A. Bradford, C. DuBois, A.T. McDaniel, D. Cobble, L. Schroeder, W. Tseh
University of North Carolina Wilmington,
United States
Keywords: electroencephalogram, neuroscience, fabric as a sensor
Summary:
Introduction Electroencephalography (EEG) plays a crucial role in monitoring brain activity, offering real-time insights into brain wave patterns during various states, including sleep. By detecting distinct sleep stages, EEG helps quantify the duration spent in each phase, which is vital for understanding sleep architecture. Clinicians use EEG in sleep studies to diagnose and manage disorders like insomnia, narcolepsy, and sleep apnea. Epidemiological studies highlight the rising prevalence of sleep disorders, affecting around 30% of adults globally. Recent innovations aim to integrate EEG with other technologies for real-time monitoring of sleep-related issues. This study seeks to validate a novel EEG prototype developed by Nuream, Inc., utilizing a "pillow array" of silver-chloride transducers to process non-traditional EEG data via Fabric as a Sensor (FaaS) technology. Materials and Methods Fifteen college students participated in the initial phase of the study. Participants reported demographic information and their history of traumatic brain injury (TBI) before completing three-hour trials. During each trial, participants lay on a bed with their head resting on the pillow array, while a traditional adhesive EEG electrode was placed at the center of their forehead for comparison. Reference electrodes were positioned on either side of the forehead to serve as baseline for the “ActiveTwo” data acquisition system (DAQ). Data from both the 64-channel pillow array and traditional EEG transducer were sampled at 512Hz and transmitted to a computer for monitoring and saving via a custom LabVIEW program. A webcam tracked participant movement and head position during each trial. Results Data processing was performed using custom MATLAB routines. Initial pre-processing involved detrending to remove gradual changes and applying a second-order Butterworth filter (2-35 Hz) to eliminate noise outside typical EEG frequency ranges. Each 30-second data segment was shaped with a Hamming window and converted to the frequency domain using fast Fourier transform (FFT). Frequency distributions within each segment were analyzed to identify sleep stage trends and anticipate signals like Alpha waves. Further processing combined active channels to compare results with the reference signal, minimizing interference from inactive channels. Discussion Figure 3 presents a comparison of frequency trends between the reference and test datasets. In this example, data from eight sensors in the middle of the pillow array were grouped to form the test set, reducing interference from less-active sensors. Frequency values for each 30-second interval were plotted, alongside 5-minute rolling averages. The correlation coefficient between the rolling averages of the test and reference datasets was R=0.36. As the study progresses, this processing method will be expanded to include multiple trials. New techniques, such as automated active channel selection and targeted waveform isolation, will further refine the analysis.