Anesthesia depth multi-level classification information evaluation method and system

Abstract

The invention relates to the technical field of medical signal processing, and discloses an anesthesia depth multi-level classification information assessment method and system, and the method comprises the steps: collecting an electroencephalogram signal, and carrying out the preprocessing operation of the electroencephalogram signal, and obtaining signals of five frequency bands; global field power is calculated for all electroencephalogram signals of each frequency band, and a peak topographic map is identified; performing clustering analysis to extract a micro-state template, and matching and reconstructing a micro-state time sequence; and establishing a feature screening and classification identification model, inputting each frequency band signal into the feature screening and classification identification model, screening the extracted micro-state features, carrying out multi-level classification on the anesthesia depth, and finally outputting an anesthesia depth evaluation result. The system comprises an electroencephalogram collecting and processing module, a clustering reconstruction module and an anesthesia depth evaluation module. According to the method, the preprocessed electroencephalogram signals are decomposed through Hilbert-Huang transform, and anesthesia depth evaluation is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical signal processing, and particularly to a method and system for evaluating multi-level classification information of anesthetic depth. Background Art

[0002] During the surgical anesthesia process, accurately evaluating the anesthetic depth of a patient is crucial for ensuring surgical safety. Insufficient anesthetic depth may cause intraoperative awareness in the patient, resulting in serious psychological trauma; while excessive anesthesia may cause adverse reactions such as blood pressure reduction and respiratory depression, increasing the surgical risk. Currently, there are mainly three methods for clinical anesthetic depth evaluation: clinical sign observation, electroencephalogram (EEG) signal analysis, and drug concentration monitoring. Clinical sign observation mainly relies on changes in vital signs such as blood pressure and heart rate, as well as responses such as pupil reflex and eyelid reflex. However, this method has obvious disadvantages of strong subjectivity and large lag. Although EEG signal analysis methods (such as BIS index, entropy index, etc.) provide quantitative indicators, they have limitations in being difficult to reflect the non-linear characteristics of the signal, ignoring spatial features, and being easily affected by external interference. Although drug concentration monitoring can be achieved through a target-controlled infusion system (TCI) and blood drug concentration detection, it cannot directly reflect the individual differences in sensitivity to anesthetic drugs. Research shows that significant changes occur in the brain functional network during anesthesia, and these changes are mainly reflected in the non-linear dynamic characteristics and spatial configuration of EEG signals. However, existing evaluation methods are difficult to comprehensively capture these features, which limits the accuracy and reliability of anesthetic depth evaluation.

[0003] Prior art one, a Chinese patent with the patent number 202411952158.9 discloses an EEG signal classification model, method, and computer system, including: extracting the power spectrum features of the original EEG signal; extracting the differential entropy features of the original EEG signal; fusing the power spectrum features and the original EEG signal to obtain a fused EEG signal; performing multiple convolution pooling operations on the fused EEG signal to extract the features of the fused EEG signal, obtaining a high-dimensional feature vector, and then mapping the high-dimensional feature vector to a low-dimensional space through a linear transformation to obtain a low-dimensional feature vector; reducing the dimension and splicing the differential entropy features of the original EEG signals in different frequency bands, and splicing them with the dimension-reduced EEG signal to obtain EEG data integrating multiple features; using a self-attention mechanism to capture key features from the EEG data integrating multiple features to obtain a final output vector; using a classifier to classify the final output vector to obtain the classified EEG signal. Although it can more comprehensively capture the key information in the EEG signal, the algorithm complexity is too high, resulting in a large computational cost.

[0004] Prior Art Two, a Chinese patent with the patent number 202411919204.5, provides a method, device, equipment, and medium for intention recognition based on brain-computer signals, belonging to the technical field of intention classification. The method includes: determining the target number of the initial visual stimulus object corresponding to the control-state task of the target user and the target number of the initial non-visual stimulus object corresponding to the target user; determining the target classification model corresponding to the target user according to the target number and the target number; determining the initial coding sequence corresponding to the initial visual stimulus object, and displaying the visual stimulus information corresponding to the initial visual stimulus object in the target interface according to the initial coding sequence; collecting the target electroencephalogram data of the target user gazing at the target interface in real time; performing control-state task recognition on the target electroencephalogram data according to the target classification model to obtain the first predicted visual stimulus object corresponding to the target electroencephalogram data. Although the target intention corresponding to the target user is determined according to the first predicted visual stimulus object, there is a lack of evaluation tools, resulting in large differences in evaluation methods and indicators.

[0005] Prior Art Three, a Chinese patent with the patent number 202411811707.0, discloses an intelligent infusion pump controller for anesthesia target control based on electroencephalogram parameter feedback, which relates to the technical field of medical anesthesia. This application includes a drug injection pump body, a TCI controller is fixedly connected to the top of the drug injection pump body, a signal amplifier is fixedly connected to one end of the TCI controller, a wire is fixedly connected to the input end of the signal amplifier, one end of the wire is fixedly connected to a scalp electrode, and a splint is installed at one end of the drug injection pump body. Although the electroencephalogram signal of the patient is transmitted to the TCI controller for analysis by using the scalp electrode in cooperation with the wire and the signal amplifier, so that the TCI controller can control the dose input into the patient's body by the drug injection pump body in real time, and at the same time, the temperature control component is used to adjust according to the temperature of the drug output inside the syringe, so that the drug pump speed and the temperature when input into the patient's body reach the best state, thereby effectively improving the accuracy of drug administration of the device and improving the anesthesia quality during the operation; however, the adaptability to special populations is insufficient, resulting in difficulty in capturing the anesthesia state.

[0006] Currently, Prior Art One, Prior Art Two, and Prior Art Three have problems such as excessive algorithm complexity, lack of evaluation tools, and insufficient adaptability to special populations. To solve the above problems, the present invention provides an anesthesia depth multi-level classification information evaluation method and system. Summary of the Invention

[0007] The main purpose of the present invention is to provide an anesthesia depth multi-level classification information evaluation method and system to solve the problems of excessive algorithm complexity, lack of evaluation tools, and insufficient adaptability to special populations in the prior art.

[0008] To achieve the above purpose, the present invention provides the following technical solutions:

[0009] A method for evaluating multi-level classification information of anesthesia depth, the method for evaluating multi-level classification information of anesthesia depth includes:

[0010] Collect electroencephalogram (EEG) signals, and perform preprocessing operations such as baseline drift correction, artifact removal, and signal segmentation on the EEG signals; decompose the preprocessed EEG signals to obtain signals in five frequency bands: delta, theta, alpha, beta, and gamma.

[0011] Calculate the global field power for all EEG signals in each frequency band, and identify the topographic maps at the peak moments; perform cluster analysis on these topographic maps at the global field power peak moments to extract representative microstate templates; match the topographic map at each time point with the microstate templates (considering polarity inversion) to reconstruct a continuous microstate time series.

[0012] Among them, the extracted microstate parameters include characteristic parameters such as microstate duration and occurrence frequency.

[0013] Establish a feature screening and classification recognition model, input the signals in each frequency band into the feature screening and classification recognition model, screen the extracted microstate features and perform multi-level classification of anesthesia depth, and finally output the anesthesia depth evaluation result.

[0014] As a further improvement of the present invention, the process of obtaining signals in the five frequency bands of delta, theta, alpha, beta, and gamma includes the following steps:

[0015] Collect EEG signals, and perform baseline drift correction on the collected EEG signals; perform artifact processing on the signals to remove the influence of electrooculogram and electromyogram artifacts; perform signal segmentation on the signals to divide the continuous signals into data segments of a preset length.

[0016] Perform empirical mode decomposition on the preprocessed EEG signals to decompose the EEG signals into several intrinsic mode function components and a residual term; and perform Hilbert transform on each intrinsic mode function component to obtain the analytic signal and instantaneous frequency.

[0017] According to the calculation results, decompose the EEG signals into five characteristic frequency bands, corresponding to the 0.5 - 4Hz delta, 4 - 8Hz theta, 8 - 13Hz alpha, 13 - 30Hz beta, and 30 - 45Hz gamma frequency bands respectively.

[0018] As a further improvement of the present invention, the process of obtaining the analytic signal and instantaneous frequency includes the following steps:

[0019] The preprocessed EEG signals are decomposed by empirical mode decomposition into a number of intrinsic mode function components and a residual term;

[0020] The Hilbert transform is performed on each intrinsic mode function component to obtain the analytic signal and the instantaneous frequency;

[0021] The sampling frequency is selected for each electrode signal, and the instantaneous frequency is obtained through the Hilbert transform; the instantaneous frequency is restricted within the frequency band, and each intrinsic mode function component is adjusted to the corresponding frequency band to generate the intrinsic mode function components of different frequency bands.

[0022] As a further improvement of the present invention, the process of reconstructing the microstates after clustering analysis includes the following steps:

[0023] The global field power is calculated for all channel signals in each frequency band; for each time point t, the spatial variation intensity of all electrode voltages is calculated;

[0024] Clustering analysis is performed on the microstates: the topographic maps at the peak moments of the global field power are clustered, and the optimal microstate categories are determined by optimizing the following indexes:

[0025] The microstate sequence is reconstructed: by calculating the similarity between the topographic map at each global field power peak moment and the microstate template, it is marked as the most similar microstate category to generate a continuous microstate time series;

[0026] Among them, the microstate time series parameters include: microstate duration, microstate occurrence frequency, and microstate transition probability.

[0027] As a further improvement of the present invention, the process of generating a continuous microstate time series includes the following steps:

[0028] The target topographic map is extracted from the topographic maps at the peak moments of the global field power through clustering analysis to obtain the microstate template; the topographic map at each global field power peak moment is compared with the microstate template, and the similarity between the topographic map at each global field power peak moment and all microstate templates is compared in detail;

[0029] For the similarity calculated for each global field power peak moment, the microstate category with the highest similarity to this moment is selected; the microstate categories corresponding to each peak moment are sorted in chronological order to form a continuous microstate time series;

[0030] Based on the generated microstate time series, the duration of each microstate in the time series, the number of times each microstate appears in the time series, and the probability of transitioning from one microstate to another are extracted.

[0031] As a further improvement of the present invention, the process of finally outputting the anesthesia depth evaluation result includes the following steps:

[0032] Perform preprocessing operations of denoising, filtering, and artifact monitoring on each extracted microstate feature; and extract features related to anesthesia depth to obtain a data set;

[0033] Among them, the features related to anesthesia depth include mean absolute power, standard deviation, power spectral density of different frequency bands, and the fusion of time series, fractal features, and spectral features;

[0034] Divide the data set into a training set, a validation set, and a test set; establish a feature screening and classification recognition model, and use the training set to train the feature screening and classification recognition model; input the microstate features into the feature screening and classification recognition model for classification evaluation; and evaluate the screened features and the classification recognition model through a confusion matrix;

[0035] Assign weights to each feature according to the evaluation results, gradually remove the feature with the lowest weight, repeat training and evaluate the performance until the preset number of features is reached; and sort the features, and use the feature subset with the preset ranking as the final output.

[0036] As a further improvement of the present invention, among them, the architecture of the feature screening and classification recognition model is:

[0037] Feature extraction and screening layer:

[0038] Basic feature network, process signals of each frequency band through multiple layers of one-dimensional convolution and pooling layers, and extract and optimize time-frequency features including mean absolute power, power spectral density, etc.;

[0039] Feature fusion module, use the self-attention mechanism to integrate the basic feature network and the extracted microstate features, and perform feature optimization and dimensionality reduction through an autoencoder;

[0040] Time series processing layer:

[0041] Recurrent neural network, capturing the dynamic process of anesthesia depth change;

[0042] Classification recognition layer:

[0043] Fully connected layer, fusing the features output by the recurrent neural network layer;

[0044] Output layer: Use the softmax function for multi-classification and output the evaluation result of anesthesia depth.

[0045] As a further improvement of the present invention, the process of evaluating the screened features and the classification recognition model through a confusion matrix includes the following steps:

[0046] Divide the dataset into a training set, a validation set, and a test set; construct a feature screening and classification recognition model based on the historical electroencephalogram processing database, and use the training set to train the feature screening and classification recognition model;

[0047] Use the training set to train the model. During the training process, extract features through the basic feature network, and the feature fusion module integrates and deeply learns the features extracted by the basic feature network and the extracted microstate features; after the feature learning is completed, perform variable screening in a recursive manner, and based on the results of each training, eliminate variables with weights less than the preset threshold, and finally obtain an optimized feature screening and classification recognition model;

[0048] Use the test set to evaluate the feature screening and classification recognition model, compare the true labels of the test set with the labels of the feature screening and classification recognition model, and generate a confusion matrix; input the features extracted by the basic feature network and the microstate features into the model for classification evaluation.

[0049] As a further improvement of the present invention, the process of taking the feature subset with a preset ranking as the final output includes the following steps:

[0050] According to the classification evaluation results of the feature screening and classification recognition model, comprehensively analyze the features extracted by the basic feature network and the microstate features, evaluate the contribution value of each feature to the classification performance through the self-attention mechanism of the feature fusion module, and assign weights accordingly;

[0051] Select the feature with the highest weight from all features as the initial feature subset, and add features in sequence; retrain the feature screening and classification recognition model and evaluate it after each adjustment until the preset number of features is reached;

[0052] When the preset number of features is reached, re-rank the remaining features according to their weights; calibrate the top k features with the highest weights as the final feature subset, preset the anesthesia depth level, and output the anesthesia depth level corresponding to the final feature subset; output the corresponding anesthesia depth level to the visualization device.

[0053] To achieve the above object, the present invention also provides the following technical solutions:

[0054] An anesthesia depth multi-level classification information evaluation system, which is applied to the anesthesia depth multi-level classification information evaluation method. The anesthesia depth multi-level classification information evaluation system includes:

[0055] The EEG acquisition and processing module is used to acquire raw EEG signals and perform a series of preprocessing operations on the signals, including baseline drift correction, artifact detection and removal (such as interference from electrooculogram, electromyogram, etc.), and signal segmentation processing; after the preprocessing is completed, the EEG signals are decomposed into frequency bands to obtain signals in five frequency bands: delta, theta, alpha, beta, and gamma;

[0056] The clustering and reconstruction module is used to calculate the global field power for all EEG signals in each frequency band and identify the topographic maps at the peak moments; perform clustering analysis on these topographic maps at the peak moments of the global field power to extract representative microstate templates; match the topographic map at each time point with the microstate templates (considering polarity inversion) to reconstruct a continuous microstate time series;

[0057] Finally, extract microstate parameters, including characteristic parameters such as the duration and occurrence frequency of each microstate.

[0058] The anesthesia depth assessment module is used to establish a feature screening and classification recognition model, input the signals in each frequency band into the feature screening and classification recognition model, screen the extracted microstate features and perform multi-level classification of the anesthesia depth, and finally output the anesthesia depth assessment result.

[0059] The present invention improves the signal quality by acquiring EEG signals and performing preprocessing operations such as baseline drift correction, artifact removal, and segmentation processing, providing reliable basic data for subsequent analysis. Calculate the global field power, extract microstate parameters (such as duration and frequency), and perform clustering analysis and reconstruction on microstates; reveal the dynamic change law of EEG signals and provide key characteristic parameters for anesthesia depth assessment. Establish a feature screening and classification recognition model, input the signals in each frequency band into the model for microstate feature screening and anesthesia depth classification; through deep learning or machine learning algorithms, achieve multi-level classification of the anesthesia depth, and finally output the assessment result, thereby improving the scientificity and accuracy of anesthesia management. Preprocess the acquired EEG signals; decompose the preprocessed EEG signals using the Hilbert-Huang transform to obtain signals in five frequency bands: delta, theta, alpha, beta, and gamma; adopt the microstate analysis method to calculate the global field power and extract microstate time series parameters; establish a feature screening and classification recognition model, compare and analyze the microstate parameter characteristics under different anesthesia states, and achieve the assessment of the anesthesia depth. Description of the Drawings

[0060] Figure 1 It is a schematic diagram of the step flow of an embodiment of the multi-level classification information assessment method for anesthesia depth of the present invention;

[0061] Figure 2Schematic diagram of the steps for obtaining signals in five frequency bands of delta, theta, alpha, beta, and gamma in an embodiment of the anesthesia depth multi-level classification information evaluation method of the present invention;

[0062] Figure 3 Schematic diagram of the steps for obtaining the analytic signal and instantaneous frequency in an embodiment of the anesthesia depth multi-level classification information evaluation method of the present invention;

[0063] Figure 4 Principle diagram of the EEG signal frequency band decomposition by Hilbert-Huang transform in the anesthesia depth multi-level classification information evaluation method of the present invention;

[0064] Figure 5 Principle diagram of the EEG signal frequency band distribution in the anesthesia depth multi-level classification information evaluation method of the present invention;

[0065] Figure 6 Schematic diagram of the steps for reconstructing the microstates after cluster analysis in an embodiment of the anesthesia depth multi-level classification information evaluation method of the present invention;

[0066] Figure 7 Principle diagram of the microstate analysis process in the anesthesia depth multi-level classification information evaluation method of the present invention;

[0067] Figure 8 Schematic diagram of the steps for generating a continuous microstate time series in an embodiment of the anesthesia depth multi-level classification information evaluation method of the present invention;

[0068] Figure 9 Schematic diagram of the steps for finally outputting the anesthesia depth evaluation result in an embodiment of the anesthesia depth multi-level classification information evaluation method of the present invention;

[0069] Figure 10 Schematic diagram of the steps for evaluating and screening features and classification recognition models through a confusion matrix in an embodiment of the anesthesia depth multi-level classification information evaluation method of the present invention;

[0070] Figure 11 Schematic diagram of the steps for taking the feature subset with a preset ranking as the final output in an embodiment of the anesthesia depth multi-level classification information evaluation method of the present invention;

[0071] Figure 12 Schematic diagram of the functional modules in an embodiment of the anesthesia depth multi-level classification information evaluation system of the present invention;

[0072] Figure 13 Schematic diagram of the structure in an embodiment of the anesthesia depth multi-level classification information evaluation system of the present invention;

[0073] Figure 14This is the schematic diagram of the anesthesia depth multi-level classification information evaluation system of the present invention;

[0074] Figure 15 This is the structural schematic diagram of an embodiment of the electronic device of the present invention;

[0075] Figure 16 This is the structural schematic diagram of an embodiment of the storage medium of the present invention. Detailed implementation manners

[0076] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0077] The terms "first", "second", and "third" in the present invention are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first", "second", and "third" may explicitly or implicitly include at least one of such features. In the description of the present invention, the meaning of "a plurality" is at least two, such as two, three, etc., unless otherwise specifically defined. All directional indications (such as up, down, left, right, front, back...) in the embodiments of the present invention are only used to explain the relative positional relationship and movement conditions between components in a specific posture (as shown in the drawings). If the specific posture changes, the directional indications will also change accordingly. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products, or devices.

[0078] Referring to "embodiments" herein means that the specific features, structures, or characteristics described in connection with the embodiments can be included in at least one embodiment of the present invention. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0079] As Figure 1 shown, this embodiment provides an embodiment of the anesthesia depth multi-level classification information evaluation method. In this embodiment, the anesthesia depth multi-level classification information evaluation method specifically includes the following steps:

[0080] Step S1: Collect EEG signals and perform preprocessing operations such as baseline drift correction, artifact removal, and segmentation on the EEG signals; decompose the preprocessed EEG signals to obtain signals in five frequency bands: delta, theta, alpha, beta, and gamma.

[0081] Step S2: Calculate the global field power for all EEG signals in each frequency band and identify the topographic maps at the peak moments; perform cluster analysis on these topographic maps at the peak moments of the global field power to extract representative microstate templates; match the topographic map at each time point with the microstate templates to reconstruct a continuous microstate time series.

[0082] Among them, the extracted microstate parameters include characteristic parameters such as microstate duration and occurrence frequency.

[0083] Step S3: Establish a feature screening and classification recognition model, input the signals in each frequency band into the feature screening and classification recognition model, screen the extracted microstate features and perform multi-level classification of the anesthesia depth, and finally output the anesthesia depth assessment result.

[0084] Preferably, in step S1 of this embodiment, by collecting EEG signals and performing preprocessing operations such as baseline drift correction, artifact removal, and segmentation, the signal quality is improved, providing reliable basic data for subsequent analysis. In step S2, calculate the global field power, perform cluster analysis and reconstruction on the microstates, and extract microstate parameters (such as duration and frequency); reveal the dynamic change law of the EEG signals, providing key characteristic parameters for anesthesia depth assessment. In step S3, establish a feature screening and classification recognition model, input the signals in each frequency band into the model for microstate feature screening and anesthesia depth classification; through deep learning or machine learning algorithms, achieve multi-level classification of the anesthesia depth, and finally output the assessment result, thereby improving the scientificity and accuracy of anesthesia management. Preprocess the collected EEG signals; decompose the preprocessed EEG signals using Hilbert-Huang transform to obtain signals in five frequency bands: delta, theta, alpha, beta, and gamma; adopt the microstate analysis method to calculate the global field power and extract microstate time series parameters; establish a feature screening and classification recognition model, compare and analyze the microstate parameter characteristics under different anesthesia states, and realize the assessment of the anesthesia depth.

[0085] Furthermore, as Figure 2 shown, the process of obtaining the signals in the five frequency bands of delta, theta, alpha, beta, and gamma in step S1 specifically includes the following steps:

[0086] Step S11: Collect EEG signals, perform baseline drift correction on the collected EEG signals to eliminate low-frequency interference in the signals; perform artifact processing on the signals to remove the influence of artifacts such as electrooculogram and electromyogram; perform segmentation processing on the signals to divide the continuous signals into data segments of a preset length.

[0087] Step S12: Perform empirical mode decomposition on the preprocessed EEG signals to decompose the EEG signals into several intrinsic mode function components and a residual term; and perform Hilbert transform on each intrinsic mode function component to obtain the analytic signal and the instantaneous frequency.

[0088] Step S13: According to the calculation results, decompose the EEG signals into five characteristic frequency bands, corresponding to the delta (0.5 - 4 Hz), theta (4 - 8 Hz), alpha (8 - 13 Hz), beta (13 - 30 Hz), and gamma (30 - 45 Hz) frequency bands respectively.

[0089] Preferably, in step S11 of this embodiment, through baseline drift correction, low-frequency interference elimination, and artifact processing (such as electrooculogram and electromyogram artifact removal), the signal-to-noise ratio of the EEG signals is improved, thereby providing purer data for subsequent analysis; the signal quality is enhanced, the influence of noise and interference on subsequent analysis is reduced, and a foundation is laid for feature extraction and signal processing. In step S12, the empirical mode decomposition (EMD) is used to decompose the EEG signals into several intrinsic mode function components and a residual term, and the instantaneous frequency is calculated through Hilbert transform to obtain the analytic signal; different components in the signals are effectively separated, and signal features with physical significance are extracted. Its technical effect is that it can analyze the time-frequency characteristics of the signals more accurately and provide support for subsequent frequency band division and feature extraction. In step S3, the preprocessed EEG signals are divided into frequency bands according to a specific frequency range (delta, theta, alpha, beta, gamma); the signals can be decomposed into features of different frequency bands, which is convenient for further analyzing the roles of these frequency bands in cognitive activities, thereby providing an important reference basis for brain function research.

[0090] Further, as Figure 3 shown, the process of obtaining the analytic signal and the instantaneous frequency in step S12 specifically includes the following steps:

[0091] Step S121: Perform empirical mode decomposition on the preprocessed EEG signals to decompose the EEG signals into several intrinsic mode function components and a residual term.

[0092] Among them, the signal x(t) is expressed as:

[0093]

[0094] Among them, c i(t) represents the i-th intrinsic mode function component, r n (t) represents the residual term, and n represents the total number of IMF components;

[0095] Step S122: Perform Hilbert transform on each intrinsic mode function component to obtain the analytic signal and the instantaneous frequency;

[0096] Among them, the analytic signal z i (t) is calculated as follows:

[0097] z i (t) = c i (t) + jH[c i (t)]

[0098] Among them, H[c i (t)] represents the Hilbert transform of the intrinsic mode function component;

[0099] Take the instantaneous frequency ω i (t) is calculated as follows:

[0100]

[0101] Among them, arg(z i (t)) represents the phase angle of the complex signal;

[0102] Step S123: Select the sampling frequency f s for each electrode signal, and obtain the instantaneous frequency through Hilbert transform Limit the instantaneous frequency within the frequency band , that is Each intrinsic mode function component IMF i (t) is adjusted to the corresponding frequency band to generate intrinsic mode function components in different frequency bands

[0103] Among them, j represents 5 different frequency bands, namely delta (0.5 - 4 Hz), theta (4 - 8 Hz), alpha (8 - 13 Hz), beta (13 - 30 Hz) and gamma (> 30 Hz) five frequency bands.

[0104] Preferably, in this embodiment, the EEG signal is decomposed into several Intrinsic Mode Function (IMF) components and a residual term. Each IMF satisfies the condition that the difference between the number of extreme points and zero-crossing points is at most one, ensuring that the local features of the signal are fully extracted; the IMF components are iteratively decomposed until a specific condition is met (such as the residual approaching a monotonic function), thereby realizing the adaptive decomposition of the signal; the Hilbert transform is performed on each IMF component to generate an analytic signal, which contains information on the instantaneous amplitude and instantaneous frequency; the instantaneous frequency is calculated from the phase angle of the analytic signal, reflecting the frequency characteristics of the signal changing over time; the instantaneous frequency is restricted within a specific frequency band (such as the delta, theta, alpha, beta, gamma frequency bands), and the IMF components are adjusted according to the characteristics of the frequency band to generate intrinsic mode function components in different frequency bands. Through EMD and HHT, the complex EEG signal is decomposed into multiple IMF components with physical meanings, and important features such as instantaneous frequency and amplitude are extracted for subsequent analysis; decomposing the signal into different frequency bands can more accurately reflect the activity characteristics of the EEG signal in different frequency bands, providing an important basis for EEG classification, diagnosis, etc.; the EMD method can effectively separate the noise and useful information in the signal, reducing the influence of background interference on the analysis results (for the specific principle, refer to Appendix Figure 4 and Appendix Figure 5 ).

[0105] Furthermore, as Figure 6 shown, the process of reconstructing the microstates after cluster analysis in step S2 specifically includes the following steps:

[0106] Step S21: Calculate the global field power for all channel signals in each frequency band; for each time point t, calculate the spatial variation intensity of the voltages of all electrodes.

[0107] Among them, the calculation formula is:

[0108]

[0109] where N is the number of electrodes, v i (t) is the potential of the i-th electrode at time t, is the average potential of all electrodes at time t;

[0110] Step S22: Conduct cluster analysis on the microstates: Cluster the topographic maps at the peak moments of the global field power, and determine the optimal microstate category by optimizing the following indicators:

[0111] Global explained variance GEV:

[0112]

[0113] Cross-validation criterion CV:

[0114]

[0115] Step S23: Reconstruct the microstate sequence: By calculating the similarity between the topographic maps at each global field power peak moment and the microstate templates, label the topographic map at each time point as the most similar microstate category, and generate a continuous microstate time series;

[0116] Among them, the microstate time series parameters include: microstate duration, microstate occurrence frequency, and microstate transition probability.

[0117] Preferably, in step S21 of this embodiment, the global field power is calculated to analyze the volatility and potential distribution of the signal, thereby providing basic data support for subsequent microstate clustering analysis; it can quantify the dynamic change characteristics of the signal and provide a reliable basis for microstate classification. In step S22, clustering analysis is performed based on the topographic maps at the global field power peak moments, and the optimal microstate category is determined by optimizing the global explained variance (GEV) and the cross-validation criterion (CV); it can efficiently identify microstate categories with significant differences, thereby improving the accuracy and robustness of classification. In step S23, by calculating the similarity between the topographic map at each time point and the microstate template, a continuous microstate time series is generated, and parameters such as microstate duration, occurrence frequency, and transition probability are analyzed; it can reveal the dynamic relationship between microstates and their distribution law in the time series, providing an in-depth analysis basis for subsequent research (for the specific principle, refer to the appendix Figure 7 ).

[0118] Furthermore, as Figure 8 shown, the process of generating a continuous microstate time series in step S23 specifically includes the following steps:

[0119] Step S231: Extract the target topographic map from the topographic maps at the global field power peak moments through clustering analysis to obtain the microstate template; compare the topographic map at each global field power peak moment with the microstate template, and specifically compare the similarity between the topographic map at the global field power peak moment and all microstate templates;

[0120] Step S232: For the similarity calculated at each time point, select the microstate category with the highest similarity to this time point; sort the microstate categories corresponding to each time point in chronological order to form a continuous microstate time series;

[0121] Step S233: Based on the generated microstate time series, extract the duration of each microstate in the time series, the number of times each microstate appears in the time series, and the probability of transitioning from one microstate to another microstate.

[0122] Among them, the similarity quantization formula in step S231:

[0123]

[0124] In the formula: is the global field power vector at time point t; represents the topographic map of the k-th class of microstate templates; ∈ S = 10 -8 is the numerical stability term; respectively represent the spatial mean of the vectors; ‖·‖ F is the Frobenius norm orthogonal projection operator;

[0125] The classification judgment logic tree formula in step S232:

[0126]

[0127] In the formula, C t ∈ {1,..., K} is the current state class identifier; λ temp = 0.8 is the time-domain continuity coefficient; τ s = 50 ms is the time smoothing window; represents the frequency density of the j-th class in the previous τ time; is the state transition penalty coefficient;

[0128] The feature functional conversion formula in step S233:

[0129] Duration operator:

[0130]

[0131] Frequency density mapping:

[0132]

[0133] Transition probability tensor:

[0134]

[0135] In the formula: Γ D is the duration normalization factor; α F , β F , μ F are the frequency feature hyperparameters; is the trainable weight matrix; η = 100 ms is the transition time window constraint; ∈ p = 10 -5 is the probability compression coefficient; is the total duration of the k-th class within the analysis segment.

[0136] Preferably, in step S231 of this embodiment, the target topographic map is extracted from the topographic map at the global field power peak moment through cluster analysis, and a microstate template is generated. At the same time, the similarity between the topographic map at each time point and the microstate template is calculated. The data points are classified into different categories through similarity measurement, so as to extract the representative microstate template, efficiently identify and classify the microstates in the electroencephalogram, and provide a basis for subsequent analysis. In step S232, based on the calculated similarity, the microstate category most similar to each time point is selected, and these categories are arranged in chronological order to form a continuous microstate time series, which can dynamically track the changes in electroencephalogram activities and reveal the temporal evolution law of microstates. In step S233, based on the generated microstate time series, the duration, occurrence times, and transition probability of each microstate in the time series are extracted, which involves further analysis of the time series data and extracts the statistical characteristics of microstates, providing quantitative indicators for understanding the dynamic patterns of electroencephalogram activities and helping to deeply study brain functions and disease mechanisms.

[0137] Further, as Figure 9 shown, the process of finally outputting the anesthesia depth assessment result in step S3 specifically includes the following steps:

[0138] Step S31: Extract the features related to the anesthesia depth through the basic feature network, and integrate the extracted microstate features to obtain a data set;

[0139] Among them, the features related to the anesthesia depth include the mean absolute power, standard deviation, power spectral density of different frequency bands, and the fusion of time series, fractal features, and spectral features, etc.;

[0140] Step S32: Divide the data set into a training set, a validation set, and a test set; establish a feature screening and classification recognition model, and use the training set to train the feature screening and classification recognition model; input the features extracted by the basic feature network and the microstate features into the feature screening and classification recognition model for classification evaluation; and evaluate the screening features and the classification recognition model through a confusion matrix;

[0141] Step S33: Assign weights to each feature according to the evaluation results, gradually remove the feature with the lowest weight, repeat training and evaluate the performance until the preset number of features is reached; and sort the features, and use the feature subset with the preset ranking as the final output.

[0142] Among them, the feature screening and classification recognition model adopts a multi-level architecture design, mainly including three core levels: feature extraction and screening, time series processing, and classification recognition. The specific architecture is:

[0143] Feature extraction and screening layer:

[0144] The basic feature network extracts the time-frequency features of EEG signals through multiple layers of one-dimensional convolutional layers and pooling layers, including the mean absolute power (MAP) of signal power features, the standard deviation of signal variation degree, the power spectral density (PSD) of different frequency energy distributions, the time series and fractal features of signal dynamics and nonlinear characteristics, and the spectral feature analysis of signal frequency domain characteristics;

[0145] Among them, the one-dimensional convolutional layer is responsible for extracting local features from time series data and capturing local patterns and changing trends in the signals; the pooling layer reduces the data dimension through max-pooling and average-pooling operations, enhancing the model's generalization ability while retaining important feature information;

[0146] The feature fusion module uses the self-attention mechanism to integrate the features extracted by the basic feature network and the microstate features that have been extracted, and optimizes and reduces the dimensions of the features through an autoencoder. The autoencoder is used for dimensionality reduction and feature selection, learning an effective representation of the input data through training the model, removing noise and unimportant features, and only retaining the most representative features;

[0147] Time series processing layer:

[0148] The recurrent neural network (RNN) effectively processes and learns the long-term dependencies in time series data, capturing the dynamic process of the change in anesthesia depth;

[0149] Classification and recognition layer:

[0150] The fully connected layer fuses the features output by the RNN layer of the recurrent neural network, preparing for the final classification task;

[0151] Output layer: The softmax function is used for multi-classification to output the evaluation result of the anesthesia depth.

[0152] Preferably, in step S31 of this embodiment, features related to anesthesia depth (such as mean absolute power, standard deviation, power spectral density, etc.) are extracted through the basic feature network, and are fused in combination with time series, fractal features, and spectral features. At the same time, the microstate features that have been extracted are integrated to form a high-quality data set, improving the reliability of the data and the accuracy of subsequent analysis. In step S32, the data set is divided into a training set, a validation set, and a test set, and a feature screening and classification recognition model is established; the model is trained using the training set, and the features extracted by the basic feature network and the microstate features are used for classification evaluation. At the same time, the confusion matrix is combined to evaluate the effects of the screening features and the classification recognition model. In step S33, weights are assigned to each feature according to the evaluation results, the features with the lowest weights are gradually removed, and the training and evaluation are repeated until the preset number of features is reached; the features are sorted and the feature subset with the top ranking is selected as the final output.

[0153] Furthermore, as Figure 10 shown, the process of evaluating and screening features and the classification recognition model through the confusion matrix in step S32 specifically includes the following steps:

[0154] Step S321: Divide the data set into a training set, a validation set, and a test set; construct a feature screening and classification recognition model based on the historical electroencephalogram processing database, and use the training set to train the feature screening and classification recognition model;

[0155] Step S322: During the training process, analyze the electroencephalogram data and the microstate features to obtain the relationship between the electroencephalogram data and the microstate features, and learn it; after the learning is completed, recursively perform variable screening, and eliminate variables smaller than the preset value according to the results of each training to obtain the final feature screening and classification recognition model;

[0156] Step S323: Use the test set to evaluate the feature screening and classification recognition model, compare the true labels of the test set with the labels of the feature screening and classification recognition model to generate a confusion matrix; input the features extracted by the basic feature network and the microstate features into the feature screening and classification recognition model for classification evaluation.

[0157] Preferably, in step S321 of this embodiment, the data set is divided into a training set, a validation set, and a test set, and a feature screening and classification recognition model is constructed based on the historical electroencephalogram processing database, and the training set is used to train the model; the data set is divided into three parts: a training set, a validation set, and a test set to ensure the independence and effectiveness of model training, parameter tuning, and final performance evaluation; by reasonably dividing the data set, overfitting is avoided, and at the same time, sufficient sample support is provided for model training to improve the generalization ability of the model. In step S322, during the training process, analyze the relationship between the electroencephalogram data and the microstate features, learn their correlation, and recursively perform variable screening to eliminate variables smaller than the preset value, and finally obtain an optimized feature screening and classification recognition model; adopt a recursive variable screening method, combine the relationship learning between the electroencephalogram data and the microstate features, and dynamically adjust the variable weights in the model; through variable screening and recursive optimization, redundant features are reduced, and the interpretability and prediction performance of the model are improved. In step S323, use the test set to evaluate the model, generate a confusion matrix, and input the features extracted by the basic feature network and the microstate features into the model for classification evaluation; use the confusion matrix as an evaluation index, combine the true labels of the test set with the predicted labels of the model for comparison, and comprehensively evaluate the model performance; analyze key indicators such as the classification accuracy and recall rate of the model through the confusion matrix to verify the reliability and stability of the model in practical applications.

[0158] Furthermore, as Figure 11As shown in the figure, the process of using the feature subset with the preset ranking as the final output in step S33 specifically includes the following steps:

[0159] Step S331: According to the classification evaluation results of the feature screening and classification recognition model, comprehensively analyze the features extracted by the basic feature network and the micro-state features, assign weights according to the contribution value of the features to the classification performance, and gradually remove the features with the lowest weights after obtaining the feature weights.

[0160] Step S332: Select the feature with the highest weight from all features as the initial feature subset, and add features in sequence; retrain the feature screening and classification recognition model and evaluate it after each adjustment until the preset number of features is reached.

[0161] Step S333: When the preset number of features is reached, reorder the remaining features according to their weights; calibrate the top k features with the highest weights as the final feature subset, preset the anesthesia depth level, and output the anesthesia depth level corresponding to the final feature subset; output the corresponding anesthesia depth level to the visualization device.

[0162] Preferably, in step S331 of this embodiment, by analyzing the contribution value of the features to the classification performance, weights are assigned to each feature, and the features with the lowest weights are gradually removed; by screening and optimizing the feature subset, redundant information is reduced, overfitting is avoided, thereby improving the accuracy and generalization ability of the classification model; the number of features is streamlined, the consumption of computing resources is reduced, and the model training and prediction efficiency are improved. In step S332, the forward selection strategy is adopted, the feature with the highest weight is selected from all features as the initial subset, and features are gradually added. The model is retrained and the performance is evaluated after each adjustment until the preset number of features is reached; the calculation and sorting of feature weights help to understand which features are most critical to the classification task, thereby enhancing the interpretability of the model. In step S333, by reordering the remaining features and selecting the top k features with the highest weights as the final feature subset, the model performance is further optimized; the final feature subset is used to calibrate the model and output the results according to the preset anesthesia depth level. At the same time, the results are visualized for intuitive display and application; the output results are sent to the visualization device, which is convenient for medical staff to quickly obtain anesthesia depth level information and support clinical decision-making.

[0163] As Figure 12 shown, this embodiment also provides an embodiment of the anesthesia depth multi-level classification information evaluation system. In this embodiment, the anesthesia depth multi-level classification information evaluation system is applied to the anesthesia depth stage method as described in the above embodiment. The anesthesia depth multi-level classification information evaluation system includes:

[0164] The EEG acquisition and processing module 1 is used to acquire the original EEG signals and perform a series of preprocessing operations on the signals, including baseline drift correction, artifact detection and removal (such as interference from electrooculogram, electromyogram, etc.), and signal segmentation processing; after completing the preprocessing, the EEG signals are decomposed into frequency bands to obtain signals in five frequency bands: delta, theta, alpha, beta, and gamma;

[0165] The clustering and reconstruction module 2 is used to calculate the global field power for all EEG signals in each frequency band and identify the topographic maps at the peak moments; perform clustering analysis on these topographic maps at the peak moments of the global field power to extract representative microstate templates; match the topographic map at each time point with the microstate templates (considering polarity inversion) to reconstruct a continuous microstate time series;

[0166] Finally, extract microstate parameters, including characteristic parameters such as the duration and occurrence frequency of each microstate.

[0167] The anesthesia depth assessment module 3 is used to establish a feature screening and classification recognition model, input the signals in each frequency band into the feature screening and classification recognition model, screen the extracted microstate features and perform multi-level classification of the anesthesia depth, and finally output the anesthesia depth assessment result.

[0168] Preferably, in this embodiment, the EEG acquisition and processing module 1 ensures the signal quality by acquiring EEG signals and performing preprocessing operations such as baseline drift correction, artifact removal processing (such as filtering, rereferencing, artifact rejection, etc.), and segmentation processing. In addition, the module decomposes the preprocessed signals into signals in five frequency bands: delta, theta, alpha, beta, and gamma; through denoising and signal decomposition, the signal-to-noise ratio and reliability of the EEG signals are improved, providing high-quality data for subsequent analysis. The clustering and reconstruction module 2 is used to calculate the global field power for all EEG signals in each frequency band and identify the topographic maps at the peak moments; perform clustering analysis on these topographic maps to extract microstate templates; match the topographic map at each time point with the templates (considering polarity inversion) to reconstruct the microstate sequence; finally, extract microstate parameters, including characteristic parameters such as duration and occurrence frequency. This analysis method can effectively characterize the dynamic characteristics of EEG activities and provide an important basis for brain function assessment. The anesthesia depth assessment module 3 establishes a feature screening and classification recognition model, inputs the signals in each frequency band into the model, screens the microstate features, and performs multi-level classification of the anesthesia depth.

[0169] Furthermore, as Figure 13 shown, the multi-level classification information assessment system for anesthesia depth further includes:

[0170] The signal processing module 510 is responsible for preprocessing the input physiological signals and storing them, which is composed of a signal preprocessing unit and a data storage unit;

[0171] The signal decomposition module 520 includes an empirical mode decomposition unit, a Hilbert transform unit, and a frequency band recombination unit; among them, the empirical mode decomposition unit adaptively decomposes the signal, the Hilbert transform unit processes it using the Hilbert-Huang transform (HHT) algorithm, and the frequency band recombination unit recombines the processed signal into five characteristic frequency bands: delta (0.5 - 4 Hz), theta (4 - 8 Hz), alpha (8 - 13 Hz), beta (13 - 30 Hz), and gamma (30 - 45 Hz);

[0172] The microstate analysis module 530 consists of a GFP calculation unit, a clustering analysis unit, and a time series generation unit; by calculating the global field power (GFP) value, combined with the clustering analysis method, it identifies the electroencephalogram microstates and generates time series parameters reflecting the dynamic characteristics of the brain;

[0173] The classification and recognition module 540 includes a feature screening unit and a classification and recognition unit, which screens and classifies the features obtained from the microstate analysis;

[0174] The display and storage module 550 is composed of a display unit and a storage unit, responsible for real-time displaying the classification results and permanently storing the relevant data.

[0175] Preferably, in this embodiment, the signal preprocessing module is used to preprocess the collected electroencephalogram signals; the signal decomposition module is used to decompose the preprocessed electroencephalogram signals using the Hilbert-Huang transform to obtain multiple frequency band signals; the microstate analysis module is used to calculate the global field power and extract the microstate time series parameters; the classification and recognition module is used to establish a feature screening and classification and recognition model to realize the assessment of the anesthesia depth;

[0176] Among them, the signal decomposition module includes: an empirical mode decomposition unit for decomposing the signal into IMF components; a Hilbert transform unit for calculating the instantaneous frequency and instantaneous amplitude; a frequency band recombination unit for recombining the IMF components into different frequency band signals;

[0177] The microstate analysis module includes: a GFP calculation unit for calculating the global field power; a clustering analysis unit for clustering the topographic maps at the peak moments of the GFP; a time series generation unit for generating the microstate time series; it also includes a display module and a storage module, which are respectively used to real-time display the anesthesia depth assessment results and store the electroencephalogram signal data and analysis results (for the specific principle, refer to 14).

[0178] Such as Figure 15As shown, this embodiment provides an embodiment of an electronic device. In this embodiment, the electronic device 4 includes a processor 41 and a memory 42 coupled to the processor 41.

[0179] The memory 42 stores program instructions for implementing the anesthesia depth multi-level classification information evaluation system of any of the above embodiments.

[0180] The processor 41 is configured to execute the program instructions stored in the memory 42 to perform the layout of the anesthesia depth multi-level classification information evaluation.

[0181] Among them, the processor 41 can also be referred to as a CPU (Central Processing Unit, central processing unit). The processor 41 may be an integrated circuit chip with signal processing capabilities. The processor 41 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0182] Furthermore, Figure 16 This is a schematic structural diagram of a storage medium according to an embodiment of the present application. The storage medium 5 of the embodiment of the present application stores program instructions 51 that can implement all of the above methods. Among them, the program instructions 51 can be stored in the above storage medium in the form of a software product, including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs, or terminal devices such as computers, servers, mobile phones, and tablets.

[0183] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, indirect couplings or communication connections of devices or units, and can be in electrical, mechanical, or other forms.

[0184] In addition, in each embodiment of the present invention, each functional unit may be integrated into a processing unit, may exist separately as individual physical units, or two or more units may be integrated into one unit. The above-mentioned integrated units may be implemented in the form of hardware or in the form of software functional units. The above are only the embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

[0185] The specific embodiments of the invention have been described in detail above, but they are only examples. The present invention is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modification or substitution of the invention is also within the scope of the present invention. Therefore, all equal transformations, modifications, improvements, etc. made without departing from the spirit and principles of the present invention should be covered by the scope of the present invention.

Claims

1. A method for evaluating multi-level classification information of anesthesia depth, characterized in that: The method for evaluating multi-level classification information of anesthesia depth comprises: Collect EEG signals, perform baseline drift correction, artifact removal and segmentation preprocessing on the EEG signals; decompose the preprocessed EEG signals to obtain signals in five frequency bands: delta, theta, alpha, beta and gamma; Calculate the global field power of all EEG signals in each frequency band and identify the topographic map at the peak moment; perform cluster analysis on the topographic maps at the peak moment of the global field power and extract representative microstate templates; match the topographic map at each time point with the microstate template to reconstruct a continuous microstate time series; Among them, the extracted microstate parameters include characteristic parameters such as microstate duration and occurrence frequency; A feature screening and classification recognition model is established, and the signals of each frequency band are input into the feature screening and classification recognition model to screen the extracted microstate features and perform multi-level classification of the anesthesia depth, and finally output the anesthesia depth assessment results.

2. The method for evaluating the multi-level classification information of anesthesia depth according to claim 1, characterized in that: The process of obtaining signals in the five frequency bands of delta, theta, alpha, beta and gamma includes the following steps: Collect EEG signals and perform baseline drift correction on the collected EEG signals; perform artifact processing on the signals to remove the influence of electrooculogram and electromyography artifacts; perform segmentation processing on the signals to divide the continuous signals into data segments of preset lengths; Performing empirical mode decomposition on the preprocessed EEG signal to decompose the EEG signal into several intrinsic mode function components and a residual term; and performing Hilbert transformation on each intrinsic mode function component to obtain an analytical signal and an instantaneous frequency; According to the calculation results, the EEG signal is decomposed into five characteristic frequency bands, corresponding to 0.5-4Hz delta, 4-8Hz theta, 8-13Hz alpha, 13-30Hz beta and 30-45Hz gamma frequency bands.

3. The method for evaluating the multi-level classification information of anesthesia depth according to claim 2, characterized in that: The process of obtaining the analytical signal and instantaneous frequency includes the following steps: Performing empirical mode decomposition on the preprocessed EEG signal to decompose the EEG signal into several intrinsic mode function components and a residual term; Perform Hilbert transform on each intrinsic mode function component to obtain the analytical signal and instantaneous frequency; A sampling frequency is selected for each electrode signal, and the instantaneous frequency is obtained through Hilbert transform; the instantaneous frequency is limited within a frequency band, and each intrinsic mode function component is adjusted to a corresponding frequency band to generate intrinsic mode function components of different frequency bands.

4. The method for evaluating anesthesia depth multi-level classification information according to claim 1, characterized in that: The process of reconstructing the microstates after cluster analysis includes the following steps: Calculate the global field power for all channel signals in each frequency band; and identify the topographic map at the peak moment; Cluster analysis of microstates: Cluster the topographic map at the peak moment of global field power, and determine the optimal microstate category by optimizing the following indicators: Reconstruct the microstate sequence: by calculating the similarity between the topographic map at each global field power peak moment and the microstate template, mark it as the most similar microstate category and generate a continuous microstate time series; Among them, the microstate time series parameters include: microstate duration, microstate occurrence frequency and microstate transition probability.

5. The method for evaluating anesthesia depth multi-level classification information according to claim 4, characterized in that: The process of generating a continuous microstate time series includes the following steps: The target topographic map is extracted from the topographic map at the moment of the global field power peak through cluster analysis to obtain the microstate template; the topographic map at each moment of the global field power peak is compared with the microstate template, and the similarity between each topographic map at the moment of the global field power peak and all the microstate templates is compared in detail; For each similarity calculated at the moment of global field power peak, select the microstate category with the highest similarity to that moment; sort the microstate categories corresponding to each peak moment in chronological order to form a continuous microstate time series; Based on the generated microstate time series, the length of time each microstate lasts in the time series, the number of times each microstate appears in the time series, and the probability of transitioning from one microstate to another are extracted.

6. The method for evaluating anesthesia depth multi-level classification information according to claim 1, characterized in that: The process of finally outputting the anesthesia depth assessment results includes the following steps: The collected EEG signals were denoised and artifacts removed, and the Hilbert-Huang transform was used to decompose the signals to obtain EEG signals in five frequency bands: delta, theta, alpha, beta, and gamma. On this basis, features related to the depth of anesthesia were extracted, including the average absolute power, standard deviation, and power spectral density in different frequency bands. The features were then fused by combining microstate time series, fractal features, and spectral features to form a complete data set. Divide the data set into a training set, a validation set, and a test set; establish a feature screening and classification recognition model, and use the training set to train the feature screening and classification recognition model; input the microstate features into the feature screening and classification recognition model for classification evaluation; and evaluate the feature screening and classification recognition model through the confusion matrix; Assign a weight to each feature based on the evaluation results, gradually remove the features with the lowest weight, repeat the training and evaluate the performance until the preset number of features is reached; then sort the features and take the feature subset with the preset ranking as the final output.

7. The method for evaluating anesthesia depth multi-level classification information according to claim 6, characterized in that: in, The architecture of the feature screening and classification recognition model is: Feature extraction and screening layer: The basic feature network processes the signals of each frequency band through multiple layers of one-dimensional convolution and pooling layers, extracting and optimizing time-frequency features including average absolute power and power spectral density; The feature fusion module uses the self-attention mechanism to integrate the basic feature network and the extracted micro-state features, and performs feature optimization and dimensionality reduction through the autoencoder; Time series processing layer: Recurrent neural network to capture the dynamic process of changes in anesthesia depth; Classification recognition layer: The fully connected layer fuses the features output by the recurrent neural network layer; Output layer: Use the softmax function for multi-classification and output the evaluation results of the anesthesia depth.

8. The method for evaluating anesthesia depth multi-level classification information according to claim 7, characterized in that: The process of screening features and classifying recognition models is evaluated through confusion matrix, including the following steps: Divide the data set into training set, validation set and test set; build feature screening and classification recognition model based on the historical EEG processing database; The model is trained using the training set. During the training process, features are extracted through the basic feature network, and the features extracted by the basic feature network and the extracted microstate features are integrated and deeply learned by the feature fusion module. After the feature learning is completed, variables are screened recursively, and variables with weights less than a preset threshold are eliminated based on the results of each training, and finally an optimized feature screening and classification recognition model is obtained; The feature screening and classification recognition model is evaluated using the test set. The true labels of the test set are compared with the labels of the feature screening and classification recognition model to generate a confusion matrix. The features extracted by the basic feature network and the microstate features are input into the model for classification evaluation.

9. The method for evaluating anesthesia depth multi-level classification information according to claim 7, characterized in that: The process of taking the preset ranked feature subset as the final output includes the following steps: According to the classification evaluation results of the feature screening and classification recognition model, the features extracted by the basic feature network and the micro-state features are comprehensively analyzed. The contribution of each feature to the classification performance is evaluated through the self-attention mechanism of the feature fusion module, and weights are assigned accordingly. Select the feature with the highest weight from all features as the initial feature subset, and add features one by one; retrain the feature screening and classification recognition models after each adjustment and evaluate them until the preset number of features is reached; When the preset number of features is reached, the remaining features are reordered according to the weights; the first k features with the highest weights are calibrated as the final feature subset, the anesthesia depth level is preset, and the anesthesia depth level corresponding to the final feature subset is output; the corresponding anesthesia depth level is output to the visualization device.

10. A system for evaluating multi-level classification information of depth of anesthesia, applied to the method for evaluating multi-level classification information of depth of anesthesia as claimed in any one of claims 1 to 9, characterized in that: The anesthesia depth multi-level classification information evaluation system comprises: The EEG acquisition and processing module is used to acquire raw EEG signals and perform a series of preprocessing operations on the signals, including baseline drift correction, artifact detection and removal, and signal segmentation processing; after completing the preprocessing, the EEG signals are decomposed into frequency bands to obtain signals in five frequency bands: delta, theta, alpha, beta, and gamma; The clustering reconstruction module is used to calculate the global field power of all EEG signals in each frequency band and identify the topographic map at the peak moment; cluster analysis is performed on these topographic maps at the peak moment of the global field power to extract representative microstate templates; the topographic map at each time point is matched with the microstate template to reconstruct a continuous microstate time series; Finally, the microstate parameters are extracted, including characteristic parameters such as the duration and frequency of each microstate. The anesthesia depth assessment module is used to establish a feature screening and classification recognition model, input each frequency band signal into the feature screening and classification recognition model, screen the extracted microstate features, and perform multi-level classification of the anesthesia depth, and finally output the anesthesia depth assessment result.

Images