Artifacts in Evoked Potential Recordings
Alan D. Legatt
Evoked potentials (EPs) are the electrical signals produced by the nervous system in response to external stimulation. The term was initially restricted to sensory EPs, elicited most often by visual, auditory, or somatosensory stimulation (1), but has more recently been expanded (2) to include motor evoked potentials (MEPs), which are signals generated in neural tissue (such as the spinal cord) or in muscles following electrical or magnetic stimulation of the descending motor tracts in the brain (3).
The recording electrodes placed on the subject pick up a variety of other signals in addition to the desired EP. These other signals, which may be labeled “artifacts,” are the subjects of this chapter. Some of them are of physiologic origin, such as ECG, EEG, the electrooculogram (EOG), and electromyographic (EMG) activity from muscles near the recording electrodes. Others, such as the electrical stimulus artifact, are related to the stimuli used to elicit the EPs. Some degree of pickup of the power line frequency is almost inevitable. And during neurophysiologic intraoperative monitoring (NIOM) of EPs, there are many other possible sources of artifact from other equipment in the operating room. Some of these artifacts are orders of magnitude larger in amplitude than the EPs (4).
A voltage cannot be measured in isolation; all EP recordings measure voltage differences between pairs of electrodes. Some signals, such as line frequency artifact, are widely distributed over the patient’s body and therefore are picked up at approximately the same amplitude by all of the recording electrodes. Ideally, the differential amplifiers in the recording equipment will amplify the voltage difference between the two amplifiers (the differential signal), whereas the signals that are present at both electrodes (the common-mode signal) will not appear in the amplifier output (a process called in-phase cancellation). The common-mode rejection ratio (CMRR) measures how well the amplifiers in the recording equipment approach that ideal; it is the ratio between the amplifier gain for the differential signal and that for the common-mode signal. EP guidelines call for a CMRR of at least 80 dB or 10,000:1 (5).
In some cases, EPs are large enough to be clearly visible following a single stimulus, such as myogenic MEPs (recorded from muscles) (Figure 12.1). The relative sizes of visual evoked potentials (VEPs) to stroboscopic flash stimulation and the background EEG upon which they are superimposed determines whether the VEPs are visible in the raw data; when they are, they are labeled a “photic driving response” (Figure 12.2). In contrast, brainstem auditory evoked potentials (BAEPs) are a fraction of a microvolt in amplitude, and are never visible in scalp-recorded data. Signal averaging is used to extract them (Figure 12.3), as well as VEPs, somatosensory evoked potentials (SEPs), and D-waves (neurogenic MEPs recorded from the spinal cord) from the other electrical signals picked up by the recording electrodes. Each epoch or sweep of data can be regarded as the sum of the EP and of all the other unrelated electrical signals, which for the purposes of EP recording, can be regarded as “noise.” In most cases, the EP is identical from trial to trial, since the same stimulus is being given to the same nervous system, and is therefore unaltered by the averaging process. The noise, in contrast, is random. Thus, at a given poststimulus latency, the noise polarity will sometimes be positive and sometimes negative, and will tend to cancel when 330the epochs are summated. Dividing the summed epochs by the number of epochs included in the average (N) further decreases the amplitude of the residual noise. Quantitatively, the amplitude of the residual noise in the average will approximate the typical noise amplitude in a single epoch divided by the square root of N. The signal-to-noise ratio is therefore multiplied by the square root of N (4). Even if VEPs are visible in the raw data (Figure 12.2), signal averaging is used to attenuate the superimposed EEG signal and therefore make the representation of the VEP waveform more accurate.
The number of epochs per average is a compromise. Increasing the number of epochs gives “cleaner” signals with a better signal-to-noise ratio but takes longer. During extraoperative diagnostic EP studies, this will require subject cooperation for a longer period of time; during NIOM, it may cause increased delay before the surgeons are notified about significant EP changes. Measures taken to decrease the amount of noise in the raw data, such as subject relaxation to decrease EMG noise from muscle tension in head and neck muscles during a clinical BAEP study, are useful. Epochs containing an unusually large amount of noise are identified by automatic artifact rejection and are excluded from the average.
Signal averaging is not an intelligent process; it selectively attenuates signals that do not correlate with stimulus delivery, while retaining those that are time-locked with stimulus delivery, such as the EPs. If the repetition rate of the stimulus is a submultiple of the line frequency (e.g., 2, 3, 5, 6, 10, or 12 per second in North America where the line frequency is 60 Hz), then the inevitable line frequency artifact will be phase-locked with stimulus delivery and will be identical in each sweep; therefore, it will be preserved in the average just as the EP is. Timing circuits can drift, causing the stimulus repetition frequency to become a subharmonic of the line frequency. Therefore, if line frequency artifact becomes troublesome (e.g., during NIOM) and the amount of line frequency artifact in the raw data does not appear to have increased, a small change in the stimulus repetition rate may be useful (6). The same principle applies to higher frequency sinusoidal artifact, to which the stimulus repetition rate may have become harmonically entrained (Figure 12.4). Whenever a sinusoidal artifact appears in EP data, the presence of line frequency or higher harmonic artifact should be suspected. The frequency of the artifact can be determined by measuring its period or peak-to-peak latency in milliseconds, and dividing this into 1,000 (6).
The one type of EP for which signal averaging is not used is myogenic MEPs because they often differ markedly from trial to trial (Figure 12.5), most likely because a different subset of motor units is recruited each time. Fortunately, myogenic MEPs are typically orders of magnitude larger than sensory EPs, with a signal-to-noise ratio that permits them to be easily identified in unaveraged recordings.
Physiologic Artifacts Arising From Signals Originating Within the Patient’s Body
Artifact from the ECG is most prominent at electrodes close to the heart, such as SEP recording electrodes at Erb’s point or on the posterior part of the neck. ECG artifact may cause some sweeps to be rejected, but usually this is not problematic unless the stimulus repetition rate and the cardiac rate are harmonically synchronized, so that a large percentage of the epochs contain prominent ECG artifact and are rejected. If it seems that too many sweeps are being rejected because of the ECG, a small alteration in the stimulus repetition rate may eliminate that harmonic relationship and lead to more epochs being included in the average.
EP recording electrodes on the head invariably pick up EEG activity, and signal averaging is used to reduce this. In patients with high-voltage EEGs, such as normal children, adolescents, and especially young children with an epileptic encephalopathy and abnormal high-voltage EEGs (e.g., hypsarrhythmia), the amount of EEG “noise” in the unaveraged EP raw data will be greater. Consequently, EP recordings may be noisier (Figure 12.6), and averaging a larger number of epochs may be necessary to produce sufficiently clean recordings. For typical SEP filter settings, slower frequencies of the EEG (e.g., theta and delta) will be removed by the filtering; however, it is the faster frequencies that contaminate the EP waveforms. High-voltage EEG activity in the alpha and beta frequency bands (Figure 12.6B), as well as the faster frequencies contained in EEG spikes, may have periods of tens of milliseconds, and thus may appear “slow” when viewed at the time window of the EP recording (Figure 12.6A).
The retina of the eye generates a corneoretinal potential, with the cornea electropositive with respect to the retina. The magnitude of this potential, which is in the millivolt range, is affected by ambient light levels. When the light level changes, it takes 10 to 15 minutes for the potential to change (7). Such slow (low-frequency) potential changes would be removed by band-pass filtering during standard EP recordings. However, when the eyes move, the equivalent dipole of the corneoretinal potential can shift rapidly, producing artifacts in the EOG that are commonly seen in the EEG. EP recording 331electrodes on the anterior part of the head can also pick up EOG signals, though these will mostly be removed from the recordings by filtering and artifact rejection.
EMG artifact from muscle tension in the head and neck muscles may cause prominent artifact during EP recordings, especially when recording SEPs and BAEPs where the EPs are small in amplitude and the high-frequency filter setting is high enough so that the EMG artifact is not removed by filtering. Dorsal scalp electrodes may be located over an aponeurosis rather than muscle, so recordings of cortical SEPs tend to be less contaminated by muscle artifact. Increased noise levels in subcortical SEP channels can give an indication of a light depth of anesthesia during NIOM (Figure 12.7). During clinical, extraoperative recordings of SEPs and BAEPs, muscle relaxation is important to obtain “clean” recordings. In the past, sedative medications such as diazepam were used in some EP laboratories to facilitate patient relaxation. But, like the use of chloral hydrate for EEG recordings, this has largely been abandoned due to the many safety constraints and personnel and monitoring requirements for conscious sedation of patients (8,9).
Implanted Electrical Stimulators
Implanted electrical stimulators such as cardiac pacemakers and deep brain stimulation (DBS) systems can introduce artifact into the EP raw data. Like ECG artifact, cardiac pacemaker artifact is infrequent and individual epochs containing it can be eliminated by automatic artifact rejection. DBS stimulation is more pervasive and produces large artifacts that may be recorded from the head (Figure 12.8B) which can cause rejection of all epochs with the usual artifact rejection settings. Even if the artifact rejection window is widened and automatic artifact rejection is turned off, the SEP waveforms will be markedly contaminated by artifact (Figure 12.8A).
During recording of SEPs to stimulation of peripheral nerves and of MEPs to transcranial electrical stimulation, large electrical stimuli that may be many volts in amplitude are applied directly to the patient’s body. While they may be substantially attenuated by distance, when they reach the recording electrodes they are still often as large as, or larger than, the EPs being recorded. These signals appear as electrical stimulus artifacts at the beginning of the data epoch (Figures 12.9A and 12.10). Inductive or capacitive coupling between the wires carrying the stimuli to the patient and the recording electrode lead wires can also contribute to the stimulus artifacts. During BAEP recordings, no electricity is applied to the patient’s body, but the transducer that converts the electrical square-wave pulse to the acoustic click typically contains a coil, and by inductive coupling can produce electrical signals in the patient recording transmitted by the lead wires to produce a stimulus artifact (Figures 12.11A and 12.12). The use of shielded headphones and headphones with piezoelectric transducers rather than voice coil transducers can reduce the stimulus artifact in the BAEP waveforms (10).
BAEP stimuli can be compression clicks, in which a wave of increased air pressure is delivered to the patient’s ear, or rarefaction clicks, in which the electrical signal delivered to the transducer is opposite in polarity and a wave of decreased air pressure is delivered. BAEPs to rarefaction and compression clicks may differ; therefore, a single click polarity is preferred for extraoperative diagnostic BAEP studies, unless alternating polarities are necessary for canceling the electrical stimulus artifact or the cochlear microphonic. During NIOM of BAEPs, where artifacts are often more problematic, alternating stimulus polarity can be used by default to (partially) cancel the electrical stimulus artifact. It will also tend to cancel the cochlear microphonic, which otherwise might hamper the identification of wave I (11).
Audiometric headphones are typically used for stimulus delivery during extraoperative diagnostic BAEP studies, but are not practical for NIOM. The electromechanical transducer that converts the electrical square pulses to clicks is housed in a small box that is located at some distance from the patient’s ear, and conducted to the ear via plastic tube that is several centimeters long. This helps mitigate stimulus artifact problems in two ways. The BAEPs, including wave I, are delayed by the time required for the acoustic stimulus to propagate through the plastic tube (typically 0.9–1.0 milliseconds). However, the electrical stimulus artifact remains simultaneous with transducer activation and the increased temporal separation between them permits greater decay of the stimulus artifact prior to the onset of wave I. The increased separation between the electrical wires/transducer and the patient’s head also serves to reduce the amplitude of the electrical stimulus artifact (11).
Since the stimulus artifact in an SEP waveform reflects the stimulation current that reaches the patient’s body, it can be used to evaluate whether the stimulus is still being delivered. If SEPs deteriorate due to neurologic dysfunction (Figure 12.9), the stimulus artifact will remain unchanged. In contrast, the stimulus artifact will be altered if technical problems are interfering with delivery of the stimulus to the patient (Figure 12.10). With constant current stimulators, the EP recording system typically indicates an open circuit or impedance limit condition when it cannot deliver the preset stimulating current intensity, such as when lead wires are disconnected or 332stimulating electrodes are dislodged. In contrast, the SEP alterations shown in Figure 12.10 were caused by a short circuit between the stimulating lead wires. In this case, the equipment was able to deliver the preset stimulating current intensity and therefore gave no error indication.
During BAEP recordings, an unchanged stimulus artifact only indicates an absence of problems with the recording equipment, the wire coil within the transducer, and the electrical connection between them. Technical problems with the conduction of the acoustic stimulus to the patient’s ear can still alter the BAEPs without affecting the stimulus artifact (Figure 12.12).
Line Frequency Artifact
Line frequency artifact is ubiquitous, derived from the electrical wiring, the lights, and the various pieces of equipment connected to the patient or in the patient’s vicinity. The scalp potentials induced by the electromagnetic field produced by the electrical wiring and room lighting can reach 100 mV. Fortunately, this is relatively uniform, and thus can be markedly reduced or eliminated by in-phase cancellation (4). Fluid warmers, used during surgery to avoid infusing cold intravenous fluids, can generate much larger amounts of 60-Hz artifact (Figure 12.13). The fluid warmer contains a coil of wire through which alternating current (AC) electricity is passed to generate heat. This acts as the primary coil of a transformer, and prominent 60-Hz artifact can be induced in nearby conductors, including the lead wires for the NIOM electrodes. Temporarily turning off the fluid warmer can identify it as the source of the artifact if it resolves. Rotating the fluid warmer or the pole to which it is attached can reduce the artifact by altering the electromagnetic coupling between the fluid warmer coil and the lead wires. Improperly grounded equipment connected to the patient can also produce line frequency artifact due to leakage currents. During NIOM, the operating table can be a prominent source of line frequency artifact; I have encountered situations in which this occurred because the ground lug of the operating table’s power cord, or the ground lug in the socket in the operating table to which the (female polarity) end of the power cord plugged in, was broken off.
Line frequency notch filters are often used during EEG recordings to eliminate line frequency artifact. Line frequency notch filters should not be used for SEP recordings, for two reasons (6): (a) spectral analysis demonstrates that the SEP waveform itself contains frequencies around 50 to 60 Hz, thus, use of the notch filter will substantially distort the SSEPs; (b) the notch filter may prolong the electrical artifact arising from the SEP stimulus into a long-lasting oscillation which will markedly obscure the SEP waveform (12) (Figure 12.14).
A notch filter is of limited utility for BAEP recordings, in which line frequency is already markedly attenuated by the high-frequency filter setting (typically 100–150 Hz). Higher harmonics of the line frequency, which might not be removed by band-pass filtering, would also not be removed by the line frequency notch filter.
Additional Sources of Artifact in EPs During NIOM
Electrocautery uses high-frequency electrical currents to heat and cut tissue and coagulate bleeding. The cautery currents can travel widely through the patient’s body, be detected at substantially greater amplitude by the recording electrodes, and contaminate the EP waveforms (waveforms at 12:15 in Figure 12.7). They can also trigger automatic artifact rejection (Figure 12.11B), but this is not inevitable. Some EP recording systems incorporate sensors that can be placed over the lead wires from the cautery equipment, to detect when the cautery is active and block EP data acquisition even if the cautery artifact itself does not trigger automatic artifact rejection.
There are two types of electrocautery: bipolar and monopolar. In bipolar cautery, the current is passed between the two ends of a forceps that the surgeons place within the surgical field. In monopolar cautery, the surgeon uses an insulated wand with an uninsulated tip; the current return is a large pad, typically placed on the patient’s leg. With monopolar cautery, the cautery currents spread widely through the patient’s body, and are more likely to trigger artifact rejection. The artifact from monopolar cautery may be large enough to saturate the amplifier input stages of some EP recording systems, so EP acquisition is blocked for several seconds after the cautery current stops. Since direct current (DC) and low frequencies are removed by band-pass filtering, the amplifier outputs rapidly return to near zero, and the data epochs are not rejected as artifactual. This may result in the incorporation of many sweeps of essentially zero data into the averages. EP latencies will not be altered but amplitudes may appear to be significantly attenuated. To manage this problem, averaging must be paused until the amplified and digitized raw data have clearly recovered, which may take several seconds (13).
With monopolar cautery, the amount of current passing through the current return pad is the same as the current passing through the wand tip; no injury occurs under the pad because the area is large and the current density 333is low. If the pad were to become dislodged so that only a small area was contacting the patient, the current density would be high and a burn could result. Also, if the electrical connection to the current return pad was interrupted (e.g., due to broken or disconnected lead wires), the cautery current could return through other paths, such as parts of the operating table touching the patient’s body or the NIOM electrodes, causing burns there. In order to avoid this, many monopolar cautery systems incorporate a return electrode quality monitoring system, in which the return electrode is split into two halves (both of which return the high cautery current) and the resistance of the circuit for passing a small monitoring current between the two halves of the pad is measured (14). Although the monitoring current is small, it can cause artifact in recording electrodes near the ground pad (Figure 12.15).
Ultrasonic aspirator devices use ultrasonic vibrations to pulverize tissue into a form that can be washed away by saline irrigation and suctioning, without traction on underlying structures. The irrigation and suction ports are incorporated into the same surgical instrument as the ultrasonic vibrator. The ultrasonic vibrations can cause high-frequency artifact which may not be large enough to trigger automatic artifact rejection (Figure 12.11C) and thus can be incorporated in the EP data, causing high-frequency artifact (Figure 12.16), or creating an apparent lower frequency artifact because of aliasing (Figure 12.17).
Other Sources of Artifact During NIOM
Voltages are generated when dissimilar metals (e.g., surgical instruments, retractors, and NIOM electrodes) are in contact with an electrolyte (e.g., the patient’s blood and tissue fluids), forming battery half-cell potentials. Electrostatic potentials can also be generated close to or within the surgical field (e.g., by surgical gloves rubbing against drapes) or more remotely by other mechanisms (e.g., by the surgeons’ shoes rubbing on the floor) with intermittent transmission of the static electricity to the patient’s body. The latter would be many orders of magnitude smaller than the thousands of volts in a typical static shock from touching a doorknob when the absolute humidity of the air is low and thus would not be visible or tactile, yet could still be many orders of magnitude larger than an EP. Artifacts from dissimilar metal potentials and electrostatic discharges are typically brief, though they may recur, and are usually mitigated by automatic artifact rejection.
Some operating microscopes (e.g., those with arc light sources) may generate artifacts that recur at the second harmonic of the line frequency as the alternating current through the arc stops and starts in either direction. Such artifacts may be sharply contoured (Figure 12.11D), with frequency spectra containing higher harmonics, so that a line frequency notch filter would not remove them.
Drills may also produce rhythmic artifacts, as the cutting edges on a rotating drill bit or bur repetitively contact tissue, due to dissimilar metals and/or electrostatic potentials (Figure 12.18). Drilling of the skull or of the roof of the internal auditory canal may interfere with BAEP recordings in another way: the high noise levels that are bone-conducted to the cochlea can attenuate the BAEPs due to acoustic masking (15,16).
Electric shavers can cause artifacts at the line frequency (Figure 12.19) since the blades are moved using a coil through which the line frequency AC is passed. As in the case of the fluid warmer, this may form the primary coil of a transformer. Shaver artifact may be problematic while attempting to establish baselines as the anesthetized patient is being prepped for surgery, but would not occur later on during the surgery after shaving is complete.
Measures to Prevent and Ameliorate Artifacts
Several measures to reduce artifacts during EP recordings, such as patient relaxation during extraoperative diagnostic EP studies or rotation of the fluid warmer during NIOM, have already been discussed. There are additional measures that can be taken to reduce artifact during EP recordings.
Electrodes that are making poor electrical contact with the subject are another source of artifact; impedances should be checked and kept below 5 kΩ when possible. Electrode impedances should be similar across the array of recording electrodes, since mismatched electrode impedances impair the rejection of common-mode signals (17) and may increase the degree to which widespread artifact (such as line frequency artifact) is present in the EP waveforms. Using various types of recording electrodes (e.g., needles, cup electrodes with electrode paste, and sticky pads) simultaneously and recording differentially between different types of electrodes will increase the chances of impedance mismatches.
In-phase cancellation removes signals present at the two inputs to a channel, such as ECG picked up by two electrodes on the head but not appearing in the differential recording between them. As the distance between the two electrodes that are inputs to the same recording channel is increased, the 334amount of artifact in that recording channel that is not in-phase canceled will increase. This may make it desirable to use alternative recording derivations, such as recording the P14 component of the median nerve SEP between two electrodes on the head rather than between an electrode on the head and a noncephalic electrode (18) (Figure 12.20).
Artifact may be picked up by the EP recording lead wires (in contrast to artifact coming from the patient’s body) when they are acting as the secondary coil of a transformer. If the lead wires are located close to each other, the artifact they pick up will be very similar and can be reduced or eliminated by in-phase cancellation. Thus, use of twisted pair recording lead wires or braiding together of recording lead wires can help to reduce artifact.
In paired stimulation lead wires, the currents in the two wires are equal and opposite and therefore the magnetic fields that they produce will largely cancel out if the wires are located close to each other. Since stimulation lead wires can act as the primary coil of a transformer, inducing stimulus artifact in the recording lead wires, use of twisted pair stimulation lead wires can help to reduce the electrical stimulus artifact.
REPLICATION OF EP WAVEFORMS
Although signal averaging decreases noise (Figure 12.3), it does not eliminate it completely. Residual artifact in the averaged EP waveform could be misinterpreted as an EP peak. Therefore, during clinical diagnostic EP studies, all waveforms should be replicated (5) at least twice, and more times if it seems appropriate. For example, in the SEP waveforms in Figure 12.20, the N20 component in the first channel replicates nicely and represents a true EP signal; the deflections in the second channel, while equally large, do not replicate and represent artifact.
If there were no noise, the EP signals in the two replications would overlap exactly. Therefore, the spacing or difference between the two replications gives a visual estimate of the amount of residual noise remaining in the EP waveforms. Since noise peaks in the two repetitions could occasionally overlap by coincidence, replicating peaks that are as small as the difference between the two replications should be interpreted with caution.
During NIOM of EPs, serially recorded waveforms provide the replications needed to distinguish reproducible EP peaks from artifacts. If the amount of noise in the waveforms is similar to the amplitude of the EPs themselves, measures should be taken to identify the causes of the artifacts and try to reduce them. If necessary, the number of sweeps in the average can be increased. Alternatively, measures can be taken to increase the size of the EPs and therefore improve the signal-to-noise ratio. This may involve increases in the stimulus intensity, decreases in the stimulus rate, and changes in the anesthetic regimen to lessen anesthetic-related attenuation of the EPs that are prominently affected by anesthesia.
Understanding of the sources and mechanisms of artifacts and ways in which they can be eliminated or reduced can lead to an improvement in the quality of EP recordings, both during NIOM and during extraoperative diagnostic EP studies.
MEP, motor evoked potential; TES, transcranial electrical stimulation.
Source: From Legatt AD, Emerson RG, Epstein CM, et al. ACNS guideline: transcranial electrical stimulation motor evoked potential monitoring. J Clin Neurophysiol. 2016;33(1):42-50. doi:10.1097/WNP.0000000000000253, with permission.