Development and application of a novel feedback electrical stimulation device

0 Introduction

With the development of human society, people's pace of life is getting faster and faster, mental stress is getting greater and greater, and more and more people are suffering from sleep disorders. Studies have shown that the judgment of sleep effect depends not only on the length of sleep time, but more importantly on the depth of sleep.

The first choice for treating sleep disorders is undoubtedly still drugs, but the harm caused by using drugs to correct sleep disorders to the human body has attracted more and more attention. In addition to drug side effects, drug dependence is also an important problem. In recent years, non-drug methods for treating sleep disorders have received attention. At present, non-drug methods for treating sleep disorders mainly include acupoint stimulation based on traditional Chinese medicine theory, brain electromagnetic stimulation, brain sound and light stimulation, etc. The common disadvantage of these methods is that the parameters of stimulation during treatment are the same for all patients, and it is difficult to determine the appropriate stimulation parameters for specific individual patients. In addition, since the mechanism of action of these methods is still unclear, the therapeutic effects of different methods vary from individual to individual, and the exact efficacy remains to be further confirmed. At present, there is no report that an instrument can produce an adaptive stimulation pattern based on the patient's sleep process state and the individual differences shown during sleep, thereby improving sleep.

This paper proposes a neurofeedback treatment device for sleep disorders that realizes information interaction between computers and brain, attempting to develop a personalized sleep disorder correction instrument.

1 Methods

1.1 System Overview

In recent years, the research on sleep EEG has made great progress. According to the analysis of sleep EEG, the quality of sleep can be quantitatively judged. People have processed sleep EEG through different methods, such as time-frequency analysis, wavelet transform neural network, information entropy theory, etc., and found that with the increase of sleep depth, sleep EEG has significantly different characteristics. The current internationally accepted sleep classification is to divide sleep into two categories, namely rapid eye movement (REM) sleep and non-rapid eye movement (NREM) sleep, according to the characteristics of EEG. NREM sleep can be divided into four stages of different degrees, which are marked as stage 1, stage 2, stage 3 and stage 4 according to their depth. Together with the wakefulness period before sleep, sleep can be divided into six different states. Recent studies have shown that as sleep deepens, the regression complexity of sleep EEG continues to decrease. The regression complexity of sleep EEG can be used as a simple and practical indicator to measure sleep quality.

Based on the above analysis, a personalized adaptive neurofeedback electrical stimulation sleep disorder treatment device based on quantitative analysis of sleep depth is proposed. First, the 12-second EEG signal of the subject is extracted by electrodes placed on the human head, and amplified, filtered and analog-to-digital converted; second, the average value of the regression complexity of the four EEGs of the subject in different sleep periods is extracted, and the average value is the quantitative result of the sleep depth; third, according to the quantitative value of the subject's sleep depth, the stimulation mode of the next electrical stimulation is generated after analysis, including the stimulation waveform and waveform parameters; fourth, the stimulation signal in this mode acts on the brain through the same set of electrodes as the EEG acquisition, and the stimulation is stopped after 60 seconds, and the 12-second EEG is collected again and the regression complexity is calculated to understand the quantitative value of sleep quality under the new stimulation conditions, compare the impact of this stimulation and the previous stimulation on sleep, and decide on the adjustment plan of the next stimulation mode. The flow chart of the system is shown in Figure 1.

1.2 EEG extraction and amplification

The electrodes used for EEG extraction are Ag/Cl disk electrodes, which are placed at four locations F3, F4, C3, and C4 on the parietal lobe of the scalp, and the reference electrodes A1 and A2 are placed on the earlobes. With the help of a bidirectional electronic switch 74LS245, the same set of electrodes placed on the subject's scalp are used to alternately perform 12 s EEG acquisition and 60 s brain stimulation tasks. The direction of the bidirectional electronic switch 74LS245 is controlled by an input/output port of the CPU.

The first stage of the EEG amplifier is composed of an op amp circuit to form a voltage follower to improve the common mode rejection ratio and input impedance of the amplifier circuit. Considering the power supply and power consumption issues, the amplifier circuit can use the micro-power and low-voltage four-op amp operational amplifier chip LM324, which has a compact structure and low power consumption. The second stage of the amplifier circuit uses the power amplifier chip LM386. The latter stage is a bandpass filter with a passband of 0.5 to 70 Hz, which can filter out the high-frequency components of the EEG and some high-frequency interference signals. A 60 Hz power frequency filtering circuit, i.e., a power frequency notch, should be set in the circuit. The amplitude of the EEG signal after amplification and conditioning should be between 0 and 5 V.

1.3 Regression Complexity

For an original one-dimensional time series of N points, such as the EEG time series ui, a set of vectors in the phase space can be reconstructed by the time delay method:

Where: τ is the delay time; m is the embedding dimension.

The above vectors are displayed on a two-dimensional plane using the algorithm described below, and then the correlation characteristics of the sequence are analyzed to see whether there is a restoration relationship between time i and time j represented by each point in the two-dimensional graph. Such a two-dimensional graph is called a regression graph.

Mathematically, the regression graph calculation formula can be expressed as:

Where: εi is the cutoff distance; ||·|| represents the norm; @(z) is the Heavi-side function

In the calculations performed by this device, the embedding dimension m and the time delay τ are both 3, and the cutoff distance εi is 5% of the maximum value of all vector distances. After obtaining the EEG regression map according to the above method, the regression complexity is defined as the proportion of regression points on the regression map:

[page]

The value of this regression complexity is between 0 and 1. According to the relationship that regression complexity is basically proportional to the depth of sleep, it can be judged that the better the sleep quality, the smaller the regression complexity; the worse the sleep quality, the greater the regression complexity. Figure 2 (a) is a regression diagram with a regression complexity of 0.17 when the sleep quality is high; Figure 2 (b) is a regression diagram with a regression complexity of 0.71 when the sleep quality is not high.

According to the calculated regression complexity index value of the EEG signal, the depth of sleep is directly represented by the regression complexity index. That is, the average value of the regression complexity of the four EEG channels in different sleep stages of the subject is extracted, and the average value is the quantitative result of the sleep depth.

1.4 Generation of stimulus waveform

The personalized neurofeedback insomnia treatment device proposed in this paper is controlled by a single-chip microcomputer with a CPU of C8051. The C8051 single-chip microcomputer integrates a 12-bit analog/digital converter, 5 general-purpose 16-bit timers, an internal programmable oscillator, a low-power 128-byte non-volatile data storage, and rich input and output resources. The expansion system also includes 64 KB of random access memory RAM and 64 KB of read-only memory ROM. Using the digital/analog conversion circuit in the system, stimulation modes with different waveforms and parameters such as square waves, triangle waves, and sine waves can be generated through programming. The frequency of the stimulation waveform can be changed by adding a delay program to the program, and the amplitude of the stimulation waveform can be adjusted by changing the digital signal value of the digital/analog conversion.

The programmable oscillator inside the microcontroller is used as the system clock of the microcontroller, and the internal reference voltage is used as the reference voltage of the digital/analog conversion circuit. The values ​​of the registers of the digital/analog conversion circuit are set in the interrupt service program of the timer, and the conversion is started. The conversion result is amplified by the power amplifier.

1.5 Determination of stimulation mode

Different stimulation strategies can be adopted according to the quantitative evaluation value of sleep depth.

(1) If it is the first stimulation, a fixed stimulation pattern is pre-determined as the starting stimulation. The starting stimulation is determined according to the size of the regression complexity. When the regression complexity is less than 0.2, no stimulation is required; when the regression complexity is greater than or equal to 0.2 and less than 0.4, a 1 Hz, 1 V sine wave is used as the starting stimulation; when the regression complexity is greater than or equal to 0.4 and less than 0.6, a 5 Hz, 5 V sine wave is used as the starting stimulation; when the regression complexity is greater than or equal to 0.6, a 10 Hz, 10 V sine wave is used as the starting stimulation.

(2) If the stimulation currently being implemented is not the first stimulation, it is determined by comparing the sleep effect produced by this stimulation with that produced by the previous stimulation.

① If it is found that the sleep quality is improved for the first time as the stimulation parameters change, then this stimulation mode is maintained.

② If this is the second time that the stimulation mode is judged to be improved, the stimulation mode will be increased step by step. The step increase means increasing the stimulation frequency and stimulation amplitude by 1 Hz and 1 V respectively within the allowable stimulation parameter range. If the stimulation parameters used for the waveform have reached the maximum allowable value, the next waveform will be replaced and start again from 1 Hz, 1 V.

③ If it is found that the sleep quality is reduced for the first time due to the change of stimulation parameters, the parameter adjustment is canceled and the previous stimulation mode is returned.

[page]

④ If it is the second time that the sleep quality is judged to be deteriorated under the current stimulation mode, the current stimulation mode will be step-down. The step-down means reducing the stimulation frequency and stimulation amplitude by 1 Hz and 1 V respectively within the allowable stimulation parameter range. If the stimulation parameters used for the waveform have been reduced to the minimum value within the allowable range, then another waveform will be changed and restarted from the allowable maximum value.

(3) According to the determined stimulation pattern, a new round of 60 s electrical stimulation was applied to the subjects.

The above stimulation can also be illustrated by Figure 3. The stimulation mode includes waveforms and parameters. There are three waveforms used in this system, which are: square wave, sine wave, and triangle wave. Each time the parameters are adjusted, 12 seconds of EEG signals are collected 60 seconds after the stimulation to analyze and understand the impact of the stimulation mode adjustment on sleep, that is, to compare the impact of the current stimulation with the previous stimulation on sleep and decide on the next adjustment of the stimulation mode.

2 Results and Discussion

Two experimental tests show that the system can effectively improve insomnia symptoms and improve sleep quality. The reasons why the system achieves good correction effects include:

(1) The arrangement of the electrodes on the scalp takes into account multiple factors, including EEG acquisition, stimulation effect, and convenience of clinical application;

(2) The sleep quality is determined by using EEG regression complexity calculation algorithm, which is time-saving and can meet the requirements of real-time EEG processing;

(3) Brain electrical stimulation and EEG acquisition are completed through the same set of electrodes, which greatly reduces the number of electrodes placed on the subject's scalp and is more suitable for clinical application;

(4) Compare the changes in sleep quality before and after the stimulation parameters are adjusted to determine the stimulation parameter adjustment strategy. This feedback treatment method and device are completely personalized, overcoming the unscientific situation in which the stimulation methods and parameters of insomnia treatment devices were the same for everyone in the past.

(5) The method proposed in this article uses EEG to quantitatively monitor sleep quality. When sleep is good, no stimulation may be generated or the generated stimulation may further maintain the good sleep. When sleep is not good, there will be a variety of stimulation modes that can be adaptively changed, including stimulation waveforms (such as triangle waves, rectangular waves, sine waves) and stimulation parameters (including amplitude, frequency, etc.). The effect of brain stimulation is monitored by EEG at any time, and the correction effect can be guaranteed.

The anti-interference performance of this system can be guaranteed by comprehensive measures combining software and hardware. A differential amplifier circuit is used at the input stage, which has a good common mode rejection ratio, generally up to 80-90 dB. When designing the program, single-byte instructions are mostly used, and some empty instructions are artificially inserted in key places, or valid single-byte instructions are written repeatedly; single-byte empty instruction operations are inserted after double-byte and three-byte instructions to protect the program from possible erroneous transfer when the single-chip CPU is interfered with. When the program may transfer to the non-program area under interference, instruction redundancy no longer works, and the software trap can use a guide instruction to forcibly lead the captured program to the error handling program. Software traps are generally arranged in unused interrupt vector areas, unused large pieces of read-only memory ROM space, and breakpoints in the program area. As a result of these measures, the influence of interference on this system is effectively prevented.

Of course, the system proposed in this article still needs further clinical research and verification, and its further improvement is necessary.