Anesthesia state evaluation method and system based on electroencephalogram data analysis

Abstract

The present application relates to the technical field of anesthesia state evaluation, in particular to an anesthesia state evaluation method and system based on electroencephalogram data analysis, comprising: obtaining original electroencephalogram signals through an electroencephalogram acquisition device and preprocessing to obtain pure electroencephalogram data; extracting characteristic indexes reflecting electroencephalogram oscillation patterns and synchronization characteristics; analyzing dynamic causal relationships between the characteristic indexes based on a causal inference model, constructing a brain function network representing anesthesia depth evolution; calculating quantitative parameters of network global integration efficiency and local isolation degree; introducing baseline electroencephalogram data before anesthesia induction to calibrate the quantitative parameters; combining an anesthesia depth determination model to determine the anesthesia state grade, and generating an evaluation report by fusing network structure characteristics. The method improves the accuracy and consistency of anesthesia state evaluation and is suitable for clinical anesthesia monitoring scenarios.

Description

Technical Field

[0001] This invention relates to the field of anesthesia status assessment technology, and in particular to anesthesia status assessment methods and systems based on electroencephalogram (EEG) data analysis. Background Technology

[0002] In clinical anesthesia, accurate assessment of the anesthetic state is crucial for ensuring surgical safety and reducing anesthesia-related risks. Currently, the mainstream assessment methods are all based on electroencephalogram (EEG) signal analysis. Existing technologies generally involve continuously acquiring the raw EEG signals of the target subject using EEG acquisition equipment, preprocessing to remove noise and artifacts, extracting relevant indicators reflecting the characteristics of the EEG signals, and then constructing a brain functional network based on these characteristic indicators, or directly inputting the characteristic indicators into a judgment model to determine the level of anesthesia status of the target subject.

[0003] Current technologies for constructing brain functional networks primarily rely on statistical correlations such as the correlation and coherence between EEG characteristic indicators. However, they fail to capture the inherent dynamic mechanisms of interaction between these indicators and cannot accurately reflect the evolution of anesthesia depth throughout the anesthesia process. In the anesthetic state assessment stage, existing technologies do not consider the baseline differences in EEG data before anesthesia induction, directly using raw quantitative parameters for judgment. This is easily influenced by individual physiological differences, leading to biased assessment results. Therefore, a technical solution is needed that can accurately capture the dynamic relationships between EEG characteristic indicators and eliminate individual baseline differences, thereby achieving precise assessment of the anesthetic state. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies by proposing a method and system for assessing anesthesia status based on electroencephalogram (EEG) data analysis.

[0005] To achieve the above objectives, the present invention employs the following technical solution: a method for assessing anesthesia status based on electroencephalogram (EEG) data analysis, comprising: The raw brainwave signals of the target object are continuously acquired using an EEG acquisition device. The raw EEG signals are preprocessed to remove noise and artifacts, resulting in clean EEG signal data. From the purified EEG signal data, multiple feature indicators reflecting the oscillation pattern and synchronization characteristics of the EEG signal are extracted; Based on the causal inference model, the dynamic causal relationship among the multiple characteristic indicators is analyzed, and a brain functional network characterizing the evolution of anesthesia depth is constructed. Based on the aforementioned brain functional network, quantitative parameters characterizing the network's global information integration efficiency and local isolation degree are calculated; Baseline EEG data of the target subject prior to anesthesia induction were introduced, and the quantitative parameters were standardized and calibrated. Based on standardized and calibrated quantitative parameters, combined with a pre-established anesthesia depth status determination model, the current anesthesia status level of the target subject is determined. The current level of anesthesia is fused with the structural features of the brain functional network to generate an anesthesia status assessment report.

[0006] As a further aspect of the present invention, the original EEG signal is preprocessed to remove noise and artifacts, resulting in clean EEG signal data, specifically including: The raw EEG signal was bandpass filtered to retain signal components within a specific frequency band range relevant to the anesthesia state analysis; Abnormal signal segments in the raw EEG signal whose amplitude exceeds the physiological range are detected and labeled, and identified as obvious motion artifacts or noise caused by poor electrode contact. For the marked noise and artifact segments, an adaptive filtering algorithm is used in combination with signals from multiple reference channels to suppress and remove them; Independent component analysis was performed on the filtered and artifact-removed EEG signals to separate artifact components related to ECG, eye movement, and electromyography. From all the separated components, the independent components corresponding to artifact activity were identified and removed, and the remaining components were reconstructed into pure EEG signal data.

[0007] As a further aspect of the present invention, multiple feature indicators reflecting the oscillation pattern and synchronization characteristics of the brainwave signals are extracted from the purified brainwave signal data, specifically including: Time-frequency analysis was performed on the purified EEG signal data to calculate the power spectral density in different time windows and frequency bands, and the time-varying trajectory of power in each frequency band was extracted. Calculate the phase lock value of EEG signals in a specific frequency band between different brain region leads, as a characteristic indicator to measure the phase synchronization strength between brain regions; Calculate the nonlinear dynamic parameters of EEG signals, including complexity and entropy, to characterize the randomness and unpredictability of EEG activity; The burst suppression pattern was detected from the purified EEG signal data, and the burst suppression ratio and the average oscillation frequency within the burst segment were calculated. The extracted power time-varying trajectory, phase-locked value, nonlinear dynamic parameters, burst suppression ratio, and average oscillation frequency of the burst segment are collectively used to form the aforementioned multiple characteristic indicators.

[0008] As a further aspect of the present invention, the step of analyzing the dynamic causal relationships among the multiple feature indicators based on a causal inference model and constructing a brain functional network characterizing the evolution of anesthesia depth specifically includes: The aforementioned multiple feature indicators are used as time series variables and input into a causal inference framework based on transfer entropy. Calculate the directed transfer entropy value between each pair of feature indicator time series at multiple time delays to quantify the information transmission strength from one indicator to another; Based on the calculated directed transition entropy matrix, a thresholding method is applied to retain causal connections whose information transmission intensity exceeds the significance level, forming a directed connection matrix. Using the aforementioned multiple feature indicators as network nodes and the directed connection matrix as the network's adjacency matrix, an initial directed brain functional network is constructed. A time-sliding window analysis is performed on the initial directed brain functional network to update the directed connection matrix within each time window, thereby obtaining a dynamically evolving brain functional network sequence.

[0009] As a further aspect of the present invention, based on the aforementioned brain functional network, quantitative parameters characterizing the network's global information integration efficiency and local isolation degree are calculated, specifically including: For each network in the dynamically evolving brain functional network sequence, its global efficiency is calculated, which is defined as the inverse average of the shortest path lengths between all pairs of nodes in the network. Calculate the average clustering coefficient of the nodes in the network to measure the tendency of network nodes to form locally tightly connected subgroups; The small-world attribute value of the network is calculated by comparing the clustering coefficient and feature path length of the actual network with a regular random network with the same number of nodes and connections. Calculate the characteristic path length of the network, which is the average of the shortest path lengths between all pairs of nodes in the network; The calculated global efficiency, average clustering coefficient, small-world attribute value, and feature path length are used together as the quantization parameters. The small-world attribute value of the network is calculated by comparing the clustering coefficient and feature path length of the actual network with those of a regular random network with the same number of nodes and connections. Specifically, this includes: Calculate the clustering coefficient and feature path length of the current network in the dynamically evolving brain functional network sequence, and denot them as the actual clustering coefficient and the actual feature path length, respectively. Generate a regular network with the same number of nodes and the same number of connected edges as the current network, and calculate the clustering coefficient and feature path length of the regular network; Generate a random network with the same number of nodes and the same number of connected edges as the current network, and calculate the clustering coefficient and feature path length of the random network; Divide the actual clustering coefficient by the clustering coefficient of the random network to obtain the standardized clustering coefficient ratio. Divide the actual feature path length by the feature path length of the rule network to obtain the standardized feature path length ratio. The quotient obtained by dividing the ratio of the standardized clustering coefficients by the ratio of the standardized feature path lengths is the small-world attribute value.

[0010] As a further aspect of the present invention, the step of introducing baseline EEG data of the target subject before anesthesia induction and standardizing and calibrating the quantitative parameters specifically includes: Before anesthesia induction, baseline EEG signals were collected from the target subject in a conscious and resting state; Using the same steps as described in this method for raw EEG signal preprocessing, feature index extraction, and brain functional network construction, the baseline EEG signal is processed, corresponding baseline feature indices are extracted, and a baseline brain functional network is constructed. Calculate the baseline quantization parameters of the baseline brain functional network, including baseline global efficiency, baseline average clustering coefficient, baseline small-world attribute value, and baseline feature path length; The quantitative parameters of each time window calculated during anesthesia are subtracted from the corresponding baseline quantitative parameters to obtain the change relative to the baseline. Divide the change by the standard deviation of the baseline quantization parameter to complete the standardization calibration of the quantization parameter, and obtain the standardized quantization parameter.

[0011] As a further aspect of the present invention, the step of determining the current anesthesia status level of the target object based on standardized and calibrated quantitative parameters and in conjunction with a pre-established anesthesia depth status determination model specifically includes: The pre-established anesthesia depth status determination model is a classification model trained using a supervised machine learning algorithm based on standardized and calibrated quantitative parameters from historical anesthesia cases and their corresponding expert-annotated anesthesia status levels. A set of standardized and calibrated quantitative parameters calculated for the target object within the most recent time window are input into the anesthesia depth status determination model; The anesthesia depth determination model outputs a probability vector based on a set of standardized and calibrated quantitative parameters. Each element of the probability vector represents the probability that the target object belongs to a certain predefined anesthesia state level. The level of anesthesia corresponding to the highest probability value in the probability vector is selected as the current level of anesthesia for the target object. If the maximum probability value is lower than the preset confidence threshold, the current anesthesia status level will be marked as "uncertain" and an alarm will be triggered.

[0012] As a further aspect of the present invention, the current anesthesia status level is fused with the structural features of the brain functional network to generate an anesthesia status assessment report, specifically including: From the dynamically evolving brain functional network sequence, extract the structural features of the brain functional network corresponding to the current evaluation time. The structural features include the set of core nodes and the set of core connecting edges of the network. The current level of anesthesia and the structural features of the brain functional network are organized according to a preset report template, which includes three parts: level of anesthesia, core brain regions, and main information flow direction. The report uses a graphical method to illustrate the core set of connections in the brain functional network, with particular emphasis on the nodes with the highest information inflow and outflow intensity; The text of the report describes the anatomical brain regions corresponding to the core node set and their functional meaning under anesthesia. The complete report, including illustrations and text descriptions, will be integrated into a structured anesthesia status assessment report.

[0013] As a further aspect of the present invention, the directed transition entropy matrix obtained by calculation is used to apply a thresholding method to retain causal connections whose information transmission strength exceeds the significance level, forming a directed connection matrix, specifically including: For each element in the directed transition entropy matrix, i.e. the directed transition entropy value between each pair of feature indicators, a statistical significance test is performed, and its p-value is calculated. Set a significance level threshold, and consider causal connections with p-values ​​less than the significance level threshold as statistically significant valid connections; For statistically significant valid connections, their directed transition entropy values ​​are retained; for insignificant connections, their values ​​are set to zero. The permutation test method is used to generate an alternative dataset, and the distribution of directed transition entropy values ​​is calculated based on the alternative dataset to further verify the reliability of effective connections; After significance testing and reliability verification, all retained statistically significant directed transition entropy values ​​are arranged according to the original matrix structure to form the final directed connection matrix.

[0014] As a further aspect of the present invention, the present invention also includes an anesthesia state assessment system based on electroencephalogram (EEG) data analysis. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the anesthesia state assessment method based on EEG data analysis as described above.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: Based on a causal inference model, this study analyzes the dynamic causal relationships among multiple characteristic indicators extracted from pure EEG signals, reflecting the oscillation patterns and synchronization characteristics of EEG signals, and constructs a brain functional network characterizing the evolution of anesthesia depth. Compared to conventional methods that construct brain functional networks solely through statistical correlations such as correlation and coherence, this approach can accurately capture the intrinsic dynamic mechanisms of interaction between EEG characteristic indicators, clearly present the real-time evolution characteristics of the brain functional network during anesthesia, and ensure a precise correspondence between the brain functional network and changes in anesthesia depth. This avoids spurious correlation interference caused by statistical correlations and makes the brain functional network more closely reflect the actual changes in the anesthetic state.

[0016] After calculating the quantitative parameters characterizing the efficiency of global information integration and the degree of local isolation in the brain functional network, baseline EEG data of the target subject before anesthesia induction are introduced. The quantitative parameters of this network are then standardized and calibrated. Based on the calibrated quantitative parameters and an anesthesia depth assessment model, the anesthesia status level is determined. Compared to the conventional method of directly using raw quantitative parameters for anesthesia status assessment, this approach effectively eliminates individual differences in EEG baselines among different target subjects, making the quantitative parameters more individualized, reducing the interference of individual physiological differences on the assessment results, and ensuring that the anesthesia status level assessment more closely reflects the actual anesthesia situation of the target subject, thus improving the consistency and accuracy of anesthesia status assessment. Attached Figure Description

[0017] Figure 1 This is a flowchart of the anesthesia state assessment method based on electroencephalogram (EEG) data analysis described in this invention; Figure 2 A flowchart for the preprocessing of raw EEG signals; Figure 3 A flowchart illustrating the construction of a dynamic brain functional network based on causal inference; Figure 4 A graph showing the trend of quantitative parameters of brain functional networks during anesthesia. Figure 5 This is a graph showing the changes in brain functional network parameters throughout the anesthesia period. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] See Figure 1 The raw EEG signals of the target subject are continuously acquired through an EEG acquisition device. The raw EEG signals are preprocessed to remove noise and artifacts, resulting in clean EEG signal data. Multiple feature indicators reflecting the oscillation pattern and synchronization characteristics of the EEG signals are extracted from the clean EEG signal data. The dynamic causal relationship between the multiple feature indicators is analyzed based on a causal inference model, and a brain functional network characterizing the evolution of anesthesia depth is constructed. Based on this brain functional network, quantitative parameters characterizing the efficiency of global information integration and the degree of local isolation of the network are calculated. The baseline EEG data of the target subject before anesthesia induction is introduced to standardize and calibrate the quantitative parameters. Based on the standardized and calibrated quantitative parameters and combined with the pre-established anesthesia depth status judgment model, the current anesthesia status level of the target subject is determined. The current anesthesia status level is fused with the structural characteristics of the brain functional network to generate an anesthesia status assessment report.

[0021] See Figure 2In one embodiment of the present invention, the raw EEG signal is preprocessed to remove noise and artifacts to obtain clean EEG signal data. The steps include: bandpass filtering the raw EEG signal to retain signal components within a specific frequency range (e.g., 0.5 Hz to 45 Hz) relevant to anesthesia state analysis; detecting and labeling abnormal signal segments in the raw EEG signal whose amplitude exceeds the physiological range (e.g., amplitudes exceeding ±200 μV), identifying them as significant motion artifacts or noise caused by poor electrode contact; using an adaptive filtering algorithm to suppress and remove the labeled noise and artifact segments in combination with signals from multiple reference channels; performing independent component analysis on the filtered and artifact-removed EEG signal to separate artifact components related to ECG, eye movement, and electromyography (EMG) activity; identifying and removing independent components corresponding to artifact activity from all separated components, and reconstructing the remaining components into clean EEG signal data. In some embodiments, the bandpass filtering frequency range can be adjusted to accommodate different anesthetic drugs or individual differences; for example, the frequency band can be set to 0.5 Hz to 30 Hz to focus on low-frequency oscillations more relevant to the depth of anesthesia. In practice, after independent component analysis, artifact components are identified based on the correlation coefficient between the component's time progression and a known artifact template. For example, components with a correlation coefficient exceeding 0.8 with a standard eye-tracking artifact template are marked as eye-tracking artifacts and removed.

[0022] This involves extracting multiple characteristic indicators reflecting the oscillation patterns and synchronization characteristics of pure EEG signal data. Specifically, this includes performing time-frequency analysis on the pure EEG signal data, calculating the power spectral density at different time windows and frequency bands, and extracting the time-varying trajectory of power in each band. For example, the EEG signal is divided into delta band (0.5-4 Hz), theta band (4-8 Hz), alpha band (8-13 Hz), beta band (13-30 Hz), and gamma band (30-45 Hz), and the power value of each band within a sliding time window is calculated. The analysis also includes calculating the power values ​​between different brain region leads. Phase lock values ​​of EEG signals in specific frequency bands serve as characteristic indicators for measuring the strength of phase synchronization between brain regions. For example, calculating the phase lock value of the prefrontal and occipital leads in the alpha band; calculating nonlinear dynamic parameters of EEG signals, including complexity and entropy, to characterize the randomness and unpredictability of EEG activity; detecting burst suppression patterns from clean EEG signal data, calculating the burst suppression ratio and the average oscillation frequency within the burst segment; and combining the extracted power time-varying trajectories of each frequency band, phase lock values, nonlinear dynamic parameters, burst suppression ratio, and average oscillation frequency of the burst segment to form multiple characteristic indicators. In specific implementation, the phase lock value is calculated using a phase extraction method based on Hilbert transform, expressed by the following formula: Wherein: Phase Locking Value This indicates the phase synchronization strength between signals from two brain region leads. This represents the total number of time points in the time series. This indicates the first brain region lead signal at time point The instantaneous phase, This indicates the signal from the second brain region at time point. The instantaneous phase, It is the imaginary unit. This can be understood as the phase-locked value. The value ranges from 0 to 1. Optionally, the time-frequency analysis uses Morley wavelet transform to balance time and frequency resolution, with a sliding time window length of 2 seconds and an overlap ratio of 50%. In some embodiments, the entropy value in the nonlinear dynamic parameters is calculated using the sample entropy method. The calculation of sample entropy involves comparing the probability of pattern repeatability in the signal sequence, with the embedding dimension set to 2 and the tolerance threshold set to 0.2 times the standard deviation of the signal. In a specific implementation, the detection of burst suppression patterns is based on an amplitude threshold method, for example, identifying segments of EEG signal amplitude below 5 microvolts as suppression segments and segments above 20 microvolts as burst segments. The burst suppression ratio is defined as the ratio of the total duration of burst segments to the total analysis time.

[0023] See Figure 3 In one embodiment of the present invention, a causal inference model is used to analyze the dynamic causal relationships among multiple feature indicators and construct a brain functional network characterizing the evolution of anesthesia depth. The implementation steps include inputting multiple feature indicators as time-series variables into a causal inference framework based on transfer entropy. The directed transfer entropy value between each pair of feature indicator time series at multiple time delays is calculated to quantify the information transfer intensity from one indicator to another. In a specific implementation, the directed transfer entropy value... The calculation formula is expressed as follows: Wherein: the directed transition entropy value Indicates from characteristic indicators To feature indicators Information transmission intensity Indicators of characteristics At the point of time state, Indicators of characteristics At the point of time and before The state vector of each historical point Indicators of characteristics At the point of time and before The state vector of each historical point This represents the joint probability or conditional probability. It can be understood as the directed transition entropy value. The value is always non-negative; the larger the value, the better the characteristic index. To feature indicators The stronger the information flow.

[0024] Based on the calculated directed transition entropy matrix, a thresholding method is applied to retain causal connections with information transmission strength exceeding a significance level, forming a directed connection matrix. In practice, the thresholding method involves performing a statistical significance test on each element in the directed transition entropy matrix, calculating its p-value, and setting a significance level threshold, for example, 0.05. Causal connections with p-values ​​less than this threshold are considered statistically significant valid connections. For statistically significant valid connections, the corresponding directed transition entropy values ​​in the directed transition entropy matrix are retained; for insignificant connections, the corresponding directed transition entropy values ​​are set to zero. A permutation test is used to generate a substitute dataset. The permutation test disrupts the potential causal relationships by randomly shuffling the time series of feature indicators. The distribution of directed transition entropy values ​​is calculated based on the substitute dataset to further verify the reliability of valid connections. After significance testing and reliability verification, all retained statistically significant directed transition entropy values ​​are arranged according to the original matrix structure to form the final directed connection matrix. Optionally, the significance test uses the surrogate method based on alternative data to generate 1000 alternative time series to construct the null distribution of the directed transition entropy values.

[0025] An initial directed brain functional network is constructed using multiple feature indicators as network nodes and the final directed connection matrix as the network's adjacency matrix. A temporal sliding window analysis is then performed on the initial directed brain functional network. This analysis uses a fixed-length window, such as a 30-second window, to slide along the time axis, updating the directed connection matrix within each time window to obtain a dynamically evolving sequence of brain functional networks. In some embodiments, the sliding window's step size is set to 50% of the window length, for example, 15 seconds. It can be understood that each time window corresponds to a directed weighted brain functional network defined by the final directed connection matrix. Optionally, the temporal sliding window analysis allows observation of the dynamic evolution of brain functional network connectivity patterns following anesthesia administration or stimulation events.

[0026] In one embodiment of the present invention, quantitative parameters characterizing the global information integration efficiency and local isolation degree of a brain functional network are calculated based on the network. The implementation steps include calculating the global efficiency for each network in the dynamically evolving brain functional network sequence. The global efficiency is defined as the reciprocal average of the shortest path lengths between all pairs of nodes in the network. Specifically, the Floyd-Worschel algorithm or Dijkstra algorithm is first used to calculate the shortest path lengths between all pairs of nodes in the network. For disconnected pairs of nodes, the shortest path length is considered infinite, and its reciprocal is taken as zero. Then, the reciprocals of the shortest path lengths between all pairs of nodes are summed and divided by the total number of pairs of nodes to obtain the global efficiency. The average clustering coefficient of the nodes in the network is calculated to measure the tendency of network nodes to form locally tightly connected subgroups. The calculation of the average clustering coefficient involves calculating the ratio of the actual number of connections between each node's neighbors to the theoretically maximum number of connections. Then, the arithmetic mean of the clustering coefficients of all nodes is taken. The small-world attribute value of the network is calculated by comparing the clustering coefficient and feature path length of the actual network with those of a regular random network with the same number of nodes and edges. The feature path length of the network is calculated as the average of the shortest path lengths between all pairs of nodes in the network, excluding disconnected pairs of nodes. The calculated global efficiency, average clustering coefficient, small-world attribute value, and feature path length are used together as quantification parameters for the network attributes.

[0027] In practical implementation, calculating the small-world attribute values ​​of a network involves a series of specific calculation steps. First, the clustering coefficient and feature path length of the current network in the dynamically evolving brain functional network sequence are calculated and denoted as the actual clustering coefficient and actual feature path length, respectively. Second, a regular network with the same number of nodes and edges as the current network is generated. The regular network is generated using the nearest neighbor coupling method, where each node is connected to its left and right K / 2 neighbor nodes. The clustering coefficient and feature path length of the regular network are then calculated. Third, a random network with the same number of nodes and edges as the current network is generated. The random network is generated using the Erdős-Raney model, randomly connecting a given number of edges. The clustering coefficient and feature path length of the random network are then calculated. (Small-world attribute values) The calculation formula is expressed as follows: Among them: Small World attribute value The strength of a network's small-world property, which combines high clustering and short paths, is measured. This represents the actual clustering coefficient of the current network in a dynamically evolving brain functional network sequence. This represents the clustering coefficient of random networks with the same number of nodes and connected edges. This represents the actual characteristic path length of the current network in a dynamically evolving brain functional network sequence. The characteristic path length represents the length of a regular network with the same number of nodes and edges. This represents the ratio of standardized clustering coefficients. This represents the ratio of standardized feature path lengths. It can be understood that when the small-world attribute value... When the value is greater than 1, the network is considered to have small-world properties. In some embodiments, the number of neighbor connections K of a regular network is set based on the average degree of the actual network, for example, taking an even number closest to the average degree. Optionally, the generation of random networks employs a method of averaging multiple randomizations, for example, generating 20 random networks and calculating the average of their clustering coefficients and feature path lengths to enhance robustness. In specific implementations, for the calculation of feature path lengths, when disconnected node pairs exist in the network, the calculation of both the actual feature path length and the feature path length of the regular network uses the average of connected node pairs to ensure comparability.

[0028] In one embodiment of the present invention, baseline EEG data of the target subject prior to anesthesia induction is introduced to standardize and calibrate the quantification parameters. The implementation steps include collecting baseline EEG signals of the target subject in a conscious and resting state before anesthesia induction for 5 minutes. The baseline EEG signals are processed using the same steps as in the original EEG signal preprocessing, feature index extraction, and brain function network construction in this method. Corresponding baseline feature indices are extracted, and a baseline brain function network is constructed. The baseline quantification parameters of the baseline brain function network are calculated, including baseline global efficiency, baseline average clustering coefficient, baseline small-world attribute value, and baseline feature path length. In specific implementation, the quantification parameters calculated for each time window during anesthesia are subtracted from the corresponding baseline quantification parameters to obtain the change relative to the baseline. The change is divided by the standard deviation of the baseline quantification parameters to complete the standardization and calibration of the quantification parameters, resulting in the standardized quantification parameters. The formula for standardization and calibration is expressed as: Wherein: quantization parameters after standardization and calibration This represents the standardized change of a quantitative parameter relative to an individual's baseline within a specific time window during anesthesia. This represents the original quantitative parameter values ​​calculated within this time window during the anesthesia process. This represents the average value of the corresponding baseline quantized parameters calculated from the baseline brain functional network. This represents the standard deviation of the corresponding baseline quantization parameter. It can be understood that the formula uses the average value of the baseline quantization parameter... with standard deviation The calculation is typically based on statistical analysis of the results from all continuous time windows within the baseline recording period. See Table 1 for an example comparison of quantization parameter values.

[0029] Table 1: Numerical Table of Exemplary Quantization Parameter Calibration Process The current anesthesia status level of the target object is determined based on standardized and calibrated quantitative parameters combined with a pre-established anesthesia depth status determination model. In practice, this model is a classification model trained using a supervised machine learning algorithm based on standardized and calibrated quantitative parameters from historical anesthesia cases and their corresponding expert-annotated anesthesia status levels. The number of historical anesthesia cases exceeds 1000, and the expert-annotated anesthesia status levels include conscious, sedated, light anesthesia, surgical anesthesia, deep anesthesia, and burst suppression. A set of standardized and calibrated quantitative parameters calculated for the target object within the most recent time window is input into the anesthesia depth status determination model. In practice, this set of parameters includes four values: standardized and calibrated global efficiency, average clustering coefficient, small-world attribute value, and feature path length. The model outputs a probability vector based on the input standardized and calibrated quantitative parameters. Each element of the probability vector represents the probability that the target object belongs to a predefined anesthesia status level. The anesthesia status level corresponding to the highest probability value in the probability vector is selected as the target object's current anesthesia status level. If the highest probability value is lower than a preset confidence threshold, the current anesthesia status level is marked as uncertain, and an alarm is triggered. In some embodiments, the preset confidence threshold is set to 0.7. Optionally, the anesthesia depth determination model is trained using a gradient boosting decision tree algorithm, and the input features of the model are a set of four standardized and calibrated quantized parameters. It can be understood that the anesthesia depth determination model has undergone performance evaluation on an independent validation dataset before application.

[0030] See Figure 4 This is a graph showing the trend of quantitative parameters of brain functional networks during anesthesia. It illustrates the standardized and calibrated changes of four core quantitative parameters of brain functional networks throughout the complete anesthesia process, from wakefulness to burst inhibition. It visually reflects the dynamic evolution of brain functional networks as anesthesia deepens. Global efficiency decreases from nearly 0 during wakefulness to approximately -2.5 during burst inhibition, indicating a significant weakening of the brain network's global information integration ability as anesthesia deepens. The average clustering coefficient peaks in the light anesthesia stage and then gradually decreases, reflecting enhanced local connectivity in brain regions during the initial stage of anesthesia, while local functional isolation is disrupted during deep anesthesia. The small-world property value decreases from nearly 0 to approximately -2.8, indicating that the "small-world" characteristics of the brain network are severely damaged under anesthesia. The characteristic path length increases from approximately 0.5 to approximately 3.5, indicating that the information transmission paths between brain regions become longer, and network communication efficiency decreases.

[0031] In one embodiment of the present invention, the current anesthesia status level is fused with the structural features of the brain functional network to generate an anesthesia status assessment report. The implementation steps include extracting the structural features of the brain functional network corresponding to the current assessment time from a dynamically evolving brain functional network sequence. The structural features include a core node set and a core edge set. The core node set is extracted based on node centrality indices such as node strength centrality or betweenness centrality, selecting the top five nodes with the highest centrality ranking to add to the core node set. The core edge set is extracted based on the weights of the edges in the directed connection matrix, selecting the top ten directed edges with the highest weight ranking to add to the core edge set. In some embodiments, the node centrality index uses weighted node strength, and its calculation formula is expressed as: Where: weighted node strength Represents a node Importance metrics in networks Indicates from node Pointing to node The weight of the connecting edge is the corresponding directed transition entropy value. Indicates from node Pointing to node The weight of the connecting edges. This can be understood as the weighted node strength. It combines the intensity of information flows sent and received by the nodes. The weight threshold of the core connection edge set can be set as the median of the weight values ​​of all non-zero edges in the directed connection matrix.

[0032] The current anesthesia status level and the structural characteristics of the brain functional network are organized according to a preset report template. The report template includes three parts: anesthesia status level, core brain regions, and main information flow direction. In practice, the anesthesia status level section directly fills in the level label output by the judgment model, such as "surgical anesthesia"; the core brain regions section lists the standard EEG lead location and anatomical brain region name corresponding to each node in the core node set, such as node "Fp1" corresponding to "left prefrontal cortex"; the main information flow direction section lists the starting node, ending node, and weight value of each directed edge in the core connection edge set. In practice, the report template exists in the form of a structured electronic document, such as JSON or XML format. Optionally, the report template can be adjusted according to different clinical scenarios, such as adding special annotations for brain regions related to the motor cortex in neurosurgical anesthesia.

[0033] The report uses a graphical approach to visualize the core set of connections in the brain's functional network, highlighting nodes with the highest inflow and outflow intensities. This visualization involves generating a directed network topology graph, where nodes are represented by circular icons, with node size proportional to their weighted node strength. Directed connections are represented by arrows, with line thickness proportional to the edge's weight. Nodes with the highest inflow and outflow intensities are highlighted by changing their color and border; for example, the node with the highest weighted node strength is rendered red and its border is thickened. The spatial arrangement of nodes in the network topology graph can be understood as mapping to standard EEG electrode location coordinates. In some embodiments, the graph is displayed side-by-side with real-time updated time-series curves on the monitor's user interface. Optionally, the network topology graph supports user interaction, such as clicking on nodes to view detailed information.

[0034] The text section of the report describes the anatomical brain regions corresponding to the core node set and their functional significance under anesthesia. In practice, the text descriptions are generated based on a predefined brain region-function mapping database. For example, for the "Pz" lead node in the core node set, the text description is: "This node corresponds to the midline of the parietal lobe, is related to the level of consciousness and sensory information integration, and its inhibition during anesthesia is often associated with loss of consciousness." The complete report, including illustrations and text descriptions, is then integrated into a structured anesthesia status assessment report. The integration process involves merging the illustrations, text descriptions, and structured data into a single PDF document or a specific data structure.

[0035] See Figure 5 This is a graph showing the changes in brain functional network parameters throughout the anesthesia process. It illustrates the dynamic trends of three core quantitative parameters of the brain functional network during five key stages from pre-anesthesia induction to post-anesthesia recovery, intuitively reflecting the impact of anesthesia on the structure and function of the brain network and the recovery process. Global efficiency reaches its lowest point (approximately 0.45) during the anesthesia maintenance phase, recovering to near pre-induction levels (approximately 0.82) after recovery, indicating that the brain network's global information integration ability is most severely impaired at the deepest anesthesia depth. The clustering coefficient also reaches its lowest value (approximately 0.2) during the anesthesia maintenance phase, slowly rising to approximately 0.38 after recovery, reflecting the disruption of local brain region connections during anesthesia and a relatively slow recovery rate. The small-world property reaches its lowest point (approximately 1.2) during the anesthesia maintenance phase, recovering to 1.75 after recovery, close to the pre-induction value of 1.8, indicating that anesthesia significantly disrupts the efficient "small-world" topology of the brain network, but it largely recovers after recovery.

[0036] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for assessing anesthesia status based on electroencephalogram (EEG) data analysis, characterized in that, Includes the following steps: The raw brainwave signals of the target object are continuously acquired using an EEG acquisition device. The raw EEG signals are preprocessed to remove noise and artifacts, resulting in clean EEG signal data. From the purified EEG signal data, multiple feature indicators reflecting the oscillation pattern and synchronization characteristics of the EEG signal are extracted; Based on the causal inference model, the dynamic causal relationship among the multiple characteristic indicators is analyzed, and a brain functional network characterizing the evolution of anesthesia depth is constructed. Based on the aforementioned brain functional network, quantitative parameters characterizing the network's global information integration efficiency and local isolation degree are calculated; Baseline EEG data of the target subject prior to anesthesia induction were introduced, and the quantitative parameters were standardized and calibrated. Based on standardized and calibrated quantitative parameters, combined with a pre-established anesthesia depth status determination model, the current anesthesia status level of the target subject is determined. The current level of anesthesia is fused with the structural features of the brain functional network to generate an anesthesia status assessment report.

2. The method for assessing anesthesia status based on electroencephalogram (EEG) data analysis according to claim 1, characterized in that, The raw EEG signals are preprocessed to remove noise and artifacts, resulting in clean EEG signal data. Specifically, this includes: The raw EEG signal was bandpass filtered to retain signal components within a specific frequency band range relevant to the anesthesia state analysis; Abnormal signal segments in the raw EEG signal whose amplitude exceeds the physiological range are detected and labeled, and identified as obvious motion artifacts or noise caused by poor electrode contact. For the marked noise and artifact segments, an adaptive filtering algorithm is used in combination with signals from multiple reference channels to suppress and remove them; Independent component analysis was performed on the filtered and artifact-removed EEG signals to separate artifact components related to ECG, eye movement, and electromyography. From all the separated components, the independent components corresponding to artifact activity were identified and removed, and the remaining components were reconstructed into pure EEG signal data.

3. The method for assessing anesthesia status based on electroencephalogram (EEG) data analysis according to claim 2, characterized in that, From the purified EEG signal data, several feature indicators reflecting the oscillation pattern and synchronization characteristics of the EEG signals are extracted, specifically including: Time-frequency analysis was performed on the purified EEG signal data to calculate the power spectral density in different time windows and frequency bands, and the time-varying trajectory of power in each frequency band was extracted. Calculate the phase lock value of EEG signals in a specific frequency band between different brain region leads, as a characteristic indicator to measure the phase synchronization strength between brain regions; Calculate the nonlinear dynamic parameters of EEG signals, including complexity and entropy, to characterize the randomness and unpredictability of EEG activity; The burst suppression pattern was detected from the purified EEG signal data, and the burst suppression ratio and the average oscillation frequency within the burst segment were calculated. The extracted power time-varying trajectory, phase-locked value, nonlinear dynamic parameters, burst suppression ratio, and average oscillation frequency of the burst segment are collectively used to form the aforementioned multiple characteristic indicators.

4. The method for assessing anesthesia status based on electroencephalogram (EEG) data analysis according to claim 3, characterized in that, The method, based on a causal inference model, analyzes the dynamic causal relationships among the multiple characteristic indicators and constructs a brain functional network characterizing the evolution of anesthesia depth, specifically including: The aforementioned multiple feature indicators are used as time series variables and input into a causal inference framework based on transfer entropy. Calculate the directed transfer entropy value between each pair of feature indicator time series at multiple time delays to quantify the information transmission strength from one indicator to another; Based on the calculated directed transition entropy matrix, a thresholding method is applied to retain causal connections whose information transmission intensity exceeds the significance level, forming a directed connection matrix. Using the aforementioned multiple feature indicators as network nodes and the directed connection matrix as the network's adjacency matrix, an initial directed brain functional network is constructed. A time-sliding window analysis is performed on the initial directed brain functional network to update the directed connection matrix within each time window, thereby obtaining a dynamically evolving brain functional network sequence.

5. The method for assessing anesthesia status based on electroencephalogram (EEG) data analysis according to claim 4, characterized in that, Based on the aforementioned brain functional network, quantitative parameters characterizing the network's global information integration efficiency and local isolation degree are calculated, specifically including: For each network in the dynamically evolving brain functional network sequence, its global efficiency is calculated, which is defined as the inverse average of the shortest path lengths between all pairs of nodes in the network. Calculate the average clustering coefficient of the nodes in the network to measure the tendency of network nodes to form locally tightly connected subgroups; The small-world attribute value of the network is calculated by comparing the clustering coefficient and feature path length of the actual network with a regular random network with the same number of nodes and connections. Calculate the characteristic path length of the network, which is the average of the shortest path lengths between all pairs of nodes in the network; The calculated global efficiency, average clustering coefficient, small-world attribute value, and feature path length are used together as the quantization parameters. The small-world attribute value of the network is calculated by comparing the clustering coefficient and feature path length of the actual network with those of a regular random network with the same number of nodes and connections. Specifically, this includes: Calculate the clustering coefficient and feature path length of the current network in the dynamically evolving brain functional network sequence, and denot them as the actual clustering coefficient and the actual feature path length, respectively. Generate a regular network with the same number of nodes and the same number of connected edges as the current network, and calculate the clustering coefficient and feature path length of the regular network; Generate a random network with the same number of nodes and the same number of connected edges as the current network, and calculate the clustering coefficient and feature path length of the random network; Divide the actual clustering coefficient by the clustering coefficient of the random network to obtain the standardized clustering coefficient ratio. Divide the actual feature path length by the feature path length of the rule network to obtain the standardized feature path length ratio. The quotient obtained by dividing the ratio of the standardized clustering coefficients by the ratio of the standardized feature path lengths is the small-world attribute value.

6. The method for assessing anesthesia status based on electroencephalogram (EEG) data analysis according to claim 5, characterized in that, The process of introducing baseline EEG data of the target subject before anesthesia induction and standardizing and calibrating the quantitative parameters specifically includes: Before anesthesia induction, baseline EEG signals were collected from the target subject in a conscious and resting state; Using the same steps as described in this method for raw EEG signal preprocessing, feature index extraction, and brain functional network construction, the baseline EEG signal is processed, corresponding baseline feature indices are extracted, and a baseline brain functional network is constructed. Calculate the baseline quantization parameters of the baseline brain functional network, including baseline global efficiency, baseline average clustering coefficient, baseline small-world attribute value, and baseline feature path length; The quantitative parameters of each time window calculated during anesthesia are subtracted from the corresponding baseline quantitative parameters to obtain the change relative to the baseline. Divide the change by the standard deviation of the baseline quantization parameter to complete the standardization calibration of the quantization parameter, and obtain the standardized quantization parameter.

7. The method for assessing anesthesia status based on electroencephalogram (EEG) data analysis according to claim 6, characterized in that, Based on standardized and calibrated quantitative parameters, and combined with a pre-established anesthesia depth status determination model, the current anesthesia status level of the target subject is determined, specifically including: The pre-established anesthesia depth status determination model is a classification model trained using a supervised machine learning algorithm based on standardized and calibrated quantitative parameters from historical anesthesia cases and their corresponding expert-annotated anesthesia status levels. A set of standardized and calibrated quantitative parameters calculated for the target object within the most recent time window are input into the anesthesia depth status determination model; The anesthesia depth determination model outputs a probability vector based on a set of standardized and calibrated quantitative parameters. Each element of the probability vector represents the probability that the target object belongs to a certain predefined anesthesia state level. The level of anesthesia corresponding to the highest probability value in the probability vector is selected as the current level of anesthesia for the target object. If the maximum probability value is lower than the preset confidence threshold, the current anesthesia status level will be marked as "uncertain" and an alarm will be triggered.

8. The method for assessing anesthesia status based on electroencephalogram (EEG) data analysis according to claim 7, characterized in that, The current anesthesia status level is fused with the structural features of the brain functional network to generate an anesthesia status assessment report, which specifically includes: From the dynamically evolving brain functional network sequence, extract the structural features of the brain functional network corresponding to the current evaluation time. The structural features include the set of core nodes and the set of core connecting edges of the network. The current level of anesthesia and the structural features of the brain functional network are organized according to a preset report template, which includes three parts: level of anesthesia, core brain regions, and main information flow direction. The report uses a graphical method to illustrate the core set of connections in the brain functional network, with particular emphasis on the nodes with the highest information inflow and outflow intensity; The text of the report describes the anatomical brain regions corresponding to the core node set and their functional meaning under anesthesia. The complete report, including illustrations and text descriptions, will be integrated into a structured anesthesia status assessment report.

9. The method for assessing anesthesia status based on electroencephalogram (EEG) data analysis according to claim 8, characterized in that, The directed transition entropy matrix obtained from the calculation is thresholded to retain causal connections with information transmission strength exceeding the significance level, forming a directed connection matrix, specifically including: For each element in the directed transition entropy matrix, i.e. the directed transition entropy value between each pair of feature indicators, a statistical significance test is performed, and its p-value is calculated. Set a significance level threshold, and consider causal connections with p-values ​​less than the significance level threshold as statistically significant valid connections; For statistically significant valid connections, their directed transition entropy values ​​are retained; for insignificant connections, their values ​​are set to zero. The permutation test method is used to generate an alternative dataset, and the distribution of directed transition entropy values ​​is calculated based on the alternative dataset to further verify the reliability of effective connections; After significance testing and reliability verification, all retained statistically significant directed transition entropy values ​​are arranged according to the original matrix structure to form the final directed connection matrix.

10. An anesthesia status assessment system based on electroencephalogram (EEG) data analysis, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the anesthesia state assessment method based on EEG data analysis as described in any one of claims 1 to 9.

Images