmogorov-Smirnov test, p > 0.47).

Latencies for the negative wave were constant at 150 (j.s after stimulation onset. Also the latencies for the positive wave remained stable at ca. 250 |j,s.

B. VEP to baseline stimulation Only VEP was measured for this experiment, as the

animal was freely moving. In Fig. 3, the N-P voltage showed an increasing trend with time, but the differences were insignificant (2-sample KS-test, p > 0.05). The recording at the 27* minute was marred severely by saturation in the recording and was excluded as an outlier.

Latencies for N and P waves did not change significantly over 30 minutes. P wave latency oscillated between 220 and 240 |j,s. N wave latency was constant at 122.5 |j,s.

IV. DISCUSSION

We are studying closed-loop feedback with VEP for a vestibular neuroprosthesis. VEP needs to be characterized to serve as a real-time metric of stimulation efficacy and eventually help tune stimulation parameters on-line to improve response magnitude and reduce erroneous eye movement directions induced by current spread (e.g., vertical eye movement when lateral canal is stimulated). The key element would be the use of VEP to predict VOR. In earlier work, we reported preliminary VEP characterization to single stimulation pulses [3]. These results delivered initial information on VEP and related it to evoked eye movement. In this report, we extended VEP characterization to include responses to pulse trains and baseline stimulation in one guinea pig.

A. Pulse trains Three points are highlighted here: first, the non-

proportional increase in peak eye position and velocity for 1-10 pulses, second, the eye movement response to 20 pulses, and third the VEP characteristics.

First, peak eye positions for 5-20 pulses were less than corresponding multiples of the 2-pulse response. This points to a non-linear effect, such as saturation, and was observable in the velocity signal. The maximum value for both 5 and 10 pulses was 4.5°/s at 15.2 ms after train onset. This indicates that the number of recruited ocular motor neurons and their innervated muscle fibers saturated and thus could not exert more force or acceleration. The inertia of eye was negligible during these small saccades [6]. Following the peak for 10 pulses, velocity decreased slowly as some muscle fibers recovered, but could not exert maximum force, because of the preceding twitch. Twitch time in monkeys is typically 6-8 ms with a more gradual decline [6]. As soon as the train ceased, the velocity dropped sharply.

Second, the response to 20 pulses was remarkably different. If saturation had been the reason, then we would have expected similar responses for 10 and 20 pulses. But the eye position for 20 pulses had a lower rise than for 10, and peak velocity was also lower for 20 than 10 pulses. One explanation might be insufficient recovery time for the eye position signal and eye muscle fibers after 20 pulses. The train was 80 ms long, and the next burst started at 500 ms. Extrapolating the 20-pulse position signal gave a zero crossing at 600 ms. This could also explain the smaller-than-expected difference between peak positions (0.13° vs. 0.16° for 10 and 20 pulses, respectively). Thus, bursts with 20 pulses and more at this pulse rate should not be considered for their non-linear response.

Third, we knew that VEP to single pulses recovered within 1 ms [3]. Similarly, vestibular evoked compound action potentials in rhesus monkey returned to baseline within that time [7]. Here, the pulses were applied every 4 ms, which gave sufficient recovery time. This was reflected in the stable N-P voltage. Although differences were insignificant, the downward trend from 1 to 5 pulses could be

857

viewed as a transition from an impulse response to a steady state. The stable latencies for N and P waves further indicate that neither the shape of V E P changed with pulse number.

B. Baseline stimulation V O R was not recorded in this experiment, as the animal

was tested unrestrained for two hours. Recordings were done in four 30-minute blocs, and the implant-to-interface connection was checked in-between. The blocs were not concatenated, since the checks took a couple of minutes during which stimulation was off.

Although no eye movements were measured here, it has been shown that they attenuate with a time constant of 5.0 minutes [8]. In that study, guinea pigs were also subjected to 250 pps baseline stimulation. Initial peak-to-peak eye movement was 0.01° with an average peak velocity of 8.1°/s. Within 30 minutes that reduced to 0.002° and 1.6°/s. A tiny nystagmus was observed at stimulation onset in our test. We can assume smaller initial values since our pulses delivered less charge. Phase widths here were considerably shorter at 25 ! s compared to 200 ! s , current amplitudes were 230 ! A vs. 60-125 ! A . Thus, one phase in our experiment injected 5.8 nC compared to 12 or 25 nC in the other study.

Following the same line of reasoning as above—that V E P returns to baseline within 1 ms—evoked potentials did not show significant changes over a period of 30 minutes in terms of magnitude or wave latencies. The recording time of 765 ms every three minutes should have had no major impact on the nerve state (the recording delivered 90 pulses in the same period that 191 baseline pulses would be expected). This is roughly equivalent to a small rotation with a duration of 765 ms in the inhibitory direction every three minutes. Total recording time constituted 8.4 s or less than 0.5% of stimulation time.

A significant effect was found when V E P measurements were saturated. The exact source of saturation was unknown and could have been due to strong animal head movement. A vestibular implant could switch into a fail-safe mode in such cases (e.g., turn off feedback control).

It is believed that the mechanism of attenuation is habituation rather than adaptation [8]. No after-effect was seen in this animal when stimulation was turned off. Our findings also add further evidence that the site of habituation is likely to be the central processing circuitry of V O R and not peripheral [9, 10], as V E P was not changing over time.

C. Impact on closed-loop control design V E P characteristics did not change significantly in the

two experiments. In contrast, eye movement showed alterations for pulse trains and also for baseline stimulation based on an earlier study [8]. Therefore, V O R prediction must also consider these two known parameters.

V . CONCLUSION

We are investigating V E P as feedback signal. Predicting V O R from V E P is a key element and also requires the incorporation of system parameters. Further characterization of V E P at baseline stimulation levels has to be carried out to eventually test a closed-loop prosthesis that promises better

performance than current open-loop prototypes.

ACKNOWLEDGMENT

We thank D . M . Merfeld from the Jenks Vestibular Physiology Lab at the Mass. Eye and Ear Infirmary, Boston, U S A , for contributing his comments on the experiments and manuscript. We are also very grateful to C . Harburcakova from the same laboratory for all her efforts with animal care.

REFERENCES

[1] D . M . Merfeld and R . F . Lewis, "Replacing semicircular canal function with a vestibular implant," Current Opinion in Otolaryngology & Head and Neck Surgery, vol. 20, pp. 386-392 10.1097/MOO.0b013e328357630f, 2012.

[2] J . -P . Guyot, A . Sigrist, M . Pelizzone, and M . I . Kos, "Adaptation to Steady-State Electrical Stimulation of the Vestibular System in Humans," Annals of Otology Rhinology and Laryngology, vol. 120, pp. 143-149, March 2011.

[3] T .A . Khoa Nguyen, W . Gong, J. DiGiovanna, W . Poppendieck, and S . Micera, "Investigating Vestibular Evoked Potentials as Feedback Signal in a Vestibular Neuroprosthesis: Relation to Eye Movement Velocity," in Converging Clinical and Engineering Research on Neurorehabilitation. vol. 1, J. L . Pons, D . Torricelli, and M . Pajaro, Eds., ed: Springer Berlin Heidelberg, 2013, pp. 1325-1329.

[4] W . Poppendieck, A . Sossalla, M . - O . Krob, C . Welsch, T .A .K. Nguyen, W . Gong, J. DiGiovanna, D . Merfeld, S . Micera, and K. -P . Hoffmann, "Development, Manufacturing and Application of Double-Sided Flexible Implantable Microelectrodes," Transactions on Neural Systems and Rehabilitation Engineering, vol. (submitted), 2013.

[5] A . Bahmer, O . Peter, and U . Baumann, "Recording and analysis of electrically evoked compound action potentials (ECAPs) with M E D -E L cochlear implants and different artifact reduction strategies in Matlab," Journal of Neuroscience Methods, vol. 191, pp. 66-74, August 2010.

[6] A . F . Fuchs and E . S . Luschei, "Development of isometric tension in simian extraocular muscle," The Journal of Physiology, vol. 219, pp. 155-166, December 1, 1971 1971.

[7] J .O . Phillips, S.J. Shepherd, A . L . Nowack, L . Ling, S . M . Bierer, C . R . S . Kaneko, C . M . T . Phillips, K . Nie, and J .T . Rubinstein, "Longitudinal performance of a vestibular prosthesis as assessed by electrically evoked compound action potential recording," in Engineering in Medicine and Biology Society (EMBC), 2012 Annual International Conference of the IEEE, 2012, pp. 6128-6131.

[8] M . A . Saginaw, G . Wangsong, C . Haburcakova, and D . M . Merfeld, "Attenuation of Eye Movements Evoked by a Vestibular Implant at the Frequency of the Baseline Pulse Rate," Biomedical Engineering, IEEE Transactions on, vol. 58, pp. 2732-2739, 2011.

[9] J .H. Courjon, W . Precht, and D . W . Sirkin, "Vestibular nerve and nuclei unit responses and eye movement responses to repetitive galvanic stimulation of the labyrinth in the rat," Experimental Brain Research, vol. 66, pp. 41-48, 1987/03/01 1987.

[10] S .G.T . Balter, R.J . Stokroos, R . M . A . Eterman, S . A . B . Paredis, J . Orbons, and H . Kingma, "Habituation to galvanic vestibular stimulation," Acta Oto-laryngologica, vol. 124, pp. 941-945, 2004.

858