Cochlear implant artifact attenuation in late auditory evoked potentials: A single channel approach

Highlights

• Single channel approach to cochlear implant artifact attenuation in cortical evoked potentials.

• CI related artifact contains a high frequency component caused by the stimulation pulses and a DC or pedestal artifact.

• DC artifact shows a non-linear time varying relationship with the stimulation pulse amplitude.

• The relationship between DC artifact and pulse amplitude is well described by a bivariate polynomial.

Abstract

Recent evidence suggests that late auditory evoked potentials (LAEP) provide a useful objective metric of performance in cochlear implant (CI) subjects. However, the CI produces a large electrical artifact that contaminates LAEP recordings and confounds their interpretation. Independent component analysis (ICA) has been used in combination with multi-channel recordings to effectively remove the artifact. The applicability of the ICA approach is limited when only single channel data are needed or available, as is often the case in both clinical and research settings. Here we developed a single-channel, high sample rate (125 kHz), and high bandwidth (0–100 kHz) acquisition system to reduce the CI stimulation artifact. We identified two different artifacts in the recording: 1) a high frequency artifact reflecting the stimulation pulse rate, and 2) a direct current (DC, or pedestal) artifact that showed a non-linear time varying relationship to pulse amplitude. This relationship was well described by a bivariate polynomial. The high frequency artifact was completely attenuated by a 35 Hz low-pass filter for all subjects (n = 22). The DC artifact could be caused by an impedance mismatch. For 27% of subjects tested, no DC artifact was observed when electrode impedances were balanced to within 1 kΩ. For the remaining 73% of subjects, the pulse amplitude was used to estimate and then attenuate the DC artifact. Where measurements of pulse amplitude were not available (as with standard low sample rate systems), the DC artifact could be estimated from the stimulus envelope. The present artifact removal approach allows accurate measurement of LAEPs from CI subjects from single channel recordings, increasing their feasibility and utility as an accessible objective measure of CI function.

Introduction

Advances in cochlear implant (CI) technology now mean that a typical recipient of a modern CI can expect to understand speech in a quiet listening environment (for a review see Zeng et al., 2008). In spite of these advances there remains a large amount of variability in performance across users. Behavioral methods such as speech perception tests or non-speech based listening tests (Fu, 2002; Henry and Turner, 2003; Henry et al., 2005; Won et al., 2007) can be used to quantify this variability. However, behavioral methods are often not suitable for pediatric CI users and speech-based tests may not be the best way to assess the performance of new CI recipients while they are still learning to understand speech heard through their implants. Neural based objective metrics of performance may provide a useful alternative to behavioral testing for both these user groups. In addition to potentially improving the standard of treatment received by an individual CI user, the development of neural objective metrics of CI performance may also advance our understanding of the origins of the performance variability, by giving information on the underlying neural mechanisms. However, the development of such neural metrics has been hampered by the large CI related electrical artifact, which contaminates evoked potential recordings in these subjects.

Firszt et al. (2002) found that cortical evoked potentials may be useful for predicting speech perception outcomes for CI. However, to minimize the artifact, this study used very short simple stimuli which are unable to fully probe the complex processing that takes place in the auditory system. Gilley et al. (2006) proposed a method for attenuating the artifact caused by longer duration stimuli. They showed how independent component analysis (ICA) could be used to recover late auditory evoked potentials (LAEP) from multi-channel data. Utilizing the multi-channel ICA approach, two recent studies by Zhang et al. (2010, 2011) showed how LAEPs obtained using a mismatch negativity paradigm can provide useful information on CI functionality and that this information can be related to behavioral outcomes such as speech perception. One drawback of the ICA approach is that multi-channel data must be acquired, even when, as with the two studies by Zhang et al., most of the results and conclusions are based on artifact-free single-channel data. Having to acquire multi-channel data necessitates the purchase of expensive multi-channel acquisition systems, increases subject preparation time, as a full EEG cap must be attached and, for CI subjects, adds to the difficulty of positioning the EEG cap over the behind-the-ear processor and magnetic link. For most clinical applications and many research questions, single-channel data are sufficient and subject preparation time much shorter. These practical considerations limit the applicability of the ICA-based artifact attenuation approach and led us to develop a single-channel based artifact attenuation approach.

To better understand the origin of the CI related artifact in LAEPs we developed a high-sample-rate, high-bandwidth, single-channel acquisition system with a temporal resolution high enough to clearly resolve each stimulation pulse. Here, we used this acquisition system to show that LAEPs recorded from CI subjects are generally composed of three components: a neural response component and two artifact components. Based on this signal composition, we proposed a three-stage artifact attenuation strategy (Fig. 1). The high frequency artifact (HFA) was found to be a direct representation of the stimulation pulses and was completely attenuated by a low-pass filter (stage 1). The low frequency or DC artifact (DCA), often referred to as a ‘pedestal’ artifact, could be accentuated by an electrode impedance mismatch and in some subjects could be attenuated by balancing the impedance of the recording electrodes (stage 2). Based on the assumption that the DCA was caused by the stimulation pulses, we developed a mathematical framework to obtain an estimate of the DCA and remove it from the LAEP (stage 3). Finally, we demonstrated how this single-channel approach could be also be applied with low sample rate data (commercial systems) and that it could be used to measure N1–P1 amplitude growth functions for CI users.