Intelligent monitoring method and system for anesthesia depth index based on electroencephalogram signals

Abstract

This invention relates to the field of medical monitoring equipment technology, and discloses an intelligent monitoring method and system for anesthesia depth index based on electroencephalogram (EEG) signals. The method includes: acquiring and preprocessing multi-channel EEG signals; applying a biophysical constraint source separation algorithm to separate the signals and automatically label them as cortical sources and artifact sources; performing source domain integrity verification to generate a quantified source purity index; performing inter-source dynamic calibration, using artifact sources as templates to update the model parameters of the separation algorithm; when the source purity index meets the standard, extracting features only from cortical sources and synthesizing the anesthesia depth index; and co-displaying the anesthesia depth index and the source purity index on the user interface. This invention comprehensively improves the accuracy, robustness, and safety of anesthesia depth monitoring through source separation to resist interference, dynamic calibration to improve stability, and co-output of the source purity index to increase reliability.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring equipment technology, specifically to an intelligent monitoring method and system for anesthesia depth index based on electroencephalogram (EEG) signals. Background Technology

[0002] In modern surgery, precise control of general anesthesia is crucial for ensuring patient safety, reducing intraoperative complications, and optimizing postoperative recovery. Anesthesia depth monitoring indices, generated from electroencephalogram (EEG) signal processing, have become a standardized auxiliary tool for clinical anesthesiologists to assess patients' levels of consciousness and sedation. By converting complex EEG waveforms into intuitive numerical values, this type of technology aims to provide objective evidence for the titration of anesthetic drugs.

[0003] In current technologies, mainstream methods for monitoring the depth of anesthesia typically involve acquiring electroencephalogram (EEG) signals using a few electrodes placed on the patient's forehead. The acquired signals are then fed into a processing module that applies a series of signal processing algorithms, such as Fast Fourier Transform, bispectral analysis, or information entropy calculation, to extract specific physiological features from the mixed EEG signals. Finally, these features are mapped to a standardized anesthesia depth index ranging from 0 to 100 using a pre-defined mathematical model and displayed on a monitor for clinicians' reference.

[0004] While existing technologies offer a means of quantifying the depth of anesthesia by processing electroencephalogram (EEG) signals, some limitations remain: First, the accuracy of existing monitoring technologies drops significantly in the presence of artifacts. This is because the limited number of electrodes they employ results in insufficient spatial sampling density. This configuration physically fails to provide the algorithm with enough information to effectively distinguish target cortical source signals from spatially overlapping artifact source signals. Therefore, when electromyographic signals from facial muscles or electromagnetic interference from electrosurgical devices occur, the energy of these artifact signals is inevitably processed along with the real EEG signals. Traditional algorithms, lacking source separation mechanisms, incorrectly include these artifact features in the synthesis of the exponent, leading to the exponent being contaminated by artifacts and becoming inaccurate.

[0005] Secondly, existing monitoring technologies lack adaptability and stability to artifact changes. Their built-in signal processing algorithms and model parameters are typically fixed, and these models are designed based on general artifact characteristics, failing to cope with the high variability of artifact morphology in clinical practice. Different patients have different electromyographic patterns, and different types of electrosurgical units produce different interference characteristics, which also change dynamically during surgery. Because existing technologies lack a mechanism for feedback and self-optimization based on current real-time artifact characteristics, their fixed inhibition performance decreases when encountering atypical or changing artifacts, leading to unstable output of the monitoring index.

[0006] Furthermore, existing technologies fail to provide a quantitative assessment of the reliability of the output monitoring index. The system presents the user with a single numerical result without indicating whether the calculation is based on pure EEG signals or contaminated signals. This black-box output makes it impossible for clinicians to determine whether an abnormal or fluctuating index truly reflects changes in the patient's physiological state or is merely a calculation error caused by artifacts. This lack of reliability forces clinicians to interpret the index with uncertainty, especially during periods of strong artifact interference, which can lead to misjudgments and affect the safety of clinical decisions. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides an intelligent monitoring method and system for the anesthesia depth index based on electroencephalogram (EEG) signals. This solves the problems of inaccurate, unstable, and unreliable monitoring results caused by the inability to effectively separate artifacts, lack of adaptive capability to artifact changes, and failure to provide a quantifiable assessment of the index's reliability in existing anesthesia depth monitoring technologies.

[0008] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of this invention provides an intelligent monitoring method for anesthesia depth index based on electroencephalogram (EEG) signals, the method comprising the following steps: Acquire multi-channel EEG signals and perform signal preprocessing on the multi-channel EEG signals; The preprocessed multi-channel EEG signal is separated into multiple independent source signals by applying a real-time biophysical constraint source separation algorithm. Based on the preset source signal spatial projection topographic map constraint and time series statistical characteristic constraint, the source signals are automatically labeled as cortical sources or artifact sources. Perform source domain integrity verification to generate a quantified source purity index; Perform dynamic calibration between sources, using the pseudo-trace source as a calibration template, and update the dynamic constraint model parameters in the real-time biophysical constraint source separation algorithm. Determine whether the source purity index is higher than a preset quality threshold. If it is, extract physiological features only from the cortical source and synthesize them into an anesthesia depth index. On the user interface, the anesthesia depth index and the source purity index are displayed together.

[0009] Preferably, the source signal spatial projection topographic map constraint includes: presetting a prior spatial probability map based on a standard head anatomy model for cortical sources, and presetting a prior spatial probability map based on the physical characteristics of the pseudo-trace sources for pseudo-trace sources; The time series statistical characteristic constraints include: predefining prior probability density functions with different distribution characteristics for cortical sources and artifact sources respectively.

[0010] Preferably, the step of performing inter-source dynamic calibration specifically includes: Real-time monitoring of the signal energy of the pseudo-trace source; When the signal energy exceeds the preset calibration energy threshold, the corresponding pseudo-trace source signal segment is used as the calibration template. The parameters of the dynamic constraint model are re-estimated using the calibration template and the maximum likelihood estimation algorithm, and the real-time biophysical constraint source separation algorithm is adjusted with the updated parameters.

[0011] Preferably, the step of performing source domain integrity verification specifically includes: The thoroughness of separation is assessed by calculating the residual mutual information between the cortical source and the artifact source. The degree of frequency band contamination is assessed by calculating the energy leakage ratio of the cortical source within the artifact characteristic frequency range. The results of the residual mutual information and energy leakage ratio are fused to generate the source purity index.

[0012] Preferably, the step of fusing the results of the residual mutual information and the energy leakage ratio includes: The residual mutual information and energy leakage ratio are compared with preset thresholds and converted into standardized independence mass scores and leakage mass scores. The source purity index is obtained by weighted summation of the independence mass fraction and the leakage mass fraction.

[0013] Preferably, the physiological characteristics include: Trans-frequency phase-amplitude coupling characteristics characterizing the coupling strength between different neural oscillation frequency components; The waveform morphology of the source signal that characterizes the dominant non-sinusoidal oscillation. Source domain burst suppression ratio calculated on the cortical source.

[0014] Preferably, the step of synthesizing the anesthesia depth index includes: The physiological features are combined into a multidimensional feature vector, and the multidimensional feature vector is input into a fusion model obtained by offline training with clinical anesthesia data to output the anesthesia depth index.

[0015] Preferably, the collaborative display step includes: The anesthesia depth index and the source purity index are displayed simultaneously on the user interface, and the source purity index characterizes the reliability of the anesthesia depth index.

[0016] Preferably, the method further includes: When the source purity index falls below a preset warning threshold due to artifact interference, a warning is issued on the user interface, indicating that the reliability of the currently output anesthesia depth index has decreased.

[0017] A second aspect of the present invention provides an intelligent monitoring system for anesthesia depth index based on electroencephalogram (EEG) signals, using any of the methods described above, the system comprising: Multi-channel electrode array for acquiring multi-channel EEG signals; A signal acquisition unit, electrically connected to the multi-channel electrode array, is used to process the acquired signals to output multi-channel digital signals; Central processing unit, connected to the signal acquisition unit, the central processing unit comprising: The signal preprocessing module is used to perform signal preprocessing on the multi-channel digital signal; The real-time BCSS module is used to apply a real-time biophysical constraint source separation algorithm to separate the preprocessed multi-channel EEG signals into cortical sources and artifact sources. The source domain integrity verification engine is used to perform source domain integrity verification to generate a quantified source purity index. The inter-source dynamic calibration module is used to perform inter-source dynamic calibration, using the pseudo-trace source as a calibration template, and updating the real-time BCSS module accordingly. An anesthesia depth index calculation module is used to extract physiological characteristics from the cortical source and synthesize the anesthesia depth index when the source purity index is higher than a preset quality threshold. A user interface, connected to the central processing unit, is used to collaboratively display the anesthesia depth index and the source purity index.

[0018] This invention provides a method and system for intelligent monitoring of anesthesia depth index based on electroencephalogram (EEG) signals. It has the following beneficial effects: 1. This invention employs a real-time biophysical constraint source separation algorithm to separate multi-channel EEG signals into cortical and artifact sources in real time. Physiological features are extracted only from the labeled pure cortical sources to synthesize an anesthesia depth index. This method fundamentally solves the problem in traditional methods where artifact signals from electromyography (EMG), electrosurgery, etc., are mixed with real EEG signals, thus contaminating and distorting the index calculation, thereby improving the accuracy and reliability of anesthesia depth monitoring.

[0019] 2. This invention performs dynamic inter-source calibration, using successfully separated high-energy artifact sources as real-time calibration templates to form a feedback adjustment closed loop that continuously optimizes the internal model parameters of the source separation algorithm. This design enables the monitoring system to have adaptive capabilities, actively learning and adapting to artifact morphologies from specific patients or changing with the surgical process. This allows it to maintain continuous and efficient suppression of interference signals in complex clinical environments, ensuring long-term stability and robustness of monitoring performance.

[0020] 3. This invention generates a quantified source purity index by performing source domain integrity verification, and then displays this index in conjunction with the final anesthesia depth index on the user interface. This provides clinicians with a real-time, quantitative assessment of the reliability of the current anesthesia depth reading, solving the problem in existing technologies where index reliability cannot be measured and clinicians can only blindly trust it. When the system encounters strong interference that causes a decline in separation quality, clinicians can be immediately notified, thus avoiding misinterpretation of inaccurate indices and improving the safety of clinical decision-making. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the system architecture of the present invention; Figure 3 This is a schematic diagram of the signal acquisition and preprocessing process of the present invention; Figure 4 This is a schematic diagram of the real-time biophysical constraint source separation functional module of the present invention; Figure 5 This is a schematic diagram of the source domain integrity verification function module of the present invention. Detailed Implementation

[0022] The technical solutions in 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.

[0023] Please see the appendix Figure 1 , Figure 1 This is a schematic flowchart of a method according to an embodiment of the present invention. The present invention provides a method for monitoring anesthesia depth based on source separation and dynamic calibration, which may include the following steps: Step S110: Multi-channel EEG signals are acquired by a multi-channel electrode array placed on the patient's forehead. This array provides the spatial sampling information required for subsequent source separation.

[0024] Step S120: Perform signal preprocessing on the acquired multi-channel EEG signals. This preprocessing removes slow baseline drift in the signal through bandpass filtering and eliminates power frequency interference through notch filtering.

[0025] Step S130: Apply a real-time biophysical constraint source separation algorithm to demix the preprocessed multi-channel signal into a set of independent source signals. During the solution process, this real-time biophysical constraint source separation algorithm, based on preset constraints of the source signal spatial projection topographic map and time series statistical characteristics, labels the separated source signals as either cortical sources or artifact sources.

[0026] Step S140: Perform source domain integrity verification. This verification generates a quantified source purity index by calculating the residual mutual information between signals labeled as cortical sources and artifact sources, and by calculating the energy leakage ratio of cortical sources within the artifact characteristic frequency range.

[0027] Step S150: Perform inter-source dynamic calibration. This calibration uses the pseudo-trace source separated from step S130, whose signal energy exceeds a preset calibration threshold, as a calibration template. This template is analyzed to update the dynamic constraint model parameters in the biophysical constraint source separation algorithm, forming a feedback adjustment loop.

[0028] Step S160: Determine whether the source purity index generated in step S140 is higher than the preset quality threshold. If yes, proceed to step S170; if no, do not perform subsequent calculations and wait for the next analysis cycle.

[0029] Step S170: When the source purity index is confirmed to be higher than the quality threshold, only a set of physiological features related to the depth of anesthesia are extracted from the signals labeled as cortical sources. This set of features includes the phase-amplitude coupling strength across frequency bands, the waveform symmetry and steepness of the signal, and the burst suppression ratio in the source signal.

[0030] In step S180, the multidimensional feature vector containing multiple physiological features extracted in step S170 is input into a pre-trained fusion model for processing, thereby synthesizing a single, standardized anesthesia depth index.

[0031] In step S190, the anesthesia depth index generated in step S180 and the source purity index generated in step S140 are displayed together on the user interface. The former reflects the anesthesia state, and the latter reflects the calculation reliability of the former.

[0032] Please see the appendix Figure 2 , Figure 2 This is a schematic diagram of a system architecture according to an embodiment of the present invention. The present invention provides an anesthesia depth monitoring system based on source separation and dynamic calibration, which may include: A multi-channel electrode array is used to acquire multi-channel electroencephalogram (EEG) signals from the patient's forehead.

[0033] The signal acquisition unit, which is electrically connected to the multi-channel electrode array, is used to amplify, synchronously sample, and convert the acquired analog signals to digital signals to output multi-channel digital signals.

[0034] The central processing unit (CPU), connected to the signal acquisition unit, receives and processes multi-channel digital signals. This CPU integrates several functional modules: The signal preprocessing module is used to perform bandpass filtering and notch filtering on the received digital signal.

[0035] The real-time BCSS module is used to run the biophysically constrained source separation algorithm, which decomposes the preprocessed signal into multiple labeled source signals, including cortical sources and artifact sources.

[0036] The source domain integrity verification engine is used to evaluate the quality of the separation results of the real-time BCSS module and calculate the output source purity index.

[0037] The inter-source dynamic calibration module is used to capture calibration templates from the pseudo-trace sources output by the real-time BCSS module and feed back updates to the internal model parameters of the real-time BCSS module.

[0038] The anesthesia depth index calculation module is used to extract features from the cortical source output by the real-time BCSS module, provided that the source purity index meets the standard, and calculate the final anesthesia depth index through a fusion model.

[0039] The user interface, connected to the central processing unit, receives and displays the anesthesia depth index and source purity index calculated by the central processing unit.

[0040] Please see the appendix Figure 3 , Figure 3 This is a schematic diagram of signal acquisition and preprocessing functions according to an embodiment of the present invention.

[0041] In step S110 of the method of the present invention, signal acquisition is performed using a multi-channel electrode array disposed on the patient's forehead. This multi-channel electrode array comprises at least... One acquisition electrode, as well as the necessary reference electrode and ground electrode.

[0042] In one specific embodiment, the following is adopted: The ≥8 multi-channel configuration is to provide a sufficient spatial sampling rate to meet the spatial information requirements of the high-precision source separation algorithm (such as BCSS) in the subsequent step S130. The traditional 3-4 electrode configuration is difficult to effectively separate multiple spatially adjacent or overlapping cortical sources and artifact sources due to its insufficient spatial sampling density.

[0043] The signal acquisition unit is electrically connected to the electrode array. This signal acquisition unit is responsible for... Simultaneous sampling and analog-to-digital (A / D) conversion of simulated target physiological signals from each channel. Sampling rate. Preset as (e.g.) =500Hz or =1000Hz), to ensure the capture of complete information of target physiological signals and high-frequency artifacts (such as EMG, ESU).

[0044] This step S110 outputs a... ×1 dimensional multichannel digital signal vector Its mathematical representation is: ; In the formula, For a moment The multi-channel digital signal column vector represents the set of signal sample values ​​from all channels at the same time. This is a time variable, representing the specific moment when the signal was sampled; For the first Each electrode channel at time... The digital signal sample values ​​were acquired and converted from analog to digital, where The value range is 1 to ; The total number of acquisition electrodes set in a multi-channel electrode array on the patient's forehead; This is the matrix transpose operator.

[0045] In step S120, the acquired raw signal The signal is then sent to the signal preprocessing module for processing. The purpose of this step is to eliminate non-physiological noise interference and prepare data for subsequent source separation calculations.

[0046] The preprocessing step S120 may specifically include: S121: Perform bandpass filtering. Apply a digital bandpass filter (e.g., an IIR Butterworth filter or an FIR filter) to the filter. Each channel is processed. The passband range of the filter can be set (e.g., 0.5 Hz to 100 Hz). The lower cutoff frequency (0.5 Hz) of this range is used to filter out slow baseline drift caused by changes in electrode-skin contact potential; the upper cutoff frequency (100 Hz) is used to filter out excessively high frequency noise that is outside the range of the target physiological signal and the main artifact signal.

[0047] S122: Perform notch filtering. After (or before) step S121, one or more digital notch filters are applied. The notch filter is used to precisely eliminate power frequency interference introduced by operating room equipment. For example, in one embodiment, the notch frequency can be set to 50Hz and its first harmonic to 100Hz; in another embodiment, the notch frequency can be set to 60Hz and its harmonic to 120Hz.

[0048] After processing in step S120 (including S121 and S122), the preprocessed product is obtained. Channel signal vector, denoted as .Should This will serve as the input signal for the subsequent real-time BCSS module.

[0049] Please see the appendix Figure 4 , Figure 4 This is a schematic diagram of a real-time biophysical constraint source separation (BCSS) functional module according to an embodiment of the present invention.

[0050] Step S130 is executed by the real-time BCSS module. The purpose of this step is to convert the output of step S120... Channel preprocessing signal Real-time separation An independent source signal In one embodiment, = That is, the number of separated sources equals the number of electrode channels. The key to this step is that, during separation, it utilizes pre-existing biophysical knowledge to analyze the separated sources. Individual source signal To perform automated physiological labeling.

[0051] Step S130 may specifically include the following sub-steps: S131: Establishing the mixing and separation model. This method is based on a linear instantaneous mixing model. This model can be described in words as: assuming the observed... Channel signal It is by Statistically independent source signals After an unknown × Mixed matrix It is formed by linear superposition and is subject to additive noise. The impact. The goal of source separation is to find a × Separation matrix Through this separation matrix right Linear unmixing is performed to obtain the estimated source signal. to make it as The best estimate. Let the separation matrix be denoted as... This is to indicate that the matrix can be adjusted and optimized in real time by the subsequent step S150 (inter-source dynamic calibration). Traditional blind source separation (BSS) algorithms (such as independent component analysis (ICA)) are in the process of solving... Since it does not rely on prior knowledge, its separation results are somewhat blind (i.e., it cannot automatically identify). This corresponds to whether the source is cortical or myogenic and permutation uncertainty (i.e., the output order of the source is random).

[0052] S132: Applying biophysical constraints. To address the problem mentioned in S131, the method of this invention solves... During the process, at least two types of biophysical prior constraints were imposed through a unified optimization objective function: The first type of constraint is a spatial constraint. This constraint is based on the physical projection pattern of the source signal onto the scalp. (Mixture matrix) The List (corresponding to) The (Column) represents the first Individual Source exist Spatial projection topographic maps on each electrode. This method pre-defines prior spatial probability maps for different types of sources. In one embodiment, the prior spatial probability map of cortical sources It can be generated based on standard head anatomy models (such as MNI standard head models or boundary element head models). Define the dipole source originating from the prefrontal cortex in Typical spatial energy distribution on a channel electrode array. In one embodiment, the prior spatial probability map of a pseudo-trace source (e.g.) Corresponding to electromyography, (Corresponding to electrooculography) can be defined based on its physical characteristics, such as the spatial diagram of ocular power sources. It has a specific distribution of electrodes concentrated near the eye socket.

[0053] The second type of constraint is a dynamic constraint. This constraint is based on the time-series statistical characteristics of different source signals. This method applies to different types of sources. Its prior probability density function (PDF) is predefined. The function consists of a set of parameters. Description. In one embodiment, the dynamic characteristics of cortical sources It is modeled as having neural oscillatory properties, with its power transfer function (PDF) approximating a Gaussian distribution or a slightly sub-Gaussian distribution. Dynamic characteristics of the myoelectric power source. It exhibits high-frequency, spike-like, non-oscillating random discharges, whose power vector (PDF) is modeled as a strongly super-Gaussian distribution (e.g., a Laplace distribution or a parameterized generalized Gaussian distribution can be used). Dynamic characteristics of the electrosurgical source. It can be modeled as a non-Gaussian process with specific spectral properties. This set of parameters... (in particular and The feedback update will be performed by the inter-source dynamic calibration module in step S150.

[0054] S133: Performs real-time solution and automatic labeling. Separate matrix. The solution is achieved by optimizing a unified cost function. In one embodiment, this optimization objective can be based on maximizing the posterior probability framework. Separation matrix By maximizing the following objective function Please provide a solution: ; In the formula, Let be the objective function to be optimized, which represents the separation matrix given the observed signal and prior constraints. The optimal separation matrix can be obtained by maximizing the logarithm of the posterior probability. for × The separation matrix of dimension is the objective variable for solving this optimization problem, and its function is to demix the observed signal; It is the natural logarithm function; Let be the likelihood function of the data, representing the likelihood of a given separation matrix. Source signal model parameters Under these conditions, the preprocessed signal data was observed. The probability of; After preprocessing in step S120 Channel EEG signal data; A set of model parameters describing the dynamic characteristics of each source signal, for example, ; Separation matrix The prior probability represents the probability given the prior constraints in the given space. Under the condition of separation matrix The probability of occurrence, this term integrates spatial constraints into the optimization objective; A set of prior spatial probability maps describing the spatial distribution characteristics of each source signal, for example, .

[0055] It is the log-likelihood of the data, in = Under the assumption of source independence, it can be expressed as: ; In the formula, This represents the total number of data sampling points included within an analysis time window; Separation matrix The absolute value of the determinant, which is the Jacobian determinant generated during the variable substitution process from the observed signal to the source signal; For the first The probability density function of a source signal, which is given by the parameter set At any moment The definition represents a priori constraints on the dynamic characteristics of the source signal; In order to be in The estimated time of the first The amplitude of the source signal; For at any time Description of the The probability density function of a source signal requires a set of parameters, which can be updated in real time by step S150.

[0056] Is applied to Priors on, and their relationship with the mixture matrix spatial constraints Related, for example In the formula, The sign indicates a linear relationship between the logarithmic prior probability on the left and the summation term on the right. For the first The spatial likelihood function of a source signal represents the probability of a given prior spatial map. Under these conditions, its estimated spatial projection topographic map The probability of occurrence is used to quantify the degree of matching between the estimation result and the spatial prior. Separation matrix inverse matrix The Column, this column vector represents the first column. Individual source signal in Spatial projection topographic map on each electrode; For the preset first Prior space probability map of source-like signals (such as cortical sources and myogenic sources).

[0057] This optimization problem can be solved iteratively using, for example, an improved natural gradient method or other nonlinear optimization algorithms, to update in real time. .

[0058] By solving this constrained optimization problem, the system not only obtains An estimated source signal Moreover, due to space constraints and dynamic constraints With forced guidance, the system can achieve automatic and stable source labeling. For example, matching cortical source priors ( and )of Automatically labeled as cortical source Matching muscle power prior ( and )of Automatically marked as a muscle power source ; Matching electrosurgical source prior ( and )of Automatically marked as an electrosurgical source This automatic labeling overcomes the blindness and permutation uncertainty of the traditional BSS algorithm, and is a prerequisite for realizing the automated processing of subsequent steps S140 to S170.

[0059] Please see the appendix Figure 5 , Figure 5 This is a schematic diagram of a source domain integrity verification function module according to an embodiment of the present invention.

[0060] Step S140 is performed by the source domain integrity verification engine. This step does not directly analyze the raw mixed EEG signals, but rather assesses the quality of the separation results of the real-time BCSS module from step S130. Its purpose is to prevent incomplete separation of the BCSS module under extreme interference or model mismatch, thereby ensuring the quality of cortical source signals subsequently used to calculate the anesthesia depth index. It has high purity. This step ultimately outputs a source purity index, the range of which in one embodiment is [value range missing].

[0061] Step S140 may specifically include the following sub-steps: S141: Perform source independence check. This check is used to examine cortical sources. Does it still contain pseudo-sources (e.g., muscle power sources)? The residual components of the two source signals. In one embodiment, this verification is achieved by calculating the residual mutual information of the two source signals within an analysis time window (e.g., 2 seconds).

[0062] Mutual information is a standard metric in information theory used to measure the statistical dependence between two random variables. The calculation process can be summarized as: by analyzing within a time window... and The signal values ​​are analyzed, their joint probability distribution and individual marginal probability distributions are estimated, and the amount of information about one signal contained in another signal is quantified based on these probability distributions. The calculated... If the value exceeds the preset independence threshold This indicates that the separation was incomplete and the cortical source was contaminated by the myogenic source.

[0063] S142: Perform an energy leakage check. This check is used to inspect cortical sources. The check examines whether artifact energy, which should not belong to the frequency band, has appeared. Taking electrosurgical unit (ESU) interference as an example, the specific implementation of this check is as follows: Define the characteristic frequency range of the electrosurgical unit In one embodiment, the range may be set according to the operating frequency of common electrosurgical devices (e.g., 150Hz, 450Hz).

[0064] Computation of cortical sources Power spectral density within the analysis time window This calculation can be performed, for example, using the Welch average periodogram method.

[0065] Calculation in Energy leakage ratio within range : ; In the formula, Energy leakage ratio is a dimensionless index that quantifies the proportion of energy of cortical source signals within a specific artifact frequency band. For frequency variables; The characteristic frequency range of predefined electrosurgical artifacts, such as 150Hz, 450Hz; Cortical source signal In frequency Power spectral density at; The sampling rate of the signal acquisition unit; This indicates the cortical source signal within the characteristic frequency range of electrosurgical treatment. Total energy within; This indicates the cortical source signal from DC to the Nyquist frequency ( The total energy across the entire frequency range of ).

[0066] The molecule represents the signal energy of the cortical source within the characteristic frequency range of electrosurgical treatment.

[0067] The denominator represents the total energy of the cortical source across the entire Nyquist frequency range.

[0068] Calculated If the value exceeds the preset leakage threshold This indicates that the electrosurgical signal contaminated the cortical source.

[0069] S143: Generate the Source Purity Index (SPI). This step combines the multiple verification metrics calculated in S141 and S142 ( , (etc.) are integrated into a single, standardized quality indicator. This fusion is achieved through a pre-defined mapping function. Finish.

[0070] In one embodiment, This can be implemented as a weighted combination function. First, each verification index is converted into a single quality score of 0-1: Independence quality score Leakage mass fraction Then, the final result is obtained by weighted summation. : ; In the formula, Source purity index; and These are preset weighting coefficients used to adjust the relative importance of different quality scores when calculating the SPI, and satisfy the following conditions: + =1; The independence quality score is derived from the source independence verification result and is used to quantify the degree of separation between cortical sources and artifact sources. The leakage mass fraction is determined by energy leakage verification (e.g., energy leakage ratio). The results are derived from this and are used to quantify the extent to which pseudo-trace source energy leaks into the cortical source frequency band.

[0071] The final generated The result is output to step S160 for judgment, and displayed in step S190 together with the ADI generated in step S180. =100 indicates extremely high separation quality (purity). =0 indicates that the separation has completely failed (unreliable).

[0072] Step S150 is performed by the inter-source dynamic calibration module. This step is the core of the invention's adaptive capability. Its function is to use the pseudo-trace source separated in step S130 (BCSS) as a calibration signal to form a feedback closed loop, so as to continuously optimize the separation performance of the real-time BCSS module.

[0073] This step S150 may specifically include the following sub-steps: S151: Performs real-time capture of artifact templates. The inter-source dynamic calibration module monitors in real-time signals identified as artifact sources, such as myoelectric power, output by the real-time BCSS module. and electrosurgical source .

[0074] muscle power source For example, this inter-source dynamic calibration module operates within a sliding time window. Calculate its average power or energy. : ; In the formula, For the current moment The average power or energy of the muscle power source was calculated. The preset length of the sliding time window used for integration calculation, for example, 1 second; The current moment; For the past moment The amplitude of the muscle power signal; In the time window The time variable for integration within the time interval.

[0075] This module 134 will calculate the... With a preset calibration energy threshold Comparison. When At that time, the system determines that it is within the current time window. Internal muscle power signals It is a high signal-to-noise ratio, high-purity electromyography artifact template. The logic that triggers this condition is that a high-energy, successfully separated artifact source is itself a high-quality sample of this type of artifact.

[0076] S152: Perform feedback updates to the BCSS model. Once the pseudo-template (i.e., ...) is captured in S151... time (Signal segment), the inter-source dynamic calibration module uses this template signal to update the internal model parameters of the real-time BCSS module.

[0077] Specifically, this template is used to re-estimate the dynamic constraint model defined in step S132. Taking the muscle power source as an example, its dynamic constraint model is a prior probability density function. , from parameter set (For example, the shape and scale parameters of the generalized Gaussian distribution) are defined. The inter-source dynamic calibration module utilizes the captured template. This method uses the principle of maximum likelihood estimation to solve for a new set of parameters. The objective of this solution process can be expressed as: ; In the formula, This is a new set of parameters obtained through maximum likelihood estimation, used to update the dynamic constraint model of the muscle power source; It is an optimization operator, meaning it finds the parameter that maximizes the subsequent expression. The value; These are a set of variable parameters representing the probability density function model of the muscle source, such as the shape and scale parameters of the generalized Gaussian distribution; The set of discrete time points contained in the captured electromyographic artifact template signal segment; It is a logarithmic function; A parameter set used to describe the dynamic characteristics of muscle power sources Defined prior probability density function; For within the artifact template The sampled value of the muscle power signal at the corresponding moment; For set An index in; For index The specific sampling time corresponding to that time.

[0078] This set of updated parameters The feedback is fed back to the BCSS optimizer (e.g., the natural gradient algorithm) in step S133.

[0079] The BCSS optimizer (belonging to the real-time BCSS module 132) will use this updated dynamic constraint in subsequent iterations to solve the separation matrix. This results in a separation matrix. The process is adjusted or reconverged to better match the current (or newly emerging) electromyographic artifact characteristics. This update process can be represented as: ; In the formula, For the next time step The separation matrix obtained after the update; This is a function that represents the process by which the BCSS optimizer performs an iterative update using the new parameters; For the current moment The separation matrix; This is the set of new model parameters learned from the artifact template and fed back to the optimizer.

[0080] Through this feedback loop formed by S151 and S152, the system of the present invention can learn and adapt to patient-specific or time-varying artifact morphologies (e.g., interference from different intensities of electromyographic tension or different types of electrosurgical units), thereby continuously maintaining the high separation accuracy of the BCSS algorithm and ensuring the cortical source Purity.

[0081] Step S170 is executed by the anesthesia depth index calculation module. This step has a prerequisite: the source purity index must have been determined in step S160. Quality exceeding the preset threshold .

[0082] When this condition is met, the system determines that the cortical source isolated in step S130 is pure and reliable. Subsequently, the system uses only the pure cortical source to extract anesthesia depth-related physiological characteristics. All signals marked as artifacts in step S130 are discarded and not included in the calculation of this step.

[0083] Step S170 may specifically include the following sub-steps: S171: Extract cross-frequency phase-amplitude coupling features. This feature is used to quantify the coupling strength between different neural oscillation frequency components in cortical sources, which is related to brain consciousness and information integration levels.

[0084] The cortical source signal is passed through at least two digital bandpass filters. In one embodiment, the first filter is used to extract the low-frequency phase signal. (For example, (Band, 0.5Hz, 4Hz); the second filter is used to extract high-frequency amplitude signals. (For example, (Bands, 8Hz, 13Hz).

[0085] Applying the principle of Hilbert transform, respectively from... Extract its instantaneous phase and from Extract its instantaneous amplitude envelope .

[0086] Within an analysis time window (Include sampling points Within ), calculate the phase-amplitude modulation index. In one embodiment, This can be obtained by calculating the average length of the composite vector: ; In the formula, The phase-amplitude coupling modulation index is a scalar value used to quantify the modulation strength of the phase of the low-frequency oscillation on the amplitude envelope of the high-frequency oscillation. The total number of sampling points within an analysis time window is a positive integer; The index of discrete time points within the analysis time window ranges from 1 to... ; For high-frequency amplitude signals in the first... Time points The instantaneous amplitude value; For the low-frequency phase signal at the 1st Time points The instantaneous phase value, in radians; is the base of the natural logarithm; It is the imaginary unit.

[0087] S172: Extracting source signal waveform morphology features. This feature is used to analyze the non-sinusoidal waveform characteristics of the dominant oscillations in cortical sources. These morphological features (such as asymmetry) are related to changes in neuronal firing patterns induced by anesthetic drugs (such as GABAergic agonists).

[0088] The cortical source signal (or the filtered component of its dominant oscillation) is detected to locate continuous waveform periods and identify the peaks and troughs of each period.

[0089] Within an analysis time window, at least two morphological features are calculated: Rise and fall time symmetry: quantifies the ratio between the time it takes for a waveform to rise from a trough to a peak (rise time) and the time it takes for it to fall from a peak to a trough (fall time).

[0090] Crest-trough steepness: quantifies the ratio or difference between the curvature (sharpness) of a waveform near its crest and the curvature (smoothness) near its trough.

[0091] S173: Extract source domain burst suppression ratio. This feature is used to quantify the degree of inhibition of cortical activity under deep anesthesia.

[0092] An amplitude threshold is set on the pure cortical source signal. (Suppression threshold). In one embodiment, this threshold may be set to (e.g., ±5 μV).

[0093] Detection The signal amplitude is consistently lower than The time period.

[0094] Filtering for durations exceeding a minimum inhibition duration The time period was identified as the inhibition period.

[0095] In a total analysis window Within this period, the total time of all identified inhibition periods is summed up and denoted as . .

[0096] Through the following calculations : ; In the formula, Source domain burst suppression ratio is a percentage value that represents the proportion of time that brain electrical activity is in a suppressed state over a relatively long analysis period. The total cumulative duration of all identified inhibition periods within the total analysis window, in seconds; This is the length of the total analysis window used to calculate the burst suppression ratio.

[0097] Because of this It is calculated on a pure cortical source with the myoelectric power source removed, avoiding the problem of myoelectric artifacts being incorrectly identified as burst activity in traditional methods, thus accurately reflecting the level of EEG inhibition under deep anesthesia.

[0098] At the end of S170, the multiple features extracted in S171 to S173, along with other source domain features (such as source domain spectral entropy), are combined into a multidimensional feature vector. , used for the exponential synthesis of S180.

[0099] Step S180 is performed by the anesthesia depth index calculation module. This module receives a multidimensional feature vector from step S170 that combines multiple source domain physiological features. .

[0100] In step S180, the feature vector Input into a pre-trained fusion model In this context, a single, standardized anesthesia depth index is synthesized: ; In the formula, This is an index of the depth of anesthesia. It is a multidimensional feature vector; This is a pre-trained fusion model.

[0101] The fusion model It was obtained through offline training using a large amount of clinical anesthesia data (including synchronously acquired multi-channel EEG signals and corresponding clinical anesthesia depth scores, such as OAA / S scores).

[0102] In one embodiment, the fusion model It is a multivariate linear regression model. The calculation method of this model can be described in words as: [the model consists of] feature vectors. Each source domain feature in Assign a pre-trained weight coefficient Then, the products of all features and their corresponding weights are summed in a weighted manner, and a bias term is added on top of this sum. Thus, the final result is obtained. value.

[0103] In another embodiment, the fusion model It is a non-linear model, such as a support vector machine regression model or a gradient boosting tree model. Such non-linear models are able to capture features. The complex nonlinear relationships between them, mapped to the final value.

[0104] The output of step S180 The values ​​are normalized to a preset range (e.g., 0-100). Within this range, 100 corresponds to a fully conscious state, 40-60 corresponds to an appropriate depth of general anesthesia, and 0 corresponds to complete suppression of brain activity.

[0105] In step S190, the system of the present invention simultaneously presents two co-working indices to the clinician on a user interface (e.g., a monitor screen): Anesthesia Depth Index: (from step S180) Reflects the patient's anesthesia depth in real time.

[0106] Source purity index: (from step S140) reflects the current purity in real time. Reliability of the readings.

[0107] This dual-exponential collaborative output method solves the problem of the inability to quantify the reliability of exponents in existing technologies.

[0108] when When the value is high, it indicates that the calculations in steps S170 and S180 are based on a pure cortical source, and the ADIADI reading is highly reliable.

[0109] when The system can alert clinicians when a transient decrease occurs due to strong artifacts (e.g., electrosurgical initiation or strong patient movement). This alert may include, for example, [the system will alert clinicians to the presence of artifacts]. The readings are displayed in a specific color (such as yellow or red) or made to flash.

[0110] This warning serves as a reminder to clinicians that currently... The readings appear to have been contaminated by artifacts, reducing their accuracy; therefore, they should be interpreted with caution.

[0111] Meanwhile, as in step S150, the inter-source dynamic calibration module is triggered when a high-energy artifact source is detected, and begins to adaptively adjust the model parameters of the real-time BCSS module to restore the system's ability to suppress the interference as quickly as possible, thereby enabling... Rebound to high levels, ensuring Long-term accuracy of readings.

[0112] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent monitoring of anesthesia depth index based on electroencephalogram (EEG) signals, characterized in that, The method includes the following steps: Acquire multi-channel EEG signals and perform signal preprocessing on the multi-channel EEG signals; The preprocessed multi-channel EEG signal is separated into multiple independent source signals by applying a real-time biophysical constraint source separation algorithm. Based on the preset source signal spatial projection topographic map constraint and time series statistical characteristic constraint, the source signals are automatically labeled as cortical sources or artifact sources. Perform source domain integrity verification to generate a quantified source purity index; Perform dynamic calibration between sources, using the pseudo-trace source as a calibration template, and update the dynamic constraint model parameters in the real-time biophysical constraint source separation algorithm. Determine whether the source purity index is higher than a preset quality threshold. If it is, extract physiological features only from the cortical source and synthesize them into an anesthesia depth index. On the user interface, the anesthesia depth index and the source purity index are displayed together.

2. The intelligent monitoring method for anesthesia depth index based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The source signal spatial projection topographic map constraint includes: a preset prior spatial probability map generated based on a standard head anatomy model for cortical sources, and a preset prior spatial probability map defined based on the physical characteristics of the pseudo-trace sources for pseudo-trace sources. The time series statistical characteristic constraints include: predefining prior probability density functions with different distribution characteristics for cortical sources and artifact sources respectively.

3. The intelligent monitoring method for anesthesia depth index based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The steps for performing inter-source dynamic calibration specifically include: Real-time monitoring of the signal energy of the pseudo-trace source; When the signal energy exceeds the preset calibration energy threshold, the corresponding pseudo-trace source signal segment is used as the calibration template. The parameters of the dynamic constraint model are re-estimated using the calibration template and the maximum likelihood estimation algorithm, and the real-time biophysical constraint source separation algorithm is adjusted with the updated parameters.

4. The intelligent monitoring method for anesthesia depth index based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The steps for performing source domain integrity verification specifically include: The thoroughness of separation is assessed by calculating the residual mutual information between the cortical source and the artifact source. The degree of frequency band contamination is assessed by calculating the energy leakage ratio of the cortical source within the artifact characteristic frequency range. The results of the residual mutual information and energy leakage ratio are fused to generate the source purity index.

5. The intelligent monitoring method for anesthesia depth index based on electroencephalogram (EEG) signals according to claim 4, characterized in that, The steps for fusing the results of the residual mutual information and the energy leakage ratio include: The residual mutual information and energy leakage ratio are compared with preset thresholds and converted into standardized independence mass scores and leakage mass scores. The source purity index is obtained by weighted summation of the independence mass fraction and the leakage mass fraction.

6. The intelligent monitoring method for anesthesia depth index based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The physiological characteristics include: Trans-frequency phase-amplitude coupling characteristics characterizing the coupling strength between different neural oscillation frequency components; The waveform morphology of the source signal that characterizes the dominant non-sinusoidal oscillation. Source domain burst suppression ratio calculated on the cortical source.

7. The intelligent monitoring method for anesthesia depth index based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The steps for synthesizing the anesthesia depth index include: The physiological features are combined into a multidimensional feature vector, and the multidimensional feature vector is input into a fusion model obtained by offline training with clinical anesthesia data to output the anesthesia depth index.

8. The intelligent monitoring method for anesthesia depth index based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The steps of the collaborative display include: The anesthesia depth index and the source purity index are displayed simultaneously on the user interface, and the source purity index characterizes the reliability of the anesthesia depth index.

9. The intelligent monitoring method for anesthesia depth index based on electroencephalogram (EEG) signals according to claim 8, characterized in that, The method further includes: When the source purity index falls below a preset warning threshold due to artifact interference, a warning is issued on the user interface, indicating that the reliability of the currently output anesthesia depth index has decreased.

10. An intelligent monitoring system for anesthesia depth index based on electroencephalogram (EEG) signals, characterized in that, The intelligent monitoring method for anesthesia depth index based on electroencephalogram signals according to any one of claims 1-9, the system comprising: Multi-channel electrode array for acquiring multi-channel EEG signals; A signal acquisition unit, electrically connected to the multi-channel electrode array, is used to process the acquired signals to output multi-channel digital signals; Central processing unit, connected to the signal acquisition unit, the central processing unit comprising: The signal preprocessing module is used to perform signal preprocessing on the multi-channel digital signal; The real-time BCSS module is used to apply a real-time biophysical constraint source separation algorithm to separate the preprocessed multi-channel EEG signals into cortical sources and artifact sources. The source domain integrity verification engine is used to perform source domain integrity verification to generate a quantified source purity index. The inter-source dynamic calibration module is used to perform inter-source dynamic calibration, using the pseudo-trace source as a calibration template, and updating the real-time BCSS module accordingly. An anesthesia depth index calculation module is used to extract physiological characteristics from the cortical source and synthesize the anesthesia depth index when the source purity index is higher than a preset quality threshold. A user interface, connected to the central processing unit, is used to collaboratively display the anesthesia depth index and the source purity index.

Images