Clinical Validation of zEEG, a Wireless EEG Recording System for Pediatric and Adult Patients With Epilepsy

Research Hypothesis:

The hypothesis is that signals recorded by zEEG do not differ significantly from signals recorded by a standard clinical EEG system (clinical EEG). In this study we will use the Xltek® (Natus, Inc.) clinical-grade FDA-approved long-term video monitoring EEG system and the Nicolet ® (Natus, Inc.) clinical-grade FDA approved ambulatory EEG system. zEEG will be compared against both systems. Minor difference in signal is expected due to the non-overlapping positions of the 6 electrodes. The 1-2 cm position difference between the electrodes is expected to generate discernible difference between the two signals.

Background:

The purpose of the study is to validate the zEEG® by ZETO, Inc. (zEEG), a wearable wireless headset and data acquisition (DAQ) system on both pediatric and adult patient populations diagnosed with epilepsy. The study compares (i) the technical parameters of the two systems, (ii) the sensitivity and specificity of the zEEG against standard clinical EEG DAQ systems used for detecting abnormal epileptiform discharges, (iii) the correlation with intracranial recordings, (iv) and the discriminability of the two systems by trained neurologists in a double-blind paradigm. Since zEEG device has received a 510(k) clearance from FDA in 2018, the study does not concern with the safety of the device. The device poses no significant risk, as evaluated by the FDA, and qualifies as Class II category. The goal of the study is to evaluate the utility of a dry, easy to use, wireless system in clinical electrography by a quantitative assessment of the overall quality of the signal relative to a traditional clinical EEG system.

Design and methodology:

Spec Aim 1: Quantify and compare the signal quality between the zEEG and clinical EEG system in simultaneous recording sessions performed for diagnostic purpose in patients under clinical setting, one session per patient. Specifically, we will be testing signal overlap (amplitude, sampling frequency, alpha waves, sharp waves, focal slowing, spike and wave patterns, and sleep spindles) and spectral correlation between the two systems. We will use a multi-factorial ANOVA design, where the independent variables are arranged into 4 factors: age-group (pediatric, adult), system (zEEG, cEEG), task (1-5), and recording sites (10-20 system). The dependent variable is the spectral signal to noise ratio (SNR) computed as the spectral mean (divided by the spectral standard deviation.

Spec Aim 2: Determine the sensitivity and specificity of the two systems for epileptiform activity in simultaneous recordings. Although Spec Aim 1 will provide us an estimate of the sensitivity of zEEG relative to the clinical EEG system, it does not provide an estimate on the specificity of the parameters for electrographic events that are relevant for diagnosis. For example, if the zEEG with its high-impedance dry electrodes turns out to be a lot more sensitive to electric potential changes than the standard clinical device, that might increase its sensitivity for movement artifacts too, which would contaminate neuronal signals. Sensitivity without specificity does not mean improvement. Therefore, we will test whether the difference in sensitivity is associated with a difference in specificity as well. A dry electrode-based system will surpass the standard wet electrode clinical systems if it provides at least the same sensitivity and specificity for the representation of clinically relevant EEG events. Therefore, we will compare the separation between myogenic artifacts, teeth clenching, head movements, eye movements, and EEG between the two systems, zEEG and the clinical EEG system. In this Specific Aim, we will only compare Xltek® (Natus, Inc.) a clinical-grade FDA approved long-term monitoring EEG system with zEEG, as opposed to using two different clinical EEG systems (Nicolet® and Xltek) to ensure the homogeneity of samples (Appendix, Figure 2A). Most importantly, we will compare the separation between baseline and epileptiform activity. Instances of epileptiform activity will be identified by trained neurologists among our key co-investigators. We will compare the magnitude of amplitude- and power spectra-change from baseline EEG to movement artifacts and epileptiform activity between the two systems. Spectral density statistics will be computed between 1 and 180 Hz.

The recordings for Spec Aim 1 and Spec Aim 2 will be performed on different subjects using different clinical amplifiers as comparator systems: Nicolet® designed for ambulatory data acquisition, and an Xltek® designed for long term monitoring. Each patient will participate in the study only once.

Spec Aim 3: Testing the concordance of clinical evaluations made by 16 expert neurologists based on EEG records obtained by the two systems. We will compare the clinical conclusions made about relevant features of epileptiform activity captured in simultaneous zEEG and clinical EEG samples.

The purpose of this aim is to evaluate the clinical validity of the records made by using zEEG relative to the records made by using standard clinical EEG systems. We are testing whether the information conveyed by samples of the two different systems are concordant when evaluated by trained neurologists. This aim compares how clinically informative the two systems are, regardless of their technical parameters. Ultimately, the new EEG system is informative only if the clinical conclusions drawn from the two types of samples are concordant. Note, that this Aim does not concern which system is better. Instead, it addresses whether the zEEG is good enough to replace the clinical EEG system, given its advantage of ease of use, set up time, wearability and comfort. Nor does the Aim assume that the two systems are indistinguishable. Expert neurologists will not compare the two systems directly to each other. Instead, they will provide clinical assessments based on the samples. Those samples will be pre-selected by other expert neurologists (the investigators of this study). There will be no overlap between CO-Investigators and expert subject neurologists. The clinical assessments provided by the subject neurologists will be blindly scored by a computer program, and the PI and the author of the study will evaluate the concordance of those scores. Based on the concordance, the investigators conclude whether the two EEG systems convey the same clinical information or not. This procedure ensures the double-blind paradigm and minimizes biases.

Data Analysis:

Spec Aim 1: For the statistical evaluation we segment the data into equal blocks of 256 points (512 ms) and perform FFT (Fast Fourier Transform). We will use the sliding-window multi-taper spectral decomposition to avoid edge effects between the blocks. We will perform these analyses in each channel separately. Using multiple blocks, we will be able to compute statistics on 351 samples from 3 m recording. The spectral difference will be estimated with a confidence of 0.05 at different frequency bands and using Bonferrioni correction for the multiple comparisons.

Finally, we will compute the correlation coefficient between the zEEG and clinical EEG spectra under all 5 conditions

Spec Aim 2: Our neurologists will identify, and clip prerecorded simultaneous zEEG and cEEG samples during normal resting state, during movements, and during electrographic events based on video replay and EEG. We will select 60 s intervals during each type of events, as many as we can obtain during a 1 hr EEG session/subject. We apply the same filter setting and sample rate (500 Hz) for both systems. Next we compute sample correlation between the zEEG and cEEG within 1.25 s sliding windows. We compare power spectral density change within the same system between resting state EEG, movement artifacts and electrographic events (inter-ictal spikes, seizures, spikes, spike and wave, other epileptic discharges). We quantify the spectral difference between these events by computing a discriminability index using ROC analysis. ROC analysis tests how an 'ideal observer' would be able to discriminate between those samples based on the spectral differences relative to the grand truth (based on the classification by our expert neurologists) without knowing what period those samples were extracted from.

For statistical an analysis, a four-way ANOVA will be used, where the 4 factors are: age group (pediatric, adult), system (zEEG, cEEG), tasks (1-5), and recording sites (10-20 system). The dependent variable is the spectral signal to noise ratio (SNR) computed as the spectral mean (divided by the spectral standard deviation.

Spec Aim 3: The maximum achievable correct score per subject is 7 for the first test and 7 for the second. Next, we combine the scores from the sessions within the same system as follows. If the scores from the first set are denoted as clinical EEG [a1 a2 a3 a4 a5 a6] and zEEG [b1 b2 b3 b4 b5 b6] and in the second set as clinical EEG [a7 a8 a9 a10 a11 a12] and zEEG [b7 b8 b9 b10 b11 b12], then we compute the difference scores as DclinicalEEG-zEEG (a1- b1, a2-b2, … ,a12-b12). Our NULL is that the evaluations by the two experts will be concordant between the two systems and the D scores will not be different from zero. The alternative is that the expert evaluations of the two systems are discordant; hence the distribution of D scores is significantly different from zero. We will calculate a one-sample student T-test to compare the deviation of distribution from zero.

We will ask 16 experts to provide their responses on testing the samples. We will make the test to be performed on-line. The EEG samples will be de-identified. Gain, filtering and sampling rate will be identical for all samples and between both systems. No more than 1 min is needed per item. The total test with the 5 questions, about 10 samples (clinical EEG set and zEEG set) will take about 10 minutes to take. According to our power test, a sample size=16, STD=3 (in the range of 1 to 7) sample average D=2 the power of statistical test is 84.7%, sufficient for statistical conclusion to be drawn. Hence, the sample size = 16 suffices the desired statistical power of >80% (http://rpsychologist.com/d3/NHST/). By convention, 80% is an acceptable level of power (Ellis, 2010).