EEG Consciousness Monitoring Data Analysis Methods and Systems for Anesthesia Management

Abstract

This application provides a method and system for analyzing EEG consciousness monitoring data applied to anesthesia management. The method includes: acquiring EEG signal entropy sequence data of anesthetized subjects; performing dynamic entropy change analysis on the EEG signal entropy sequence data to obtain the entropy change feature vector of the anesthetized subjects; performing temporal phase coupling modeling on the EEG signal entropy sequence data to obtain the phase synchronization feature matrix of the anesthetized subjects; generating consciousness state fusion features of the anesthetized subjects based on the entropy change feature vector and the phase synchronization feature matrix; predicting the anesthesia state of the anesthetized subjects based on the consciousness state fusion features; and obtaining the EEG signal entropy of the anesthetized subjects at sampling time points in a second time period. This application can accurately predict the EEG signal entropy of anesthetized subjects at sampling time points in a second time period, helping to more accurately monitor the anesthesia state of anesthetized subjects so as to adjust the dosage of anesthetic drugs in a timely manner and ensure the safety and effectiveness of the surgical procedure.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for analyzing EEG consciousness monitoring data applied to anesthesia management. Background Technology

[0002] In modern surgical procedures, anesthesia management is a crucial aspect of ensuring patient safety and the smooth progress of surgery. Accurate monitoring of the patient's level of consciousness and timely adjustment of anesthetic drug dosage are essential for preventing intraoperative awareness and reducing anesthetic complications. Currently, commonly used clinical anesthesia monitoring methods rely primarily on single indicators, such as blood pressure, heart rate, and bispectral index (BPI). However, these single indicators often only reflect one aspect of the anesthetic state and cannot comprehensively and accurately assess the patient's level of consciousness. Therefore, a comprehensive and accurate anesthesia monitoring method is needed to precisely predict and monitor the anesthetic state in real time. Summary of the Invention

[0003] In view of this, this application provides a method and system for analyzing EEG consciousness monitoring data applied to anesthesia management. The technical solution of this application is implemented as follows:

[0004] On one hand, this application provides a method for analyzing EEG consciousness monitoring data applied to anesthesia management, comprising: acquiring EEG signal entropy sequence data of an anesthetized subject, wherein the EEG signal entropy sequence data includes multiple EEG signal entropies of the anesthetized subject arranged according to sampling time nodes in a first time period; performing dynamic entropy change analysis on the EEG signal entropy sequence data to obtain an entropy change feature vector of the anesthetized subject, wherein the dimensions of the entropy change feature vector include bi-frequency exponent, state entropy, response entropy, and long-range correlation, and the entropy change feature vector characterizes the dynamic change characteristics of the EEG signal entropy of the anesthetized subject due to the evolution of sampling time nodes; and performing dynamic entropy change analysis on the EEG signal entropy sequence data to obtain an entropy change feature vector of the anesthetized subject. The EEG signal entropy sequence data is used for temporal phase coupling modeling to obtain the phase synchronization feature matrix of the anesthetized subject. The phase synchronization feature matrix represents the topological structure of the phase relationship between multi-channel EEG signals and indicates the long-range synchronicity between EEG signal entropies at different sampling time nodes. Based on the entropy change feature vector and the phase synchronization feature matrix, the consciousness state fusion feature of the anesthetized subject is generated. Based on the consciousness state fusion feature, the anesthesia state of the anesthetized subject is predicted to obtain the EEG signal entropy of the anesthetized subject at the sampling time node in the second time period, which is located after the first time period.

[0005] On the other hand, this application provides a computer system including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the program to implement the steps in the method described above.

[0006] After obtaining the EEG signal entropy sequence data of the anesthetized subject, this application performs dynamic entropy change analysis on the EEG signal entropy sequence data to obtain the entropy change feature vector of the anesthetized subject, performs temporal phase coupling modeling on the EEG signal entropy sequence data to obtain the phase synchronization feature matrix of the anesthetized subject, generates consciousness state fusion features based on the entropy change feature vector and the phase synchronization feature matrix, and predicts the EEG signal entropy of the sampling time sequence node of the anesthetized subject in the second time period based on the consciousness state fusion features. Among them, dynamic entropy change analysis analyzes EEG signal entropy sequence data from the perspective of EEG signal entropy fluctuation. The entropy change feature vector obtained by dynamic entropy change analysis can characterize the dynamic change characteristics of the EEG signal entropy of the anesthetized subject due to the evolution of sampling time nodes. Temporal phase coupling modeling can analyze EEG signal entropy sequence data from the perspective of long-range synchronicity between EEG signal entropy at different sampling time nodes. The phase synchronization feature matrix obtained by temporal phase coupling modeling can characterize the long-range synchronicity between EEG signal entropy at different sampling time nodes in the anesthetized subject. By combining the entropy change feature vector and the phase synchronization feature matrix, the EEG signal entropy of the anesthetized subject at sampling time nodes in the second time period can be accurately predicted, which helps to more accurately monitor the anesthesia status of the anesthetized subject so as to adjust the dosage of anesthetic drugs in a timely manner and ensure the safety and effectiveness of the surgical procedure.

[0007] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of this application. Attached Figure Description

[0008] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application.

[0009] Figure 1 This is a schematic diagram illustrating the implementation process of an EEG consciousness monitoring data analysis method for anesthesia management, provided in an embodiment of this application.

[0010] Figure 2 This is a schematic diagram of the composition structure of a brainwave consciousness monitoring data analysis device provided in an embodiment of this application.

[0011] Figure 3 This is a schematic diagram of the hardware entity of a computer system provided in an embodiment of this application. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application are further described in detail below with reference to the accompanying drawings and embodiments. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] This application provides a method for analyzing EEG consciousness monitoring data for anesthesia management, which can be executed by a processor of a computer system. The computer system can refer to a monitoring device during surgery, such as a laptop, tablet, or desktop computer with data processing capabilities, or an EEG monitoring device integrating the code provided in this application's embodiments into a storage medium.

[0014] Figure 1 This application provides a schematic diagram of the implementation process of an EEG consciousness monitoring data analysis method for anesthesia management, as illustrated in the embodiments of this application. Figure 1 As shown, the method includes the following steps:

[0015] Step 100: Obtain the EEG signal entropy sequence data of the anesthetized subject. The EEG signal entropy sequence data includes multiple EEG signal entropies of the anesthetized subject arranged according to the sampling time sequence nodes in the first time period.

[0016] In step 100, the EEG signal entropy sequence data of the anesthetized subject can be obtained by preprocessing the basic EEG signal entropy sequence data of the anesthetized subject. In other words, the basic EEG signal entropy sequence data of the anesthetized subject can be obtained. The basic EEG signal entropy sequence can include multiple basic EEG signal entropies arranged according to the sampling time sequence nodes in the first time period. The basic EEG signal entropy sequence can be preprocessed to obtain the EEG signal entropy sequence data. EEG, or electroencephalogram, is a spontaneous and rhythmic electrical activity of brain cell groups of anesthetized subjects recorded by electrodes. EEG signal entropy is a measure of the complexity and disorder of EEG signals, reflecting the state of brain neural activity. The sampling time sequence node refers to each time point in the first time period where the EEG signal is sampled at certain time intervals. For example, with an interval of 1 second, there will be 600 sampling time sequence nodes in the first time period of 10 minutes.

[0017] Entropy is calculated on the preprocessed EEG signal. Entropy calculation methods include, for example, sample entropy (SampEn) and approximate entropy (ApEn). Taking sample entropy as an example, its calculation formula is as follows: Where m is the embedding dimension, r is the similarity tolerance, N is the data length, and A(m, r) represents the number of data pairs that satisfy specific similarity conditions in the m-dimensional space. Within the first time period, the sample entropy of the EEG signal is calculated point-by-point or segment-by-segment according to the sampling time sequence nodes, thereby obtaining the EEG signal entropy corresponding to each sampling time sequence node.

[0018] Finally, the EEG signal entropies corresponding to each sampling time point within the first time period are arranged according to the sampling order to form EEG signal entropy sequence data. For example, within the aforementioned 10-minute first time period, a sequence containing 600 EEG signal entropies will be obtained, with each EEG signal entropy corresponding to a specific sampling time point. In this way, the EEG signal entropy sequence data of the anesthetized subject is obtained, providing basic data for subsequent dynamic entropy change analysis and temporal phase coupling modeling.

[0019] The preprocessing mentioned above can include missing data interpolation and outlier data processing. The methods can be found in existing technologies and will not be elaborated here.

[0020] Step 200: Perform dynamic entropy change analysis on the EEG signal entropy sequence data to obtain the entropy change feature vector of the anesthetized subject. The dimensions of the entropy change feature vector include dual-frequency exponent, state entropy, response entropy and long-range correlation. The entropy change feature vector characterizes the dynamic change characteristics of the EEG signal entropy of the anesthetized subject due to the evolution of sampling time sequence nodes.

[0021] In step 200, dynamic entropy change analysis is performed on the obtained EEG signal entropy sequence data of the anesthetized subject to obtain the entropy change feature vector of the anesthetized subject. The dimensions of the entropy change feature vector include dual-frequency exponent, state entropy, response entropy and long-range correlation. Its function is to characterize the dynamic change characteristics of the EEG signal entropy of the anesthetized subject due to the evolution of sampling time sequence nodes.

[0022] First, it is necessary to obtain the neural oscillation spectrum corresponding to the EEG signal entropy sequence data. This spectrum contains multiple time-frequency fluctuations of the anesthetized subject arranged according to the sampling time sequence nodes during the first time period. Neural oscillations are the synchronous electrical activity of a group of neurons in the brain, and the time-frequency fluctuations reflect the changes in neural oscillations in time and frequency. The neural oscillation spectrum can be obtained by converting the EEG signal from the time domain to the time-frequency domain using time-frequency analysis methods, such as wavelet transform. The formula for wavelet transform is: Where f(t) is the EEG signal, (t) is the wavelet basis function, a is the scaling parameter, and b is the translation parameter.

[0023] In this embodiment, the dynamic change characteristics may include several aspects: (1) the entropy change characteristics of the EEG signal entropy of the anesthetized object due to the evolution of the sampling time sequence node; (2) the abnormal oscillation of the EEG signal entropy of the anesthetized object due to the evolution of the sampling time sequence node; (3) the influence of the past entropy change characteristics of the EEG signal entropy of the anesthetized object due to the evolution of the sampling time sequence node on the subsequent entropy change characteristics.

[0024] For each level of characteristics, there are multiple ways to perform dynamic entropy change analysis to obtain the entropy change feature vector. The first method is to use a target dynamic entropy change predictor, such as a generalized autoregressive conditional heteroscedasticity model or an autoregressive conditional heteroscedasticity model, to predict the anesthetized subject based on the neural oscillation fluctuation spectrum, obtain the time-frequency fluctuation of the anesthetized subject at the sampling time sequence node in the second time period, and use it as the entropy change feature vector.

[0025] The second approach involves first decomposing the neural oscillation spectrum into multiple neural oscillation segments based on the time windows corresponding to preset sampling time nodes. Within each segment, a reference time-frequency fluctuation and a sample time-frequency fluctuation are determined. The reference time-frequency fluctuation is the time-frequency fluctuation of the last one or more sampling time nodes in the neural oscillation segment, while the sample time-frequency fluctuation is the time-frequency fluctuation of the remaining sampling time nodes excluding the last one or more sampling time nodes. Then, a target dynamic entropy change predictor infers the inferred time-frequency fluctuation corresponding to the reference time-frequency fluctuation in each neural oscillation segment based on the sample time-frequency fluctuation. The oscillation error result corresponding to each neural oscillation segment is determined based on the error between the reference time-frequency fluctuation and the inferred time-frequency fluctuation. These oscillation error results are used as entropy change feature vectors, which characterize the abnormal oscillations in the EEG signal entropy of the anesthetized subject caused by the evolution of the sampling time nodes.

[0026] The third approach involves repeatedly predicting the parameters of a dynamic entropy change predictor based on multiple neural oscillation segments. The parameters obtained from these repeated predictions are used as an entropy change feature vector. This vector characterizes the influence of past entropy changes caused by the evolution of sampling time nodes on subsequent entropy changes in the EEG signal entropy of the anesthetized subject. Assuming the neural oscillation spectrum is decomposed into x neural oscillation segments, the parameters of the basic dynamic entropy change predictor will be predicted x times. The b-th prediction is executed with the dynamic entropy change predictor obtained from the a-th prediction as a prerequisite, where 1 ≤ a ≤ x, and a = b - 1.

[0027] In the b-th prediction of the parameter, the reference time-frequency fluctuation and the sample time-frequency fluctuation are determined in the b-th neural oscillation segment. The dynamic entropy change predictor obtained in the a-th prediction predicts the inference time-frequency fluctuation corresponding to the reference time-frequency fluctuation based on the sample time-frequency fluctuation, and obtains the confidence that the inference time-frequency fluctuation is the reference time-frequency fluctuation. Then, with the goal of maximizing this confidence, the parameter obtained in the a-th prediction is iterated to obtain the parameter obtained in the b-th prediction.

[0028] Through dynamic entropy change analysis using the above different methods, the entropy change feature vector of the anesthetized subject is finally obtained. This vector contains information in dimensions such as dual-frequency index, state entropy, response entropy, and long-range correlation. It can accurately reflect the dynamic change characteristics of the EEG signal entropy of the anesthetized subject during the evolution of sampling time nodes, providing an important basis for subsequent anesthesia state prediction.

[0029] Step 300: Perform temporal phase coupling modeling on the EEG signal entropy sequence data to obtain the phase synchronization feature matrix of the anesthetized subject. The phase synchronization feature matrix represents the topological structure of the phase relationship between multi-channel EEG signals and indicates the long-range synchronicity between EEG signal entropies at different sampling time nodes.

[0030] In step 300, temporal phase coupling modeling is performed on the EEG signal entropy sequence data of the anesthetized subject to obtain the phase synchronization feature matrix of the anesthetized subject. This matrix is ​​used to characterize the topological structure of the phase relationship between multi-channel EEG signals and to indicate the long-range synchronization between the EEG signal entropies of different sampling time nodes.

[0031] The EEG signal entropy sequence data contains the EEG signal entropy of y sampling time nodes in the first time period, where y ≥ 2. Temporal phase coupling modeling is performed by a target temporal phase coupling modeler, such as LSTM or RNN, which contains y temporal phase coupling modeling layers that are matched one by one with the EEG signal entropy of the y sampling time nodes.

[0032] In performing temporal phase coupling modeling, the EEG signal entropy at the corresponding sampling time node and the EEG signal entropy at the sampling time node preceding the corresponding sampling time node are first modeled using each temporal phase coupling modeling layer. This yields the temporal phase coupling modeling result for each layer. The temporal phase coupling here reflects the degree of phase correlation between EEG signals at different times, demonstrating the synchronicity of neural activity in different brain regions. For example, during anesthesia, if there is strong phase coupling between EEG signals at two specific moments, it indicates that neural activity in certain brain regions is synchronized at those two moments.

[0033] In the specific modeling process, taking the f-th time-domain phase coupling modeling layer out of y time-domain phase coupling modeling layers as an example, it is responsible for performing time-domain phase coupling modeling on the EEG signal entropy of the f-th sampling time-series node and the EEG signal entropy of the e sampling time-series nodes preceding the f-th sampling time-series node, where 1≤e≤y and e=f-1. The characteristics of the EEG signal entropy of the previous e sampling time-series nodes are maintained in the cell state of the e-th time-domain phase coupling modeling layer, and the cell state and time-domain phase coupling modeling results of the e-th time-domain phase coupling modeling layer are loaded into the f-th time-domain phase coupling modeling layer.

[0034] The temporal phase coupling modeling process of the f-th temporal phase coupling modeling layer is relatively complex. First, based on the temporal phase coupling modeling result of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node, a forgetting gate is applied to the cell state of the e-th temporal phase coupling modeling layer to determine the first cell state to be maintained. The forgetting gate determines which information needs to be discarded from the cell state, and its calculation formula can be expressed as:

[0035] ;

[0036] in, It is the output of the forget gate. It is the sigmoid function. It is the weight matrix of the forget gate, h t-1 It is the hidden state from the previous moment, x t It is the input at the current moment, b f It is the bias term of the forget gate.

[0037] Next, based on the same input, a memory gate is applied to the f-th temporal phase coupling modeling layer to determine the second cell state that should be incorporated into the cell state of the f-th temporal phase coupling modeling layer from the temporal phase coupling modeling result of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node. The memory gate is used to determine which new information needs to be added to the cell state, and its calculation formula is as follows:

[0038] ;

[0039] ;;

[0040] Among them, i t It is the output of the memory gate. It is a candidate cell state. It is a weight matrix. It is a bias term.

[0041] Then, the first and second cell states are merged to obtain the cell states of the f-th temporal phase coupling modeling layer. The merging formula is: Among them, C t It is the current state of the cell. This indicates element-wise multiplication.

[0042] Finally, based on the temporal phase coupling modeling results of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node, output analysis is performed on the cell state of the f-th temporal phase coupling modeling layer to obtain the temporal phase coupling modeling results of the f-th temporal phase coupling modeling layer. The output analysis determines the output weights based on the temporal phase coupling modeling results of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node. Based on these output weights, the information to be output from the cell state of the f-th temporal phase coupling modeling layer is used as the temporal phase coupling modeling result. The calculation formula is as follows:

[0043] ;

[0044] ;

[0045] in, It is the output of the output gate, h t It is the hidden state at the current moment, W o It is the weight matrix of the output gate, b o It is the bias term of the output gate.

[0046] After completing the modeling of each temporal phase coupling modeling layer, there are two ways to determine the phase synchronization feature matrix. One method is to use the temporal phase coupling modeling result of the y-th temporal phase coupling modeling layer as the phase synchronization feature matrix. This matrix characterizes the synchronization of the EEG signal entropy of the y-th sampling time-series node with the EEG signal entropy of the y-1 sampling time-series nodes preceding the y-th sampling time-series node. The other method is to use the statistical result of the temporal phase coupling modeling results of the y-th temporal phase coupling modeling layers as the phase synchronization feature matrix. This matrix characterizes the comprehensive synchronization of the EEG signal entropy of each sampling time-series node with the EEG signal entropy of each sampling time-series node preceding it. Through this temporal phase coupling modeling, the long-range synchronicity between the EEG signal entropies of different sampling time-series nodes of the anesthetized subject can be accurately captured, providing an important basis for subsequent generation of consciousness state fusion features and prediction of anesthesia state.

[0047] Step 400: Generate the consciousness state fusion features of the anesthetized subject based on the entropy-change feature vector and the phase synchronization feature matrix.

[0048] In step 400, a fusion feature of the consciousness state of the anesthetized subject is generated based on the entropy change feature vector and the phase synchronization feature matrix. The entropy change feature vector is obtained by performing dynamic entropy change analysis on the entropy sequence data of the anesthetized subject's EEG signal, which characterizes the dynamic change characteristics of the EEG signal entropy due to the evolution of sampling time sequence nodes. The phase synchronization feature matrix is ​​obtained through temporal phase coupling modeling and is used to characterize the long-range synchronicity between the EEG signal entropies of different sampling time sequence nodes. Fusing these two features can comprehensively reflect the consciousness state of the anesthetized subject during the anesthesia process.

[0049] There are several ways to generate consciousness state fusion features. One approach is to concatenate the entropy-varying feature vector and the phase-synchronization feature matrix, combining their elements sequentially to form a new feature vector, which is the consciousness state fusion feature. Another approach is to perform a weighted average, assigning different weights to the entropy-varying feature vector and the phase-synchronization feature matrix, then multiplying their corresponding elements by their respective weights and summing the results to obtain the consciousness state fusion feature. Convolution is also a viable method. A sliding convolution kernel can be used to perform a convolution operation on the entropy-varying feature vector and the phase-synchronization feature matrix, extracting their feature information through convolution calculations to generate the consciousness state fusion feature.

[0050] Before generating the consciousness state fusion features, feature filtering can be performed. The first influence weight of each entropy change feature vector element in the entropy change feature vector on the EEG signal entropy of the sampling time-series nodes in the second time period is obtained. Entropy change feature vector elements whose first influence weights satisfy the weight threshold are identified, resulting in a determined entropy change feature vector. Similarly, the second influence weight of each phase synchronization feature matrix element in the phase synchronization feature matrix on the EEG signal entropy of the sampling time-series nodes in the second time period is obtained. Phase synchronization feature matrix elements whose second influence weights satisfy the weight threshold are identified, resulting in a determined phase synchronization feature matrix. Then, based on these two determined features, the consciousness state fusion features are generated. Through these operations, the information from the entropy change feature vector and the phase synchronization feature matrix can be more accurately integrated to generate fusion features that effectively reflect the consciousness state of the anesthetized subject, providing strong support for subsequent anesthesia state prediction.

[0051] Step 500: Based on the consciousness state fusion features, predict the anesthesia state of the anesthetized subject and obtain the EEG signal entropy of the sampling time sequence node of the anesthetized subject in the second time period, which is after the first time period.

[0052] In step 500, the anesthesia state of the anesthetized subject is predicted based on the consciousness state fusion features, thereby obtaining the EEG signal entropy of the anesthetized subject at the sampling time sequence nodes in the second time period, where the second time period is after the first time period. The consciousness state fusion features are obtained by processing the entropy change feature vector and the phase synchronization feature matrix. It integrates the dynamic change characteristics of the anesthetized subject's EEG signal entropy and the long-range synchronicity between the EEG signal entropies of different sampling time sequence nodes, and can comprehensively reflect the consciousness state of the anesthetized subject during the anesthesia process.

[0053] Various machine learning or deep learning models can be used to predict the state of anesthesia. For example, neural network models such as multilayer perceptrons (MLPs) can be used. A multilayer perceptron is a feedforward artificial neural network model consisting of an input layer, hidden layers, and an output layer. In this scenario, the consciousness state fusion features serve as the input to the input layer, and the model's goal is to predict the EEG signal entropy at each sampling time point in the second time period. These predicted values ​​constitute the output of the output layer. The hidden layers are responsible for performing nonlinear transformations and feature extraction on the input information to learn the complex mapping relationship between the consciousness state fusion features and the EEG signal entropy.

[0054] Specifically, the forward propagation process of a multilayer perceptron can be represented by the following formula. Let the input layer have n neurons, corresponding to the n dimensions of the consciousness state fusion feature; the hidden layer have m neurons; and the output layer have k neurons, corresponding to the predicted EEG signal entropy values ​​of k sampling time nodes in the second time period. The weight matrix from the input layer to the hidden layer is W1, with a dimension of m×n, and the bias vector is b1, with a dimension of m; the weight matrix from the hidden layer to the output layer is W2, with a dimension of k×m, and the bias vector is b2, with a dimension of k. Then the output h of the hidden layer can be calculated using the following formula: h = f(W1x + b1), where x is the consciousness state fusion feature vector, and f is the activation function. Possible activation functions include the Sigmoid function and the ReLU function. Taking the ReLU function as an example, its expression is f(z) = max(0,z).

[0055] The output y of the output layer, i.e., the predicted EEG signal entropy vector, can be calculated using the following formula: y = W2h + b2. To train this multilayer perceptron model, a large amount of historical data is required. This historical data includes consciousness state fusion features and the corresponding actual EEG signal entropy. During training, the weight matrices W1 and W2 and the bias vectors b1 and b2 are continuously adjusted to minimize the error between the model's predicted output and the actual EEG signal entropy. A feasible error function is the mean squared error (MSE), which is calculated using the following formula:

[0056] ;

[0057] Where N is the number of training samples, y i It is the actual EEG signal entropy. These are the model's predicted values.

[0058] Besides multilayer perceptrons, models such as support vector regression (SVR) can also be used to predict anesthesia status. SVR works by finding an optimal hyperplane that minimizes the error between the predicted and actual values ​​within a certain range. When using SVR, it's necessary to choose an appropriate kernel function, such as a linear kernel or radial basis function, to handle different types of data.

[0059] Through the above prediction process, the EEG signal entropy of the anesthetized subject at the sampling time point in the second time period can be accurately predicted using the consciousness state fusion features. These prediction results serve as a reference for doctors to understand the anesthetized state of the subject. Doctors can adjust the dosage of anesthetic drugs based on the predicted EEG signal entropy to ensure that the subject is at an appropriate depth of anesthesia during surgery, thereby improving the safety and effectiveness of the surgery.

[0060] As one implementation of step 200, dynamic entropy change analysis is performed on the EEG signal entropy sequence data to obtain the entropy change feature vector of the anesthetized object. This includes: acquiring the neural oscillation spectrum corresponding to the EEG signal entropy sequence data, wherein the neural oscillation spectrum includes multiple time-frequency fluctuations of the anesthetized object arranged according to the sampling time sequence nodes in the first time period; predicting the anesthetized object based on the neural oscillation spectrum using a target dynamic entropy change predictor to obtain the time-frequency fluctuations of the anesthetized object at the sampling time sequence nodes in the second time period; using the time-frequency fluctuations of the anesthetized object at the sampling time sequence nodes in the second time period as the entropy change feature vector; the entropy change feature vector characterizes the entropy change characteristics of the anesthetized object's EEG signal entropy due to the evolution of the sampling time sequence nodes.

[0061] In the above embodiments, the neural oscillation spectrum corresponding to the entropy sequence data of the EEG signal is obtained. This spectrum contains multiple time-frequency fluctuations of the anesthetized subject arranged according to the sampling time sequence nodes during the first time period. Neural oscillations are the synchronous electrical activities of a group of neurons in the brain, and the time-frequency fluctuations reflect the changes in neural oscillations in time and frequency. The neural oscillation spectrum can be obtained by converting the EEG signal from the time domain to the time-frequency domain using time-frequency analysis methods, such as wavelet transform. For example, during the first time period of anesthesia of the anesthetized subject, EEG signals are collected at 1-second sampling intervals. Wavelet transform can be used to obtain the time-frequency fluctuations corresponding to each sampling time sequence node, and these time-frequency fluctuations constitute the neural oscillation spectrum.

[0062] Next, the target dynamic entropy change predictor predicts the anesthetized subject based on the neural oscillation spectrum, obtaining the time-frequency fluctuation of the sampled time-series nodes in the second time period. The target dynamic entropy change predictor can be a generalized autoregressive conditional heteroscedasticity model or an autoregressive conditional heteroscedasticity model. Taking the generalized autoregressive conditional heteroscedasticity model as an example, since it is an existing model, its formula will not be elaborated in detail here; the conditional variance can be used as the time-frequency fluctuation.

[0063] Then, the time-frequency fluctuation of the sampling time nodes of the anesthetized subject in the second time period is used as the entropy change feature vector. This vector characterizes the entropy change characteristics of the anesthetized subject's EEG signal entropy due to the evolution of the sampling time nodes. The entropy change characteristic reflects the dynamic change of the EEG signal entropy over time. By analyzing the entropy change feature vector, the stability and changing trend of the brain's neural activity in the anesthetized subject during anesthesia can be understood. For example, if the time-frequency fluctuation in the entropy change feature vector gradually increases in the second time period, it indicates that the EEG signal entropy of the anesthetized subject has become more unstable, which may mean that the anesthetized state of the anesthetized subject has changed, and the doctor needs to adjust the dosage of anesthetic drugs in a timely manner.

[0064] In practical applications, the target dynamic entropy change predictor needs to be trained and validated to ensure its predictive accuracy. The training process can use historical data, taking historical neural oscillation spectra as input and the corresponding actual time-frequency fluctuations as output, minimizing prediction errors by adjusting model parameters. The validation process uses independent validation data to evaluate the model's performance on unseen data. Feasible evaluation metrics include mean squared error (MSE) and mean absolute error (MAE). By continuously adjusting the model parameters to optimize these evaluation metrics, a high-performance target dynamic entropy change predictor can be obtained, accurately predicting the time-frequency fluctuations of anesthetized subjects at sampling time points in the second time period, providing strong support for the monitoring and adjustment of anesthesia status.

[0065] As another implementation of step 200, dynamic entropy change analysis is performed on the EEG signal entropy sequence data to obtain the entropy change feature vector of the anesthetized subject. This includes: acquiring the neural oscillation spectrum corresponding to the EEG signal entropy sequence data, wherein the neural oscillation spectrum includes multiple time-frequency fluctuations of the anesthetized subject arranged according to sampling time nodes in the first time period; decomposing the neural oscillation spectrum into multiple neural oscillation segments according to the time window corresponding to the preset sampling time nodes; determining the reference time-frequency fluctuation and the sample time-frequency fluctuation in each neural oscillation segment, wherein the reference time-frequency fluctuation is the time-frequency fluctuation of the last one or more sampling time nodes in the neural oscillation segment, and the sample time-frequency fluctuation is... The fluctuation amount is the time-frequency fluctuation amount of the remaining sampling time-series nodes in the neural oscillation segment, excluding the last one or more sampling time-series nodes. Based on the sample time-frequency fluctuation amount in each neural oscillation segment, the target dynamic entropy change predictor infers the inferred time-frequency fluctuation amount corresponding to the reference time-frequency fluctuation amount in each neural oscillation segment. Based on the error between the reference time-frequency fluctuation amount and the corresponding inferred time-frequency fluctuation amount in each neural oscillation segment, the oscillation error result corresponding to each neural oscillation segment is determined. The oscillation error result corresponding to each neural oscillation segment is used as the entropy change feature vector, where the entropy change feature vector represents the abnormal oscillation of the EEG signal entropy of the anesthetized subject caused by the evolution of the sampling time-series nodes.

[0066] Specifically, by acquiring the neural oscillation spectrum corresponding to the entropy sequence data of the EEG signal, which includes multiple time-frequency fluctuations of the anesthetized subject arranged according to the sampling time sequence nodes in the first time period, neural oscillations are the synchronous electrical activity of a group of neurons in the brain, reflecting the information exchange and collaborative work between different areas of the brain. The time-frequency fluctuations reflect the changing characteristics of neural oscillations in the time and frequency dimensions. Time-frequency analysis methods, such as wavelet transform, can be used to acquire the neural oscillation spectrum. Wavelet transform can perform localized analysis of the signal in time and frequency. By integrating wavelets at different scales and locations, the time-frequency fluctuations corresponding to each sampling time sequence node can be obtained, thereby constructing the neural oscillation spectrum. For example, in the first time period of anesthesia (assuming it is the first 30 minutes after the start of surgery), EEG signals are acquired at 1-second sampling intervals. After wavelet transform, the time-frequency fluctuations corresponding to these 1800 sampling time sequence nodes are obtained. These data constitute the neural oscillation spectrum of the anesthetized subject in this time period.

[0067] Next, based on the time windows corresponding to the preset sampling time nodes, the neural oscillation spectrum is decomposed into multiple neural oscillation segments. The time window is set to divide the continuous neural oscillation spectrum into segments of a certain length, allowing for independent analysis of each segment later. Assuming a preset time window of 10 seconds, within the first time period of 30 minutes, the neural oscillation spectrum will be decomposed into 180 neural oscillation segments, each containing the time-frequency fluctuations of 10 sampling time nodes. This decomposition method helps capture the local characteristics and variation patterns of neural oscillations across different time periods.

[0068] Then, reference time-frequency fluctuations and sample time-frequency fluctuations are determined in each neural oscillation segment. The reference time-frequency fluctuation is the time-frequency fluctuation of the last one or more sampling time-series nodes in the neural oscillation segment, while the sample time-frequency fluctuation is the time-frequency fluctuation of the remaining sampling time-series nodes excluding the last one or more sampling time-series nodes. For example, in a neural oscillation segment containing 10 sampling time-series nodes, if the time-frequency fluctuations of the last two sampling time-series nodes are selected as the reference time-frequency fluctuations, then the time-frequency fluctuations of the first eight sampling time-series nodes are the sample time-frequency fluctuations. The reference time-frequency fluctuation represents the current or future state of the segment, while the sample time-frequency fluctuation is used to infer the reference time-frequency fluctuation. By comparing the two, abnormal changes in the neural oscillation can be detected.

[0069] The target dynamic entropy change predictor infers the inferred time-frequency fluctuation corresponding to the reference time-frequency fluctuation in each neural oscillation segment based on the sample time-frequency fluctuation in each neural oscillation segment. The target dynamic entropy change predictor can be a generalized autoregressive conditional heteroscedasticity model or an autoregressive conditional heteroscedasticity model, etc., which have been introduced above and will not be repeated here.

[0070] Subsequently, based on the error between the reference time-frequency fluctuation and the corresponding inferred time-frequency fluctuation in each neural oscillation segment, the oscillation error result for each neural oscillation segment is determined. The oscillation error result reflects the degree of difference between the reference time-frequency fluctuation and the inferred time-frequency fluctuation. If the error is large, it indicates that there is an abnormality in the neural oscillation of that segment. A feasible error calculation method is the mean squared error (MSE), and its calculation formula can be found in the aforementioned content or existing technology, and will not be elaborated here.

[0071] Finally, the oscillation error results corresponding to each neural oscillation segment are used as the entropy change feature vector. This vector characterizes the abnormal oscillations in the EEG signal entropy of the anesthetized subject caused by the evolution of the sampling time sequence. Each element in the entropy change feature vector corresponds to the oscillation error result of a neural oscillation segment. By analyzing this vector, the stability and abnormalities of the brain's neural activity during anesthesia can be understood. For example, if some elements in the entropy change feature vector have large values, it indicates that the corresponding neural oscillation segment has large abnormal fluctuations, which may mean that the anesthetized subject's anesthetic state is unstable, requiring timely adjustment of the anesthetic drug dosage or other intervention measures.

[0072] As one implementation method, the method provided in this application embodiment further includes a training process for a dynamic entropy change predictor, which specifically may include:

[0073] Step 10: Obtain the basic dynamic entropy change predictor and training templates. The training templates include tuning templates and validation templates.

[0074] Step 20: Based on the calibration template, predict the parameters of the basic dynamic entropy change predictor to obtain the first dynamic entropy change predictor. The parameters include autoregressive variance weight and autoregressive squared error weight. The autoregressive variance weight represents the influence of past variance on subsequent variance, and the autoregressive squared error weight represents the influence of past squared error on subsequent squared error.

[0075] Step 30: Perform matching verification between the first dynamic entropy change predictor and the verification template.

[0076] Specifically, if the first dynamic entropy change predictor matches the verification template, the first dynamic entropy change predictor is used as the target dynamic entropy change predictor.

[0077] If the first dynamic entropy change predictor does not match the validation template, a new dynamic entropy change predictor is obtained. Based on the calibration template, the parameters of the new dynamic entropy change predictor are predicted to obtain a second dynamic entropy change predictor. Matching and validation are performed between the second dynamic entropy change predictor and the validation template. If the second dynamic entropy change predictor matches the validation template, it is used as the target dynamic entropy change predictor. If the second dynamic entropy change predictor does not match the validation template, the dynamic entropy change predictor is updated until the predicted dynamic entropy change predictor matches the validation template. The dynamic entropy change predictor that matches the validation template is used as the target dynamic entropy change predictor. The new dynamic entropy change predictor is of a different type than the basic dynamic entropy change predictor.

[0078] In this embodiment, the training template includes a calibration template and a validation template. The basic dynamic entropy change predictor is an initial model used to predict the entropy change of EEG signals in anesthetized subjects. As mentioned earlier, it can be a generalized autoregressive conditional heteroscedasticity model or an autoregressive conditional heteroscedasticity model, etc. The training template is a dataset used to train and validate the basic dynamic entropy change predictor. The calibration template is used to adjust the parameters of the basic dynamic entropy change predictor, and the validation template is used to verify whether the adjusted predictor is effective. For example, EEG signal entropy sequence data of the anesthetized subject in previous anesthesia processes can be collected, and a portion of the data can be used as the calibration template, while another portion can be used as the validation template. Specifically, a basic dynamic entropy change predictor and a training template (specifically, a first training template) can be obtained. The first training template can include multiple sample EEG signal entropies of the anesthetized subject arranged according to sampling time nodes in the first time period. The training template can be decomposed into multiple sub-training templates according to the time window corresponding to the preset sampling time nodes. These multiple sub-training templates can be divided into two categories: calibration templates and validation templates. In other words, the training template can include calibration templates and validation templates. Based on the tuning template, the parameters of the basic dynamic entropy change predictor can be predicted to obtain the first dynamic entropy change predictor. Then, matching and verification can be performed between the first dynamic entropy change predictor and the validation template. When predicting the parameters of the basic dynamic entropy change predictor and performing matching and verification between the predicted dynamic entropy change predictor and the validation template, the decomposition of multiple sub-training templates into tuning and validation templates is not restricted in terms of the decomposition method. For example, multiple sub-training templates can be directly decomposed into two distinct sets, one as the tuning template and the other as the validation template.

[0079] In step 20, based on the calibration template, the parameters of the basic dynamic entropy change predictor are predicted to obtain the first dynamic entropy change predictor. The parameters include autoregressive variance weights and autoregressive squared error weights. The autoregressive variance weights represent the influence of past variance on subsequent variance, and the autoregressive squared error weights represent the influence of past squared error on subsequent squared error. Maximum likelihood estimation can be used to predict the values ​​of the parameters in the model. The basic idea of ​​maximum likelihood estimation is to find a set of parameter values ​​that maximizes the probability of observing data in the calibration template under these parameters. Specifically, a likelihood function needs to be defined. ,in It is a parameter vector. These are the observation data in the calibration template. In the parameter variable Down The probability density function is obtained. By maximizing the likelihood function, the optimal estimate of the parameter can be obtained, thus obtaining the first dynamic entropy change predictor.

[0080] In step 30, a matching validation is performed between the first dynamic entropy-change predictor and the validation template. The purpose of the matching validation is to examine the prediction performance of the first dynamic entropy-change predictor on data not used in training (i.e., the validation template). Various evaluation metrics can be used for validation, such as mean squared error (MSE) and mean absolute error (MAE). The first dynamic entropy-change predictor is applied to the validation template, and the mean squared error or mean absolute error between the predicted and actual values ​​is calculated. If the error is within an acceptable range, the first dynamic entropy-change predictor is considered to be a match for the validation template.

[0081] Furthermore, a verification matching method is provided. Specifically, the verification template can include multiple verification samples. In each verification sample, a reference time-frequency fluctuation and a sample time-frequency fluctuation are determined. The method for determining the time-frequency fluctuation is as described above, and it is a general calculation method. The predicted dynamic entropy change predictor, based on the sample time-frequency fluctuation in each verification sample, infers the inferred time-frequency fluctuation corresponding to the reference time-frequency fluctuation in each neural oscillation segment. Then, based on the error between the reference time-frequency fluctuation and the corresponding inferred time-frequency fluctuation of each verification sample, the oscillation error result corresponding to each verification sample is determined. The oscillation error results corresponding to each verification sample form an error set. The error set is analyzed. For example, if the error set follows a Gaussian distribution, it is determined that the predicted dynamic entropy change predictor matches the verification template; if the error set does not follow a Gaussian distribution, it is determined that the predicted dynamic entropy change predictor does not match the verification template. Alternatively, autocorrelation verification is performed on the error set. If the error set has autocorrelation, it is determined that the predicted dynamic entropy change predictor matches the verification template; if the error set does not have autocorrelation, it is determined that the predicted dynamic entropy change predictor does not match the verification template. The predicted dynamic entropy change predictor is matched and validated with the validation template to ensure a high degree of matching between the dynamic entropy change predictor and the anesthetized subject. The dynamic entropy change predictor is more suitable for dynamic entropy change analysis of the EEG signal entropy sequence of the anesthetized subject, thereby increasing the accuracy of dynamic entropy change analysis and improving the accuracy of subsequent EEG signal entropy.

[0082] If the first dynamic entropy change predictor matches the validation template, it is designated as the target dynamic entropy change predictor. This means that after calibration and validation, the first dynamic entropy change predictor can accurately predict the entropy change of the EEG signal in anesthetized subjects and can be used for actual anesthesia state monitoring and prediction. For example, if the mean square error of the first dynamic entropy change predictor on the validation template is less than a preset threshold, it is determined as the target dynamic entropy change predictor, and this predictor can be used subsequently to predict the entropy change of the EEG signal in the anesthetized subject during the current surgical anesthesia process.

[0083] If the first dynamic entropy change predictor does not match the validation template, a new dynamic entropy change predictor is obtained. Based on the calibration template, the parameters of the new dynamic entropy change predictor are predicted to obtain a second dynamic entropy change predictor. The new dynamic entropy change predictor is of a different type than the basic dynamic entropy change predictor. For example, if the basic dynamic entropy change predictor is a generalized autoregressive conditional heteroscedasticity model, then the new dynamic entropy change predictor could be an autoregressive conditional heteroscedasticity model. The parameters of the new dynamic entropy change predictor are estimated again using the maximum likelihood estimation method, similar to step 20. The optimal values ​​of the parameters are determined by maximizing the likelihood function, thus obtaining the second dynamic entropy change predictor.

[0084] Then, a matching verification is performed between the second dynamic entropy change predictor and the verification template. The verification method is the same as in step 30, that is, the mean square error or mean absolute error of the second dynamic entropy change predictor on the verification template is calculated, and it is determined whether the error is within an acceptable range.

[0085] If the second dynamic entropy change predictor matches the validation template, it is used as the target dynamic entropy change predictor. This indicates that after model replacement and recalibration, the new dynamic entropy change predictor meets the prediction requirements and can be used in practical applications.

[0086] If the second dynamic entropy change predictor does not match the validation template, the dynamic entropy change predictor continues to be updated until the predicted dynamic entropy change predictor matches the validation template. The dynamic entropy change predictor that matches the validation template is then used as the target dynamic entropy change predictor. Different types of dynamic entropy change predictors can be tried, such as other improved generalized autoregressive conditional heteroscedasticity models or neural network models. After each model change, steps 20-30 are repeated, continuously adjusting the parameters and performing validation, until a target dynamic entropy change predictor that can accurately predict the entropy change of the EEG signal in anesthetized subjects is found.

[0087] Throughout the training process, cross-validation can be used to improve the model's generalization ability. Cross-validation involves dividing the training data into multiple subsets, using one subset as the validation set and the remaining subsets as the training set in turn, training and validating the model multiple times, and finally taking the average evaluation metric as the model's performance indicator. This allows for a more comprehensive evaluation of the model's performance on different data subsets, avoiding overfitting. Furthermore, the search space for parameters can be optimized, for example, by using grid search or random search to find the optimal combination of parameters within a certain range, improving the accuracy and efficiency of parameter estimation. Simultaneously, the training process can be monitored, recording the evaluation metric and parameter values ​​for each training session to analyze the model's training status and performance trends, providing a reference for subsequent model optimization. For example, a curve can be plotted showing the mean squared error changing with the number of training iterations to observe whether the error gradually decreases and stabilizes. If the error fluctuates or fails to converge, it indicates that the training strategy may need to be adjusted or the model replaced. Through these measures, a high-performance target dynamic entropy change predictor can be trained, providing strong support for the accurate monitoring and prediction of anesthesia states.

[0088] As another implementation of step 200, dynamic entropy change analysis is performed on the EEG signal entropy sequence data to obtain the entropy change feature vector of the anesthetized subject. This includes: acquiring the neural oscillation spectrum corresponding to the EEG signal entropy sequence data, the neural oscillation spectrum including multiple time-frequency fluctuations of the anesthetized subject arranged according to the sampling time sequence nodes in the first time period; decomposing the neural oscillation spectrum into multiple neural oscillation segments according to the time window corresponding to the preset sampling time sequence nodes; repeatedly predicting the parameters of the dynamic entropy change predictor based on the multiple neural oscillation segments; using the parameters obtained from the repeated prediction as the entropy change feature vector; the entropy change feature vector characterizes the influence of the past entropy change characteristics of the anesthetized subject's EEG signal entropy caused by the evolution of the sampling time sequence nodes on the subsequent entropy change characteristics.

[0089] In another embodiment of step 200, the neural oscillation spectrum corresponding to the EEG signal entropy sequence data is obtained. This spectrum contains multiple time-frequency fluctuations of the anesthetized subject arranged according to the sampling time sequence nodes during the first time period. Using time-frequency analysis methods, such as wavelet transform, the EEG signal of the anesthetized subject is converted from the time domain to the time-frequency domain to obtain the neural oscillation spectrum. Then, according to the time window corresponding to the preset sampling time sequence nodes, the neural oscillation spectrum is decomposed into multiple neural oscillation segments. Assuming the preset time window is 30 seconds, then within this 60-minute first time period, the neural oscillation spectrum will be decomposed into 120 neural oscillation segments, each containing time-frequency fluctuations of 30 sampling time sequence nodes. This decomposition facilitates detailed analysis of the local characteristics of neural oscillations at different time periods.

[0090] Next, based on multiple neural oscillation segments, the parameters of the dynamic entropy change predictor are repeatedly predicted. Here, it is assumed that the neural oscillation spectrum is decomposed into x neural oscillation segments, and the parameters of the basic dynamic entropy change predictor are predicted x times, with each of these x predictions matching one of the x neural oscillation segments, x ≥ 2. The b-th prediction of the parameters (1 ≤ b ≤ x) is performed with the dynamic entropy change predictor obtained in the a-th prediction as a prerequisite, where a = b - 1. Specifically, in the b-th prediction of the parameters, the reference time-frequency fluctuation and the sample time-frequency fluctuation are first determined in the b-th neural oscillation segment. The reference time-frequency fluctuation is the time-frequency fluctuation of the last one or more sampling time-series nodes in this segment, and the sample time-frequency fluctuation is the time-frequency fluctuation of the remaining sampling time-series nodes excluding the last one or more sampling time-series nodes. Then, using the dynamic entropy change predictor obtained in the a-th prediction, the inference time-frequency fluctuation corresponding to the reference time-frequency fluctuation is predicted based on the sample time-frequency fluctuation, and the confidence level that the inference time-frequency fluctuation is the reference time-frequency fluctuation is obtained. Confidence can be measured by calculating the similarity between the predicted and actual values, for example, using a probability density function. Then, with the goal of maximizing this confidence, the parameters obtained from the *a*-th prediction are iterated. The iterative process can employ optimization algorithms such as gradient descent to continuously adjust the values ​​of the parameters, gradually increasing the confidence.

[0091] Then, the parameters obtained from repeated predictions are used as the entropy-change feature vector. This vector characterizes the influence of past entropy changes caused by the evolution of sampling time nodes on subsequent entropy changes in the EEG signal entropy of the anesthetized subject. After x predictions, the parameters obtained from the x-th prediction constitute the entropy-change feature vector. For example, after 120 predictions, the parameters of the generalized autoregressive conditional heteroscedasticity model are... and The values ​​of these parameters form the entropy change feature vector. By analyzing this vector, we can understand the pattern of EEG signal entropy changes during anesthesia and the impact of past entropy changes on subsequent ones. A significant change in a parameter within the entropy change feature vector may indicate a change in the anesthetized state, necessitating timely adjustment of the anesthetic drug dosage.

[0092] In practice, the prediction results must be evaluated each time the parameters are predicted to ensure accuracy. Mean squared error (MSE) can be used as an evaluation metric. If the MSE is large, it indicates inaccurate prediction. The iteration step size should be adjusted or the parameters re-initialized, and the iteration repeated until the MSE meets the preset threshold.

[0093] In one implementation, the neural oscillation spectrum is decomposed into x neural oscillation segments. Repeated prediction means that the parameters of the basic dynamic entropy change predictor are predicted x times. The x predictions are matched one by one with the x neural oscillation segments, and x ≥ 2. The b-th prediction in the x predictions is performed with the dynamic entropy change predictor obtained from the a-th prediction as a prerequisite. The parameters obtained from repeated predictions are the parameters obtained from the x-th prediction, 1 ≤ a ≤ x, and a = b - 1.

[0094] The parameters for the b-th prediction include: determining the reference time-frequency fluctuation and the sample time-frequency fluctuation in the b-th neural oscillation segment out of x neural oscillation segments; predicting the inference time-frequency fluctuation corresponding to the reference time-frequency fluctuation based on the sample time-frequency fluctuation using the dynamic entropy change predictor obtained from the a-th prediction, and obtaining the confidence that the inference time-frequency fluctuation is the reference time-frequency fluctuation; and iterating through the parameters obtained from the a-th prediction with the goal of maximizing the confidence to obtain the parameters obtained from the b-th prediction.

[0095] In the b-th neural oscillation segment out of x segments, the reference time-frequency fluctuation and the sample time-frequency fluctuation are determined. The neural oscillation segment is obtained by decomposing the neural oscillation spectrum corresponding to the entropy sequence data of the anesthetized subject's EEG signal based on a preset sampling time window. Assuming a preset time window of 30 seconds, EEG signals are collected at 1-second sampling intervals within the first 60 minutes after the start of surgery. The neural oscillation spectrum is obtained through wavelet transform and then decomposed into 120 neural oscillation segments. In the b-th neural oscillation segment, the reference time-frequency fluctuation is the time-frequency fluctuation of the last one or more sampling time nodes in the segment, and the sample time-frequency fluctuation is the time-frequency fluctuation of the remaining sampling time nodes excluding the last one or more sampling time nodes. For example, if the time-frequency fluctuation of the last 5 sampling time nodes is selected as the reference time-frequency fluctuation, then the time-frequency fluctuation of the first 25 sampling time nodes in the segment is the sample time-frequency fluctuation. The reference time-frequency fluctuation represents the current or future state of the segment, while the sample time-frequency fluctuation is used to infer the reference time-frequency fluctuation.

[0096] The dynamic entropy change predictor obtained from the a-th prediction predicts the inference time-frequency fluctuation corresponding to the reference time-frequency fluctuation based on the sample time-frequency fluctuation, and obtains the confidence level that the inference time-frequency fluctuation is the reference time-frequency fluctuation. Assuming a generalized autoregressive conditional heteroscedasticity model is used as the dynamic entropy change predictor, a set of parameters has already been obtained in the a-th prediction. and The value of is then substituted into the generalized autoregressive conditional heteroscedasticity model obtained from the a-th prediction, and the inference time-frequency fluctuation corresponding to the reference time-frequency fluctuation is calculated according to the model's recursive formula.

[0097] To obtain the confidence level that the inference time-frequency fluctuation is the same as the reference time-frequency fluctuation, the probability density function method can be used. Assuming that the inference and reference time-frequency fluctuations follow a normal distribution, the probability density value of the inference time-frequency fluctuation near the reference time-frequency fluctuation can be calculated and used as the confidence level. For example, if the inference time-frequency fluctuation is... The reference time-frequency fluctuation is r, and their mean is The standard deviation is The confidence level can be expressed by the probability density function of the normal distribution. Calculate the confidence level by substituting x=r into the formula. The confidence level reflects the degree of similarity between the inferred time-frequency fluctuation and the reference time-frequency fluctuation; the higher the confidence level, the more reliable the inference result.

[0098] To maximize the confidence score, the parameters obtained from the *a*th prediction are iterated to obtain the parameters from the *b*th prediction. Gradient descent can be used for this iterative process. Gradient descent is an optimization algorithm that continuously updates the values ​​of the parameters along the negative gradient of the objective function, gradually decreasing the value of the objective function. In this scenario, the objective function is the negative value of the confidence score; therefore, to maximize the confidence score, we must minimize the negative value of the confidence score.

[0099] Let the objective function be ,in It is a parameter vector The iterative formula for the gradient descent algorithm is: ,in This is the current parameter value. These are the updated parameter values. It's the learning rate, which controls the step size of each iteration. Is the objective function in The gradient at that point.

[0100] First, calculate the gradient of the objective function at the parameter obtained in the a-th prediction. Then, update the parameter value according to the iterative formula of the gradient descent algorithm. After each update, use the updated parameter again to predict the inference time-frequency fluctuation corresponding to the reference time-frequency fluctuation based on the sample time-frequency fluctuation, and calculate the new confidence level. Repeat this process until the confidence level no longer increases or reaches the preset number of iterations. The parameter obtained at this point is the parameter obtained in the b-th prediction.

[0101] Through the above steps, the parameters of the dynamic entropy change predictor can be continuously optimized, making it better adaptable to the changes in EEG signal entropy during anesthesia. As the number of predictions increases, the performance of the dynamic entropy change predictor gradually improves, enabling it to more accurately predict the EEG signal entropy changes of the anesthetized subject, providing a more reliable basis for adjusting the dosage of anesthetic drugs. For example, if, after multiple predictions, the value of the autoregressive squared error weight gradually increases, it indicates that the influence of past squared errors on the current time-frequency fluctuation is increasing. This suggests that the anesthetized state of the subject may be becoming more unstable, requiring close monitoring and timely adjustment of the anesthesia plan.

[0102] As one implementation method, the EEG signal entropy sequence data includes the EEG signal entropy of y sampling time-series nodes in the first time period, where y ≥ 2; temporal phase coupling modeling is performed through a target temporal phase coupling modeler, which includes y temporal phase coupling modeling layers, and the EEG signal entropy of each of the y temporal phase coupling modeling layers is matched one by one with the EEG signal entropy of the y sampling time-series nodes; step 300, temporal phase coupling modeling is performed on the EEG signal entropy sequence data to obtain the phase synchronization feature matrix of the anesthetized object, including: through each temporal phase coupling modeling layer, temporal phase coupling modeling is performed on the EEG signal entropy of the corresponding sampling time-series node and the EEG signal entropy of the sampling time-series nodes before the corresponding sampling time-series node to obtain The temporal phase coupling modeling results of each temporal phase coupling modeling layer; the temporal phase coupling modeling result of the y-th temporal phase coupling modeling layer in the y temporal phase coupling modeling layers is used as the phase synchronization feature matrix. The phase synchronization feature matrix represents the EEG signal entropy of the y-th sampling time sequence node and the synchronization status of the EEG signal entropy of the y-1 sampling time sequence nodes before the y-th sampling time sequence node; or, the statistical results of the temporal phase coupling modeling results of the y temporal phase coupling modeling layers are used as the phase synchronization feature matrix. The phase synchronization feature matrix represents the EEG signal entropy of each sampling time sequence node in the y sampling time sequence nodes and the comprehensive synchronization status of the EEG signal entropy of the sampling time sequence nodes before each sampling time sequence node.

[0103] The EEG signal entropy sequence data contains the EEG signal entropy of y sampling time nodes in the first time period, where y ≥ 2. Temporal phase coupling modeling is performed by a target temporal phase coupling modeler, such as an LSTM (Long Short-Term Memory) network or an RNN (Recurrent Neural Network), which contains y temporal phase coupling modeling layers that are matched one by one with the EEG signal entropy of the y sampling time nodes.

[0104] By performing temporal phase coupling modeling on the EEG signal entropy of the corresponding sampling time node and the EEG signal entropy of the sampling time node preceding the corresponding sampling time node, the temporal phase coupling modeling results of each temporal phase coupling modeling layer are obtained. Temporal phase coupling reflects the degree of phase correlation between EEG signals at different times, reflecting the synchronicity of neural activity in different brain regions. For example, during anesthesia, if there is strong phase coupling in the temporal domain between the EEG signals at two certain times, it indicates that the neural activity in certain brain regions at those two times is synchronous.

[0105] Taking LSTM as an example, in the f-th temporal phase coupling modeling layer (1≤f≤y), it is responsible for performing temporal phase coupling modeling on the EEG signal entropy of the f-th sampling time node and the EEG signal entropy of the e sampling time nodes preceding the f-th sampling time node, where 1≤e≤y and e=f-1. The characteristics of the EEG signal entropy of the previous e sampling time nodes are maintained in the cell state of the e-th temporal phase coupling modeling layer, and the cell state and temporal phase coupling modeling results of the e-th temporal phase coupling modeling layer are loaded into the f-th temporal phase coupling modeling layer.

[0106] The core of LSTM is the cell state, which can preserve long-term information during sequence processing. In the modeling process of the f-th temporal phase coupling modeling layer, a forgetting gate is first performed. The forgetting gate determines which information needs to be discarded from the cell state; its calculation formula is as follows: ; where f t It is the output of the forget gate. It's the sigmoid function, which maps input values ​​to the interval (0, 1). W f It is the weight matrix of the forget gate, h t-1 It is the hidden state of the previous time step (i.e., the e-th temporal phase coupling modeling layer), x t It is the EEG signal entropy at the current time (i.e., the f-th sampling time node), b f It is the bias term of the forget gate. The forget gate allows us to decide which information in the cell state to retain or discard based on the current input and the hidden state of the previous time step.

[0107] Next comes the memory gate processing, which determines which new information needs to be added to the cell state. The calculation formula is as follows:

[0108] ;

[0109] ;

[0110] Among them, i tThe output of the memory gate is also mapped to the (0, 1) interval using the sigmoid function. This represents the candidate cell state. The tanh function maps the input values ​​to the interval (-1, 1). W i W C It is the weight matrix, b i b C This is the bias term. Through a memory gate, candidate cell states are generated based on the current input and the hidden state from the previous time step, and the number of candidate cell states to add to the cell state is determined. Then, the first cell state (i.e., the cell state retained after the forget gate) and the second cell state (i.e., the candidate cell states added as determined by the memory gate) are merged to obtain the cell state of the f-th temporal phase coupling modeling layer. The merging formula is: Where C_{t} is the cell state at the current moment, This represents element-wise multiplication. In this way, the cell state can be updated based on the current input and historical information.

[0111] Finally, based on the temporal phase coupling modeling results of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node, output analysis is performed on the cell state of the f-th temporal phase coupling modeling layer to obtain the temporal phase coupling modeling results of the f-th temporal phase coupling modeling layer. The output analysis determines the output weights based on the temporal phase coupling modeling results of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node. Based on these output weights, the information to be output from the cell state of the f-th temporal phase coupling modeling layer is used as the temporal phase coupling modeling result. The calculation formula is as follows:

[0112] ;

[0113] ;

[0114] Among them, o t It is the output of the output gate, h t It is the hidden state at the current moment, that is, the temporal phase coupling modeling result of the f-th temporal phase coupling modeling layer, W o It is the weight matrix of the output gate, b o It is the bias term of the output gate.

[0115] Next, the phase synchronization feature matrix is ​​determined. There are two methods. One is to use the temporal phase coupling modeling result of the y-th temporal phase coupling modeling layer as the phase synchronization feature matrix. This matrix characterizes the synchronization of the EEG signal entropy of the y-th sampling time-series node with the EEG signal entropy of the y-1 sampling time-series nodes preceding it. For example, in an EEG signal entropy sequence containing 100 sampling time-series nodes, the phase synchronization relationship between the EEG signal entropy of the 100th sampling time-series node and the EEG signal entropy of the previous 99 sampling time-series nodes is reflected through the output of the 100th temporal phase coupling modeling layer; this output is the phase synchronization feature matrix. The other method is to use the statistical result of the temporal phase coupling modeling results of the y-th temporal phase coupling modeling layers as the phase synchronization feature matrix. It characterizes the overall synchronization of the EEG signal entropy of each of the y sampling time-series nodes with the EEG signal entropy of the sampling time-series nodes preceding it. Statistical operations such as summing and averaging can be performed on the y time-domain phase coupling modeling results to obtain a comprehensive matrix as the phase synchronization feature matrix. For example, adding each time-domain phase coupling modeling result element-wise and then dividing by y yields the comprehensive phase synchronization feature matrix.

[0116] This temporal phase coupling modeling accurately captures the long-range synchronicity between EEG signal entropies at different sampling time points in anesthetized subjects. The phase synchronization feature matrix provides an important basis for subsequent generation of consciousness state fusion features and prediction of anesthesia status. For example, doctors can use the phase synchronization feature matrix to determine the coordinated operation of different brain regions during anesthesia. If abnormal changes occur in the phase synchronization of certain regions, it may indicate a problem with the anesthesia status, requiring timely adjustment of the anesthetic drug dosage.

[0117] In one implementation, the f-th time-domain phase coupling modeling layer in the y time-domain phase coupling modeling layers is used to perform time-domain phase coupling modeling on the EEG signal entropy of the f-th sampling time-series node and the EEG signal entropy of the e sampling time-series nodes preceding the f-th sampling time-series node, to obtain the time-domain phase coupling modeling result of the f-th time-domain phase coupling modeling layer; the characteristics of the EEG signal entropy of the e sampling time-series nodes are maintained in the cell state of the e-th time-domain phase coupling modeling layer, and the cell state of the e-th time-domain phase coupling modeling layer and the time-domain phase coupling modeling result of the e-th time-domain phase coupling modeling layer are loaded into the f-th time-domain phase coupling modeling layer; 1≤e≤y, e=f-1. The temporal phase coupling modeling process of the f-th temporal phase coupling modeling layer includes: based on the temporal phase coupling modeling results of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node, performing forget gate processing on the cell states of the e-th temporal phase coupling modeling layer to determine the first cell state to be maintained in the cell states of the e-th temporal phase coupling modeling layer; and based on the temporal phase coupling modeling results of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node, performing memory gate processing on the f-th temporal phase coupling modeling layer to determine the e-th temporal phase coupling modeling layer. The second cell state of the cell state of the f-th time-domain phase coupling modeling layer is to be incorporated into the temporal phase coupling modeling result of the phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node; the first cell state and the second cell state are merged to obtain the cell state of the f-th time-domain phase coupling modeling layer; based on the temporal phase coupling modeling result of the e-th time-domain phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node, the cell state of the f-th time-domain phase coupling modeling layer is output analyzed to obtain the temporal phase coupling modeling result of the f-th time-domain phase coupling modeling layer.

[0118] Specifically, based on the temporal phase coupling modeling results of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node, a forgetting gate is applied to the cell state of the e-th temporal phase coupling modeling layer to determine the first cell state to be maintained in the e-th temporal phase coupling modeling layer. Here, it is assumed that LSTM is used as the temporal phase coupling modeler. The core of LSTM is the cell state, which can preserve long-term information during sequence processing. The forgetting gate determines which information needs to be discarded from the cell state to avoid accumulating too much useless information, thus ensuring that the model can focus on the currently important information. The calculation process can be referred to the aforementioned introduction to LSTM, and will not be repeated here. For example, suppose that the EEG signal features of the anesthetized subject at the e-th sampling time node contain some information about the early anesthesia induction stage, while at the f-th sampling time node, the anesthetized subject has entered a stable anesthesia maintenance stage, and the information from the early anesthesia induction stage may no longer be important for the current modeling. Through the forgetting gate, a forgetting coefficient f is calculated based on the results of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node. t Each element of this forgetting coefficient is between (0, 1). Then, this forgetting coefficient is multiplied element-wise with the cell state of the e-th temporal phase-coupled modeling layer to obtain the first cell state. If an element is close to 0, it indicates that the corresponding information will be forgotten; if it is close to 1, it indicates that the information will be retained.

[0119] Next, based on the temporal phase coupling modeling results of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node, a memory gate is applied to the f-th temporal phase coupling modeling layer to determine the second cell state to be incorporated into the cell state of the f-th temporal phase coupling modeling layer from the temporal phase coupling modeling results of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node. For example, at the f-th sampling time node, some new features appear in the EEG signal of the anesthetized subject. These features may be related to the dosage adjustment of the anesthetic drug or the physiological response of the anesthetized subject. Through the memory gate, based on the results of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node, it is determined which new features should be added to the cell state. If an element of the memory gate is close to 1, it means that the corresponding candidate cell state element will be completely added to the cell state; if it is close to 0, it means that the element will not be added.

[0120] Then, the first and second cell states are merged to obtain the cell state of the f-th temporal phase coupling modeling layer. For example, the first cell state retains some stable EEG signal features of the anesthetized subject at previous sampling time nodes, while the second cell state contains new features at the f-th sampling time node. The merged cell state contains both historical information and new information, and can more comprehensively reflect the current EEG state of the anesthetized subject.

[0121] Finally, based on the temporal phase coupling modeling results of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node, output analysis is performed on the cell state of the f-th temporal phase coupling modeling layer to obtain the temporal phase coupling modeling results of the f-th temporal phase coupling modeling layer. The purpose of output analysis is to determine the information to be output based on the current cell state and input information. For example, at the f-th sampling time node, the output of the output gate is determined based on the results of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node. If an element of the output gate is close to 1, it indicates that the corresponding cell state element will be output completely; if it is close to 0, it indicates that the element will not be output. In this way, information related to the current modeling can be selectively output based on the current input and cell state to obtain the temporal phase coupling modeling results of the f-th temporal phase coupling modeling layer. This result reflects the phase coupling relationship between the EEG signal entropy of the f-th sampling time node and the EEG signal entropy of the previous sampling time nodes.

[0122] In practical applications, the weight matrix and bias terms of the LSTM need to be trained to optimize the model's performance. A feasible training method is the backpropagation algorithm, which continuously adjusts the values ​​of the weight matrix and bias terms by minimizing the error between the predicted and actual results. Simultaneously, to avoid overfitting, regularization methods, such as L1 or L2 regularization, can be used to constrain the model. Through these steps and techniques, the temporal phase coupling modeling of the EEG signal entropy of anesthetized subjects can be accurately performed, providing strong support for the monitoring and adjustment of the anesthesia state. The output analysis specifically involves determining the output weights based on the temporal phase coupling modeling result of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node. The information required to be output in the cell state of the f-th temporal phase coupling modeling layer is then used as the temporal phase coupling modeling result of the f-th temporal phase coupling modeling layer based on these output weights.

[0123] As one implementation, the entropy-change feature vector includes multiple entropy-change feature vector elements, and the phase synchronization feature matrix includes multiple phase synchronization feature matrix elements. Therefore, before step 400, generating the consciousness state fusion feature of the anesthetized subject based on the entropy-change feature vector and the phase synchronization feature matrix, the method provided in this application embodiment may further include: obtaining the first influence weight of each entropy-change feature vector element in the entropy-change feature vector on the EEG signal entropy of the sampling time-series node in the prediction of the second time period; determining the entropy-change feature vector elements in the entropy-change feature vector whose first influence weight satisfies a weight threshold, and obtaining the determined entropy-change feature vector; obtaining the second influence weight of each phase synchronization feature matrix element in the phase synchronization feature matrix on the EEG signal entropy of the sampling time-series node in the prediction of the second time period; determining the phase synchronization feature matrix elements in the phase synchronization feature matrix whose second influence weight satisfies a weight threshold, and obtaining the determined phase synchronization feature matrix.

[0124] Based on this, step 400 generates consciousness state fusion features of the anesthetized subject based on the entropy change feature vector and the phase synchronization feature matrix, including: generating consciousness state fusion features based on the determined entropy change feature vector and the determined phase synchronization feature matrix.

[0125] Specifically, the first influence weight of each entropy change feature vector element in the entropy change feature vector is obtained in relation to the predicted EEG signal entropy at the sampling time sequence node in the second time period. The entropy change feature vector elements whose first influence weights satisfy the weight threshold are identified, thus obtaining the determined entropy change feature vector. The entropy change feature vector is obtained by performing dynamic entropy change analysis on the EEG signal entropy sequence data of the anesthetized subject. It contains multiple entropy change feature vector elements, each representing different aspects of entropy change information. The first influence weight indicates the importance of each element in predicting the EEG signal entropy in the second time period. Various methods can be used to calculate the first influence weight, such as a linear regression model. In linear regression, it is assumed that the entropy change feature vector is... If the entropy of the EEG signal in the second time period is y, then the linear regression model can be expressed as: ,in It refers to the weight of the i-th entropy-change eigenvector element. These are the error terms. The weights are estimated by minimizing the sum of squares of the error terms. The weight threshold is a pre-defined value used to filter out elements that have a significant impact on the prediction. For example, setting the weight threshold to 0.1 will identify elements in the entropy-change feature vector whose absolute weights are greater than 0.1, and these elements will form the determined entropy-change feature vector. This step helps to remove information with less impact on the prediction, improving the accuracy of subsequent predictions.

[0126] Then, the second influence weight of each element of the phase synchronization feature matrix on the predicted EEG signal entropy of the sampling time-series nodes in the second time period is obtained. The phase synchronization feature matrix elements whose second influence weights satisfy the weight threshold are determined, thus obtaining the determined phase synchronization feature matrix. The phase synchronization feature matrix is ​​obtained through temporal phase coupling modeling, reflecting the long-range synchronization between the EEG signal entropies of different sampling time-series nodes. Similarly, the second influence weight can be calculated using methods such as linear regression. Assume the phase synchronization feature matrix is ​​M, where the elements are m. ij The weight of each element in predicting the entropy of the EEG signal in the second time period is obtained through a method similar to linear regression. Then, based on a pre-set weight threshold, elements whose weights meet the requirements are selected to form the determined phase synchronization feature matrix. For example, in a 10×10 phase synchronization feature matrix, after calculation and selection, only some elements may have weights that meet the threshold requirements; these elements constitute the determined phase synchronization feature matrix.

[0127] Based on the determined entropy-change feature vector and the determined phase synchronization feature matrix, a consciousness state fusion feature is generated. There are several ways to generate this feature. One method is concatenation, where the determined entropy-change feature vector and the determined phase synchronization feature matrix are unfolded into vectors and then concatenated sequentially to form a new feature vector as the consciousness state fusion feature. Another method is weighted averaging. Different weights are assigned to the determined entropy-change feature vector and the determined phase synchronization feature matrix, and then the weighted averages are summed to obtain the consciousness state fusion feature. Alternatively, convolution can be performed, using a convolution kernel to perform a sliding convolution operation on the determined entropy-change feature vector and the determined phase synchronization feature matrix to extract their feature information and generate the consciousness state fusion feature. Furthermore, feature fusion and feature cross processing are also feasible methods. Feature fusion uses a specific algorithm to organically fuse the features of both, while feature cross processing cross-combines the elements of both to uncover potential relationships. These methods enable the comprehensive utilization of information from the determined entropy change feature vector and the determined phase synchronization feature matrix to generate fusion features that effectively reflect the consciousness state of the anesthetized subject, providing strong support for subsequent anesthesia state prediction.

[0128] Based on the foregoing embodiments, this application provides a brainwave consciousness monitoring data analysis device. The various units and modules included in the device can be implemented by a processor in a computer device. Figure 2 This application provides a schematic diagram of the composition of an EEG consciousness monitoring data analysis device, as shown in the embodiments of the present application. Figure 2As shown, the EEG consciousness monitoring data analysis device 200 includes:

[0129] Data acquisition module 210 is used to acquire EEG signal entropy sequence data of anesthetized subjects. The EEG signal entropy sequence data includes multiple EEG signal entropies of anesthetized subjects arranged according to sampling time nodes in the first time period.

[0130] The entropy change analysis module 220 is used to perform dynamic entropy change analysis on EEG signal entropy sequence data to obtain the entropy change feature vector of the anesthetized object. The dimensions of the entropy change feature vector include dual-frequency exponent, state entropy, response entropy and long-range correlation. The entropy change feature vector characterizes the dynamic change characteristics of the EEG signal entropy of the anesthetized object due to the evolution of sampling time sequence nodes.

[0131] Phase modeling module 230 is used to perform temporal phase coupling modeling on EEG signal entropy sequence data to obtain the phase synchronization feature matrix of the anesthetized subject. The phase synchronization feature matrix characterizes the topological structure of the phase relationship between multi-channel EEG signals and indicates the long-range synchronicity between EEG signal entropies at different sampling time nodes.

[0132] The feature fusion module 240 is used to generate fusion features of the consciousness state of the anesthetized subject based on the entropy-change feature vector and the phase synchronization feature matrix.

[0133] The state prediction module 250 predicts the anesthesia state of the anesthetized subject based on the consciousness state fusion features, and obtains the EEG signal entropy of the sampling time sequence node of the anesthetized subject in the second time period, which is located after the first time period.

[0134] The descriptions of the apparatus embodiments above are similar to those of the method embodiments above, and have similar beneficial effects. In some embodiments, the functions or modules included in the apparatus provided in this application can be used to perform the methods described in the method embodiments above. For technical details not disclosed in the apparatus embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0135] Figure 3 A hardware entity diagram of a computer system provided in an embodiment of this application is shown below. Figure 3 As shown, the hardware entity of the computer system 1000 includes a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program that can run on the processor 1001, and the processor 1001 executes the program to implement the steps in the method of any of the above embodiments.

[0136] The memory 1002 stores computer programs that can run on the processor. The memory 1002 is configured to store instructions and applications that can be executed by the processor 1001. It can also cache data to be processed or already processed (e.g., image data, audio data, voice communication data, and video communication data) of the processor 1001 and various modules in the computer system 1000. It can be implemented by flash memory or random access memory (RAM).

[0137] When the processor 1001 executes the program, it implements any of the steps of the above-mentioned method for analyzing EEG consciousness monitoring data applied to anesthesia management. The processor 1001 typically controls the overall operation of the computer system 1000.

[0138] This application provides a computer storage medium storing one or more programs that can be executed by one or more processors to implement the steps of the EEG consciousness monitoring data analysis method for anesthesia management as described in any of the above embodiments.

[0139] It should be noted that the descriptions of the storage medium and device embodiments above are similar to those of the method embodiments above, and have similar beneficial effects. For technical details not disclosed in the storage medium and device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding. The processor described above can be at least one of the following: Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that the electronic device implementing the above processor function can also be other types, and this application does not specifically limit the specific types.

[0140] The aforementioned computer storage media / memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM), etc.; or it can be various terminals that include one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0141] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0142] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence or the part that contributes to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, magnetic disks, or optical disks.

[0143] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for analyzing EEG consciousness monitoring data applied to anesthesia management, characterized in that, include: Obtain EEG signal entropy sequence data of the anesthetized subject, wherein the EEG signal entropy sequence data includes multiple EEG signal entropies of the anesthetized subject arranged according to sampling time nodes in the first time period; Dynamic entropy change analysis is performed on the EEG signal entropy sequence data to obtain the entropy change feature vector of the anesthetized object. The dimensions of the entropy change feature vector include dual-frequency exponent, state entropy, response entropy and long-range correlation. The entropy change feature vector characterizes the dynamic change characteristics of the EEG signal entropy of the anesthetized object due to the evolution of sampling time sequence nodes. Temporal phase coupling modeling is performed on the EEG signal entropy sequence data to obtain the phase synchronization feature matrix of the anesthetized subject. The phase synchronization feature matrix characterizes the topological structure of the phase relationship between multi-channel EEG signals and indicates the long-range synchronicity between EEG signal entropies at different sampling time nodes. Based on the entropy change feature vector and the phase synchronization feature matrix, the consciousness state fusion feature of the anesthetized subject is generated; Based on the consciousness state fusion features, the anesthesia state of the anesthetized subject is predicted, and the EEG signal entropy of the sampling time sequence node of the anesthetized subject in the second time period is obtained, wherein the second time period is after the first time period.

2. The method according to claim 1, characterized in that, The step of performing dynamic entropy change analysis on the EEG signal entropy sequence data to obtain the entropy change feature vector of the anesthetized subject includes: Obtain the neural oscillation spectrum corresponding to the EEG signal entropy sequence data, wherein the neural oscillation spectrum includes multiple time-frequency fluctuations of the anesthetized subject arranged according to the sampling time sequence nodes in the first time period; The target dynamic entropy change predictor predicts the anesthetized object based on the neural oscillation fluctuation spectrum to obtain the time-frequency fluctuation of the anesthetized object at the sampling time sequence node in the second time period; The time-frequency fluctuation of the sampling time sequence node of the anesthetized object in the second time period is used as the entropy change feature vector; the entropy change feature vector characterizes the entropy change characteristics of the EEG signal entropy of the anesthetized object due to the evolution of the sampling time sequence node.

3. The method according to claim 1, characterized in that, The step of performing dynamic entropy change analysis on the EEG signal entropy sequence data to obtain the entropy change feature vector of the anesthetized subject includes: Obtain the neural oscillation spectrum corresponding to the EEG signal entropy sequence data, wherein the neural oscillation spectrum includes multiple time-frequency fluctuations of the anesthetized subject arranged according to the sampling time sequence nodes in the first time period; Based on the time window corresponding to the preset sampling timing node, the neural oscillation spectrum is decomposed into multiple neural oscillation segments; In each of the neural oscillation segments, a reference time-frequency fluctuation and a sample time-frequency fluctuation are determined, wherein the reference time-frequency fluctuation is the time-frequency fluctuation of the last one or more sampling timing nodes in the neural oscillation segment, and the sample time-frequency fluctuation is the time-frequency fluctuation of the remaining sampling timing nodes in the neural oscillation segment excluding the last one or more sampling timing nodes. The target dynamic entropy change predictor infers the inferred time-frequency fluctuation corresponding to the reference time-frequency fluctuation in each neural oscillation segment based on the sample time-frequency fluctuation in each neural oscillation segment. Based on the error between the reference time-frequency fluctuation and the corresponding inferred time-frequency fluctuation in each neural oscillation segment, the oscillation error result corresponding to each neural oscillation segment is determined; The oscillation error results corresponding to each of the neural oscillation segments are used as the entropy change feature vector; the entropy change feature vector characterizes the abnormal oscillation of the EEG signal entropy of the anesthetized object caused by the evolution of the sampling time sequence nodes.

4. The method according to claim 2 or 3, characterized in that, The method further includes: Obtain a basic dynamic entropy change predictor and a training template, wherein the training template includes a tuning template and a validation template; Based on the tuning template, the parameters of the basic dynamic entropy change predictor are predicted to obtain the first dynamic entropy change predictor. Matching and verification are performed between the first dynamic entropy change predictor and the verification template; If the first dynamic entropy change predictor matches the verification template, the first dynamic entropy change predictor is used as the target dynamic entropy change predictor. If the first dynamic entropy change predictor does not match the verification template, a new dynamic entropy change predictor is obtained, and the parameters of the new dynamic entropy change predictor are predicted based on the tuning template to obtain a second dynamic entropy change predictor. Matching verification is performed between the second dynamic entropy change predictor and the verification template; If the second dynamic entropy change predictor matches the verification template, the second dynamic entropy change predictor is used as the target dynamic entropy change predictor. If the second dynamic entropy change predictor does not match the verification template, the dynamic entropy change predictor continues to be updated until the predicted dynamic entropy change predictor matches the verification template. The dynamic entropy change predictor that matches the verification template is then used as the target dynamic entropy change predictor. The new dynamic entropy change predictor is of a different type than the basic dynamic entropy change predictor.

5. The method according to claim 1, characterized in that, The step of performing dynamic entropy change analysis on the EEG signal entropy sequence data to obtain the entropy change feature vector of the anesthetized subject includes: Obtain the neural oscillation spectrum corresponding to the EEG signal entropy sequence data, wherein the neural oscillation spectrum includes multiple time-frequency fluctuations of the anesthetized subject arranged according to the sampling time sequence nodes in the first time period; Based on the time window corresponding to the preset sampling timing node, the neural oscillation spectrum is decomposed into multiple neural oscillation segments; Based on the multiple neural oscillation segments, the parameters of the dynamic entropy change predictor are repeatedly predicted; The parameters obtained from repeated predictions are used as the entropy change feature vector; the entropy change feature vector characterizes the influence of the past entropy change characteristics of the EEG signal entropy of the anesthetized object caused by the evolution of sampling time sequence nodes on the subsequent entropy change characteristics.

6. The method according to claim 5, characterized in that, The neural oscillation spectrum is decomposed into x neural oscillation segments. The repeated prediction means that the parameters of the basic dynamic entropy change predictor are predicted x times. The x predictions are matched one by one with the x neural oscillation segments, and x ≥ 2. The b-th prediction in the x predictions is executed with the dynamic entropy change predictor obtained from the a-th prediction as a prerequisite. The parameters obtained from the repeated predictions are the parameters obtained from the x-th prediction, 1 ≤ a ≤ x, a = b - 1. The parameters for the b-th prediction include: In the b-th neural oscillation segment out of the x neural oscillation segments, determine the reference time-frequency fluctuation and the sample time-frequency fluctuation; The dynamic entropy change predictor obtained by the a-th prediction predicts the inference time-frequency fluctuation corresponding to the reference time-frequency fluctuation based on the sample time-frequency fluctuation, and obtains the confidence that the inference time-frequency fluctuation is the reference time-frequency fluctuation. With the goal of maximizing the confidence level, the parameters obtained from the a-th prediction are iterated to obtain the parameters obtained from the b-th prediction.

7. The method according to claim 1, characterized in that, The EEG signal entropy sequence data includes the EEG signal entropy of y sampling time nodes in the first time period, where y ≥ 2; the temporal phase coupling modeling is performed through a target temporal phase coupling modeler, which includes y temporal phase coupling modeling layers, and the y temporal phase coupling modeling layers are matched one by one with the EEG signal entropy of the y sampling time nodes; the temporal phase coupling modeling of the EEG signal entropy sequence data to obtain the phase synchronization feature matrix of the anesthetized object includes: Through each of the time-domain phase coupling modeling layers, time-domain phase coupling modeling is performed on the EEG signal entropy of the corresponding sampling time sequence node and the EEG signal entropy of the sampling time sequence node before the corresponding sampling time sequence node, so as to obtain the time-domain phase coupling modeling result of each of the time-domain phase coupling modeling layers. The phase synchronization feature matrix is ​​obtained by using the temporal phase coupling modeling result of the y-th temporal phase coupling modeling layer among the y temporal phase coupling modeling layers. The phase synchronization feature matrix represents the EEG signal entropy of the y-th sampling time sequence node and the synchronization status of the EEG signal entropy of the y-1 sampling time sequence nodes preceding the y-th sampling time sequence node. Alternatively, the phase synchronization feature matrix is ​​obtained by using the statistical result of the temporal phase coupling modeling result of the y temporal phase coupling modeling layers. The phase synchronization feature matrix represents the EEG signal entropy of each sampling time sequence node among the y sampling time sequence nodes and the comprehensive synchronization status of the EEG signal entropy of the sampling time sequence nodes preceding each sampling time sequence node.

8. The method according to claim 7, characterized in that, The f-th time-domain phase coupling modeling layer in the y-th time-domain phase coupling modeling layer is used to perform time-domain phase coupling modeling on the EEG signal entropy of the f-th sampling time-series node and the EEG signal entropy of the e sampling time-series nodes preceding the f-th sampling time-series node, to obtain the time-domain phase coupling modeling result of the f-th time-domain phase coupling modeling layer; the characteristics of the EEG signal entropy of the e sampling time-series nodes are maintained in the cell state of the e-th time-domain phase coupling modeling layer, and the cell state of the e-th time-domain phase coupling modeling layer and the time-domain phase coupling modeling result of the e-th time-domain phase coupling modeling layer are loaded into the f-th time-domain phase coupling modeling layer; 1≤e≤y, e=f-1; The time-domain phase coupling modeling process of the f-th time-domain phase coupling modeling layer includes: Based on the temporal phase coupling modeling result of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling timing node, the cell state of the e-th temporal phase coupling modeling layer is subjected to forget gate processing to determine the first cell state to be maintained in the cell state of the e-th temporal phase coupling modeling layer. Based on the temporal phase coupling modeling result of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node, a memory gate is applied to the f-th temporal phase coupling modeling layer to determine the second cell state that is to be incorporated into the cell state of the f-th temporal phase coupling modeling layer from the temporal phase coupling modeling result of the e-th temporal phase coupling modeling layer and the EEG signal entropy of the f-th sampling time node; The first cell state and the second cell state are merged to obtain the cell state of the f-th temporal phase coupling modeling layer; Based on the time-domain phase coupling modeling result of the e-th time-domain phase coupling modeling layer and the EEG signal entropy of the f-th sampling time-series node, the cell state of the f-th time-domain phase coupling modeling layer is output analyzed to obtain the time-domain phase coupling modeling result of the f-th time-domain phase coupling modeling layer.

9. The method according to claim 1, characterized in that, The entropy-change feature vector includes multiple entropy-change feature vector elements, and the phase synchronization feature matrix includes multiple phase synchronization feature matrix elements. Before generating the consciousness state fusion features of the anesthetized subject based on the entropy change feature vector and the phase synchronization feature matrix, the method further includes: Obtain the first influence weight of each entropy change feature vector element in the entropy change feature vector on the predicted EEG signal entropy of the sampling time sequence node in the second time period, determine the entropy change feature vector elements in the entropy change feature vector whose first influence weight satisfies the weight threshold, and obtain the determined entropy change feature vector. Obtain the second influence weight of each phase synchronization feature matrix element in the phase synchronization feature matrix on the EEG signal entropy of the sampling timing node in the second time period, determine the phase synchronization feature matrix elements in the phase synchronization feature matrix whose second influence weights satisfy the weight threshold, and obtain the determined phase synchronization feature matrix. The process of generating the consciousness state fusion feature of the anesthetized subject based on the entropy change feature vector and the phase synchronization feature matrix includes: Based on the determined entropy change feature vector and the determined phase synchronization feature matrix, the consciousness state fusion feature is generated.

10. A computer system comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 9.

Images