Deep learning-based electroencephalogram (EEG) data model inversion anesthesia brain state evaluation method

Abstract

The invention discloses a deep learning-based brain anesthesia state inversion evaluation method based on an EEG data model. Firstly, a neuron group model suitable for an anesthesia state is constructed, and the neuron group model is based on a Jansen-Rit model, covers multiple neuron groups and has a specific kinetic equation. Then determining a value range of model parameters, sampling to generate a combination, and generating a simulated EEG signal to construct a training set; designing a deep learning model for training, and learning a parameter mapping relation; in practical application, a tested EEG signal is collected and preprocessed and then is input into a trained model inversion parameter, and then the anesthesia depth index ADI is calculated according to the parameter so as to evaluate the anesthesia brain state. The method has high accuracy in evaluation of anesthesia depth, and provides an effective means for anesthesia monitoring.

Description

Technical Field

[0001] The present invention relates to the fields of biomedical engineering, neuroscience, and anesthesia monitoring, and particularly relates to a method for evaluating anesthetic brain state by inverting an EEG data model based on deep learning. Background Art

[0002] Traditional methods for evaluating anesthetic depth mainly rely on the experience of anesthesiologists and physiological indicators of the subjects, such as heart rate, blood pressure, etc. These methods have certain subjectivity and lag, and may not accurately reflect the brain state of the subjects.

[0003] Electroencephalogram (EEG), as a non-invasive neuroelectrophysiological monitoring means, can record the electrical activities of the cerebral cortex in real time and is considered an effective tool for evaluating anesthetic depth. Currently, there are some anesthetic depth monitoring devices based on EEG signals, such as Bispectral Index (BIS) monitors. However, these devices are mostly based on empirical formulas and statistical methods, lacking in-depth understanding of neurophysiological mechanisms and may produce misjudgments in specific situations.

[0004] The Neural Mass Model (NMM) is a mathematical model that describes brain activities from a neurophysiological perspective and can simulate the collective dynamic behaviors of a large number of neuron populations. By adjusting the model parameters, EEG signals in different brain states can be simulated. However, traditional model parameter estimation methods, such as optimization algorithms, have problems such as high computational complexity and being easily trapped in local optima, making it difficult to meet the requirements of real-time monitoring.

[0005] In recent years, deep learning has achieved remarkable results in the fields of signal processing and pattern recognition. Extracting features from EEG signals using deep learning models and performing classification and regression analysis have become research hotspots. However, combining deep learning with the neural mass model to invert model parameters from EEG signals and achieve the evaluation of anesthetic brain state is still a challenge. Summary of the Invention

[0006] The purpose of the present invention is to overcome the deficiencies of the above-mentioned prior art and provide a method for evaluating anesthetic brain state by inverting an EEG data model based on deep learning.

[0007] The present invention is implemented as follows. A method for evaluating anesthetic brain state by inverting an EEG data model based on deep learning includes the following steps:

[0008] Step S1: Construct a neural mass model applicable to brain electrical activities under anesthetic state;

[0009] Step S2: Determine the value range of the model parameters, generate multiple parameter combinations; generate simulated EEG signals according to the parameter combinations to form a training dataset;

[0010] Step S3: Design and train a deep learning model for the model parameter space, and learn the parameter mapping relationship from the simulated EEG signals;

[0011] Step S4: Apply the trained deep learning model for the model parameter space to the actual EEG data for model parameter inversion;

[0012] Step S5: Calculate the anesthesia depth index according to the estimated model parameters to achieve real-time evaluation of the anesthetic brain state;

[0013] Preferably, in step S1, the neural population model applicable to electroencephalogram activities under anesthesia is based on the Jansen-Rit model and consists of an excitatory neural population and an inhibitory neural population, including a pyramidal cell population PC, an excitatory interneuron population EI, and an inhibitory interneuron population II.

[0014] Preferably, in step S1, the description of the model dynamic equation of the neural population model applicable to electroencephalogram activities under anesthesia includes a postsynaptic membrane potential transfer function, the average firing rate of the neural population, and a state equation:

[0015] The postsynaptic membrane potential transfer function includes the postsynaptic membrane potential response functions h e (t) and h i (t) of the excitatory neural population and the inhibitory neural population, specifically as follows:

[0016]

[0017] Among them, A and B are the excitatory synaptic gain and the inhibitory synaptic gain respectively; a and b are the reciprocals of the time constants of the excitatory neural population and the inhibitory neural population; t represents the time; exp() represents the natural exponential function;

[0018] Average firing rate function:

[0019] The average firing rate S(v) of the neural population converts the postsynaptic membrane potential v into the average impulse density of action potentials:

[0020]

[0021] Among them, e0 is the maximum firing rate, v0 is the membrane potential corresponding to the half-maximum firing rate, and r is the slope parameter;

[0022] The state equation is specifically as follows:

[0023]

[0024] Where y1 is the membrane potential of the pyramidal cell group, y2 and y3 are the membrane potentials of the excitatory neuron group and the inhibitory neuron group, respectively; y4, y5, and y6 are the derivatives corresponding to y1, y2, and y3, that is, the rate of change of their membrane potentials; p(t) is the external input; C1 and C2 represent the average connectivity coefficients of the incoming and outgoing excitatory circuits, respectively; C3 and C4 represent the average connectivity coefficients of the incoming and outgoing inhibitory circuits, respectively; represents the derivative of x.

[0025] Preferably, in step S2, each parameter is randomly selected within the value range of the model parameter to generate multiple parameter combinations; the parameters selected in different parameter combinations are different from each other;

[0026] The model parameters include excitatory synaptic gain A, inhibitory synaptic gain B, the inverse time constant a of the excitatory neuron group, the inverse time constant b of the inhibitory neuron group, the average connection coefficient C1 of the excitatory circuit, the average connection coefficient C2 of the excitatory circuit, the average connection coefficient C3 of the inhibitory circuit, and the average connection coefficient C4 of the inhibitory circuit;

[0027] Preferably, in step S2, for each parameter combination, the dynamic equation is numerically integrated to generate a corresponding simulated EEG signal:

[0028] V i (t) = y 2i (t)-y 3i (t) Formula (5)

[0029] where y 2i (t), y 3i (t) represents the membrane potential of the excitatory neuron group and the inhibitory neuron group in the i-th group of simulated EEG;

[0030] Preferably, in step S2, the training data set is formed by simulated EEG signals and corresponding model parameters;

[0031] Preferably, in step S2, the simulated EEG signal needs to be preprocessed before being input into the deep learning model of the model parameter space;

[0032] Preferably, in step S3, the deep learning model of the model parameter space includes an input layer, a convolutional layer, a pooling layer, a recurrent layer and a fully connected layer;

[0033] The input layer receives EEG signals;

[0034] The convolutional layer adopts a one-dimensional convolutional neural network (CNN) and uses multiple convolution kernels to extract local temporal features from the input;

[0035] The pooling layer uses max pooling to reduce the feature dimension of the output of the convolutional layer;

[0036] The recurrent layer adopts a long short-term memory network layer to capture the long-term dependence relationship of the time series of the output of the pooling layer;

[0037] The fully connected layer unfolds the features extracted by the recurrent layer and maps them to the model parameter space to obtain the estimated model parameters

[0038] The output layer outputs the estimated model parameters

[0039] The loss function of the deep learning model in the model parameter space is as follows:

[0040]

[0041] where θ i represents the true value of the model parameters.

[0042] Preferably, step S4 is specifically:

[0043] Collect the EEG signals of the patient and preprocess the EEG signals;

[0044] Input the preprocessed EEG signal segment into the deep learning model in the trained model parameter space to obtain the estimated model parameters

[0045] Preferably, step S5 is specifically:

[0046] Calculate the anesthesia depth index ADI according to the estimated model parameters:

[0047]

[0048] where w1, w2, w3 are weight coefficients;

[0049] Divide the anesthesia depth into different levels according to the anesthesia depth index ADI.

[0050] The beneficial effects of the present invention at least include the following:

[0051] Based on EEG data and the neural mass model, the present invention calculates the anesthesia depth index by inversely calculating the model parameters from EEG signals through deep learning, which can effectively improve the evaluation accuracy and more accurately reflect the anesthetic brain state.

[0052] The neural mass model adopted by the present invention describes brain activities from the perspective of neurophysiology, can simulate EEG signals in different brain states, combines deep learning to realize the inversion of model parameters, deeply integrates neurophysiological mechanisms, and enhances the evaluation reliability.

[0053] The deep learning model designed by the present invention has a reasonable structure. By using convolutional layers, pooling layers, recurrent layers, fully connected layers, etc. to extract features and map the parameter space, it can improve the calculation efficiency while ensuring the accuracy of parameter estimation, and can evaluate the anesthesia brain state in real time. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0055] Figure 1 is a flowchart of the anesthesia brain state evaluation method provided by the embodiment of the present invention.

[0056] Figure 2 is a schematic diagram of the model structure of the neural population model applicable to electroencephalogram (EEG) activities under anesthesia.

[0057] Figure 3 is a comparison chart of each parameter and bis. (a) is a correlation chart of the parameters related to excitatory connections and bis, (b) is a correlation chart of the parameters related to inhibitory connections and bis, (c) is a correlation chart of excitatory synaptic gain and bis, (d) is a correlation chart of inhibitory synaptic gain and bis, (e) is a correlation chart of excitatory time constant and bis, and (f) is a correlation chart of inhibitory time constant and bis.

[0058] Figure 4 is the SVM classification result in binary classification. (a) is the classification result of the model, and (b) is the true reference. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0059] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention.

[0060] As Figure 1 shown, the embodiment of the present invention provides an EEG data model inversion anesthesia brain state evaluation method. The brain state evaluation here is not for treatment purposes or diagnostic purposes. The method includes the following steps:

[0061] Step S1: Construct a neural population model applicable to EEG activities under anesthesia;

[0062] Step S2: Determine the value range of the model parameters, generate multiple parameter combinations; generate simulated EEG signals according to the parameter combinations to form a training data set;

[0063] Step S3: Design and train a deep learning model for the model parameter space to learn the parameter mapping relationship from the simulated EEG signals; Step S4: Apply the trained deep learning model of the model parameter space to the actual EEG data for model parameter inversion;

[0064] Step S5: Calculate the anesthesia depth index (ADI) based on the estimated model parameters to achieve real-time assessment of the anesthetic brain state;

[0065] Specifically, in Step S1, construct a neural population model suitable for electroencephalogram activities under anesthesia, such as Figure 2 the Jansen-Rit model. This model describes the interactions between cortical excitatory neural populations, inhibitory neural populations, and thalamic inputs, and can simulate the generation process of EEG signals. It consists of excitatory and inhibitory neural populations, including pyramidal cell populations (PC), excitatory interneuron populations (EI), and inhibitory interneuron populations (II).

[0066] The model dynamic equation description of the neural population model suitable for electroencephalogram activities under anesthesia includes the postsynaptic membrane potential transfer function, the average firing rate of the neural population, and the state equation:

[0067] The postsynaptic membrane potential transfer function includes the postsynaptic membrane potential response functions h e (t) and h i (t) as follows:

[0068]

[0069] where A and B are the excitatory synaptic gain and inhibitory synaptic gain respectively; a and b are the reciprocals of the time constants of the excitatory and inhibitory neural populations; t represents the time; exp() represents the natural exponential function;

[0070] Average firing rate function:

[0071] The average firing rate S(v) of the neural population is represented by the Sigmoid function:

[0072]

[0073] where e0 is the maximum firing rate, v0 is the membrane potential corresponding to the half-maximum firing rate, and r is the slope parameter;

[0074] The state equation is as follows:

[0075]

[0076] Among them, y1 is the membrane potential of the pyramidal cell population, y2 and y3 are the membrane potentials of the excitatory neuron population and the inhibitory neuron population respectively, y4, y5, and y6 are the derivatives of y1, y2, and y3 respectively, that is, the change rate of their membrane potentials, p(t) is the external input, and C1 - C4 are the connection strength coefficients. C1 and C2 represent the average connection coefficients connecting to and out of the excitatory circuit respectively, and C3 and C4 represent the average connection coefficients connecting to and out of the inhibitory circuit respectively; Represents the derivative of x;

[0077] Specifically, step S2 is specifically:

[0078] 2.1 Parameter setting

[0079] According to the influence of anesthesia on neuron activity, set the value range of the model parameters. The value range of the excitatory synaptic gain A is 3 to 8 mV; the value range of the inhibitory synaptic gain B is 20 to 100 mV. The value ranges of the time constants a and b are 50 to 100 s; the value ranges of the coupling coefficients C1 and C2 are 0.8 to 1.2. Other parameters such as the maximum firing rate e0, the membrane potential v0 corresponding to the half - maximum firing rate, and the slope parameter r are fixed at 2.5 s, 6 mV, and 0.56 mV respectively.

[0080] 2.2 Simulation data generation

[0081] Adopt the Latin hypercube sampling method to generate N groups of parameter combinations θ i ={A i , B i , a i , b i , C 1i , C 2i}, where i = 1, 2, …, N. For each group of parameters, use the fourth - order Runge - Kutta method to numerically integrate the kinetic equation, set the time step as Δt = 0.001 s, and the total time as T = 10 s to generate the corresponding simulated EEG signal:

[0082] V i (t)=y 2i (t)-y 3i (t).

[0083] Among them, y 2i (t) and y 3i (t) represent the membrane potentials of the excitatory neuron population and the inhibitory neuron population in the i - th group of simulated EEG;

[0084] Preprocess the simulated EEG signals, including band-pass filtering (0.5 Hz to 45 Hz) to remove low-frequency drift and high-frequency noise, normalization (zero mean, unit variance) to eliminate the influence of dimension, and signal segmentation (e.g., each segment is 5 seconds, sampling rate 256 Hz). A training dataset is formed from the simulated EEG signals and the corresponding model parameters.

[0085] Specifically, step S3 is specifically as follows:

[0086] 3.1 Training data preparation

[0087] The training data consists of simulated EEG signal segments and the corresponding model parameters. The input data is an EEG signal segment with a shape of (T, 1), where T = 1280 (corresponding to a 5-second signal length, sampling rate 256 Hz); the target output is the corresponding model parameter θ = {A, B, a, b, C1, C2}.

[0088] 3.2 Model structure design

[0089] Design a deep learning model that can extract features from EEG signals and map them to the model parameter space. The specific structure of the model includes an input layer, a convolutional layer, a pooling layer, a recurrent layer, and a fully connected layer. The input layer accepts EEG signal segments; the convolutional layer uses a one-dimensional convolutional neural network (CNN) with multiple convolutional kernels to extract local temporal features; the pooling layer uses max pooling to reduce the feature dimension; the recurrent layer adds a long short-term memory network (LSTM) layer to capture the long-term dependencies of the time series; the fully connected layer unfolds the extracted features and maps them to the model parameter space; the output layer outputs the estimated model parameters. Use a linear activation function.

[0090] 3.3 Model training

[0091] The deep learning model in the model parameter space uses the mean squared error (MSE) as the loss function:

[0092]

[0093] where θ i represents the true value of the model parameter.

[0094] Adopt the Adam optimizer, set the initial learning rate to 0.001, and the batch size to 32. To prevent overfitting, use a Dropout layer with a ratio of 0.5, and monitor the loss of the validation set during training, using the early stopping strategy. When the loss of the validation set no longer decreases for a consecutive number of epochs (e.g., 10 epochs), stop training. After multiple iterative trainings, the mean squared error of the parameter estimation of the model on the test set reaches a low level, and the estimation accuracy of each parameter is relatively high.

[0095] Specifically, step S4 is as follows:

[0096] 4.1 EEG Data Acquisition and Preprocessing

[0097] During the surgical anesthesia process, a high-precision EEG device is used to collect the patient's electroencephalogram signals with a sampling rate of 256 Hz. The EEG signals are filtered (from 0.5 Hz to 45 Hz) to remove low-frequency and high-frequency noises, and the independent component analysis (ICA) method is used to remove artifacts such as electrooculogram and electromyogram. The continuous EEG signals are segmented into windows of 5 seconds, and the overlapping part can be set to 50% to increase the data volume and improve the time resolution.

[0098] 4.2 Parameter Inversion

[0099] The preprocessed EEG signal segments are input into the trained deep learning model to obtain the estimated model parameters. Since the model is trained on simulated data, to adapt to the actual data, model fine-tuning may be required. Use a small amount of labeled actual data to retrain some layers of the model to improve the estimation accuracy.

[0100] Specifically, step S5 is as follows:

[0101] According to the estimated model parameters, calculate the Anesthesia Depth Index (ADI):

[0102]

[0103] Among them, w1, w2, w3 are weight coefficients, which can be determined through clinical data. Determining the weight coefficients requires using a certain number of clinical samples to establish a regression model using the actual anesthesia depth (such as the BIS index or clinical score) and the estimated model parameters.

[0104] According to the value of ADI, set thresholds τ1, τ2, τ3 to divide the anesthesia depth into different levels:

[0105] Awake: ADI ≥ τ1.

[0106] Light anesthesia: τ2 ≤ ADI < τ1.

[0107] Appropriate anesthesia: τ3 ≤ ADI < τ2.

[0108] Deep anesthesia: ADI < τ3.

[0109] By analyzing the change of ADI over time, the anesthesia depth can be evaluated in real time.

[0110] This embodiment also provides an anesthesia brain state assessment device, including:

[0111] A data acquisition module, which is responsible for acquiring EEG signals and preprocessing them;

[0112] An anesthetic brain state assessment module, which is responsible for inputting EEG signals into a deep learning model in the trained model parameter space for model parameter inversion; according to the estimated model parameters, calculating anesthetic depth index to realize real-time assessment of anesthetic brain state.

[0113] In addition, this embodiment conducts experimental research on the method for inverting anesthetic brain state assessment based on a deep learning EEG data model, aiming to verify its effectiveness and reliability in the field of anesthetic monitoring.

[0114] The data of this embodiment is from the public database VitalDB (http: / / vitaldb.net / data-bank). To verify the correlation between the estimated parameters and clinical anesthetic depth indicators, the Pearson correlation coefficient between key parameters and BIS index is calculated, as shown in Figure 3 (a) - Figure 3 (f) in the figure. The results show that the correlation coefficient between the excitatory synaptic gain A and BIS index reaches 0.85, showing a significant positive correlation. This further proves that excitatory parameters can be used as important indicators for anesthetic depth assessment.

[0115] In the four-classification task, the overall accuracy of the model is about 80%, and the main misclassifications occur between the "light anesthesia period" and the "recovery period", probably because the physiological characteristics in these two states are relatively similar.

[0116] To improve the classification accuracy, the classification task is simplified to a binary classification of "awake period" and "moderate anesthesia period", as shown in Figure 4 (a) - Figure 4 (b) in the figure, and the classification accuracy reaches more than 95%, indicating that the model can accurately distinguish significantly different anesthetic states.

[0117] The above is the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. A method for evaluating the anesthetic brain state by inverting an EEG data model based on deep learning, characterized in that It includes the following steps: Step S1: Construct a neuron population model applicable to electroencephalogram (EEG) activities under anesthesia; Step S2: Determine the value range of model parameters, generate multiple parameter combinations; generate simulated EEG signals according to the parameter combinations to form a training data set; Step S3: Design and train a deep learning model in the model parameter space, and learn the parameter mapping relationship from the simulated EEG signals; Step S4: Apply the trained deep learning model in the model parameter space to actual EEG data for model parameter inversion; Step S5: Calculate the anesthesia depth index according to the estimated model parameters to achieve real-time assessment of the anesthetized brain state.

2. The method according to claim 1, wherein In step S1, the neuron population model applicable to EEG activities under anesthesia is based on the Jansen-Rit model and consists of an excitatory neuron population and an inhibitory neuron population, including a pyramidal cell population PC, an excitatory interneuron population EI, and an inhibitory interneuron population II.

3. The method according to claim 2, wherein In step S1, the model dynamic equation description of the neuron population model applicable to EEG activities under anesthesia includes a postsynaptic membrane potential transfer function, the average firing rate of the neuron population, and a state equation: The postsynaptic membrane potential transfer function includes the postsynaptic membrane potential response functions h e (t) and h i (t), which are specifically as follows: where A and B are the excitatory synaptic gain and the inhibitory synaptic gain respectively; a and b are the reciprocals of the time constants of the excitatory neuron population and the inhibitory neuron population; t represents the time; exp() represents the natural exponential function; Average firing rate function: The average firing rate S(v) of the neuron population converts the postsynaptic membrane potential v into the average impulse density of action potentials: where e0 is the maximum firing rate, v0 is the membrane potential corresponding to the half-maximum firing rate, and r is the slope parameter; The state equation is specifically as follows: Among them, y1 is the membrane potential of the pyramidal cell population, y2 and y3 are the membrane potentials of the excitatory neuron population and the inhibitory neuron population respectively, y4, y5, and y6 are the derivatives of y1, y2, and y3 respectively, that is, the change rate of their membrane potentials, p(t) is the external input, C1 and C2 respectively represent the average connection coefficients connecting to and connecting out of the excitatory circuit, and C3 and C4 respectively represent the average connection coefficients connecting to and connecting out of the inhibitory circuit; represents the derivative of x.

4. The method according to claim 1, wherein In step S2, random selection of each parameter is performed within the value range of the model parameters to generate multiple parameter combinations; the parameters taken within different parameter combinations are different from each other; The model parameters include the excitatory synaptic gain A, the inhibitory synaptic gain B, the reciprocal of the time constant a of the excitatory neuron population, the reciprocal of the time constant d of the inhibitory neuron population, the average connection coefficient C1 connecting to the excitatory circuit, the average connection coefficient C2 connecting out of the excitatory circuit, the average connection coefficient C3 connecting to the inhibitory circuit, and the average connection coefficient C4 connecting out of the inhibitory circuit.

5. The method according to claim 4, wherein In step S2, for each group of parameter combinations, numerical integration is performed on the dynamic equation to generate corresponding simulated EEG signals: V i V(t) = y 2i V(t) - y 3i V(t) Equation (5) where y 2i (t), y 3i (t) represents the membrane potentials of the excitatory neuron population and the inhibitory neuron population in the i-th group of simulated EEG signals.

6. The method according to claim 4, characterized in that In step S2, the training data set is formed by the simulated EEG signals and the corresponding model parameters.

7. The method according to claim 6, wherein In step S2, the simulated EEG signals need to be preprocessed before being input into the deep learning model in the model parameter space.

8. The method according to claim 1, wherein In step S3, the deep learning model in the model parameter space includes an input layer, a convolutional layer, a pooling layer, a recurrent layer, and a fully connected layer; The input layer receives EEG signals; The convolutional layer uses a one-dimensional convolutional neural network (CNN) and uses multiple convolutional kernels to extract local time features from the input; The pooling layer uses max pooling to reduce the feature dimension of the output of the convolutional layer; The recurrent layer uses a long short-term memory network layer to capture the long-term dependence relationship of the time series for the output of the pooling layer; The fully connected layer unfolds the features extracted by the recurrent layer and maps them to the model parameter space to obtain the estimated model parameters The output layer outputs the estimated model parameters The loss function of the deep learning model in the model parameter space is as follows: where θ i represents the true value of the model parameter.

9. The method according to claim 1, characterized in that, Specifically, step S4 is as follows: Collect the EEG signals of the patient and preprocess the EEG signals; Input the preprocessed EEG signal segments into the deep learning model in the trained model parameter space to obtain the estimated model parameters 10. The method according to claim 9, wherein Specifically, step S5 is as follows: Calculate the anesthesia depth index ADI according to the estimated model parameters: where w1, w2, and w3 are weight coefficients; Divide the anesthesia depth into different levels according to the anesthesia depth index ADI.

Images