WO2019127558A1

WO2019127558A1 - Electroencephalogram-based depth of anesthesia monitoring method and device

Info

Publication number: WO2019127558A1
Application number: PCT/CN2017/120350
Authority: WO
Inventors: 金星亮; 罗汉源; 何先梁; 张宁玲; 叶志刚; 李明; 姚祖明
Original assignee: 深圳迈瑞生物医疗电子股份有限公司
Priority date: 2017-12-29
Filing date: 2017-12-29
Publication date: 2019-07-04
Also published as: CN110267590A

Abstract

An electroencephalogram-based depth of anesthesia monitoring method, comprising: acquiring an electroencephalogram signal via a sensor (S101); performing denoising of the electroencephalogram signal (S102); extracting signal features from the electroencephalogram signal after denoising (S103); and determining, on the basis of the signal features and a predetermined depth of anesthesia prediction model, a depth of anesthesia value (S104). The depth of anesthesia prediction model is trained by a depth of anesthesia database. The depth of anesthesia database comprises a plurality of electroencephalogram signals of an anesthesia process and anesthesia states corresponding to each section of the electroencephalogram signal of the anesthesia process. The depth of anesthesia database is used to train the depth of anesthesia prediction model and depth of anesthesia values are determined by the depth of anesthesia prediction model, avoiding overfitting in the case of a large amount of data.

Description

EEG-based anesthesia depth monitoring method and device

Technical field

The invention relates to the field of inducing electrophysiological signal processing, in particular to an electrocardiogram based anesthesia depth monitoring method and device.

Background technique

In surgery, proper depth of anesthesia is a necessary condition for surgery. Different clinical needs have different requirements for the depth control of anesthesia. For example, for a large-surgery operation, a deeper degree of anesthesia is required to ensure that the patient does not have an anesthesia accident such as intraoperative awareness due to surgical stimulation; and for a small stimulation operation, it is necessary to effectively control the amount of the anesthetic to make the patient fit. The depth of anesthesia can quickly wake up after the end of surgery, while saving the amount of anesthetic use.

In the prior art, the need for controllable and visualized depth of anesthesia in the clinic is primarily met by an anesthesia depth monitoring instrument.

Anesthesia Depth Monitoring Instrument In the prior art, the method of feature weighted summation is used to obtain the depth of anesthesia. For example, the anesthesia depth is obtained by weighted summation using a Beta Ratio, a Bispectral Index (BIS), a Burst Suppression (BSR), and a Suppression Index (QUAZI). However, the depth of anesthesia obtained by this weighted summation method is not accurate enough. Specifically, the anesthetic depth values that are fed back under different drug types for patients of different age groups are different from the actual anesthesia depth levels.

In addition, an anesthesia depth monitoring instrument also has a method of calculating the depth of anesthesia. It obtains the current anesthesia status (divided into awake, shallow hemp, medium hemp, deep anesthesia and excessive anesthesia) through the C4.5 decision tree, and then obtains the depth of anesthesia according to different anesthesia states. The C4.5 decision tree has a clear logical relationship and is a classic single tree structure in the decision tree algorithm. However, in the field of anesthesia depth, different patient types, different drug types, and corresponding anesthesia depths require a large amount of data to establish a correspondence. Using the C4.5 decision tree with a large amount of data is prone to over-fitting problems, even if pruning or the like does not solve the problem well. Therefore, the use of the C4.5 decision tree to determine the state of anesthesia cannot be adapted to the complex clinical anesthesia needs.

Summary of the invention

The present invention has been made in view of the above circumstances, and an object thereof is to provide an electroencephalogram-based anesthesia depth monitoring method and apparatus capable of avoiding over-fitting problems in a large data amount.

To this end, an aspect of the present invention provides an electroencephalogram-based anesthesia depth monitoring method, including: acquiring an electroencephalogram signal by a sensor; performing denoising processing on the electroencephalogram signal; and extracting the brain after denoising processing A signal characteristic of the electrical signal; and determining an anesthesia depth value based on the signal characteristic and a predetermined anesthesia depth prediction model.

In the present invention, the anesthesia depth prediction model is trained by an anesthesia depth database comprising a plurality of anesthesia full-course EEG signals and an anesthesia state corresponding to each segment of the EEG signals in each anesthesia.

In the present invention, the anesthesia depth prediction model is trained using the anesthesia depth database, and the anesthesia depth value is confirmed by the collected electroencephalogram signal and the trained anesthesia depth prediction model. In this case, over-fitting problems can be avoided in the case of large data volumes.

In an electroencephalogram-based anesthesia depth monitoring method according to an aspect of the present invention, the method further includes displaying the anesthesia depth value. Thereby, the anesthesia depth value can be obtained intuitively.

In an EEG-based anesthesia depth monitoring method according to an aspect of the present invention, the acquired EEG signal is a simulated EEG signal, and the simulated EEG signal is before the EEG signal is denoised. A pre-amplification process and an analog-to-digital conversion process are performed to obtain a digital EEG signal. Thereby, the storage of the EEG signal is facilitated.

In an electroencephalogram-based anesthesia depth monitoring method according to an aspect of the invention, the denoising process includes processing a physiological interference signal and a non-physiological interference signal. Thereby, the interference signal is removed, and the anesthesia depth value can be obtained more accurately.

In an electroencephalogram-based anesthesia depth monitoring method according to an aspect of the invention, the signal characteristics include time domain features, frequency domain features, and nonlinear domain features. Thereby, the anesthesia depth value can be obtained according to the time domain feature, the frequency domain feature and the nonlinear domain feature, and the preset anesthesia depth prediction model.

In an electroencephalogram-based anesthesia depth monitoring method according to an aspect of the invention, the time domain characteristic includes an explosion suppression ratio, the frequency domain characteristic includes an electroencephalogram related energy ratio, and the nonlinear domain characteristic includes information entropy . Thereby, the anesthesia depth value can be obtained based on the burst suppression ratio, the EEG-related energy ratio, and the information entropy and the preset anesthesia depth prediction model.

In an electroencephalogram-based anesthesia depth monitoring method according to an aspect of the present invention, the anesthesia state corresponding to each segment of the EEG signal includes awake, sedation, general anesthesia, deep anesthesia, excessive anesthesia, and no brain electrical activity. One or more. As a result, the data in the anesthesia depth database is more comprehensive.

In an electroencephalogram-based anesthesia depth monitoring method according to an aspect of the present invention, the anesthesia depth database further includes one or more of population information, drug information, and a type of surgery corresponding to an anesthesia full-course EEG signal. Thereby, the correspondence relationship between the EEG signal and the anesthesia depth in the anesthesia depth prediction model can be made wider.

In an EEG-based anesthesia depth monitoring method according to an aspect of the present invention, the predetermined anesthesia depth prediction model has a decision tree number of at least 200, and each decision tree has no more than three layers. Thereby, it is possible to avoid an over-fitting problem in the case where the amount of data is large.

Another aspect of the present invention provides an electroencephalogram-based anesthesia depth monitoring apparatus, comprising: an acquisition module that acquires an electroencephalogram signal through a sensor; a denoising module that performs denoising processing on the EEG signal; and an extraction module And extracting a signal characteristic of the electroencephalogram signal after the denoising process; and a calculation module determining the anesthesia depth value according to the signal characteristic and the preset anesthesia depth prediction model.

In the present invention, the anesthesia depth prediction model is obtained from an anesthesia depth database comprising a plurality of anesthesia full-course EEG signals and an anesthetic state corresponding to each segment of the EEG signals in each anesthesia.

In the present invention, the calculation module confirms the anesthesia depth value by the acquired EEG signal and the trained anesthesia depth prediction model, and the anesthesia depth prediction model is obtained by training the anesthesia depth database. In this case, over-fitting problems can be avoided in the case of large data volumes.

In an electroencephalogram-based anesthesia depth monitoring apparatus according to another aspect of the present invention, a display module for displaying the anesthesia depth value is further included. Thereby, the anesthesia depth value can be obtained intuitively through the display module.

In an electroencephalogram-based anesthesia depth monitoring apparatus according to another aspect of the present invention, in the acquisition module, the acquired EEG signal is a simulated EEG signal, and the EEG signal is denoised Previously, the analog EEG signal was subjected to preamplification processing and analog-to-digital conversion processing to obtain a digital EEG signal. Thereby, the anesthesia depth value can be obtained by the digital EEG signal and the preset anesthesia depth prediction model.

In an electroencephalogram-based anesthesia depth monitoring apparatus according to another aspect of the present invention, in the denoising module, the denoising process includes processing a physiological interference signal and a non-physiological interference signal. Thereby, the interference signal is removed, and the anesthesia depth value can be obtained more accurately.

In an electroencephalogram-based anesthesia depth monitoring apparatus according to another aspect of the present invention, the signal characteristics include time domain features, frequency domain features, and nonlinear domain features. Thereby, the anesthesia depth value can be obtained according to the time domain feature, the frequency domain feature and the nonlinear domain feature, and the preset anesthesia depth prediction model.

In an electroencephalogram-based anesthesia depth monitoring apparatus according to another aspect of the present invention, the time domain characteristic includes an explosion suppression ratio, the frequency domain characteristic includes an electroencephalogram related energy ratio, and the nonlinear domain characteristic includes information entropy. Thereby, the anesthesia depth value can be obtained based on the burst suppression ratio, the EEG-related energy ratio, and the information entropy and the preset anesthesia depth prediction model.

In an electroencephalogram-based anesthesia depth monitoring device according to an aspect of the present invention, the anesthesia state corresponding to each segment of the EEG signal includes awake, sedation, general anesthesia, deep anesthesia, excessive anesthesia, and no EEG activity. One or more. As a result, the data in the anesthesia depth database is more comprehensive.

In an electroencephalogram-based anesthesia depth monitoring apparatus according to another aspect of the present invention, the anesthesia depth database further includes one or more of population information, medication information, and a type of surgery corresponding to an anesthesia global electroencephalogram signal. Thereby, the correspondence relationship between the EEG signal and the anesthesia depth in the anesthesia depth prediction model can be made wider.

In an electroencephalogram-based anesthesia depth monitoring apparatus according to another aspect of the present invention, the predetermined anesthesia depth prediction model has a decision tree number of at least 200, and each decision tree has no more than three layers. . Thereby, it is possible to avoid an over-fitting problem in the case where the amount of data is large.

Another aspect of the present invention provides an electroencephalogram-based anesthesia depth monitoring apparatus including: a sensor for collecting an electroencephalogram signal; a memory for storing the collected electroencephalogram signal; and a processor The method is configured to: extract a signal characteristic of the EEG signal; determine an anesthesia depth value according to the signal characteristic and a preset anesthesia depth prediction model.

In the present invention, the anesthesia depth prediction model is obtained from an anesthesia depth database comprising a plurality of anesthesia full-course EEG signals and an anesthetic state corresponding to each segment of the EEG signals in each anesthesia. In this case, over-fitting problems can be avoided in the case of large data volumes.

In an electroencephalogram-based anesthesia depth monitoring apparatus according to another aspect of the present invention, the signal characteristic includes one or more of a time domain characteristic, a frequency domain characteristic, and a nonlinear domain characteristic. Thereby, the anesthesia depth value can be obtained according to the time domain feature, the frequency domain feature and the nonlinear domain feature, and the preset anesthesia depth prediction model.

In an electroencephalogram-based anesthesia depth monitoring apparatus according to another aspect of the present invention, the processor further performs denoising processing on the acquired electroencephalogram signal before performing the step. Thereby, the interference signal is removed, and the anesthesia depth value can be obtained more accurately.

DRAWINGS

FIG. 1 is a schematic structural view showing an electroencephalogram-based anesthesia depth monitoring apparatus according to the present embodiment.

FIG. 2 is a schematic diagram showing the structure of an acquisition module of an electroencephalogram-based anesthesia depth monitoring apparatus according to the present embodiment.

3 is a schematic structural view showing a denoising module of an electroencephalogram-based anesthesia depth monitoring device according to the present embodiment.

4 is a schematic view showing the structure of another electroencephalogram-based anesthesia depth monitoring device according to the present embodiment.

FIG. 5 is a schematic structural view showing another electroencephalogram-based anesthesia depth monitoring apparatus according to the present embodiment.

Fig. 6 is a flow chart showing a method for monitoring anesthesia depth based on electroencephalogram according to the present embodiment.

Fig. 7 is a flow chart showing another method of monitoring anesthesia depth based on electroencephalogram according to the present embodiment.

Detailed ways

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description, the same components are denoted by the same reference numerals, and the description thereof will not be repeated. In addition, the drawings are merely schematic views, and the ratio of the dimensions of the components to each other or the shape of the components may be different from the actual ones.

FIG. 1 is a schematic structural view showing an electroencephalogram-based anesthesia depth monitoring apparatus according to the present embodiment. FIG. 2 is a schematic diagram showing the structure of an acquisition module of an electroencephalogram-based anesthesia depth monitoring apparatus according to the present embodiment. 3 is a schematic structural view showing a denoising module of an electroencephalogram-based anesthesia depth monitoring device according to the present embodiment. 4 is a schematic diagram showing the configuration of a calculation module of an electroencephalogram-based anesthesia depth monitoring apparatus according to the present embodiment. The present embodiment discloses an electroencephalogram-based anesthesia depth monitoring device 1 that can be used for anesthesia monitoring.

In the present embodiment, as shown in FIG. 1 , the electroencephalogram-based anesthesia depth monitoring apparatus 1 may include an acquisition module 10 , a denoising module 20 , an extraction module 30 , and a calculation module 40 .

In this embodiment, the acquisition module 10 can acquire an EEG signal through a sensor. For example, the acquisition module 10 can acquire an EEG signal through the electrode pads. The acquisition module 10 can also acquire an EEG signal through an EEG acquisition device or the like. In addition, the EEG signal collected by the sensor may be a simulated EEG signal. The acquired EEG signals may include various EEG signals of the patient throughout the anesthesia.

In the embodiment, the acquisition module 10 may further include an amplification unit 110. The amplification unit 110 may be a preamplifier. The amplifying unit 110 can perform preamplification processing on the analog EEG signal. Thereby, the simulated EEG signal can be amplified to improve the power and signal quality of the simulated brain signal.

In this embodiment, as shown in FIG. 2, the acquisition module 10 may further include an analog to digital conversion unit 120. The analog to digital conversion unit 120 may be an analog to digital (A/D) converter. The analog to digital conversion unit 120 can perform analog to digital conversion processing on the analog EEG signal. That is, the analog to digital conversion unit 120 can convert the amplified analog EEG signal into a digital EEG signal. Specifically, the analog-to-digital conversion mainly samples the analog EEG signal and then quantizes it into a digital EEG signal. Thereby, the storage of the EEG signal is facilitated.

In the present embodiment, the electroencephalogram-based anesthesia depth monitoring device 1 may further include a storage module (not shown). The storage module can be used to receive and store the EEG signals output by the acquisition module 10. The storage module can be a memory such as a RAM, a memory stick, or the like.

In the present embodiment, as shown in FIG. 1 , the electroencephalogram-based anesthesia depth monitoring device 1 may further include a denoising module 20 . The denoising module 20 can be a processor such as a central processing unit (CPU), a micro processor (MPU), an application specific integrated circuit (ASIC), or the like. However, the embodiment is not limited thereto, and the denoising module 20 may also be a filter.

In the present embodiment, the denoising module 20 can identify and process the interfering signals in the EEG signals. That is, the denoising module 20 can be used to perform denoising processing on the EEG signal. The interference signal may include a physiological interference signal and a non-physiological interference signal.

In the embodiment, the denoising module 20 performs denoising processing on the EEG signal, and may perform denoising processing on the digital EEG signal. Specifically, the acquisition module 10 can transmit the digital EEG signal to the denoising module 20, and the denoising module 20 can receive the digital EEG signal from the interface, and then identify and process the physiological and non-physiological factors in the digital EEG signal. The resulting interference signal.

In the present embodiment, as shown in FIG. 3, the denoising module 20 may include identifying and removing the physiological interference signal unit 210. Removing the physiological interference signal unit 210 can remove the physiological interference signal. The physiological interference signal may include an eye movement interference signal and a myoelectric interference signal in the brain electrical signal. Among them, the eye movement interference signal is mainly the low frequency waveform caused by blinking and eyeball rotation, and the myoelectric interference signal is mainly the high frequency waveform caused by muscle discharge. Thereby, the interference signal is removed, and the anesthesia depth value can be obtained more accurately.

In the present embodiment, as shown in FIG. 3, the denoising module 20 may further include a non-physiological interference signal unit 220. Removing the non-physiological interference signal unit 220 may remove the non-physiological interference signal. Non-physiological interference signals may include abnormal signal amplitudes in the EEG signals, abnormal signal slopes, and electrosurgical interference. The interference signal caused by the signal amplitude and the abnormal slope is mainly the interference signal caused by the external pressing of the touch sensor. The electric knife interference is mainly caused by the impact of the high-frequency electric knife on the impact of the acquisition system during the operation.

In the present embodiment, as shown in FIG. 1, the EEG signal processing apparatus 1 may further include an extraction module 30. The extraction module 30 can be a processor such as a central processing unit (CPU), a micro processor (MPU), an application specific integrated circuit (ASIC), or the like.

In the present embodiment, the extraction module 30 can be used to extract signal characteristics of the EEG signal after the denoising process. The signal characteristics can include at least one of a time domain feature, a frequency domain feature, and a non-linear domain feature. The time domain feature can include an outbreak suppression ratio. The burst suppression ratio can be obtained by dividing the length of the suppression signal by the length of the entire segment of the signal. The frequency domain characteristics may include an EEG-related energy ratio. The energy ratio is the proportion of energy in different frequency bands of the EEG signal to the total energy. Nonlinear domain features may include information entropy. The method of calculating information entropy is to calculate the probability that each waveform amplitude in a length of EEG waveform signal appears in the entire segment of the signal.

In the present embodiment, as shown in FIG. 1, the EEG signal processing apparatus 1 may further include a calculation module 40. The computing module 40 can be a processor such as a central processing unit (CPU), a micro processor (MPU), an application specific integrated circuit (ASIC), or the like.

In the present embodiment, the calculation module 40 can be configured to determine an anesthesia depth value based on the signal characteristics and a predetermined anesthesia depth prediction model. The anesthesia depth prediction model is trained by the anesthesia depth database. The anesthesia depth database contains a plurality of anesthesia full-encephalic EEG signals and anesthesia status corresponding to each segment of EEG signals in each anesthesia.

In the present embodiment, the anesthesia depth database may include different EEG signals, signal characteristics of the EEG signals, and correspondence between each segment of the EEG signals and the anesthetic state (ie, the anesthesia depth database includes various segments of EEG signals). Correspondence between signal characteristics and anesthesia status).

In the present embodiment, the signal characteristic of the EEG signal in the anesthesia depth database may be at least one of a time domain feature, a frequency domain feature, and a nonlinear domain feature. Therefore, the correspondence between the signal characteristics of different EEG signals and the anesthetic state may be a correspondence between at least one of the time domain characteristics, the frequency domain characteristics, and the nonlinear domain characteristics of different EEG signals and the anesthesia state. In the present embodiment, the anesthesia state in the anesthesia depth database may be that the anesthesiologist indicates the current anesthesia state based on the electroencephalogram waveform and the clinical anesthesia state corresponding to each segment of the EEG signal. The anesthesia status corresponding to each segment of EEG signals includes one or more of awake, sedation, general anesthesia, deep anesthesia, excessive anesthesia, and no EEG activity.

In addition, in the present embodiment, the anesthesia depth database further includes one or more of crowd information, medication information, and type of surgery corresponding to the anesthesia full-course EEG signal.

In the present embodiment, the calculation module 40 can obtain an anesthesia depth prediction model through the above-described anesthesia depth database training. Specifically, the calculation module 40 can train the anesthesia depth prediction model through the signal characteristics of the anesthesia full-encephalic EEG signal in the anesthesia depth database and the anesthesia state corresponding to each segment of the EEG signal in each anesthesia. In some examples, computing module 40 may train anesthesia depth prediction by correspondence between at least one of time domain features, frequency domain features, and non-linear domain features of the different EEG signals in the anesthesia depth database and anesthesia status model. Thereby, the correspondence relationship between the EEG signal and the anesthesia depth in the anesthesia depth prediction model can be obtained more accurately.

In this embodiment, the predetermined number of decision trees of the anesthesia depth prediction model may include at least 200 trees, and the number of layers of each decision tree does not exceed 3 layers. However, the embodiment is not limited thereto, and for example, the number of decision trees may include at least 300 trees. In addition, for each decision tree in the random forest, an anesthetic depth value of 0 to 100 is obtained. There is a correspondence between anesthesia status and anesthesia value, as shown in Table 1.

In this case, the signal characteristics of the extraction module 30 may include one or more of a time domain feature, a frequency domain feature, and a non-linear domain feature. Thus, the calculation module 40 is capable of obtaining an anesthesia depth value based on one or more of the time domain features, the frequency domain features, and the non-linear domain features and the predetermined anesthesia depth prediction model. The time domain feature may include an explosion suppression ratio, the frequency domain feature may include an EEG related energy ratio, and the nonlinear domain feature may include information entropy. Thus, the calculation module 40 can obtain an anesthesia depth value based on the burst suppression ratio, the EEG-related energy ratio, and the information entropy and the preset anesthesia depth prediction model.

In the present embodiment, as shown in FIG. 4, the EEG signal processing apparatus 1 may further include a display module 50. Display module 50 can be a screen or other device. For example, display module 50 can be a plurality of indicator lights. Different indicators indicate different anesthesia status. The display module 50 can also be a display screen, such as a liquid crystal display device, an LED display device, or an LCD display device.

In the present embodiment, the display module 50 is configured to display an anesthesia depth value. Thereby, the anesthesia depth value can be obtained intuitively through the display module. The present embodiment is not limited to this, and an anesthesia state may be displayed, and both the anesthesia depth value and the anesthesia state may be displayed.

In the present embodiment, the anesthesia depth value is confirmed by the acquired EEG signal and the trained anesthesia depth prediction model, and the anesthesia depth prediction model is obtained by training the anesthesia depth database. In this case, over-fitting problems can be avoided in the case of large data volumes.

The following are EEG signal processing devices in some examples. FIG. 5 is a schematic structural view showing another electroencephalogram-based anesthesia depth monitoring apparatus according to the present embodiment.

In some examples, as shown in FIG. 5, the EEG signal processing apparatus 1 may include a sensor 11, a memory 12, a processor 13, and a display module 14.

In the present embodiment, the sensor 11 can collect an EEG signal. In some examples, sensor 11 can be an electrode sheet. The EEG signal acquired by the sensor 11 may be a simulated EEG signal.

In the present embodiment, the electroencephalogram signal processing device 1 may further include a preamplifier (not shown) and an analog-to-digital (A/D) converter (not shown). Since the quality of the simulated EEG signal collected by the sensor 11 may be poor, the acquired EEG signal can be preamplified by the preamplifier. Thereby, the EEG signal can be amplified to improve the power and signal quality of the EEG signal. An analog-to-digital (A/D) converter can simulate the conversion of EEG signals into digital EEG signals. Thereby, the storage of the EEG signal is facilitated.

In the present embodiment, as shown in FIG. 5, the memory 12 can store the collected EEG signals. The memory can be RAM, FIFO, memory stick, and the like.

In the present embodiment, as shown in FIG. 5, the processor 13 can extract the signal characteristics of the EEG signal. The signal characteristics include one or more of a time domain feature, a frequency domain feature, and a nonlinear domain feature. The time domain feature can include an outbreak suppression ratio. The burst suppression ratio can be obtained by dividing the length of the suppression signal by the length of the entire segment of the signal. The frequency domain characteristics may include an EEG-related energy ratio. The energy ratio is the proportion of energy in different frequency bands of the EEG signal to the total energy. Nonlinear domain features may include information entropy. The method of calculating information entropy is to calculate the probability that each waveform amplitude in a length of EEG waveform signal appears in the entire segment of the signal.

In the present embodiment, the processor 13 may also determine an anesthesia depth value based on the signal characteristics and a predetermined anesthesia depth prediction model. The anesthesia depth prediction model is trained by the anesthesia depth database. The anesthesia depth database contains a plurality of anesthesia full-encephalic EEG signals and anesthesia status corresponding to each segment of EEG signals in each anesthesia. The anesthesia status corresponding to each segment of the EEG signal includes one or more of awake, sedation, general anesthesia, deep anesthesia, excessive anesthesia, and no EEG activity. In addition, the anesthesia depth database also includes one or more of population information, medication information, and type of surgery corresponding to the anesthesia signal. In some examples, the predetermined anesthesia depth prediction model has a decision tree number of at least 200 and the number of layers per decision tree does not exceed three. Processor 13 is similar to computing module 40 described above.

In this embodiment, the processor 13 can also perform denoising processing on the acquired EEG signals. The denoising process of the processor 13 can denoise the acquired EEG signals by an algorithm. Of course, in addition to the processor 13, the denoising of the EEG signal can be realized by the hardware circuit. For the specific denoising process, the denoising process in the denoising module 20 described above may be referred to.

In the present embodiment, as shown in FIG. 5, the EEG signal processing apparatus 1 may further include a display module 14. The display module 14 is similar to the display module 50 described above.

Hereinafter, the electroencephalogram-based anesthesia depth monitoring method according to the present embodiment will be described in detail with reference to FIGS. 6 and 7.

In the present embodiment, as shown in FIG. 6, the electroencephalogram-based anesthesia depth monitoring method includes acquiring an electroencephalogram signal by a sensor (step S101).

In step S101, the EEG signal can be acquired by the sensor, and the EEG signal of the patient can also be collected by the EEG device. The EEG device can be any of a variety of devices known to those skilled in the art for collecting EEG signals.

In the present embodiment, the EEG signal collected by the sensor may be a simulated EEG signal. Simulated EEG signals can usually be converted into digital EEG signals.

In the present embodiment, a pre-amplification process and an analog-to-digital conversion process are performed on the simulated EEG signal to obtain a digital EEG signal. The preamplification process can amplify the simulated EEG signal and improve the power and signal quality of the simulated EEG signal. The analog to digital conversion process can convert the amplified analog EEG signal into a digital EEG signal. Thereby, the storage of the EEG signal is facilitated.

In the present embodiment, as shown in FIG. 6, the electroencephalogram-based anesthesia depth monitoring method may further include performing denoising processing on the electroencephalogram signal (step S102).

In step S102, the denoising process includes identifying and processing the physiological interference signal and the non-physiological interference signal. Denoising the EEG signal can improve the quality of the EEG signal. Thereby, the interference signal is removed, and the anesthesia depth value can be obtained more accurately.

In the present embodiment, the physiological interference signal may include an eye movement interference signal and a myoelectric interference signal.

In the present embodiment, the eye movement interference signal is usually a low frequency waveform caused by blinking and eyeball rotation. For example, the eye movement interference signal may be a frequency band below 10hz in the EEG signal. For the eye movement interference signal frequency band, it can be identified by analyzing the abnormal low frequency signal in the EEG waveform, and the segment data is deleted after the eye movement interference is recognized.

In the present embodiment, the myoelectric interference signal is usually a high frequency mutation caused by muscle discharge. For example, the myoelectric interference signal can be a frequency band above 50hz in the EEG signal. For the frequency band of the myoelectric interference signal, it can be identified by analyzing the high frequency abnormal mutation in the EEG waveform, and the data is removed after the high frequency interference of the myoelectric is recognized.

In the present embodiment, the non-physiological interference signal may include abnormal signal amplitude, abnormal signal slope, and electric knife interference.

In the present embodiment, the abnormal signal amplitude and the abnormality of the signal slope are usually caused by the external pressing of the touch sensor. The signal amplitude usually refers to the maximum value of the signal fluctuation over a period of time. The slope of the signal is usually the maximum of the difference between two adjacent sampling points. For the abnormal signal amplitude and abnormal signal slope, the segment data can be eliminated after the amplitude and slope abnormal interference are recognized.

In the present embodiment, the electrosurgical interference is usually caused by an impact of the high-frequency electric knife on the acquisition system during the operation. The electric knife interference can be identified by abnormal fluctuations in the high frequency band. The high frequency band generally refers to the frequency band above 500hz. The segment data is rejected after the electric knife interference is recognized.

In step S102, the denoising process is performed on the EEG signal, and the digital EEG signal may be denoised to obtain an effective digital EEG signal.

In the present embodiment, as shown in FIG. 6, the electroencephalogram-based anesthesia depth monitoring method may further include extracting a signal characteristic of the electroencephalogram signal after the denoising process (step S103).

In this embodiment, the signal feature may include at least one of a time domain feature, a frequency domain feature, and a nonlinear domain feature.

In the present embodiment, the time domain feature can calculate the burst suppression ratio (BSR). Burst suppression is a waveform that occurs when the brain is under deep anesthesia. It is characterized by a small amplitude suppression waveform alternating with a large burst waveform. When the EEG signal fluctuation range is less than 5 uv and the duration exceeds 0.5 second, it can be considered as a suppression signal.

In the present embodiment, the burst suppression ratio can be obtained by dividing the length of the suppression signal by the length of the entire segment of the signal. The burst suppression ratio is a quantitative indicator for evaluating the burst suppression waveform. The calculation formula for the burst suppression ratio is as follows:

However, the embodiment is not limited thereto. For example, the time domain feature may further include a signal amplitude, a maximum and minimum value, a difference between adjacent points of the signal (slope of the waveform), an extreme value of the adjacent point difference of the signal, and a zero crossing of the waveform. The situation (that is, the number of times the waveform crosses the zero point per unit time).

In the present embodiment, the frequency domain characteristic may be an EEG-related energy ratio. The energy ratio is the proportion of energy in different frequency bands of the EEG signal to the total energy.

In this embodiment, different frequency bands may include a delta wave with a frequency range of 0.5 to 3 hz, a theta wave with a frequency range of 4 to 7 hz, an alpha wave with a frequency range of 8 to 13 hz, and a frequency range of 14 to 30 hz. Beta wave. The total energy frequency can range from 0.5 to 47 hz.

In this embodiment, energy of different frequency bands can be calculated by the following steps:

First, perform a Fourier transform on a piece of EEG signal according to formula (2):

Then, according to the Parseval theorem, the energy is equal to the sum of the squares of the Fourier transforms, ie the energy is calculated according to equation (3):

By calculating the energy of each frequency band of δ, θ, α, and β by the above two steps, the energy level of the corresponding frequency band can be obtained, and the energy ratio of each frequency band is obtained by comparing the energy level of the corresponding frequency band with the total energy.

However, the embodiment is not limited thereto. For example, the frequency domain feature may further include a frequency characteristic of the signal, a frequency ratio of the different frequency bands, a power spectrum of the specific frequency band, and a ratio thereof, and typically includes a 95% edge frequency (Spectral Edge Frequency: SEF). , 50% Median Frequency (MF), Total Power (TP), Peak Power Frequency (PPF), etc.

In the present embodiment, the nonlinear domain feature may be information entropy. The calculation method of information entropy is to calculate the probability Pi of each waveform amplitude in the EEG waveform signal of length n in the whole segment of the signal, as shown in formula (4).

However, the embodiment is not limited thereto. For example, the nonlinear domain feature may further include spectral entropy, complexity, sample entropy, and the like.

In the present embodiment, as shown in FIG. 6, the electroencephalogram-based anesthesia depth monitoring method may further include determining an anesthesia depth value based on the signal characteristics and the preset anesthesia depth prediction model (step S104). The anesthesia depth prediction model is trained by the anesthesia depth database. The anesthesia depth database contains a plurality of anesthesia full-encephalic EEG signals and anesthesia status corresponding to each segment of the EEG signal.

In the present embodiment, the signal characteristic of the EEG signal in the anesthesia depth database may be at least one of the features involved in step S103. That is, the signal characteristics may be at least one of a time domain feature, a frequency domain feature, and a nonlinear domain feature. Therefore, the correspondence between the signal characteristics of different EEG signals and the anesthetic state may be a correspondence between at least one of the time domain characteristics, the frequency domain characteristics, and the nonlinear domain characteristics of different EEG signals and the anesthesia state.

In the present embodiment, the anesthesia state in the anesthesia depth database may be that the anesthesiologist indicates the current anesthesia state based on the electroencephalogram waveform and the clinical anesthesia state corresponding to each segment of the EEG signal. For example, in an anesthesia full-course EEG signal, the anesthesia status corresponding to each segment of the EEG signal may include one or more of awake, sedation, general anesthesia, deep anesthesia, excessive anesthesia, and no EEG activity. As a result, the data in the anesthesia depth database is more comprehensive.

In the present embodiment, the anesthesia depth database further includes one or more of population information, medication information, and type of surgery corresponding to the anesthesia full-course EEG signal. Thereby, it is possible to solve the problem of the patient with anesthesia depth and the coverage of the drug/surgery type.

In the present embodiment, the population coverage of the database refers to the coverage of the whole age group, which is divided into children (0 to 13 years old), adults (13 to 60 years old), and the elderly (over 60 years old). The drug coverage is mainly Refers to anesthesia data collection for sedative, analgesic, and muscle relaxants used in mainstream venous and gas anesthesia, including but not limited to the following types: propofol, etomidate, midazolam, dextromethorphan, and Fluoroether, sevoflurane, desflurane, fentanyl, remifentanil, alfentanil, sufentanil, rocuronium bromide, vecuronium bromide, atracurium, etc. General surgery.

In the present embodiment, the signal characteristics of the different EEG signals and the corresponding anesthesia state can be stored in the anesthesia depth database. The anesthesia depth prediction model was trained using signal characteristics and anesthesia status in the anesthesia depth database.

In the present embodiment, the predetermined number of decision trees for the anesthesia depth prediction model is at least 200, and the number of layers per decision tree does not exceed three. However, the embodiment is not limited thereto, and for example, the number of decision trees may include at least 300 trees. Thereby, it is possible to avoid the problem of over-fitting problems and insufficient model extensibility in the case where the amount of data is large.

In addition, in the present embodiment, an anesthesia depth value of 0 to 100 is obtained for each decision tree in the random forest. There is a correspondence between anesthesia status and anesthesia value, as shown in Table 1.

Table 1 Correspondence table between anesthesia depth value and anesthesia depth status

麻醉状态Anesthesia	麻醉深度值范围Range of anesthesia depth values
清醒wide awake	100-80100-80
镇静、睡眠状态Calm, sleep state	79-6079-60
一般麻醉General anes	59-4059-40
深度麻醉Deep anesthesia	39-2039-20
过度麻醉Over-anesthesia	19-119-1
无脑电活动No brain activity	00

In the present embodiment, the output anesthesia depth values of all decision trees in the anesthesia depth prediction model are averaged, and the final output anesthesia depth value can be obtained.

In this embodiment, each random forest decision tree can be regarded as a weak classifier. The anesthesia depth prediction model is a combination of a large number of weak classifiers. For example, the anesthesia depth prediction model has 300 decision trees. That is, the anesthesia depth prediction model is a combination of 300 weak classifiers. The results of the 300 decision trees are averaged, which is equivalent to a plurality of weak classifiers forming a strong classifier. This makes the classification results more accurate, and it is more accurate to reflect the depth of anesthesia. Moreover, such a structure is very robust and scalable, and is well suited for a variety of drugs and anesthesia depths covered by different populations.

In this embodiment, a single decision tree is trained to generate by:

In the first step, for N samples, there are N samples randomly selected (note that each time a sample is randomly selected, the sample is put back for the next random selection), and the selected N samples are used for the training decision tree. . The sample may be a collection of anesthesia depth database sequences that have been stored in the order of "feature-label". The feature may be at least one of the signal characteristics described above. The annotation may be one of the anesthesia states corresponding to the above-mentioned various segments of the EEG signal under the corresponding feature. In the second step, each sample has M attributes. When the decision tree node needs to be split, m attributes are randomly selected from the M attributes, and then information gain is used from the m attributes to select an attribute as the classification attribute of the node. . The attribute can be a characteristic of the signal in the set of anesthesia depth database sequences. In the third step, each node in the decision tree formation process is split according to the second step. There are two stopping conditions for the split. First, the split feature selected in this split is the parent node feature of the node, and the split ends. Second, the number of layers of the decision tree after the split reaches the specified number of layers, and the split stops.

In the present embodiment, a large number of decision trees are established in accordance with the above three steps to form an anesthesia depth prediction model.

In this embodiment, the signal characteristics may include one or more of a time domain feature, a frequency domain feature, and a non-linear domain feature. Thereby, the anesthesia depth value can be obtained according to one or more of the time domain feature, the frequency domain feature, and the nonlinear domain feature and the preset anesthesia depth prediction model. Additionally, the time domain feature can include an explosion suppression ratio, the frequency domain feature can include an EEG related energy ratio, and the nonlinear domain feature can include information entropy. Thereby, the anesthesia depth value can be obtained based on the burst suppression ratio, the EEG-related energy ratio, and the information entropy and the preset anesthesia depth prediction model.

In the present embodiment, as shown in FIG. 7, the electroencephalogram-based anesthesia depth monitoring method may further include displaying an anesthesia depth value (step S105).

In the present embodiment, step S105 may directly display the anesthesia depth value, or may display an anesthesia state, or may display both the anesthesia depth value and the anesthesia state. Thereby, the depth of anesthesia or the state of anesthesia can be obtained intuitively.

In the present embodiment, the anesthesia depth prediction model is trained using the anesthesia depth database, and the anesthesia depth value is confirmed by the collected electroencephalogram signal and the trained anesthesia depth prediction model. In this case, over-fitting problems can be avoided in the case of large data volumes.

While the invention has been described in detail with reference to the drawings and embodiments, it is understood that The present invention may be modified and changed as needed by those skilled in the art without departing from the spirit and scope of the invention, and such modifications and variations are within the scope of the invention.

Claims

An electroencephalogram-based anesthesia depth monitoring method, characterized in that

include:

Obtaining an EEG signal through a sensor;

Denoising the EEG signal;

Extracting signal characteristics of the EEG signal after the denoising process; and

Anesthesia depth values are determined based on the signal characteristics and a predetermined anesthesia depth prediction model.
The method of claim 1 wherein

The anesthesia depth prediction model is obtained from an anesthesia depth database comprising a plurality of anesthesia full-course EEG signals and an anesthetic state corresponding to each segment of the EEG signals in each anesthesia.
Method according to claim 1 or 2, characterized in that

The acquired EEG signal is a simulated EEG signal.

Before denoising the EEG signal, performing preamplification processing and analog-to-digital conversion processing on the simulated EEG signal to obtain a digital EEG signal.
Method according to claim 1 or 2, characterized in that

The denoising process includes processing the physiological interference signal and the non-physiological interference signal.
Method according to claim 1 or 2, characterized in that

The signal features include time domain features, frequency domain features, and non-linear domain features.
The method of claim 5 wherein:

The time domain feature includes an explosion suppression ratio, the frequency domain feature including an EEG related energy ratio, and the nonlinear domain characteristic includes information entropy.
Method according to claim 1 or 2, characterized in that

The anesthesia state corresponding to each segment of the EEG signal includes one or more of awake, sedation, general anesthesia, deep anesthesia, excessive anesthesia, and no EEG activity.
Method according to claim 1 or 2, characterized in that

The anesthesia depth database also includes one or more of population information, medication information, and type of surgery corresponding to an anesthetic full-course EEG signal.
Method according to claim 1 or 2, characterized in that

The predetermined anesthesia depth prediction model has a decision tree number of at least 200 trees, and the number of layers of each decision tree does not exceed three layers.
An electroencephalogram-based anesthesia depth monitoring device, characterized in that

include:

An acquisition module that acquires an EEG signal through a sensor;

a denoising module that performs denoising processing on the EEG signal;

An extraction module that extracts signal characteristics of the EEG signal after denoising processing;

A calculation module that determines an anesthesia depth value based on the signal characteristics and a predetermined anesthesia depth prediction model.
The device of claim 10 wherein:

The anesthesia depth prediction model is obtained from an anesthesia depth database comprising a plurality of anesthesia full-course EEG signals and an anesthetic state corresponding to each segment of the EEG signals in each anesthesia.
Device according to claim 10 or 11, characterized in that

In the acquisition module, the acquired EEG signal is a simulated EEG signal, and

Before denoising the EEG signal, performing preamplification processing and analog-to-digital conversion processing on the simulated EEG signal to obtain a digital EEG signal.
Device according to claim 10 or 11, characterized in that

In the denoising module, the denoising process includes processing a physiological interference signal and a non-physiological interference signal.
Device according to claim 10 or 11, characterized in that

The signal features include time domain features, frequency domain features, and non-linear domain features.
The device of claim 14 wherein:

The time domain feature includes an explosion suppression ratio, the frequency domain feature including an EEG related energy ratio, and the nonlinear domain characteristic includes information entropy.
Device according to claim 10 or 11, characterized in that

The anesthesia state corresponding to each segment of the EEG signal includes one or more of awake, sedation, general anesthesia, deep anesthesia, excessive anesthesia, and no EEG activity.
Device according to claim 10 or 11, characterized in that

The anesthesia depth database also includes one or more of population information, medication information, and type of surgery corresponding to an anesthetic full-course EEG signal.
Device according to claim 10 or 11, characterized in that

The predetermined anesthesia depth prediction model has a decision tree number of at least 200 trees, and the number of layers of each decision tree does not exceed three layers.
An electroencephalogram-based anesthesia depth monitoring device, characterized in that

include:

Sensor, collecting EEG signals;

a memory storing the collected EEG signals;

Processor, perform the following steps:

Extracting signal characteristics of the EEG signal;

Anesthesia depth values are determined based on the signal characteristics and a predetermined anesthesia depth prediction model.
The device according to claim 19, characterized in that

The signal characteristics include one or more of a time domain feature, a frequency domain feature, and a non-linear domain feature.
The method of claim 20 wherein:

The time domain feature includes an explosion suppression ratio, the frequency domain feature including an EEG related energy ratio, and the nonlinear domain characteristic includes information entropy.
The apparatus according to claim 19, wherein said processor further performs denoising processing on said acquired electroencephalogram signal before said step of extracting said signal characteristic of said electroencephalogram signal.
The method of claim 19 wherein:

The anesthesia depth prediction model is obtained from an anesthesia depth database comprising a plurality of anesthesia full-course EEG signals and an anesthetic state corresponding to each segment of the EEG signals in each anesthesia.
The method of claim 19 wherein:

The anesthesia depth database also includes one or more of population information, medication information, and type of surgery corresponding to an anesthetic full-course EEG signal.
The method of claim 19 wherein:

The predetermined anesthesia depth prediction model has a decision tree number of at least 200 trees, and the number of layers of each decision tree does not exceed three layers.