Search

Analysis of sleep stages using electroencephalograms (EEGs), a critical procedure in health monitoring, has been researched extensively. Scoring of the sleep stages is highly dependent on experts’ knowledge or hand-crafted features created by experts. Recent advancements in deep learning have, nevertheless, successfully automated the scoring process with accuracy comparable to human performance. Previous methods were less convenient as they employed specific EEG devices designed for the experiments. Such devices are usually expensive and require a special environment and arrangements for their use. In this study, we propose an automatic sleep-stage scoring system using a patch-type wearable EEG sensor based on deep learning. The proposed sensor, namely Haru, is lightweight, inexpensive, and easy to use. As for the training architecture, we implement a model based on the state-of-the-art deep learning algorithm specifically designed for automatic sleep staging. We record three channels of sleep EEGs from 30 subjects at a sampling rate of 250 Hz with the proposed system and a polysomnogram (PSG) device simultaneously. Training datasets are created on the basis of the annotated samples in PSG measurement. In the experiment, we train our model and evaluate the results using the leave-one-out cross-validation method. The results indicate that the proposed Haru Sleep is capable of scoring sleep stages with a scoring accuracy of 78.6% and an F1 score of 73.4%, which is comparable to that obtained by clinical PSG devices.

Flexible and transparent electrodes are highly useful in wearable optoelectronic systems for healthcare and biosensing applications for conducting multimodal assessments with electrophysiological and optical measures. In such systems, the electrodes should exhibit a low sheet resistance, high visible transmittance, and small feature size, for reliable electrical sensing, optical observation of attached objects, and integration of devices for mapping local biology events, respectively. Herein, fine‐printed, flexible, and transparent microelectrodes that allow biosensing and device integration are reported. The microelectrodes containing cross‐aligned silver nanowire (AgNW) networks are patterned via a selective wetting deposition technique on a 1 μm‐thick polymer substrate. A low sheet resistance of 25 Ω sq−1 and a visible transmittance of 96%–99% are achieved with a small pattern width of 25 μm. The biosensing application is demonstrated by detecting the leaf electric potential; the leaf cells under the microelectrodes are observed because of visible transparency. Furthermore, device integration is demonstrated by electronic circuits with ultrathin and transparent organic transistors. The transistors exhibit white‐light illumination stability and mechanical stability to bending stress. These demonstrations provide a basis for developing ultraimperceptible wearable sensor systems with ultrathinness and high visible transmittance.

Purpose: To evaluate the reproducibility and reliability of a novel electromyogram (EMG) device with a flexible sheet sensor for measuring muscle activity related to mastication and swallowing. Methods: We developed a new EMG device made of elastic sheet electrodes to measure the masseter and digastric muscle activities for evaluating mastication and swallowing. To examine the measurement reproducibility of the new EMG device, masseter muscle activity was analyzed using the intraclass correlation coefficient (ICC). Further, we measured the maximum amplitude, duration, integrated value, and signal-to-noise ratio (SNR) using the new EMG device and conventional EMG devices and evaluated the reliability using ICC and Bland-Altman analysis. Results: We confirmed high ICC (1,1) and ICC (2,1) scores (0.92 and 0.88, respectively) while measuring the reproducibility of the new EMG device. When compared to the active electrode EMG device, we found a high correlation for the maximum amplitude (0.90), duration (0.99), integrated values (0.90), and SNR (0.75), with no observation of significant fixed errors. Moreover, the regression coefficient was not significant for any of the evaluation items and no proportional error was observed. Compared with the passive electrode EMG device, the maximum amplitude and duration were highly correlated (0.73 and 0.89). In addition, the SNR exhibited a significant fixed error. In contrast, the regression coefficient was not significant for any of the evaluation items and no proportional error was observed. Conclusions: Our results suggest that the new EMG device can be used to reliably and reproducibly evaluate muscle activity during mastication and swallowing. [ABSTRACT FROM AUTHOR]