Anesthesia state monitoring method, system and device based on multi-source physiological signal analysis

Abstract

This invention provides a method, system, and device for monitoring anesthesia status based on multi-source physiological signal analysis, relating to the field of medical monitoring technology. The method includes: acquiring electroencephalogram (EEG) signals, real-time heart rate, heart rate variability coefficient, and real-time mean arterial pressure of a patient under anesthesia; performing preliminary filtering of the EEG signals using high-pass and low-pass filters, and normalizing the amplitude of the EEG signals to obtain new EEG signals; acquiring several modal functions; extracting multidimensional features from each modal function; forming a feature set; training a machine learning model using the feature set, and using the trained machine learning model to predict the patient's anesthesia status in real time, outputting anesthesia stage classification results; determining the comprehensive stability coefficient of the anesthesia status; determining the anesthesia depth instability fluctuation coefficient; and performing anesthesia status monitoring. According to this invention, the accuracy of anesthesia status monitoring can be improved.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring technology, and in particular to a method, system and device for monitoring anesthesia status based on multi-source physiological signal analysis. Background Technology

[0002] In related technologies, anesthesia status monitoring mostly relies on simple single physiological parameter measurements (such as heart rate and blood pressure), lacking an objective, continuous, and multi-dimensional quantitative system. This makes it difficult to provide anesthesiologists with reasonable and accurate adjustment suggestions. In other words, related technologies are unable to improve the effectiveness and accuracy of anesthesia status monitoring.

[0003] The information disclosed in the background section of this application is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0004] This invention provides a method, system, and device for monitoring anesthesia status based on multi-source physiological signal analysis, which can solve the technical problem that related technologies are unable to improve the effectiveness and accuracy of anesthesia status monitoring.

[0005] According to a first aspect of the present invention, a method for monitoring anesthesia status based on multi-source physiological signal analysis is provided, comprising: At multiple points during the monitoring period, the patient's electroencephalogram (EEG) signals, real-time heart rate, heart rate variability coefficient, and real-time mean arterial pressure were collected while the patient was under anesthesia. High-pass and low-pass filters were used to initially filter out EEG signals, and amplitude normalization was performed on the EEG signals to obtain new EEG signals. The new EEG signal was subjected to variational mode decomposition using a hierarchical frequency decomposition method to obtain several mode functions; Extract multidimensional features from each mode function; The extracted multidimensional features are fused and filtered to form a feature set; The machine learning model is trained using a feature set, and the trained machine learning model is used to predict the patient's anesthesia status in real time, outputting the classification results of the anesthesia stage. Based on the real-time heart rate, the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results, the comprehensive stability coefficient of the anesthesia state is determined. The anesthesia depth instability fluctuation coefficient is determined based on the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results. Anesthesia status is monitored based on the comprehensive stability coefficient of the anesthesia status and the unstable fluctuation coefficient of the depth of anesthesia.

[0006] According to the present invention, a high-pass filter and a low-pass filter are used to initially filter out EEG signals, and the amplitude of the EEG signals is normalized to obtain new EEG signals, including: A high-pass filter was used to remove baseline drift signals from the electroencephalogram (EEG) signal; A low-pass filter was used to remove high-frequency interference from the EEG signal; The amplitude of the EEG signal after removing baseline drift and high-frequency interference was normalized to obtain a new EEG signal.

[0007] According to the present invention, a comprehensive stability coefficient of anesthetic state is determined based on the real-time heart rate, the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification result, including: Based on the real-time heart rate and the real-time mean arterial pressure, a first dynamic weight, a second dynamic weight, and a third dynamic weight are determined. Determine the target anesthesia stage; Based on the target anesthesia stage and the anesthesia stage classification results, the probability of adequate anesthesia is determined; Based on the appropriate anesthesia probability, determine the standard deviation of the appropriate anesthesia probability within a preset time window; Obtain the baseline value of the individual's heart rate variability coefficient and the ideal real-time mean arterial pressure under anesthesia; The comprehensive stability coefficient of the anesthesia state is determined based on the first dynamic weight, the second dynamic weight, the third dynamic weight, the standard deviation of the probability of adequate anesthesia, the baseline value of the individual heart rate variability coefficient, the real-time mean arterial pressure of ideal anesthesia, the real-time mean arterial pressure, and the heart rate variability coefficient.

[0008] According to the present invention, determining a first dynamic weight, a second dynamic weight, and a third dynamic weight based on the real-time heart rate and the real-time mean arterial pressure includes: Obtain the first basic weight, the second basic weight, and the third basic weight; Obtain the lower limit of safe mean arterial pressure and the lower limit of dangerous mean arterial pressure; Get the safe upper heart rate limit and the dangerous upper heart rate limit; Based on the first basic weight, the second basic weight, the third basic weight, the real-time heart rate, the real-time mean arterial pressure, the lower limit of safe mean arterial pressure, the lower limit of dangerous mean arterial pressure, the upper limit of safe heart rate, and the upper limit of dangerous heart rate, the first dynamic weight, the second dynamic weight, and the third dynamic weight are determined.

[0009] According to one embodiment of the present invention, determining a first dynamic weight, a second dynamic weight, and a third dynamic weight based on a first basic weight, a second basic weight, a third basic weight, a real-time heart rate, a real-time mean arterial pressure, a lower limit of safe mean arterial pressure, a lower limit of dangerous mean arterial pressure, an upper limit of safe heart rate, and an upper limit of dangerous heart rate includes: according to the formula: Determine the first dynamic weight at time i of the monitoring period. Second dynamic weight and the third dynamic weight ,in, The first dynamic transition weight is the weight at the i-th time point of the monitoring period. The second dynamic transition weight is the weight at the i-th time point of the monitoring period. The third dynamic transition weight is the value at time i of the monitoring period. To preset the minimum first dynamic transition weight threshold, , and Here, is the preset coefficient, and max is the function to find the maximum value. As the first basic weight, As the second basic weight, As the third basic weight, To ensure a safe lower limit of mean arterial pressure, To measure the real-time mean arterial pressure at time i of the monitoring cycle, This is the lower limit of the dangerous mean arterial pressure. To ensure a safe upper limit of heart rate, The upper limit of the dangerous heart rate, This represents the real-time heart rate at the i-th moment of the monitoring cycle.

[0010] According to the present invention, a comprehensive stability coefficient of anesthetic state is determined based on the first dynamic weight, the second dynamic weight, the third dynamic weight, the standard deviation of the probability of adequate anesthesia, the baseline value of the individual heart rate variability coefficient, the real-time mean arterial pressure of ideal anesthesia, the real-time mean arterial pressure, and the heart rate variability coefficient, including: according to the formula: Determine the comprehensive stability coefficient of the anesthesia state at time i of the monitoring period. Where min is the function for finding the minimum value, and max is the function for finding the maximum value. The first dynamic weight at the i-th moment of the monitoring period, The second dynamic weight is the weight at the i-th time point of the monitoring period. The third dynamic weight is the value at the i-th time point of the monitoring period. Let $\frac{i}{i}$ be the standard deviation of the probability of adequate anesthesia within a preset time window at time $i$ during the monitoring period. Let be the heart rate variability coefficient at time i in the monitoring cycle. This is the baseline value for the individual's heart rate variability coefficient. To prevent the coefficient of variability from deviating from a preset threshold, To measure the real-time mean arterial pressure at time i of the monitoring cycle, The mean arterial pressure at real time is the ideal anesthesia level. This is a preset threshold for pulsating pressure deviation.

[0011] According to the present invention, the anesthesia depth instability fluctuation coefficient is determined based on the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification result, including: Based on the classification results of the anesthesia stages, the probability of deep anesthesia and the probability of awakening from light anesthesia are determined. Based on the probability of the awakening phase of light anesthesia, the standard deviation of the probability of deep anesthesia and the standard deviation of the probability of the awakening phase of light anesthesia within a preset time window are determined. The anesthesia depth instability fluctuation coefficient is determined based on the probability of deep anesthesia, the probability of awakening during light anesthesia, the standard deviation of the probability of deep anesthesia, the standard deviation of the probability of awakening during light anesthesia, the real-time heart rate, and the real-time mean arterial pressure.

[0012] According to the present invention, the anesthesia depth instability fluctuation coefficient is determined based on the probability of deep anesthesia, the probability of light anesthesia awakening stage, the standard deviation of the probability of deep anesthesia, the standard deviation of the probability of light anesthesia awakening stage, the real-time heart rate, and the real-time mean arterial pressure, including: According to the formula Determine the anesthesia depth instability fluctuation coefficient at time i of the monitoring period. ,in, , , , and Here, is the preset coefficient, and max is the function to find the maximum value. Let the sum of the probabilities of the light anesthesia awakening phase at time i of the monitoring period be denoted as . Let the sum of the probabilities of the light anesthesia awakening phase at time i-1 of the monitoring cycle be denoted as . For the time point corresponding to the i-th moment of the monitoring period, This refers to the time point corresponding to the (i-1)th moment of the monitoring period. To measure the probability and standard deviation of the light anesthesia awakening phase within a preset time window at time i of the monitoring period. To determine the probability of deep anesthesia at time i in the monitoring cycle. To determine the probability of deep anesthesia at time i-1 of the monitoring period, Let be the standard deviation of the probability of deep anesthesia within a preset time window at the i-th moment of the monitoring period. Let be the heart rate variability coefficient at time i in the monitoring cycle. This is the baseline value for the individual's heart rate variability coefficient. To prevent the coefficient of variability from deviating from a preset threshold, To measure the real-time mean arterial pressure at time i of the monitoring cycle, The mean arterial pressure at real time is the ideal anesthesia level. This is a preset threshold for pulsating pressure deviation.

[0013] According to a second aspect of the present invention, an anesthesia state monitoring system based on multi-source physiological signal analysis is provided, comprising: The data acquisition module is used to collect the patient's electroencephalogram (EEG) signals, real-time heart rate, heart rate variability coefficient, and real-time mean arterial pressure at multiple points during the monitoring cycle while the patient is under anesthesia. The signal processing module is used to initially filter out EEG signals using high-pass and low-pass filters, and to perform amplitude normalization processing on the EEG signals to obtain new EEG signals. The mode decomposition module is used to perform variational mode decomposition on new EEG signals using a hierarchical frequency decomposition method to obtain several mode functions; The feature extraction module is used to extract multidimensional features from each mode function; The feature set module is used to fuse and filter the extracted multidimensional features to form a feature set; The classification results module is used to train the machine learning model using the feature set, and to use the trained machine learning model to predict the patient's anesthesia status in real time, outputting the classification results of the anesthesia stage. The stability coefficient module is used to determine the comprehensive stability coefficient of the anesthesia state based on the real-time heart rate, the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results. The fluctuation coefficient module is used to determine the anesthesia depth instability fluctuation coefficient based on the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results. The anesthesia monitoring module is used to monitor the anesthesia status based on the comprehensive stability coefficient of the anesthesia status and the unstable fluctuation coefficient of the depth of anesthesia.

[0014] According to a third aspect of the present invention, an anesthesia state monitoring device based on multi-source physiological signal analysis is provided. The device includes a memory storing a computer program and a processor for executing the computer program. When the computer program is executed by the processor, it implements the steps of the anesthesia state monitoring method based on multi-source physiological signal analysis.

[0015] Technical Effects: According to the present invention, anesthesia status can be predicted based on the processed new EEG signals, and anesthesia stage classification results can be obtained. Based on the anesthesia stage classification results, real-time heart rate, heart rate variability coefficient, and real-time mean arterial pressure, the current stable anesthesia status and the risk development trend under unstable status can be assessed, and the comprehensive stability coefficient of anesthesia status and the instability fluctuation coefficient of anesthesia depth can be determined. Furthermore, based on the comprehensive stability coefficient of anesthesia status and the instability fluctuation coefficient of anesthesia depth, anesthesia status monitoring can be performed, improving the accuracy of anesthesia status monitoring. When determining the first dynamic weight, the second dynamic weight, and the third dynamic weight, the first basic weight, the second basic weight, the third basic weight, real-time heart rate, real-time mean arterial pressure, lower limit of safe mean arterial pressure, lower limit of dangerous mean arterial pressure, upper limit of safe heart rate, and upper limit of dangerous heart rate can be used to determine the first dynamic weight, the second dynamic weight, and the third dynamic weight. During the calculation process, the first dynamic weight, the second dynamic weight, and the third dynamic weight can be set according to the risk status of tachycardia and the risk status of low mean arterial pressure, respectively, for the three aspects of EEG signals, mean arterial pressure, and heart rate, improving the accuracy of the first dynamic weight, the second dynamic weight, and the third dynamic weight. When determining the comprehensive stability coefficient of the anesthetic state, it can be determined based on the first dynamic weight, the second dynamic weight, the third dynamic weight, the standard deviation of the probability of moderate anesthesia, the baseline value of the individual heart rate variability coefficient, the real-time mean arterial pressure under ideal anesthesia, and the real-time mean arterial pressure and heart rate variability coefficient. During the calculation process, the contribution of anesthesia depth to stability and the contribution of vital signs to safety can be fully analyzed, improving the comprehensiveness and accuracy of the comprehensive stability coefficient of the anesthetic state. When determining the anesthesia depth instability fluctuation coefficient, it can be determined based on the probability of deep anesthesia, the probability of light anesthesia awakening, the standard deviation of the probability of deep anesthesia, the standard deviation of the probability of light anesthesia awakening, real-time heart rate, and real-time mean arterial pressure. During the calculation process, the risk of EEG trends can be fully analyzed, and the risk amplification factor can be determined based on the degree of matching between physiological deviations and EEG trends. Furthermore, based on the EEG trend risk and the risk amplification factor, the anesthesia depth instability fluctuation coefficient is determined, improving the comprehensiveness of the anesthesia depth instability fluctuation coefficient.

[0016] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Other features and aspects of the invention will become clearer from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort. Figure 1 A schematic flowchart of an anesthesia state monitoring method based on multi-source physiological signal analysis according to an embodiment of the present invention is shown as an example. Figure 2 An exemplary schematic diagram illustrating the acquisition of new electroencephalogram (EEG) signals according to an embodiment of the present invention is shown; Figure 3 A schematic diagram illustrating the determination of the comprehensive stability coefficient of the anesthetic state according to an embodiment of the present invention is shown; Figure 4 A schematic diagram illustrating the determination of the unstable fluctuation coefficient of anesthesia depth according to an embodiment of the present invention is shown; Figure 5 A block diagram of an anesthesia status monitoring system based on multi-source physiological signal analysis according to an embodiment of the present invention is shown as an example. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0020] Figure 1 An exemplary flowchart illustrates a method for monitoring anesthesia status based on multi-source physiological signal analysis according to an embodiment of the present invention, the method comprising: Step S1: At multiple points during the monitoring cycle, collect the patient's electroencephalogram (EEG) signals, real-time heart rate, heart rate variability coefficient, and real-time mean arterial pressure while the patient is under anesthesia. Step S2: Use high-pass and low-pass filters to initially filter out EEG signals, and perform amplitude normalization on the EEG signals to obtain new EEG signals; Step S3: Use the hierarchical frequency decomposition method to perform variational mode decomposition on the new EEG signal to obtain several mode functions; Step S4: Extract multidimensional features from each modal function; Step S5: The extracted multidimensional features are fused and filtered to form a feature set; Step S6: Train the machine learning model using the feature set, and use the trained machine learning model to predict the patient's anesthesia status in real time, and output the anesthesia stage classification results. Step S7: Determine the comprehensive stability coefficient of the anesthesia state based on the real-time heart rate, the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results; Step S8: Determine the anesthesia depth instability fluctuation coefficient based on the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results; Step S9: Monitor the anesthesia status based on the comprehensive stability coefficient of the anesthesia status and the unstable fluctuation coefficient of the depth of anesthesia.

[0021] The anesthesia state monitoring method based on multi-source physiological signal analysis according to embodiments of the present invention can predict the anesthesia state based on the processed new electroencephalogram (EEG) signals, obtain the anesthesia stage classification results, and assess the current stable anesthesia status and the risk development trend under unstable status based on the anesthesia stage classification results, real-time heart rate, heart rate variability coefficient, and real-time mean arterial pressure, thereby determining the comprehensive stability coefficient of the anesthesia state and the anesthesia depth instability fluctuation coefficient. Furthermore, the anesthesia state monitoring is performed based on the comprehensive stability coefficient of the anesthesia state and the anesthesia depth instability fluctuation coefficient, thus improving the accuracy of anesthesia state monitoring.

[0022] According to one embodiment of the present invention, in step S1, at multiple times during the monitoring cycle, the patient's electroencephalogram (EEG) signal, real-time heart rate, heart rate variability coefficient, and real-time mean arterial pressure are collected while the patient is under anesthesia.

[0023] For example, a non-invasive method can be used to collect electroencephalogram (EEG) signals from patients under anesthesia. This involves placing electrodes on the patient's scalp to record the signals. A multi-electrode system is used to collect EEG signals. The electrodes are fixed with standard electrode pads. The sampling frequency for collecting EEG data is 256 Hz. The system consists of four electrode pads, which are placed on the patient's forehead area. Real-time heart rate, heart rate variability coefficient, and real-time mean arterial pressure are obtained from the patient under anesthesia using specialized monitoring equipment.

[0024] According to one embodiment of the present invention, in step S2, a high-pass filter and a low-pass filter are used to initially filter out the EEG signal, and the amplitude of the EEG signal is normalized to obtain a new EEG signal.

[0025] Figure 2 An exemplary schematic diagram of obtaining new electroencephalogram (EEG) signals according to an embodiment of the present invention is shown.

[0026] According to an embodiment of the present invention, step S2 includes: Step S21: Use a high-pass filter to remove baseline drift signals from the EEG signal; Step S22: Use a low-pass filter to remove high-frequency interference from the EEG signal; Step S23: The EEG signal after removing baseline drift signal and high frequency interference is subjected to amplitude normalization processing to obtain a new EEG signal.

[0027] For example, baseline drift is a low-frequency interference that increases the amplitude of EEG signals. Baseline drift is mainly caused by the patient's breathing, movement, and electrode movement. The patient's breathing causes fluctuations in their body, resulting in unavoidable baseline drift. Additionally, surgical procedures may require the surgeon to move the patient, which can also cause significant baseline drift. Furthermore, due to the lengthy surgery and complex operating room environment, external factors such as patient sweating, decreased electrode adhesiveness, and accidental contact with the leads can cause the electrodes used to collect EEG signals to drift. A certain degree of positional change can cause baseline drift. Therefore, a 0.5Hz high-pass filter is used to remove baseline drift signals from the EEG signal, and a 50Hz low-pass filter is used to remove high-frequency interference. The combination of high-pass and low-pass filters ensures that the filtered signal is mainly retained in the 0.5Hz to 50Hz frequency band, which is also the main frequency range of EEG activity. To further reduce amplitude variations caused by differences in electrode contact or other environmental factors during signal acquisition, amplitude normalization is performed on the signal. The amplitude normalization process adjusts the signal amplitudes of different channels to the same range, ensuring the comparability of signals between different electrodes. The normalized signal amplitude is standardized to the 0 to 1 interval, ensuring the accuracy of subsequent decomposition processes.

[0028] According to an embodiment of the present invention, in step S3, a variational mode decomposition method is used to perform variational mode decomposition on the new EEG signal to obtain several mode functions.

[0029] For example, the specific process of variational mode decomposition of a new EEG signal using the hierarchical frequency decomposition method is as follows: 1. Perform a first-level decomposition on the new EEG signal to obtain k first-order mode functions; 2. Perform a second-level decomposition on the k first-order mode functions to obtain k×L second-order mode functions, where L represents the decomposition of each first-order mode function into l second-order mode functions. Update the frequency distribution of the k×L second-order mode functions and calculate the frequency characteristics and amplitude characteristics of the k×L second-order mode functions; 3. Using the first-order mode function... The frequency is used as the reference frequency for baseline drift signal screening. From k-1 first-order mode functions, p first-order low-frequency mode functions containing the reference frequency are selected. The p×L second-order mode functions following the selected p first-order low-frequency mode functions are all decomposed into third-order mode functions using a third-level decomposition. The frequency distribution of the third-order mode functions is updated, and the frequency and amplitude characteristics of the third-order mode functions are calculated. 4. Introduce the baseline drift signal standard frequency. From the p×L second-order mode functions, select the standard frequency containing the baseline drift signal... The second-order low-frequency mode function is obtained, and the third-order mode function below the selected second-order low-frequency mode function is decomposed into a fourth-level decomposition to obtain a fourth-order mode function. The frequency distribution of the fourth-order mode function is updated, and the frequency characteristics and amplitude characteristics of the fourth-order mode function are calculated. 5. Continue to select the third-order low-frequency mode function containing the standard frequency of the baseline drift signal from the third-order mode function, and perform the next level decomposition on the fourth-order mode function below the selected third-order low-frequency mode function. Repeat step 4 above until the nth-order low-frequency mode function is obtained. If none of the (n+1)th order mode functions in the decomposition of the (n+1)th layer contain the standard frequency of the baseline drift signal, then the decomposition stops. The frequency distribution of the (n+1)th order mode functions is updated, and the frequency and amplitude characteristics of the (n+1)th order mode functions are calculated. 6. The first-order mode function, the first-order low-frequency mode function, the second-order low-frequency mode function, the third-order low-frequency mode function, and the nth-order low-frequency mode function of the first layer decomposition are selected as candidate modes for baseline drift signal reconstruction. 7. Baseline drift signal reconstruction is performed using the candidate modes, as shown in the following formula: ,in, Represents signal time. This represents the baseline drift signal obtained from the reconstruction. These are the low-frequency mode functions of each order after hierarchical decomposition. This represents the total number of low-frequency mode functions. By superimposing these low-frequency mode functions, a complete baseline drift signal can be obtained. 8. Subtract the reconstructed baseline drift signal from the new EEG signal to obtain the purified EEG signal, as shown in the following formula: ,in, This indicates a new EEG signal. In the purified signal, baseline drift and other low-frequency interferences have been removed, preserving the patient's actual EEG activity. Based on the above steps, each low-frequency mode function is further decomposed into several multi-mode functions during the hierarchical decomposition process. As the hierarchical decomposition proceeds, the frequency components of each multi-mode function gradually decrease, and the amplitude characteristics become more defined. This allows for precise identification of which multi-mode functions are the main components of baseline drift and which are other interference signals. After hierarchical decomposition, the frequency distribution and amplitude characteristics of each multi-mode function need to be analyzed. Generally, during prolonged surgery, the amplitude of baseline drift often increases significantly due to the patient's breathing, changes in body position, and electrode movement. Therefore, in the sub-modes after hierarchical decomposition, we select those modes with larger amplitudes as candidate modes for baseline drift. Furthermore, baseline drift is a very low-frequency signal, typically below 1 Hz, and may even be below 0.1 Hz. The purpose of hierarchical decomposition is to further isolate these low-frequency components in order to accurately identify baseline drift. Through hierarchical decomposition and mode optimization, the extraction accuracy of the baseline drift signal is significantly improved. The multi-level processing of hierarchical decomposition ensures that low-frequency interference signals are gradually stripped away, while the optimized mode selection ensures that only those low-frequency modes highly correlated with baseline drift are retained.

[0030] According to one embodiment of the present invention, in step S4, multidimensional features are extracted from each modal function.

[0031] For example, modal functions include first-order, second-order, third-order, fourth-order, and n+1-order modal functions. Multidimensional features include: the center frequency and bandwidth of each modal function, reflecting the main frequency characteristics of the new EEG signal; the frequency domain energy distribution and time-frequency energy matrix of each modal function, used to quantify the time-frequency characteristics of the new EEG signal; higher-order statistics of each modal function, including kurtosis and skewness, used to assess the nonlinearity and complexity of the new EEG signal; and the coherence and frequency overlap ratio between modal functions, used to describe the information correlation and independence between modal functions. In the feature extraction and calculation stage, the intermediate calculation results and modal features obtained during the decomposition process are fully utilized to construct a comprehensive and effective feature set. First, based on the modal functions obtained from each level of decomposition, the center frequency, bandwidth, and frequency domain energy distribution of each multi-order modal function are extracted. Bandwidth and frequency domain energy distribution are conventional methods and will not be elaborated here. The `mne` package in Python contains tools for calculating the frequency domain energy distribution, so their calculation will not be described in detail here.

[0032] According to one embodiment of the present invention, in step S5, the extracted multidimensional features are fused and filtered to form a feature set.

[0033] For example, feature fusion and selection employs a highly innovative feature engineering method, combining advanced mathematical models and feature fusion algorithms from the machine learning field to construct a more accurate and adaptable feature set. Unlike traditional recursive feature elimination (RFE), this step uses a multi-level feature fusion strategy, combined with automated feature generation technology and deep feature representation methods, to ensure that the extracted features can capture the dynamic changes of the anesthesia state to the greatest extent, while improving the model's generalization ability and robustness. The specific process of fusing and selecting the extracted multidimensional features is as follows: 1. Using nonlinear feature mapping technology to map the extracted multidimensional features to high-dimensional features. The invention employs a multi-dimensional feature learning method to capture the complex nonlinear relationships between multi-dimensional features. This involves: 1) using a sparse coding algorithm to sparsely represent multi-dimensional features in a high-dimensional feature space; 2) using an adaptive feature weighting method to dynamically learn the contribution of each multi-dimensional feature under different anesthesia states and adjust the weights of each feature; 3) introducing a multi-view feature learning method to construct feature subspaces for each modality function and jointly training these subspaces using a co-training algorithm to mine the implicit information between different modality functions; and 4) obtaining a fused feature set after the above four steps and using an ensemble feature selection algorithm to filter the fused feature set. Furthermore, to further improve the efficiency of feature fusion, this invention introduces a multi-view feature learning method. This method treats the different frequency modalities obtained from decomposition as multi-view information, constructing a feature subspace for each modality. Then, it jointly trains these subspaces using a co-training algorithm. The core idea of ​​co-training is that feature subspaces under different modalities can complement each other, improving the model's generalization ability. Multi-view feature learning, by fusing features from different modalities, can fully explore the implicit cross-modal correlations in EEG signals. Finally, an ensemble feature selection algorithm is used to filter the fused feature set. This algorithm combines multiple feature selection methods, including recursive feature elimination and Lasso regression, and ultimately determines the feature set with the highest diagnostic value through weighted voting. The ultimate goal of ensemble feature selection is to improve the stability of the model and the reliability of the selection results by combining different selectors. Through the above feature fusion method, the entire process from high-dimensional nonlinear mapping, sparse coding, weighted feature fusion, multi-view learning to ensemble feature selection is optimized, significantly improving the monitoring accuracy of anesthesia and the model's generalization ability.

[0034] According to an embodiment of the present invention, in step S6, a machine learning model is trained using a feature set, and the trained machine learning model is used to predict the patient's anesthesia status in real time, and output the anesthesia stage classification result.

[0035] For example, this embodiment uses multiple machine learning models to classify and predict anesthesia states. Comparative experiments are conducted using models such as Support Vector Machines (SVMs), Random Forests, and Neural Networks to find the optimal classification scheme for anesthesia states. SVMs, by constructing a hyperplane that maximizes the classification boundary, can accurately classify EEG signals at different anesthesia stages. Random Forests, through ensemble learning, combine the results of multiple decision trees to improve classification stability and anti-interference capabilities. Neural Networks, through a multilayer perceptron structure, perform deep learning on extracted complex nonlinear features, capturing the nonlinear dynamic changes at different anesthesia stages. Finally, the machine learning model is trained and classified based on the input feature set, outputting prediction results for four different stages: awake, light anesthesia, moderate anesthesia, and deep anesthesia—that is, the anesthesia stage classification results (four discrete probabilities, e.g., 90% in moderate anesthesia).

[0036] According to an embodiment of the present invention, in step S7, a comprehensive stability coefficient of the anesthesia state is determined based on the real-time heart rate, the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification result.

[0037] Figure 3 A schematic diagram illustrating the determination of the comprehensive stability coefficient of the anesthetic state according to an embodiment of the present invention is shown.

[0038] According to an embodiment of the present invention, step S7 includes: Step S71: Determine the first dynamic weight, the second dynamic weight, and the third dynamic weight based on the real-time heart rate and the real-time mean arterial pressure. Step S72: Determine the target anesthesia stage; Step S73: Determine the probability of adequate anesthesia based on the target anesthesia stage and the anesthesia stage classification results; Step S74: Determine the standard deviation of the appropriate anesthesia probability within a preset time window based on the appropriate anesthesia probability. Step S75: Obtain the baseline value of the individual heart rate variability coefficient and the ideal real-time mean arterial pressure under anesthesia; Step S76: Determine the comprehensive stability coefficient of the anesthesia state based on the first dynamic weight, the second dynamic weight, the third dynamic weight, the standard deviation of the probability of adequate anesthesia, the baseline value of the individual heart rate variability coefficient, the real-time mean arterial pressure of ideal anesthesia, the real-time mean arterial pressure, and the heart rate variability coefficient.

[0039] For example, dynamic weights are set for three aspects—EEG signal, blood pressure, and heart rate—based on real-time heart rate and real-time mean arterial pressure, namely, the first dynamic weight, the second dynamic weight, and the third dynamic weight. The target anesthesia stage represents the anesthesia stage that the patient needs to be in for the current surgery. Based on the target anesthesia stage and the anesthesia stage classification results, the probability of adequate anesthesia is determined. For example, when the target anesthesia stage is moderate anesthesia, the probability of adequate anesthesia is set to the probability of moderate anesthesia in the anesthesia stage classification results. Based on the probability of adequate anesthesia, the standard deviation of the probability of adequate anesthesia within a preset time window is determined. The length of the preset time window is dynamically set: a short window (e.g., 10 seconds) is used during the induction / emergence phase because the anesthetic state changes rapidly and requires high sensitivity; a long window (e.g., 60 seconds) is used during the maintenance phase because stable monitoring is required and noise must be avoided. To mitigate acoustic interference, the window is automatically shortened (e.g., 2 seconds) in high-risk conditions (e.g., severe blood pressure fluctuations, abnormal EEG trends) to accelerate response. The data sampling interval is set to 0.5 seconds, and the standard deviation of the appropriate anesthesia probability is calculated for multiple appropriate anesthesia probabilities within the preset time window. The baseline value of the individual heart rate variability coefficient and the ideal real-time mean arterial pressure during anesthesia are statistically obtained based on historical data from populations similar to the patient's physical condition and surgical type, representing the ideal values ​​of the individual heart rate variability coefficient and real-time mean arterial pressure during anesthesia. The stability of anesthesia is assessed and a comprehensive stability coefficient of the anesthesia state is determined based on the first dynamic weight, the second dynamic weight, the third dynamic weight, the standard deviation of the appropriate anesthesia probability, the baseline value of the individual heart rate variability coefficient, the ideal real-time mean arterial pressure during anesthesia, the real-time mean arterial pressure, and the heart rate variability coefficient.

[0040] According to an embodiment of the present invention, step S71 includes: Step S711: Obtain the first basic weight, the second basic weight, and the third basic weight; Step S712: Obtain the lower limit of safe mean arterial pressure and the lower limit of dangerous mean arterial pressure; Step S713: Obtain the safe heart rate limit and the dangerous heart rate limit; Step S714: Determine the first dynamic weight, the second dynamic weight, and the third dynamic weight based on the first basic weight, the second basic weight, the third basic weight, the real-time heart rate, the real-time mean arterial pressure, the lower limit of safe mean arterial pressure, the lower limit of dangerous mean arterial pressure, the upper limit of safe heart rate, and the upper limit of dangerous heart rate.

[0041] For example, the first, second, and third baseline weights can be set to 0.5, 0.3, and 0.2, respectively. The first baseline weight is the EEG baseline weight. When there is no serious physiological risk, 50% of the decision weight is allocated to the EEG signal by default, as it is a direct and core indicator of anesthesia depth; that is, the first baseline weight is set to 0.5. When there is no risk of hypotension, 30% of the decision weight is allocated to real-time mean arterial pressure by default, because real-time mean arterial pressure is a key indicator of circulatory perfusion (brain perfusion depends on stable blood pressure); that is, the second baseline weight is set to 0.3. When there is no risk of tachycardia, 20% of the decision weight is allocated to heart rate by default, because heart rate is a voluntary... The "sensitive indicators" of neural balance (indirectly reflecting the anesthesia state) are determined by setting the third baseline weight to 0.2; obtaining the lower limit of safe mean arterial pressure (e.g., 65 mmHg) and the lower limit of dangerous mean arterial pressure (e.g., 50 mmHg); obtaining the upper limit of safe heart rate (e.g., 100 bpm) and the upper limit of dangerous heart rate (e.g., 120 bpm); and setting dynamic weights corresponding to EEG signals, arterial pressure, and heart rate based on the first baseline weight, the second baseline weight, the third baseline weight, real-time heart rate, real-time mean arterial pressure, lower limit of safe mean arterial pressure, lower limit of dangerous mean arterial pressure, upper limit of safe heart rate, and upper limit of dangerous heart rate, respectively. That is, determining the first dynamic weight, the second dynamic weight, and the third dynamic weight.

[0042] According to an embodiment of the present invention, step S714 includes: determining the first dynamic weight at the i-th moment of the monitoring period according to formula (1). Second dynamic weight and the third dynamic weight , (1) in, The first dynamic transition weight is the weight at the i-th time point of the monitoring period. The second dynamic transition weight is the weight at the i-th time point of the monitoring period. The third dynamic transition weight is the value at time i of the monitoring period. To preset the minimum first dynamic transition weight threshold, , and Here, is the preset coefficient, and max is the function to find the maximum value. As the first basic weight, As the second basic weight, As the third basic weight, To ensure a safe lower limit of mean arterial pressure, To measure the real-time mean arterial pressure at time i of the monitoring cycle, This is the lower limit of the dangerous mean arterial pressure. To ensure a safe upper limit of heart rate, The upper limit of the dangerous heart rate, This represents the real-time heart rate at the i-th moment of the monitoring cycle.

[0043] According to one embodiment of the present invention, It is the ratio of the difference between the lower limit of safe mean arterial pressure and the real-time mean arterial pressure to the lower limit of safe mean arterial pressure and the lower limit of dangerous mean arterial pressure. Indicates taking 0 and The maximum value of the above-mentioned maximum value can be used to represent a low mean arterial pressure risk condition when the real-time mean arterial pressure is normal (greater than the lower limit of safe mean arterial pressure). The value is 0 when the real-time mean arterial pressure is too low (close to the lower limit of the danger mean arterial pressure). The value is close to 1, and similarly, It can be used to indicate the risk of tachycardia when the real-time heart rate is normal (less than the upper limit of the safe heart rate). The value is 0 when the heart rate is abnormal (close to the upper limit of the dangerous heart rate). The value is close to 1.

[0044] According to one embodiment of the present invention, As the primary weight, this indicates that in the absence of risk, 50% of the decision-making weight is allocated to EEG by default to ensure priority is given to capturing "too superficial / too deep" anesthesia states. This indicates that a first dynamic transition weight is determined based on the first baseline weight, the risk of low mean arterial pressure, and the risk of tachycardia. When the risk of low mean arterial pressure or tachycardia increases, the first dynamic transition weight decreases. This means that when physiological risks (risk of low mean arterial pressure and risk of tachycardia) occur, the absolute dominance of EEG is temporarily reduced, and more attention is paid to the safety of vital signs (blood pressure, heart rate). This is a preset coefficient used to control the degree to which physiological risks weaken brain signals. It can be set to 0.1. During anesthesia, the risk of low mean arterial pressure and tachycardia is generally moderate. Setting it to 0.1 means when When the sum is 1, A reduction of only 0.1 ensures both risk response and maintains the core position of EEG. The first dynamic transition weight at the i-th moment of the monitoring cycle needs to be greater than or equal to the preset minimum first dynamic transition weight threshold (which can be set to 0.3). The above processing can ensure that EEG monitoring is not completely ignored. That is, even if the risk is extremely high, at least 30% of the weight should still be reserved for EEG to prevent over-reliance on vital signs and loss of anesthesia depth information.

[0045] According to one embodiment of the present invention, As the second basic weight, This indicates that a second dynamic transition weight is determined based on the second baseline weight and the risk status of low mean arterial pressure. This means that as the risk status of low mean arterial pressure increases, the second dynamic transition weight increases linearly; that is, the higher the mean arterial pressure risk, the more focused the blood pressure monitoring, and the faster the response to hypotension (e.g., adjusting anesthetics, fluid resuscitation, vasopressors). This is a preset coefficient used to adjust the degree to which the risk of low mean arterial pressure is amplified in relation to the second dynamic transition weight. It can be set to 0.2. During anesthesia, the risk threshold for low mean arterial pressure typically corresponds to a 10% to 20% decrease in mean arterial pressure from baseline. Setting it to 0.2 means that when When the value is equal to 1, the second dynamic transition weight is increased to 0.5. Assuming that the first dynamic transition weight is 0.4 at this time, the first dynamic transition weight is less than or equal to the second dynamic transition weight, that is, the EEG weight is less than or equal to the mean arterial pressure weight, to ensure that circulatory safety is prioritized.

[0046] According to one embodiment of the present invention, As the third basic weight, This indicates that a third dynamic transition weight is determined based on the third baseline weight and the risk of tachycardia. As the risk of tachycardia increases, the third dynamic transition weight increases linearly. In other words, the higher the risk of tachycardia, the more focused the heart rate monitoring becomes, and the faster the identification of the triggers for tachycardia (e.g., pain, hypoxia, volume insufficiency, etc.). For preset coefficients, The magnitude of the increase in the third dynamic transition weight used to control the risk of tachycardia, and... The setting logic is similar. It can be set to 0.2 to ensure that when When the value is 1, priority should be given to heart rate-related stress and analgesia issues.

[0047] According to one embodiment of the present invention, , and This indicates that the first dynamic transition weight, the second dynamic transition weight, and the third dynamic transition weight are normalized to determine the first dynamic weight. Second dynamic weight and the third dynamic weight The above normalization settings can make the first dynamic weight Second dynamic weight and the third dynamic weight The sum is 1.

[0048] In this way, the first, second, and third dynamic weights can be determined based on the first, second, and third basic weights, the third basic weights, the real-time heart rate, the real-time mean arterial pressure, the lower limit of safe mean arterial pressure, the lower limit of dangerous mean arterial pressure, the upper limit of safe heart rate, and the upper limit of dangerous heart rate. During the calculation process, the first, second, and third dynamic weights can be set for the three aspects of EEG signal, mean arterial pressure, and heart rate, respectively, based on the risk status of tachycardia and the risk status of low mean arterial pressure, thereby improving the accuracy of the first, second, and third dynamic weights.

[0049] According to an embodiment of the present invention, step S76 includes: determining the comprehensive stability coefficient of the anesthesia state at the i-th moment of the monitoring period according to formula (2). , Where min is the function for finding the minimum value, and max is the function for finding the maximum value. The first dynamic weight at the i-th moment of the monitoring period, The second dynamic weight is the weight at the i-th time point of the monitoring period. The third dynamic weight is the value at the i-th time point of the monitoring period. Let $\frac{i}{i}$ be the standard deviation of the probability of adequate anesthesia within a preset time window at time $i$ during the monitoring period. Let be the heart rate variability coefficient at time i in the monitoring cycle. This is the baseline value for the individual's heart rate variability coefficient. To prevent the coefficient of variability from deviating from a preset threshold, To measure the real-time mean arterial pressure at time i of the monitoring cycle, The mean arterial pressure at real time is the ideal anesthesia level. This is a preset threshold for pulsating pressure deviation.

[0050] According to one embodiment of the present invention, The first dynamic weight at the i-th moment of the monitoring cycle represents the priority of anesthesia depth monitoring. The second dynamic weight is the weight at the i-th time point of the monitoring period. The third dynamic weight is the value at the i-th time point of the monitoring period. and This indicates the priority of vital sign safety monitoring.

[0051] According to one embodiment of the present invention, Let be the standard deviation of the probability of adequate anesthesia within a preset time window at time i of the monitoring period. A value close to 1 indicates that the probability of adequate anesthesia fluctuates little within the preset time window, and the depth of anesthesia is very stable. A smaller value indicates a large fluctuation in the probability of adequate anesthesia within the preset time window, and an unstable EEG state (e.g., alternating between too light and too deep anesthesia). This indicates the degree to which the stability of the depth of anesthesia is contributed.

[0052] According to one embodiment of the present invention, This is the ratio of the difference between the real-time mean arterial pressure at time i of the monitoring cycle and the ideal real-time mean arterial pressure under anesthesia, to a preset pulsation pressure deviation threshold. The preset pulsation pressure deviation threshold can be set to half the range of the ideal mean arterial pressure during anesthesia. This indicates the ideal level of real-time mean arterial pressure during anesthesia. When the mean arterial pressure is close to 0, it indicates that the real-time mean arterial pressure is within the ideal range and the circulatory perfusion is stable. A reading close to 1 indicates that the real-time mean arterial pressure deviates significantly from the ideal range, suggesting insufficient blood supply to the brain or a stress response. Similarly, This indicates the ideal degree of the individual heart rate variability coefficient during anesthesia. When the heart rate is close to 0, the individual's heart rate variability is within the ideal range, and the autonomic nervous system is in balance. When the coefficient of variability is close to 1, the individual's heart rate variability deviates significantly from the ideal range, suggesting insufficient analgesia (sympathetic nerve excitation) or circulatory inhibition (parasympathetic nerve excitation). To prevent the coefficient of variability from deviating from a preset threshold, it can be set to half the range of the ideal individual heart rate variability during anesthesia. This indicates the contribution to the safety of vital signs.

[0053] In this way, the comprehensive stability coefficient of the anesthesia state can be determined based on the first dynamic weight, the second dynamic weight, the third dynamic weight, the standard deviation of the probability of adequate anesthesia, the baseline value of the individual heart rate variability coefficient, the real-time mean arterial pressure under ideal anesthesia, and the real-time mean arterial pressure and heart rate variability coefficient. During the calculation process, the degree of contribution of the depth of anesthesia to stability and the contribution of vital signs to safety can be fully analyzed, thereby improving the comprehensiveness and accuracy of the comprehensive stability coefficient of the anesthesia state.

[0054] According to one embodiment of the present invention, in step S8, the anesthesia depth instability fluctuation coefficient is determined based on the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification result.

[0055] Figure 4 A schematic diagram illustrating the determination of the unstable fluctuation coefficient of anesthesia depth according to an embodiment of the present invention is shown.

[0056] According to an embodiment of the present invention, step S8 includes: Step S81: Based on the classification results of the anesthesia stages, determine the probability of deep anesthesia and the probability of awakening from light anesthesia. Step S82: Based on the probability of the light anesthesia awakening stage, determine the standard deviation of the probability of deep anesthesia and the standard deviation of the probability of the light anesthesia awakening stage within a preset time window. Step S83: Determine the anesthesia depth instability fluctuation coefficient based on the deep anesthesia probability, the light anesthesia awakening stage probability, the standard deviation of the deep anesthesia probability, the standard deviation of the light anesthesia awakening stage probability, the real-time heart rate, and the real-time mean arterial pressure.

[0057] For example, based on the anesthesia stage classification results, the probability of being in the deep anesthesia stage is determined, i.e., the probability of deep anesthesia, and the sum of the probabilities of being in the light anesthesia stage and the awake stage, i.e., the probability of deep anesthesia and the probability of light anesthesia and the awake stage are summed. Based on the probability of light anesthesia and the awake stage, the standard deviation of the probability of deep anesthesia and the standard deviation of the probability of light anesthesia and the awake stage within a preset time window are determined. The calculation method of the standard deviation of the probability of deep anesthesia and the probability of light anesthesia and the awake stage is similar to that of moderate anesthesia, and will not be repeated here. Based on the probability of deep anesthesia, the sum of the probabilities of light anesthesia and the awake stage, the standard deviation of the probability of deep anesthesia, the standard deviation of the probability of light anesthesia and the awake stage, real-time heart rate, and real-time mean arterial pressure, the development trend and risk under unstable fluctuations in anesthesia depth are assessed, and the anesthesia depth instability fluctuation coefficient is determined. According to an embodiment of the present invention, step S83 includes: determining the anesthesia depth instability fluctuation coefficient at the i-th moment of the monitoring period according to formula (3). , in, , , , and Here, is the preset coefficient, and max is the function to find the maximum value. Let the sum of the probabilities of the light anesthesia awakening phase at time i of the monitoring period be denoted as . Let the sum of the probabilities of the light anesthesia awakening phase at time i-1 of the monitoring cycle be denoted as . For the time point corresponding to the i-th moment of the monitoring period, This refers to the time point corresponding to the (i-1)th moment of the monitoring period. To measure the probability and standard deviation of the light anesthesia awakening phase within a preset time window at time i of the monitoring period. To determine the probability of deep anesthesia at time i in the monitoring cycle. To determine the probability of deep anesthesia at time i-1 of the monitoring period, Let be the standard deviation of the probability of deep anesthesia within a preset time window at the i-th moment of the monitoring period. Let be the heart rate variability coefficient at time i in the monitoring cycle. This is the baseline value for the individual's heart rate variability coefficient. To prevent the coefficient of variability from deviating from a preset threshold, To measure the real-time mean arterial pressure at time i of the monitoring cycle, The mean arterial pressure at real time is the ideal anesthesia level. This is a preset threshold for pulsating pressure deviation.

[0058] According to one embodiment of the present invention, This indicates the rate of change in the probability of transitioning to light anesthesia or a state of full consciousness. A value greater than 0 indicates increased risk. Indicates taking 0 and The maximum value mentioned above, which is used to determine the maximum value, can be applied to situations where only a transition to light anesthesia or a state of full consciousness is required. To measure the probability and standard deviation of the light anesthesia awakening phase within a preset time window at time i of the monitoring period. The larger the value, the greater the fluctuation in the tendency to transition to light anesthesia or a state of full consciousness, and the higher the perceived risk. This indicates the risk of shifting towards light anesthesia; similarly, This indicates the risk of a shift towards deep anesthesia. and The preset coefficients represent the weighting coefficients for the risks of light and deep anesthesia, respectively. These coefficients can be determined based on the surgical stage. For example, during the incision phase (the moment the scalpel touches the skin and the first incision is completed, plus the following few minutes to ten minutes, which is the stage with the most intense surgical trauma stress response and one of the most critical periods in anesthesia management), more precautions are needed against light anesthesia. During the stable phase (the intermediate stage between the end of the incision phase and the beginning of the suturing phase, when the surgical procedure is relatively smooth, the patient is under continuous anesthesia maintenance, vital signs fluctuate less, and the required depth of anesthesia tends to be constant), more precautions are needed against deep anesthesia. and It can be initially set to 0.5, during the skin cutting period. It can be set to 0.7 during the stable period. It can be set to 0.7), and and The sum is 1.

[0059] This indicates a risk factor related to brainwave trends.

[0060] According to one embodiment of the present invention, The change in the probability of deep anesthesia is represented by the conditional function in formula (3). The value includes the following two cases: when the change in the probability of deep anesthesia is greater than 0, it indicates that the anesthesia trend is towards deeper anesthesia, and the value of the conditional function is 1; otherwise, it indicates that the anesthesia trend is towards shallower anesthesia, and the value of the conditional function is -1.

[0061] According to one embodiment of the present invention, It indicates the degree of normalization of the heart rate deviating from the ideal value (e.g., heart rate is too high / too low; the closer the value is to 1, the more severe the deviation). It indicates the degree of normalization of mean arterial pressure deviating from the ideal value (e.g., blood pressure is too high / too low; the closer the value is to 1, the more severe the deviation). and These are preset coefficients, representing the weights of heart rate and blood pressure, which can be determined according to the surgical stage (e.g., during the skin incision phase, more attention is paid to mean arterial pressure). Larger and It can be initially set to 0.5, during the surgical incision phase. It can be set to 0.7. It can be set to 0.3), and and The sum is 1.

[0062] According to one embodiment of the present invention, Indicates the risk amplification factor. This is a preset coefficient used to control the amplification of overall risk by physiological deviations. The settings can be based on different stages of anesthesia (e.g., skin incision / trauma stage: vital signs fluctuate greatly (e.g., sudden drop in blood pressure, increased heart rate), requiring strong amplification of physiological deviations). Take a larger value, such as 0.5 to 1.0, to ensure the system responds quickly to physiological crises. During the stable period (vital signs are stable), it is necessary to slightly amplify physiological deviations. Take smaller values, such as 0.1 to 0.3, to avoid over-interfering with subtle changes in EEG trends. This indicates the degree of match between physiological deviations and EEG trends, such as when... A value of 1 indicates that the EEG trend is shifting towards deeper anesthesia. The value is If physiological indicators (heart rate variability coefficient, mean arterial pressure) also deviate from ideal values, that is, if heart rate and mean pulse pressure are abnormal, then A value greater than 0 indicates a trend of EEG shift towards deep anesthesia, which correlates with the trend of physiological indicators deviating from ideal values ​​(e.g., EEG indicates "insufficient deep anesthesia," while blood pressure / heart rate is already abnormal; both point to "increased risk," and their directions are consistent). When the EEG trend (e.g., shift towards deep anesthesia) is consistent with the physiological deviation (e.g., abnormal blood pressure / heart rate), The absolute value is larger, indicating a greater risk amplification factor. This significantly increases the risk of EEG trend indications (e.g., if the EEG indicates "insufficient deep anesthesia" while blood pressure has already dropped, the system will consider it "extremely high risk," even if the EEG deviation is small). This is especially true when the EEG trend is inconsistent with the physiological deviation direction (or the degree of physiological deviation is low). The absolute value is smaller, and the risk amplification factor has a weaker risk amplification effect on the EEG trend (e.g., if the EEG indicates "insufficient deep anesthesia" but the blood pressure / heart rate is normal, the system will consider "low risk" and there is no need to over-adjust the depth of anesthesia).

[0063] According to one embodiment of the present invention, This indicates that the unstable fluctuation coefficient of anesthesia depth is determined based on EEG trend risk and risk amplification factor.

[0064] According to one embodiment of the present invention, when A value greater than 0 indicates that the overall risk points to the risk of excessive anesthesia. A value less than 0 indicates that the overall risk points to the risk of insufficient anesthesia. The larger the absolute value, the higher the risk level.

[0065] In this way, the anesthesia depth instability fluctuation coefficient can be determined based on the probability of deep anesthesia, the probability of light anesthesia awakening stage, the standard deviation of the probability of deep anesthesia, the standard deviation of the probability of light anesthesia awakening stage, real-time heart rate, and real-time mean arterial pressure. During the calculation process, the EEG trend risk can be fully analyzed, and the risk amplification factor can be determined based on the degree of matching between physiological deviation and EEG trend. Furthermore, the anesthesia depth instability fluctuation coefficient can be determined based on the EEG trend risk and the risk amplification factor, thus improving the comprehensiveness of the anesthesia depth instability fluctuation coefficient.

[0066] According to one embodiment of the present invention, in step S9, the anesthesia status is monitored based on the comprehensive stability coefficient of the anesthesia status and the unstable fluctuation coefficient of the anesthesia depth.

[0067] For example, if the overall stability coefficient of the anesthesia state is greater than or equal to 0.9, it indicates that the anesthesia state is very ideal and stable, the depth of brain electrical activity is stable, and vital signs such as heart rate and mean pulse pressure are all within the ideal safe range. The system's dynamic weighting of various indicators is usually at a basic level, with no significant risk signals. If the overall stability coefficient of the anesthesia status is between 0.7 and 0.9, the anesthesia status is within an acceptable range, but there may be slight fluctuations. This prompts the physician to pay attention to minor trends in individual indicators, but no immediate intervention is required. When the overall stability coefficient of the anesthesia status is less than 0.7, the stability of the anesthesia status decreases, indicating a clear risk. This prompts the anesthesiologist to check the patient's condition and consider adjusting the anesthetic drugs or taking measures to stabilize circulation. When the absolute value of the anesthesia depth instability fluctuation coefficient is less than or equal to 0.1, the anesthesia depth and vital signs are balanced, with no significant trend risk. The current anesthesia protocol should be maintained, and routine monitoring should be performed. When the absolute value of the anesthesia depth instability fluctuation coefficient is between 0.1 and 0.3, it is necessary to pay attention to whether there is excessive sedation or to assess whether there is insufficient analgesia or sedation. The anesthesiologist should be reminded to check the trend curve (EEG, HR, MAP) to confirm whether the drug dosage needs to be fine-tuned. When the absolute value of the anesthesia depth instability fluctuation coefficient is greater than 0.3, an alarm should be issued immediately, automatic drug administration should be forcibly stopped, and the event time, trend graph, and patient vital sign snapshot should be recorded.

[0068] The anesthesia state monitoring method based on multi-source physiological signal analysis according to embodiments of the present invention can predict the anesthesia state based on the processed new electroencephalogram (EEG) signals, obtain the anesthesia stage classification results, and assess the current stable anesthesia status and the risk development trend under unstable status based on the anesthesia stage classification results, real-time heart rate, heart rate variability coefficient, and real-time mean arterial pressure, thereby determining the comprehensive stability coefficient of the anesthesia state and the anesthesia depth instability fluctuation coefficient. Furthermore, the anesthesia state monitoring is performed based on the comprehensive stability coefficient of the anesthesia state and the anesthesia depth instability fluctuation coefficient, thus improving the accuracy of anesthesia state monitoring. When determining the first, second, and third dynamic weights, they can be based on the first, second, and third baseline weights, real-time heart rate, real-time mean arterial pressure, lower limit of safe mean arterial pressure, lower limit of dangerous mean arterial pressure, upper limit of safe heart rate, and upper limit of dangerous heart rate. During the calculation process, the first, second, and third dynamic weights can be set based on the risk of tachycardia and the risk of low mean arterial pressure, specifically for EEG signals, mean arterial pressure, and heart rate, respectively, thus improving the accuracy of the first, second, and third dynamic weights. When determining the comprehensive stability coefficient of the anesthetic state, it can be based on the first, second, and third dynamic weights, the standard deviation of the probability of adequate anesthesia, the baseline value of the individual heart rate variability coefficient, the ideal real-time mean arterial pressure under anesthesia, the real-time mean arterial pressure, and the heart rate variability coefficient. During the calculation process, the contribution of anesthesia depth to stability and the contribution of vital signs to safety can be fully analyzed, improving the comprehensiveness and accuracy of the comprehensive stability coefficient of the anesthetic state. When determining the anesthetic depth instability fluctuation coefficient, it can be determined based on the probability of deep anesthesia, the probability of light anesthesia during the awakening stage, the standard deviation of the probability of deep anesthesia, the standard deviation of the probability of light anesthesia during the awakening stage, real-time heart rate, and real-time mean arterial pressure. During the calculation process, the EEG trend risk can be fully analyzed, and the risk amplification factor can be determined based on the degree of matching between physiological deviation and EEG trend. Furthermore, the anesthetic depth instability fluctuation coefficient can be determined based on the EEG trend risk and the risk amplification factor, thus improving the comprehensiveness of the anesthetic depth instability fluctuation coefficient.

[0069] Figure 5 An exemplary block diagram of an anesthesia state monitoring system based on multi-source physiological signal analysis according to an embodiment of the present invention is shown, the system comprising: The data acquisition module is used to collect the patient's electroencephalogram (EEG) signals, real-time heart rate, heart rate variability coefficient, and real-time mean arterial pressure at multiple points during the monitoring cycle while the patient is under anesthesia. The signal processing module is used to initially filter out EEG signals using high-pass and low-pass filters, and to perform amplitude normalization processing on the EEG signals to obtain new EEG signals. The mode decomposition module is used to perform variational mode decomposition on new EEG signals using a hierarchical frequency decomposition method to obtain several mode functions; The feature extraction module is used to extract multidimensional features from each mode function; The feature set module is used to fuse and filter the extracted multidimensional features to form a feature set; The classification results module is used to train the machine learning model using the feature set, and to use the trained machine learning model to predict the patient's anesthesia status in real time, outputting the classification results of the anesthesia stage. The stability coefficient module is used to determine the comprehensive stability coefficient of the anesthesia state based on the real-time heart rate, the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results. The fluctuation coefficient module is used to determine the anesthesia depth instability fluctuation coefficient based on the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results. The anesthesia monitoring module is used to monitor the anesthesia status based on the comprehensive stability coefficient of the anesthesia status and the unstable fluctuation coefficient of the depth of anesthesia.

[0070] According to an embodiment of the present invention, an anesthesia state monitoring device based on multi-source physiological signal analysis is provided. The device includes a memory storing a computer program and a processor for executing the computer program. When the computer program is executed by the processor, it implements the steps of the anesthesia state monitoring method based on multi-source physiological signal analysis.

[0071] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0072] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and any variations or modifications may be made to the implementation of the present invention without departing from the stated principles.

Claims

1. A method for monitoring anesthesia status based on multi-source physiological signal analysis, characterized in that, include: At multiple points during the monitoring period, the patient's electroencephalogram (EEG) signals, real-time heart rate, heart rate variability coefficient, and real-time mean arterial pressure were collected while the patient was under anesthesia. High-pass and low-pass filters were used to initially filter out EEG signals, and amplitude normalization was performed on the EEG signals to obtain new EEG signals. The new EEG signal was subjected to variational mode decomposition using a hierarchical frequency decomposition method to obtain several mode functions; Extract multidimensional features from each mode function; The extracted multidimensional features are fused and filtered to form a feature set; The machine learning model is trained using a feature set, and the trained machine learning model is used to predict the patient's anesthesia status in real time, outputting the classification results of the anesthesia stage. Based on the real-time heart rate, the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results, the comprehensive stability coefficient of the anesthesia state is determined. The anesthesia depth instability fluctuation coefficient is determined based on the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results. Anesthesia status is monitored based on the comprehensive stability coefficient of the anesthesia status and the unstable fluctuation coefficient of the depth of anesthesia.

2. The anesthesia state monitoring method based on multi-source physiological signal analysis according to claim 1, characterized in that, High-pass and low-pass filters are used to initially filter out EEG signals, and amplitude normalization is performed on the EEG signals to obtain new EEG signals, including: A high-pass filter was used to remove baseline drift signals from the electroencephalogram (EEG) signal; A low-pass filter was used to remove high-frequency interference from the EEG signal; The amplitude of the EEG signal after removing baseline drift signal and high frequency interference was normalized to obtain a new EEG signal.

3. The anesthesia state monitoring method based on multi-source physiological signal analysis according to claim 1, characterized in that, Based on the real-time heart rate, the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results, a comprehensive stability coefficient of the anesthesia state is determined, including: Based on the real-time heart rate and the real-time mean arterial pressure, a first dynamic weight, a second dynamic weight, and a third dynamic weight are determined. Determine the target anesthesia stage; Based on the target anesthesia stage and the anesthesia stage classification results, the probability of adequate anesthesia is determined; Based on the appropriate anesthesia probability, determine the standard deviation of the appropriate anesthesia probability within a preset time window; Obtain the baseline value of the individual's heart rate variability coefficient and the ideal real-time mean arterial pressure under anesthesia; The comprehensive stability coefficient of the anesthesia state is determined based on the first dynamic weight, the second dynamic weight, the third dynamic weight, the standard deviation of the probability of adequate anesthesia, the baseline value of the individual heart rate variability coefficient, the real-time mean arterial pressure of ideal anesthesia, the real-time mean arterial pressure, and the heart rate variability coefficient.

4. The anesthesia state monitoring method based on multi-source physiological signal analysis according to claim 3, characterized in that, Based on the real-time heart rate and the real-time mean arterial pressure, a first dynamic weight, a second dynamic weight, and a third dynamic weight are determined, including: Obtain the first basic weight, the second basic weight, and the third basic weight; Obtain the lower limit of safe mean arterial pressure and the lower limit of dangerous mean arterial pressure; Get the safe upper heart rate limit and the dangerous upper heart rate limit; Based on the first basic weight, the second basic weight, the third basic weight, the real-time heart rate, the real-time mean arterial pressure, the lower limit of safe mean arterial pressure, the lower limit of dangerous mean arterial pressure, the upper limit of safe heart rate, and the upper limit of dangerous heart rate, the first dynamic weight, the second dynamic weight, and the third dynamic weight are determined.

5. The anesthesia state monitoring method based on multi-source physiological signal analysis according to claim 4, characterized in that, Based on the first basic weight, the second basic weight, the third basic weight, the real-time heart rate, the real-time mean arterial pressure, the lower limit of safe mean arterial pressure, the lower limit of dangerous mean arterial pressure, the upper limit of safe heart rate, and the upper limit of dangerous heart rate, the first dynamic weight, the second dynamic weight, and the third dynamic weight are determined, including according to the formula: Determine the first dynamic weight at time i of the monitoring period. Second dynamic weight and the third dynamic weight ,in, The first dynamic transition weight is the weight at the i-th time point of the monitoring period. The second dynamic transition weight is the weight at the i-th time point of the monitoring period. The third dynamic transition weight is the value at time i of the monitoring period. To preset the minimum first dynamic transition weight threshold, , and Here, is the preset coefficient, and max is the function to find the maximum value. As the first basic weight, As the second basic weight, As the third basic weight, To ensure a safe lower limit of mean arterial pressure, To measure the real-time mean arterial pressure at time i of the monitoring cycle, This is the lower limit of the dangerous mean arterial pressure. To ensure a safe upper limit of heart rate, The upper limit of the dangerous heart rate, This represents the real-time heart rate at the i-th moment of the monitoring cycle.

6. The method for monitoring anesthesia status based on multi-source physiological signal analysis according to claim 3, characterized in that, Based on the first dynamic weight, the second dynamic weight, the third dynamic weight, the standard deviation of the probability of adequate anesthesia, the baseline value of the individual heart rate variability coefficient, the ideal real-time mean arterial pressure, the real-time mean arterial pressure, and the heart rate variability coefficient, a comprehensive stability coefficient of the anesthesia state is determined, including: according to the formula: Determine the comprehensive stability coefficient of the anesthesia state at time i of the monitoring period. Where min is the function for finding the minimum value, and max is the function for finding the maximum value. The first dynamic weight at the i-th moment of the monitoring period, The second dynamic weight is the weight at the i-th time point of the monitoring period. The third dynamic weight is the value at the i-th time point of the monitoring period. Let $\frac{i}{i}$ be the standard deviation of the probability of adequate anesthesia within a preset time window at time $i$ during the monitoring period. Let be the heart rate variability coefficient at time i in the monitoring cycle. This is the baseline value for the individual's heart rate variability coefficient. To prevent the coefficient of variability from deviating from a preset threshold, To measure the real-time mean arterial pressure at time i of the monitoring cycle, The mean arterial pressure at real time is the ideal anesthesia level. This is a preset threshold for pulsating pressure deviation.

7. The method for monitoring anesthesia status based on multi-source physiological signal analysis according to claim 1, characterized in that, Based on the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results, the anesthesia depth instability fluctuation coefficient is determined, including: Based on the classification results of the anesthesia stages, the probability of deep anesthesia and the probability of awakening from light anesthesia are determined. Based on the probability of the awakening phase of light anesthesia, the standard deviation of the probability of deep anesthesia and the standard deviation of the probability of the awakening phase of light anesthesia within a preset time window are determined. The anesthesia depth instability fluctuation coefficient is determined based on the probability of deep anesthesia, the probability of light anesthesia awakening stage, the standard deviation of the probability of deep anesthesia, the standard deviation of the probability of light anesthesia awakening stage, the real-time heart rate, and the real-time mean arterial pressure.

8. The method for monitoring anesthesia status based on multi-source physiological signal analysis according to claim 7, characterized in that, The anesthesia depth instability fluctuation coefficient is determined based on the probability of deep anesthesia, the probability of light anesthesia awakening phase, the standard deviation of the probability of deep anesthesia, the standard deviation of the probability of light anesthesia awakening phase, the real-time heart rate, and the real-time mean arterial pressure, including: According to the formula Determine the anesthesia depth instability fluctuation coefficient at time i of the monitoring period. ,in, , , , and Here, is the preset coefficient, and max is the function to find the maximum value. Let the sum of the probabilities of the light anesthesia awakening phase at time i of the monitoring period be denoted as . Let the sum of the probabilities of the light anesthesia awakening phase at time i-1 of the monitoring cycle be denoted as . For the time point corresponding to the i-th moment of the monitoring period, This refers to the time point corresponding to the (i-1)th moment of the monitoring period. To measure the probability and standard deviation of the light anesthesia awakening phase within a preset time window at time i of the monitoring period. To determine the probability of deep anesthesia at time i in the monitoring cycle. To determine the probability of deep anesthesia at time i-1 of the monitoring period, Let be the standard deviation of the probability of deep anesthesia within a preset time window at the i-th moment of the monitoring period. Let be the heart rate variability coefficient at time i in the monitoring cycle. This is the baseline value for the individual's heart rate variability coefficient. To prevent the coefficient of variability from deviating from a preset threshold, To measure the real-time mean arterial pressure at time i of the monitoring cycle, The mean arterial pressure at real time is the ideal anesthesia level. This is a preset threshold for pulsating pressure deviation.

9. An anesthesia state monitoring system based on multi-source physiological signal analysis, characterized in that, For performing the method of any one of claims 1-7, comprising: The data acquisition module is used to collect the patient's electroencephalogram (EEG) signals, real-time heart rate, heart rate variability coefficient, and real-time mean arterial pressure at multiple points during the monitoring cycle while the patient is under anesthesia. The signal processing module is used to initially filter out EEG signals using high-pass and low-pass filters, and to perform amplitude normalization processing on the EEG signals to obtain new EEG signals. The mode decomposition module is used to perform variational mode decomposition on new EEG signals using a hierarchical frequency decomposition method to obtain several mode functions. The feature extraction module is used to extract multidimensional features from each modality function; The feature set module is used to fuse and filter the extracted multidimensional features to form a feature set; The classification results module is used to train the machine learning model using the feature set, and to use the trained machine learning model to predict the patient's anesthesia status in real time, outputting the classification results of the anesthesia stage. The stability coefficient module is used to determine the comprehensive stability coefficient of the anesthesia state based on the real-time heart rate, the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results. The fluctuation coefficient module is used to determine the anesthesia depth instability fluctuation coefficient based on the heart rate variability coefficient, the real-time mean arterial pressure, and the anesthesia stage classification results. The anesthesia monitoring module is used to monitor the anesthesia status based on the comprehensive stability coefficient of the anesthesia status and the unstable fluctuation coefficient of the anesthesia depth.

10. An anesthesia state monitoring device based on multi-source physiological signal analysis, the device comprising a memory storing a computer program and a processor for executing the computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-8.

Images