Introduction
Sleep disorder, a rapidly growing field
Therefore, sleep disorders can cause both a disease in the body and a physical condition itself. 1] These disorders usually include rapid eye movement (REM) sleep behavior disorders, apnea, sleep walking, etc., which can cause serious sleep disorders [2]. Population epidemiology shows [4] that sleep deprivation and disability are a significant part of the world's population, with 30% of adults suffering from insomnia.
Conventional sleep diagnostic tool and currently studies
The general point of healthcare devices for sleep analysis is that they are hand-held [7], so that they can be worn and measured at home. However, they were only used in partial sleep analysis and simple body condition checks such as heart rate and pedometer. Wearable headband research for dry electrodes [11] is a fairly ideal home sleep analysis tool, but also lacks its own sleep analysis.
In the research of healthcare devices for sleep apnea patients [12], analytical systems [13] with their own algorithms are studied.
Conventional sleep analysis process and direction for improvement
When the reset signal is 'H', the oscillator repeats charging and discharging, the period of which is determined by the time constant of RC. 27(a) shows the Teager energy operator (TEO), which is the feature extraction circuit in the EEG sleep stage. Quan, “The AASM manual for scoring sleep and associated events: rules, terminology, and technical specifications,” 1st ed.
9] DJ, Levendowski, et al., "Accuracy, night-to-night variability, and stability of electroencephalographic biomarkers of frontopolar sleep," J.
Background
Sleep stages and classification
The EOG signal has a large amplitude due to the blinking of the eye and the frequency shows a frequency below 2 Hz. Sleep stage1 sleep stage is a transition stage in which the body and brain move from a waking stage to sleep stage2. And an important indicator of REM sleep is the amplitude tone of the EMG signal.
Therefore, the muscle's activity signal is almost flat, which is an important indicator of REM sleep.
Electrode placement for multi-biosignal
AASM's recommended electrode placement was 1 cm above and slightly lateral to the outer canthus of one eye, with a reference electrode on the ipsilateral earlobe or mastoid. The EOG channel on the other side was to be recorded from an electrode 1 cm lower and slightly lateral to the outer canthus of the fellow eye, also called the contralateral ear or mastoid.
Introduce a problem of sleep analysis of conventional PSG
The problem with commercial product for sleep analysis
The shape of the sensor is changed to The proposed work is a research of sensor interfaces. The output of the feature extraction stage is the digital value, and they go to the MCU with a serial to parallel interface on the ROIC. The decision tree algorithm rearranges and marks the sleep stage using the digital values of the input feature extraction to the MCU.
The overall structure of multiple biosignal detection ROIC and MCU with BT module is shown in the figure. The rectified signal affects the charge cycle of the integrator feedback capacitor depending on the amplitude. So the amplitude of the rectified signal changes according to the sleep phase, REM and NREM can be detected.
Depending on the feature count for an epoch, the phase classification is determined by the decision tree in Fig. 30 is a real-time monitoring of serial EEG feature extraction data processed in ROIC with multi-biosignal sensors. The multi-biosignal sensor ROIC consists of a preamplification stage and a feature extraction stage, including a specialized multi-biosignal.
In particular, the feature extraction stage can detect sleep stage features for sleep classification scoring. The feature extraction stage consists of a biosignal custom circuit for EEG, EMG and EOG.
Previous work
A skin-inspired sensor interface using customized analog circuit and system
It has the same direction as feature finding and implementation at the circuit level, which corresponds to the implementation of optimized signal processing for a multi-biosignal sensor interface system. Therefore, the ROIC is proposed to have an optimized circuit architecture for operating the three electrical characteristics of the PVDF/RGO composite e-skin device. The proposed e-skin device implements inspired receptors to detect skin pressure and temperature in an interconnected architecture.
In contrast to conventional readout circuits [15], the ROIC is implemented to be reconfigurable so that it can support the three electrical characteristic modes of the e-skin device. The feature extraction stage extracts some signals with brainwave characteristics during sleep using the AASM scoring manual. The implementation of the smart headband allows convenient checking of the state of sleep.
The power consumption of the multi-biosignal sensing interface system is 70.9 mW, among which the ROIC occupies 0.9 mW. The EEG target frequency has a configuration of 11-16 Hz, which distinguishes N2 and N3 from the sleep phase. Figure 26 shows a graph of the PGA frequency characteristics for each channel adjusted for EEG, EOG and EMG.
In addition, the current consumption for the circuit structure of the gm cell configuration is reduced. The transconductance of the input stage amplifier changes its frequency characteristics according to the amount of current. Then the signal is filtered at the frequency of the N2 stage, which outputs the sleeper finger through the comparator.
One of the datasets contains the sleep EDF dataset, which consists of PSG records with six sleep stages [28,29]. The OpenBCI interface was worn on the upper part of the head and the proposed smart headband was worn on the forehead. With the exception of the signal level detector, the output of the remaining circuit is a digital bit.
Implementation of multi-biosignal sensing interface with ROIC
Multi-biosignal sensing system for sleep analysis
Proposed multi-Biosignal sensing ROIC
When body signals are input through the module's electrodes, they are amplified and filtered according to the signal characteristics of each part of the body in the pre-amplification stage. 25 illustrates the schematic of the preamplification stage, including a low noise amplifier (LNA) with chopper stabilization and switched capacitive PGA. This process is intended to distinguish between SS and DS, and can be used to measure the essential characteristics of the sleep spindle [25].
It is waveform below 2Hz due to eye blink and when it is REM it shows the characteristic of waveform going above 2Hz. 29 with Table 2, if the feature extraction count of an epoch matches the partition content of a decision node, the epoch state goes in the direction of the left side. The data was input to the smartband module through the simulator with an amplitude of less than 50uV, the size of the brain waves.
On the right side of the figure is an oscilloscope-amplified analog waveform from the preamp stage. As a result, a multi-biosignal interface was obtained in this work, and the performance of the proposed method could be determined by the average performance of the EEG feature extraction, as shown in Table 3. However, it is shown that it is inverse from the 200-epoch part entering the DS of the EEG_FE graph.
Feature extraction measurements for an 800 epoch (6 h 40 m) (a) Schematic of EEG feature extraction data. b) Schematic of EOG feature extraction data (c) Schematic of EMG feature extraction data. 27] R, Ferri, et al., "A quantitative statistical analysis of submentalis muscle EMG amplitude during sleep in normal controls and patients with REM sleep behavior disorder", J.
Experimental results
Evaluation of the EEG analysis using sleep dataset
The data sampling rate is 100 Hz and was estimated based on the R&K rule. The progress method outputs data to the CSV file by dividing the SS and DS waveforms by epoch through the Hypnogram file of a dataset. 30(a) is an experimental figure using the sleep stage 2 signal corresponding to SS using the data set, and Fig.
As shown on the Tablet display, the output of feature extraction is different because the frequency of step 2 is relatively higher than step 3. TP indicates the true positive, TN refers to the true negative, FP indicates the false positive, and FN refers to the false negatives. The second experiment is the comparison of EEG measurement with OpenBCI and the proposed smart band.
Sleep stage verification was performed by extracting single-channel signals from F4 and F3 using the verified EEG interface [ 4 ]. The two signals received from OpenBCI were subjected to sleep classification using a custom code using the SVM classifier of MATLAB.
Sleep scoring using a rule-based decision tree algorithm
McCarter, et al., “Factors associated with harm in REM sleep behavior disorder,” Sleep Med, vol. 10] E, Fino, G, Plazzi, Filardi, M, et al., “(Not so) Smart sleep tracking through the phone: Results from a polysomnographic study testing the reliability of four sleep applications" J. 11] C-T Lin, et al., "Forehead EEG in Support of Future Possible Personal Health Solutions: Sleep Management, Headache Prevention, and Depression Treatment," IEEE Access, vol.
Patanaik, et al., “An end-to-end framework for real-time automatic sleep stage classification.” Liang, et al., “Automatic phase scoring of single-channel sleep EEG using multiscale entropy and autoregressive models,” IEEE Trans. Jiang, et al., “BrainNet: A multi-person brain-to-brain interface for direct brain collaboration,” Sci.
Valentin, et al., “Validation and benchmarking of a wearable EEG acquisition platform for real-world applications,” IEEE Trans. Ye, et al., “A Non-Invasive Sleep Analysis Approach Based on a Fuzzy Inference System and a Finite State Machine,” IEEE Access, vol. Dick, et al., “AASM standards for practice-based validation of actigraphic sleep analysis of SOMNOwatchTM versus polysomnographic sleep diagnostics show high compliance, including in subjects with sleep-disordered breathing,” Physiol.
그리고 저를 위해 애써주신 분들께 감사의 말씀을 전하고 싶습니다. 마지막까지 잘 지도해주신 교수님에게 진심으로 감사하다는 말씀 전하고 싶습니다.
Conclusion
