A method for monitoring the state of sevoflurane anesthesia
By using multi-channel EEG signal frequency band decomposition and hierarchical clustering algorithms, combined with the specific mechanism of action of sevoflurane, an improved hierarchical clustering discrimination model and dynamic feedback mechanism were constructed. This solved the accuracy problem of existing equipment in monitoring sevoflurane anesthesia, and achieved more efficient identification and stable monitoring of anesthesia status.
Patent Information
- Application Number
- CN202611195279.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-07
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2046-08-07
AI Technical Summary
Existing anesthesia depth monitoring equipment has difficulty accurately reflecting the true changes in sevoflurane anesthesia status because it uses a fixed model designed for general anesthesia scenarios and fails to adapt to the specific EEG characteristics of sevoflurane.
Multi-channel EEG signal frequency band decomposition was used to extract representative fused signals, calculate the PAC features of frequency band combinations, and perform real-time classification using a hierarchical clustering algorithm. The specific mechanism of action of sevoflurane was combined for weighted processing to construct an improved hierarchical clustering discrimination model, and a dynamic feedback weight correction mechanism was set up.
It improves the ability to identify and differentiate sevoflurane anesthesia status, enhances the correlation between the discrimination results and clinical behavioral classification, and improves the stability and real-time performance of monitoring, making it suitable for rapid and accurate assessment of sevoflurane anesthesia status.
Smart Images

Figure CN122701288B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the medical field, and more specifically, to a method for monitoring sevoflurane anesthesia. Background Technology
[0002] During general anesthesia, the core objective of monitoring the depth of anesthesia is to assess the degree of central nervous system inhibition by anesthetic drugs in real time, thereby optimizing dosing regimens, reducing the risks associated with intraoperative awareness and excessive anesthesia, and providing objective evidence for pharmacodynamic studies. In recent years, perioperative EEG monitoring has evolved from relying solely on the post-processed index to a comprehensive assessment method combining raw EEG, spectral data, and quantitative EEG parameters. Different anesthetic drugs exhibit different EEG characteristic spectra, thus targeted monitoring based on drug mechanisms of action has clear clinical significance.
[0003] Currently, the most widely used devices for monitoring the depth of anesthesia in clinical practice include the Bispectral Index (BIS) and similar processed EEG (pEEG) devices, which extract features, compress parameters, and map models from EEG signals to output a single comprehensive index that reflects the patient's level of consciousness inhibition. Existing research indicates that processed EEG indices have application value in some intravenous anesthesia and routine scenarios; however, their core parameters are mostly fixed, universal models, and their adaptability to different anesthetic drugs, patients of different ages, and different EEG trajectories remains limited.
[0004] Sevoflurane is a commonly used volatile inhaled anesthetic in clinical practice. Its anesthetic effect is not simply manifested as a "weakening" of the overall amplitude of electroencephalogram (EEG), but rather as a reconstruction of specific neural oscillation patterns and cross-frequency coupling relationships. Publicly available research shows that under sevoflurane anesthesia, the prefrontal cortex EEG not only exhibits changes in power spectrum distribution but also shows reshaping of low-frequency to high-frequency cross-frequency coupling patterns, including dynamic changes in delta-alpha, alpha-gamma, and theta-gamma coupling relationships. These changes are closely related to the drug's effect on the cortico-thalamic network and the enhancement of inhibitory neurotransmission. Therefore, if a fixed model constructed for general anesthesia scenarios is still used for monitoring, it is often difficult to accurately reflect the true changes in the anesthetic state under sevoflurane anesthesia. Summary of the Invention
[0005] This invention addresses the technical problems existing in the prior art by providing a method for monitoring the state of sevoflurane anesthesia, which can overcome the deficiencies of the prior art.
[0006] This invention provides a method for monitoring the state of sevoflurane anesthesia, comprising: Step S1: Under sevoflurane anesthesia, continuously collect EEG signals from multiple channels of the subject; Step S2: Decompose the EEG signal of each channel into frequency bands to obtain bandpass signals in multiple frequency bands, and extract the representative fusion signals of each frequency band in the multiple channel space from the bandpass signals of each frequency band in multiple channels. Step S3: Combine multiple frequency bands to obtain all frequency band combinations. Based on the representative fused signal of each frequency band, calculate the PAC feature of each frequency band combination. Step S4: Based on the specific action mechanism of sevoflurane, the PAC features of each frequency band combination are weighted to obtain a weighted PAC feature vector. Step S5: The weighted PAC feature vector is classified in real time using a hierarchical clustering algorithm to obtain the preliminary classification result of the anesthesia status in the current sliding analysis window.
[0007] This invention provides a method for monitoring sevoflurane anesthesia status. It involves frequency band decomposition of the acquired EEG signals from each channel, extracting the fused signals of each frequency band across multiple channels from the bandpass signals of each band. Based on the fused signal of each frequency band, the PAC features of each frequency band combination are calculated and weighted to obtain a weighted PAC feature vector. A hierarchical clustering algorithm is then used to classify the weighted PAC feature vector in real time, yielding a preliminary classification result of the anesthesia status within the current sliding analysis window. This invention uses phase-amplitude coupling features, which reflect the modulation relationship between low-frequency phase and high-frequency amplitude, as the core characterization indicator. Through a method of "full-band calculation + key coupling weighting and highlighting," the feature vector can more effectively characterize the most critical neural modulation patterns during sevoflurane anesthesia, thus improving the ability to identify and differentiate sevoflurane anesthesia status compared to general EEG features. Attached Figure Description
[0008] Figure 1 A flowchart of a method for monitoring sevoflurane anesthesia status according to an embodiment of the present invention; Figure 2 This is a flowchart of multi-channel full-band phase amplitude coupling feature extraction and sevoflurane-specific weighting according to an embodiment of the present invention; Figure 3 This is a flowchart of hierarchical clustering real-time discrimination and dynamic feedback correction according to an embodiment of the present invention; Figure 4 This is a schematic diagram of a sevoflurane anesthesia monitoring system provided in one embodiment of the present invention. Detailed Implementation
[0009] 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, 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. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined with each other to form feasible technical solutions. Such combinations are not constrained by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.
[0010] To address the problems in the background art, this invention provides a method for monitoring sevoflurane anesthesia status. This method does not rely on a discriminant model based on universal anesthesia depth features. Instead, it focuses on the specific effects of sevoflurane on the prefrontal cortex's neural oscillatory coupling patterns. It constructs sevoflurane-specific weighted PAC features, an improved hierarchical clustering discriminant model oriented towards sevoflurane anesthesia grading, and a dynamic feedback weight correction mechanism, thereby achieving rapid, accurate, and interpretable assessment of sevoflurane anesthesia status. As a volatile inhaled anesthetic, sevoflurane primarily enhances GABA_A receptor-mediated inhibitory neurotransmission and modulates cortico-thalamic circuit activity. During anesthesia, it exhibits a different EEG oscillatory remodeling trajectory compared to universal intravenous anesthetics. Therefore, this invention employs a targeted feature construction and discriminant strategy based on the mechanism of sevoflurane in its method design.
[0011] Figure 1 A flowchart of a method for monitoring the state of sevoflurane anesthesia according to an embodiment of the present invention is shown, as follows: Figure 1 As shown, the method includes the following steps: Step S1: Under sevoflurane anesthesia, continuously collect EEG signals from multiple channels of the subject.
[0012] First, raw EEG signals from multiple channels are continuously acquired from the subject under sevoflurane anesthesia. In this embodiment of the invention, four channels of EEG signals are acquired in real time: channel 1 (Fp1), channel 2 (Fp2), channel 3 (F7), and channel 4 (F8). The reference electrode is placed on the right earlobe (A2), and the sampling rate is preferably set to 500 Hz. The acquired raw EEG signals are stored in a 10-second circular buffer in 24-bit signed integer format via DMA.
[0013] Let the coverage duration of the circular buffer be T seconds, the number of channels be C, and the sampling rate be Fs. Then the total number of data points N in a single circular buffer is: .
[0014] Subsequently, the acquired raw EEG signals were preprocessed and artifacts were detected. Specifically, the preprocessing included digital bandpass filtering of 0.5-45 Hz and power frequency notch filtering of 50 Hz on the raw EEG signals.
[0015] In this embodiment of the invention, artifact detection includes: if the amplitude of the EEG signal of any channel at any time point in the current sliding artifact detection window is not within a preset amplitude range, or the difference between the amplitudes of the EEG signals of two adjacent time points exceeds a preset difference threshold, then the EEG signal in the current detection window is marked as invalid and removed; otherwise, it is marked as valid and retained.
[0016] Specifically, a sliding artifact detection window with a length of 250ms is used. For any detection window, as long as any sampling point in any channel satisfies... More than ±100μV or If the value exceeds 50 μV / ms, the 250 ms data segment will be marked as invalid and will not be included in subsequent feature calculations.
[0017] Artifact detection can be expressed as: ;
[0018] in, This represents the signal amplitude of channel c at time t; =1 / Fs is the sampling time interval (if the sampling rate Fs = 500Hz, then...). =2ms, at this time the "difference between adjacent points" is (t) (t 2ms) When A(t)=1, the 250ms data segment is marked as invalid and will not be included in subsequent feature calculations.
[0019] Step S2: Decompose the EEG signal of each channel into frequency bands to obtain bandpass signals in multiple frequency bands, and extract the representative fusion signals of each frequency band in the multiple channel space from the bandpass signals of each frequency band in multiple channels.
[0020] Understandably, see Figure 2 Parallel band decomposition was performed on the preprocessed EEG signals of the four channels (Fp1, Fp2, F7, F8) to obtain five standard physiological frequency band signals (called bandpass signals): Delta, Theta, Alpha, Beta, and Gamma. This provides the foundation for subsequent full-band PAC feature construction. Specifically, for the EEG signal of the c-th channel... The frequencies are decomposed into five standard physiological frequency bands by a set of digital bandpass filters: Delta (δ): 1-4 Hz, Theta (θ): 4-8 Hz, Alpha (α): 8-13 Hz, Beta (β): 13-30 Hz, and Gamma (γ): 30-45 Hz.
[0021] Let the preprocessed c-th channel EEG signal be... After frequency band decomposition, the bandpass signal of this channel in frequency band b is obtained. ,in , .
[0022] To preserve multi-channel spatial information while controlling computational complexity, a strategy of "spatial fusion first, then PAC computation" is adopted. For any frequency band b, the time series of the four channels in that frequency band are arranged column-wise into a matrix: ;
[0023] in, Indicates the first Each channel in the frequency band The time series column vector below, ; This indicates the number of sampling points within the current analysis time window. When the analysis window length is 2 seconds and the sampling rate is 500 Hz, For the matrix After removing the mean from each column, calculate its covariance matrix: ;
[0024] Seeking The largest eigenvalue corresponds to the eigenvector , The expression is: ;
[0025] in, Each component represents the four channels in the frequency band. Linear combination weights during fusion. The matrix... Projected onto feature vector Up, obtain the frequency band Representative fusion signals in multi-channel space: ;
[0026] in This represents the fused time series corresponding to frequency band b. The above process is performed on each of the five frequency bands to obtain the fused signals corresponding to the five frequency bands, which are then used for subsequent PAC feature calculation.
[0027] This serves as the representative fused signal for frequency band b in the multi-channel space. The above process is performed on each of the five frequency bands to obtain five fused signals. This operation maximizes the preservation of the oscillation modes with the largest variance shared by multiple channels, effectively suppressing channel-specific noise. The same operation is performed on other frequency bands such as Gamma, ultimately yielding a fused signal corresponding to the five frequency bands. .
[0028] Step S3: Combine multiple frequency bands to obtain all frequency band combinations. Based on the representative fused signal of each frequency band, calculate the PAC feature of each frequency band combination.
[0029] Among them, for the five frequency bands Delta, Theta, Alpha, Beta and Gamma, two frequency bands constitute a frequency band combination, generating a total of 10 frequency band combinations. Each frequency band combination includes a low frequency band and a high frequency band. For any frequency band combination, the modulation index of the frequency band combination is calculated as a PAC feature based on the representative fused signal of the low frequency band and the representative fused signal of the high frequency band using the phase binning method.
[0030] Specifically, the PAC feature is calculated only for the low-frequency phase-high-frequency amplitude combination. Let L represent the low-frequency band and H represent the high-frequency band, satisfying the relationship between the center frequency of the band or the upper limit of the frequency band. A total of C(5,2)=10 frequency band combinations of PAC features were generated.
[0031] Obtain the representative fused signal of the low-frequency band L in any frequency band combination at time t. Representative fused signal of high-frequency band H at time t ;right Perform Hilbert transform to extract the instantaneous phase sequence. and to Perform Hilbert transform to extract the instantaneous amplitude envelope sequence. .
[0032] The modulation index is calculated using the phase binning method within a 2-second sliding time window. : ;
[0033] In the phase binning method, only the low-frequency instantaneous phase is considered. Perform interval division; high-frequency instantaneous amplitude envelope Instead of dividing the data into intervals, the data is categorized into corresponding phase bins for average statistical analysis based on the principle of time synchronization. That is, the average value at a given time t... Belongs to The determined phase box , It is exist After dividing the phase range into M intervals, the corresponding intervals are... The probability distribution formed by the average value, It is its entropy value, where M is a positive integer greater than 1.
[0034] Iterate through each frequency band combination, obtain the PAC features of 10 frequency band combinations, and output 10 low-frequency phase-high-frequency amplitude PAC combinations composed of five frequency band combinations, as follows: .
[0035] Based on this, a 10-dimensional PAC feature vector is formed: ;
[0036] in The corresponding PAC coupling characteristics of Theta-Gamma.
[0037] Step S4: Based on the specific action mechanism of sevoflurane, the PAC features of each frequency band combination are weighted to obtain a weighted PAC feature vector.
[0038] Understandably, based on the known pharmacological mechanism of sevoflurane, it can inhibit and modulate the cross-frequency coupling relationship between Theta and Gamma oscillations in the prefrontal cortex by enhancing GABA. Therefore, among all PAC features, the PAC sub-features corresponding to Theta-Gamma coupling are given the highest weight and used as the core discriminant features.
[0039] Specifically, construct a weight vector by defining a weight vector w corresponding to the feature dimension. , This represents the weight coefficient of the i-th frequency band combination, where the Theta-Gamma frequency band combination has the largest weight coefficient. The weight corresponding to the Theta-Gamma coupling feature is set to 2.0, and the weights of the others are set to 1.0.
[0040] Based on the weight vector, the PAC features of the 10 frequency band combinations are weighted, and the weighted PAC feature vector is expressed as follows: ;
[0041] Where ⊙ denotes element-wise multiplication. This represents the weighted PAC eigenvector. This represents the PAC characteristic sequence of 10 frequency band combinations.
[0042] Step S5: The weighted PAC feature vector is classified in real time using a hierarchical clustering algorithm to obtain the preliminary classification result of the anesthesia status in the current sliding analysis window.
[0043] Understandably, see Figure 3 By using an improved hierarchical clustering algorithm, the weighted PAC feature vector calculated based on multi-channel fusion signals is classified in real time, and the preliminary classification results of the anesthesia state are output.
[0044] In one embodiment of the present invention, a clustering template is constructed, wherein multi-channel full-band EEG signals within multiple sliding analysis windows of the subject's history are acquired, and the multi-channel full-band EEG signals within each sliding analysis window are taken as a sample. The weighted PAC feature vector of each sample is extracted, and the anesthesia state category label of each sample is obtained through MOAA / S score. Multiple samples corresponding to each category of anesthesia state are randomly selected, and the template center of each category of anesthesia state is calculated based on the weighted PAC feature vector of multiple samples. The template center of each category of anesthesia state is also a 10-dimensional PAC feature.
[0045] Specifically, multiple training samples are acquired and divided into three categories—conscious, moderately sedated, and deeply sedated—based on the MOAA / S score. For each category, K representative samples are selected as seed points. Using these seed points as initial sub-clusters, the remaining samples are assigned to the nearest seed cluster of the same category based on weighted Euclidean distance. Subsequently, agglomerative hierarchical clustering is used within each category to progressively merge sub-clusters belonging to the same category until each category forms a principal cluster. Finally, the mean of the weighted PAC feature vectors of all samples within each of the three principal clusters is calculated to obtain the template centers of the three categories. .
[0046] Calculate the weighted PAC eigenvector of the current sliding analysis window. With each template center Weighted Euclidean distance: ;
[0047] in, This represents the weight coefficient of the i-th dimension of the PAC feature in the weighted PAC feature vector. Consistent with the weights used in feature computation, this ensures that the distance metric is also aligned with the sevoflurane mechanism. This represents the i-th dimension of the PAC feature in the weighted PAC feature vector of the current sliding analysis window. This represents the i-th dimension PAC feature in the center of the k-th template. This represents the weighted Euclidean distance between the weighted PAC eigenvector of the current sliding analysis window and the center of the k-th template. The anesthesia state category with the smallest weighted Euclidean distance to the weighted PAC feature vector of the current sliding analysis window is taken as the preliminary classification result of the anesthesia state in the current sliding analysis window. .
[0048] To ensure the stability of real-time discrimination results, this invention employs a two-level confirmation mechanism based on a sliding analysis window. The sliding analysis window length for PAC features is set to 2 seconds, and the window movement step size is 1 second. This means the system updates the analysis results every 1 second and outputs the preliminary classification result corresponding to the current sliding analysis window.
[0049] The system for the most recent consecutive A consistency check was performed on the preliminary classification results. The optimal value is 5. That is, in the nth update, the most recent consecutive... Are the preliminary classification result sequences corresponding to each sliding analysis window completely consistent? If the most recent consecutive If all preliminary classification results are completely consistent, the preliminary classification result corresponding to the current sliding analysis window is confirmed as valid; if the most recent consecutive If any of the preliminary classification results differ, the current round is considered "inconsistent," and the system retains the previously confirmed classification result and does not update the output.
[0050] Furthermore, the system sets an inconsistent persistence counter. This is used to accumulate the number of consecutive rounds of judgment where "inconsistency" occurs. When a round has the most recent consecutive... When the initial classification results of the sliding analysis windows are not completely consistent, let = +1; when a certain round of most recent consecutive When the initial classification results of the sliding analysis windows are completely consistent and form a new valid state, let =0. When Reaching the preset threshold At this point, the system determines that the current discrimination model lacks sufficient stability in discriminating real-time EEG features and triggers a subsequent feedback weighting mechanism. The preferred value is 5.
[0051] Once the feedback mechanism is triggered, the system retrieves the most recent response before the trigger. The weighted PAC feature vectors of each sliding analysis window and their corresponding preliminary classification results are used as feedback adjustment inputs, where The optimal value is 30, corresponding to the real-time analysis results from approximately the last 30 seconds. Based on this recent feature matrix, the intra-class dispersion and inter-class separability of each PAC sub-feature are further analyzed, and the weight vector is updated in a constrained and smooth manner.
[0052] Once the feedback mechanism is triggered, the system automatically executes the following optimization process in the background: extract the most recent data before the feedback was triggered. The feature matrix X_recent for each time window and its corresponding preliminary classification results distribution, where The optimal value is 30. Analyze the intra-class dispersion and inter-class separability of each feature in X_recent to quantify the contribution of each PAC feature to distinguishing different anesthetic states. Based on the analysis results, the weight vector w is adjusted in a directional and smooth manner: for features that still maintain high discriminative power in the current data, their weights are appropriately increased; for features exhibiting high intra-class dispersion or low discriminative power, their weights are appropriately decreased. The weight adjustment maintains the stability of the total weight, and the magnitude of each adjustment is limited to prevent drastic oscillations.
[0053] Specifically, let the feature matrix of recently triggered feedback be... ,in Let D be the number of recent sliding analysis windows, and D=10 be the dimension of the PAC features. Based on the preliminary classification results of each sliding analysis window, the samples are divided into several state groups. For each state group, for the i-th PAC feature, its inter-class variance is calculated. Within-class variance : ;
[0054] ;
[0055] Where K is the number of current state categories, Let k be the number of samples in the k-th class. Let be the mean of the k-th class on the i-th feature. Let be the population mean of all samples on the i-th feature. The discriminative power score for the i-th feature is defined as: ;
[0056] in, To prevent the use of tiny positive numbers with a denominator of zero. If A relatively large value indicates that this feature simultaneously possesses the properties of "large inter-class differences and small intra-class fluctuations," contributing significantly to distinguishing different anesthetic states; if A smaller value indicates that the feature has weak discriminative power at the current stage and contributes little to state differentiation. Based on this, it can be determined which PAC sub-feature contributes more and which contributes less. Subsequently, the weight vector is updated in a directional and smooth manner according to the normalized score. The smooth update formula is: ;
[0057] in, Let be the current discriminative power score of the PAC features for the i-th frequency band combination. The learning rate is small (e.g., 0.05), and D is the total number of features (10).
[0058] Using the fine-tuned new weight vector The weighted PAC feature vector of the current sliding analysis window is recalculated and fed into the discriminant model for re-discrimination, yielding a preliminary classification result of the updated anesthesia status corresponding to the current time window. After completing the weight fine-tuning, the inconsistency counter H is cleared, the feedback flag is reset, and the classification result buffer is cleared. Then, a new monitoring cycle begins with the optimized new weights.
[0059] As an output stabilization control strategy, when the final classification result undergoes a state switch, the switched final classification result is locked for at least 30 seconds. During the locked period, even if subsequent sliding analysis windows generate different preliminary classification results for anesthesia states, the current final classification result remains unchanged until the locked period ends, at which point state updates resume. This state locking mechanism works in conjunction with the aforementioned two-level monitoring mechanism and weight adjustment feedback mechanism. While allowing for adaptive adjustments within the discriminant model, it ensures the continuity of the clinical interface output, preventing rapid state reversals caused by internal algorithm adjustments or short-term signal fluctuations, thereby improving output stability. This mechanism also works in conjunction with the aforementioned dynamic feedback mechanism, ensuring the continuity of the clinical interface output while allowing for internal adaptation within the discriminant model, preventing interference to medical staff due to internal algorithm adjustments or short-term signal fluctuations.
[0060] Finally, the confirmed effective anesthesia status (State_confirmed) is used as the final judgment result. This final result, the real-time confidence score calculated based on cluster distance, and the trend curves of core biomarkers (such as Theta-Gamma PAC) together constitute the monitoring information, which is updated in real time on the local display terminal of the system and can be pushed to the central monitoring system via the network interface.
[0061] See Figure 4 The present invention also provides a sevoflurane anesthesia status monitoring system, which includes six core modules, specifically including: The EEG signal acquisition module 401 is used to continuously acquire EEG signals from multiple channels of the subject under sevoflurane anesthesia.
[0062] The EEG signal acquisition module 401 uses disposable Ag / AgCl electrodes that comply with the IEC 60601-2-26 safety standard for medical electrical equipment. The electrode impedance is monitored online and maintained below 5 kΩ. The electrode layout adopts a simplified configuration to meet the needs of sevoflurane anesthesia research: channel 1 (Fp1), channel 2 (Fp2), channel 3 (F7), channel 4 (F8), and the reference electrode is placed on the right earlobe (A2). The electrode wires use double-shielded lead wires to effectively suppress environmental electromagnetic interference and motion artifacts.
[0063] The analog front-end signal conditioning module 402 is used to amplify and filter the acquired EEG signals.
[0064] The analog front-end signal conditioning module 402 consists of three levels of circuitry: Preamplifier circuit: Employs Analog Devices' AD8422 instrumentation amplifier with a gain of 1000 (40dB) and an input impedance ≥10 GΩ. The circuit integrates right-leg drive technology, improving the system's common-mode rejection ratio to over 120dB.
[0065] Bandpass filter circuit: A 4th-order Butterworth active filter is used, with a high-pass cutoff frequency of 0.5Hz, a low-pass cutoff frequency of 45Hz, and a passband ripple of ≤0.5dB, to filter out DC offset, baseline drift, and high-frequency electromyographic noise.
[0066] Notch filter circuit: A dual-T active notch filter is used with a center frequency of 50 Hz, a notch depth of ≥40 dB, and a bandwidth of ±1.5 Hz. It effectively suppresses power frequency interference while preserving the signal characteristics of adjacent α and β frequency bands.
[0067] Analog-to-digital converter module 403 is used to convert analog EEG signals into digital EEG signals.
[0068] The analog-to-digital converter module 403 uses TI's ADS12998 24-channel Δ-Σ analog-to-digital converter chip, specifically designed for bioelectrical signals. In this embodiment, it is configured with 4 channels enabled, a sampling rate of 500 SPS, an internal programmable gain amplifier set to 12x, and input reference noise ≤1 μVpp (bandwidth 70 Hz). The chip communicates with the main control processor via an SPI interface, and data is directly stored in the memory buffer via DMA.
[0069] The data processing and control unit 404 is used to process the EEG signal, extract the fused signal of each frequency band in multiple channel spaces from the EEG signal, and obtain the PAC feature vector by weighting, and use a hierarchical clustering algorithm to identify the weighted PAC feature vector to obtain the preliminary classification result of the current anesthesia state.
[0070] The core processor of the data processing and control unit 404 uses the STM32F407ZGT6 high-performance ARM Cortex-M4 microcontroller with a main frequency of 168 MHz and is equipped with a hardware floating-point unit. The system runs the FreeRTOS real-time operating system, and the task scheduling is as follows: Data acquisition task (highest priority, 2 ms cycle): Read raw EEG data from 4 channels of ADS1299 and store it in a double-buffered circular queue (capacity 10 seconds of data).
[0071] Preprocessing task (high priority, 10 ms cycle): Performs digital bandpass filtering (0.5-45 Hz), power frequency notch filtering, and real-time artifact detection based on amplitude / gradient.
[0072] Feature extraction task (medium priority, 1s cycle): Perform multi-channel full-band signal decomposition, full-band PAC feature calculation and fusion, and apply sevoflurane-specific weights to the Theta-Gamma key features.
[0073] Classification inference task (high priority, 1s cycle): Input the constructed weighted PAC feature vector into the pre-trained hierarchical clustering model and output the current anesthesia state.
[0074] Display and Communication Task (low priority, 100 ms cycle): Refreshes the user interface and uploads data via the network interface.
[0075] The data transmission module 405 is used to transmit EEG data.
[0076] The data transmission module 405 supports dual-mode transmission. Wired transmission is based on the W5500 hardware TCP / IP protocol stack chip, and the data format conforms to the HL7 medical information exchange standard. Wireless transmission is based on the ESP32-WROOM-32 module, supports Wi-Fi and Bluetooth 5.0, and pushes data to the cloud or mobile terminal via the MQTT protocol. The transmission process uses AES-128 encryption.
[0077] The display and interaction module 406 is used to display the raw EEG signals and analysis results.
[0078] The display and interaction module 406 uses a 7-inch TFT-LCD capacitive touchscreen with a resolution of 1024×600. The user interface is divided into: ① Real-time waveform area, displaying 4 channels of raw EEG; ② Trend graph area, displaying the real-time change curve of the Theta-Gamma phase amplitude coupling modulation index (approximately 5 minutes); ③ Status indicator area, prominently displaying the current discrimination state (awake / moderately sedated / deeply sedated, three categories) and confidence level using colors (green / yellow / red) and bar graphs; ④ System status area, displaying electrode impedance, signal quality, and network connection status.
[0079] It is understood that the sevoflurane anesthesia state monitoring system provided by the present invention can realize the monitoring of sevoflurane anesthesia state. The relevant technical features of the monitoring system for monitoring the sevoflurane anesthesia state can be referred to the relevant technical features of the aforementioned sevoflurane anesthesia state monitoring method, and will not be repeated here.
[0080] To verify the effectiveness of the sevoflurane-specific weighted PAC features, improved hierarchical clustering discrimination model, and dynamic feedback mechanism proposed in this invention, the three-class discrimination performance, real-time performance, and output stability can be comprehensively evaluated under the condition of simultaneous acquisition of EEG data, MOAA / S scoring labels, and system output results.
[0081] In efficacy validation, the anesthetic state was uniformly divided into three categories: awake, moderately sedated, and deeply sedated, and the MOAA / S score was used as the gold standard for behavioral studies. It is recommended to at least calculate the overall accuracy, sensitivity and specificity of each category, macro-mean F1 score, area under the receiver operating characteristic curve (AUC), mean system delay, and state output stability rate.
[0082] The method for monitoring sevoflurane anesthesia provided in this invention has the following advantages compared to existing general anesthesia depth monitoring schemes: (1) A feature system is constructed around the mechanism of action of sevoflurane, significantly improving the specificity of discrimination. This invention does not only use general EEG power features or general statistical features, but uses phase-amplitude coupling features that can reflect the modulation relationship between low-frequency phase and high-frequency amplitude as the core characterization index. Furthermore, based on the feature of sevoflurane enhancing the GABA-A receptor-mediated inhibition and remodeling of the Theta-Gamma cross-frequency coupling relationship in the prefrontal cortex, this invention performs targeted enhancement weighting on the Theta-Gamma sub-features that are highly correlated with the transition of sevoflurane anesthesia state on the basis of full-band PAC features. Through this "full-band calculation + key coupling weighting highlighting" approach, the feature vector can more centrally characterize the most critical neural modulation patterns during sevoflurane anesthesia, thus being more conducive to improving the ability to identify and distinguish the state of sevoflurane anesthesia compared to general EEG features.
[0083] (2) An improved classifier structure for sevoflurane anesthesia grading enhances clinical consistency and interpretability. Instead of directly employing ordinary unsupervised clustering or general classification models, this invention constructs a hierarchical clustering real-time discrimination model specifically for sevoflurane. During the training phase, the model introduces the MOAA / S behavioral score as a constraint label and constructs clustering templates using seed points, ensuring that the resulting template centers stably correspond to the three sevoflurane anesthesia states: conscious, moderately sedated, and deeply sedated. In the real-time discrimination phase, a weighted distance metric consistent with the weighted PAC features is used, ensuring that the distance calculation logic within the classifier remains consistent with the design of sevoflurane-specific features. Therefore, this invention not only improves the stability of state discrimination in sevoflurane scenarios but also enhances the correspondence and interpretability between the discrimination results and the clinical behavioral grading.
[0084] (3) Establish a feedback system for the dynamic changes in sevoflurane EEG to improve real-time monitoring stability. During sevoflurane anesthesia, the EEG coupling pattern changes dynamically with the progression of anesthesia, individual differences, and transient noise. To address this characteristic, this invention sets up a dynamic feedback consistency confirmation mechanism in the real-time discrimination process: when the system detects a continuous mismatch between the current real-time features and the existing sevoflurane clustering template, it does not simply output fluctuating results, but rather performs a limited and smooth adaptive correction of the weights of each PAC sub-feature based on the intra-class dispersion and inter-class separability of the recent feature vectors, and combines a state lag control mechanism to suppress short-term flips. Through this feedback system, this invention can improve the adaptability to the current signal environment while maintaining the sevoflurane-specific discrimination logic, thereby improving output stability and clinical usability.
[0085] (4) Balancing the real-time and high temporal resolution analysis requirements in sevoflurane research scenarios. This invention employs short-window sliding analysis, lightweight multi-channel PAC feature extraction, and real-time clustering discrimination strategies, which are more conducive to rapid response to state changes during sevoflurane anesthesia compared to traditional monitoring methods that rely on long-term averaging. This design is not only suitable for clinical monitoring but also more suitable for high temporal resolution observation of the dynamic relationship between "drug administration-EEG effect-behavioral state" in sevoflurane pharmacodynamic studies.
[0086] (5) The output information is more open, facilitating the revelation of the sevoflurane anesthesia mechanism. This invention, while outputting the final anesthesia state result, can also simultaneously output core PAC characteristic trends, template matching results, and system operation information. Researchers can not only obtain conclusions about the "current anesthesia state," but also further observe the dynamic changes of key neurophysiological characteristics during sevoflurane anesthesia. Compared to closed commercial devices that only output a single comprehensive index, this invention is more suitable for serving research on the mechanism of action of sevoflurane, optimization of anesthesia depth assessment methods, and subsequent comparative studies of novel inhaled anesthetic drugs.
[0087] (6) A complete closed loop of "drug mechanism - feature construction - classification - feedback correction" is formed, with distinctive patent features. The advantage of this invention does not lie in using a single EEG feature or a single classifier, but in establishing a complete, closed-loop dedicated technical solution around the mechanism of action of sevoflurane: First, information reflecting neural oscillation remodeling is extracted through multi-channel full-band PAC calculation; second, feature weighting is implemented for key coupling relationships of sevoflurane; third, the hierarchical clustering model is improved based on behavioral label constraints; finally, real-time output stabilization is achieved through dynamic feedback and state lag control. The above steps work together to form a specific anesthesia state discrimination system and method for sevoflurane, which is also the core advantage of this invention that distinguishes it from existing general anesthesia depth monitoring technologies.
[0088] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.
[0089] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0090] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0091] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0092] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0093] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.
[0094] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A sevoflurane anesthesia status monitoring system, characterized in that, include: The EEG signal acquisition module is used to continuously acquire EEG signals from multiple channels of the subject under sevoflurane anesthesia. The data processing and control unit is used to perform frequency band decomposition on the EEG signals of each channel to obtain bandpass signals in multiple frequency bands, and to extract representative fusion signals of each frequency band in the multi-channel space from the bandpass signals of each frequency band in multiple channels. Specifically, for the EEG signals of each channel, frequency band decomposition is performed to decompose the EEG signals of each channel into bandpass signals of five frequency bands: Delta, Theta, Alpha, Beta, and Gamma. For any frequency band, the bandpass signals of that frequency band in multiple channels are arranged into a matrix by columns, and the covariance matrix is calculated after mean removal processing of each column. The eigenvector corresponding to the largest eigenvalue of the covariance matrix is obtained, and the representative fusion signal of that frequency band in the multi-channel space is calculated based on the eigenvector. Furthermore, the unit performs frequency band combination on multiple frequency bands to obtain all frequency band combinations, and calculates the PAC feature of each frequency band combination based on the representative fusion signal of each frequency band. Specifically, for Delta, Theta, Alpha, Beta, and Gamma bands... The MMA has five frequency bands, with two bands forming a band combination, resulting in 10 band combinations. Each band combination includes one low-frequency band and one high-frequency band. Based on the specific mechanism of action of sevoflurane, the PAC features of each band combination are weighted to obtain a weighted PAC feature vector. Among them, the Theta-Gamma band combination has the largest weight coefficient. The weighted PAC feature vector is also used to perform real-time classification using a hierarchical clustering algorithm to obtain a preliminary classification result of the anesthesia state in the current sliding analysis window. Specifically, based on the MOAA / S score, samples corresponding to historical sliding analysis windows are divided into three anesthesia states: awake, moderately sedated, and deeply sedated. Corresponding category template centers are constructed, and the weighted Euclidean distance between the weighted PAC feature vector of the current sliding analysis window and each template center is calculated. The weighted PAC feature vector of the current sliding analysis window is then matched with each category template center to obtain a preliminary classification result of the anesthesia state. This also includes setting up two levels of monitoring, where the first level of monitoring includes: The most recent contiguous, including the current sliding analysis window A sliding analysis window is used as the monitoring time period to obtain the most recent continuous... Each sliding analysis window corresponds to the preliminary classification results of the anesthesia state, and the most recent continuous... Are the preliminary classification results of the anesthesia status in each sliding analysis window completely consistent? If they are completely consistent, then the preliminary classification result of the anesthesia status in the current sliding analysis window shall be determined as the final classification result. If they are not completely consistent, the final classification result corresponding to the previous monitoring period remains unchanged, and secondary monitoring is triggered; the secondary monitoring includes: Set up an inconsistent persistence counter This is used to accumulate the number of consecutive rounds of judgment where "inconsistency" occurs. Each round is based on the most recent consecutive... Consistency judgment of the preliminary classification results of anesthesia status in each sliding analysis window is recorded as one judgment round. When the most recent consecutive... When the preliminary classification results of anesthesia status in the sliding analysis windows are not completely consistent, let = +1; When the most recent consecutive... When the preliminary classification results of the anesthesia status in each sliding analysis window are completely consistent and a new effective status is formed, let =0; when Reaching the preset threshold At that time, the weight adjustment feedback mechanism is automatically triggered to adjust the weight vector w of each frequency band combination.
2. The sevoflurane anesthesia status monitoring system according to claim 1, characterized in that, The EEG signal acquisition module is used to continuously acquire EEG signals from multiple channels of the subject under sevoflurane anesthesia, including: Under sevoflurane anesthesia, the EEG signals of the subjects were continuously collected from four channels: Fp1, Fp2, F7, and F8. The acquired four-channel EEG signals are stored in a circular buffer. Let the coverage duration of the circular buffer be T seconds, the number of channels be C, and the sampling rate be Fs. Then, the total number of data points N in a single circular buffer is: 。 3. The sevoflurane anesthesia status monitoring system according to claim 1, characterized in that, The data processing control unit is further configured to: The acquired EEG signals are preprocessed and artifacts are detected. The preprocessing includes digital bandpass filtering and notch filtering. The artifact detection includes: If the amplitude of the EEG signal in any channel at any time point within the current sliding artifact detection window is not within the preset amplitude range, or the difference between the amplitudes of the EEG signals at two adjacent time points exceeds the preset difference threshold, then the EEG signal in the current sliding artifact detection window is marked as invalid and removed; otherwise, it is marked as valid and retained.
4. The sevoflurane anesthesia status monitoring system according to claim 1, characterized in that, The data processing and control unit is used to perform frequency band decomposition on the EEG signals of each channel to obtain bandpass signals in multiple frequency bands, and to extract representative fused signals of each frequency band in the multiple channel space from the bandpass signals of each frequency band in multiple channels, including: Frequency band decomposition is performed on the EEG signal of each channel. The bandpass signal of channel c in frequency band b after frequency band decomposition is represented as follows: , , , These represent the Delta, Theta, Alpha, Beta, and Gamma frequency bands, respectively. For any frequency band b, obtain the pass signals of the frequency band b in the four channels, and arrange the pass signals of the frequency band b in the four channels into a matrix according to columns: ; right After removing the mean from each column, calculate its covariance matrix. : ; in, This represents the number of data sampling points within the current analysis time window. Seeking The largest eigenvalue corresponds to the eigenvector Based on the feature vector Calculate the representative fused signal of frequency band b in multi-channel space. : ; Determine the representative fused signal of the five frequency bands in the multi-channel space. .
5. The sevoflurane anesthesia status monitoring system according to claim 4, characterized in that, The process involves combining multiple frequency bands to obtain all frequency band combinations, and calculating the PAC characteristics of each frequency band combination based on the representative fused signal of each frequency band, including: For any frequency band combination, the modulation index of the frequency band combination is calculated using the phase binning method based on the representative fused signals of the low-frequency band and the representative fused signals of the high-frequency band, and is used as the PAC feature of the frequency band combination.
6. The sevoflurane anesthesia status monitoring system according to claim 5, characterized in that, For any frequency band combination, based on the representative fused signals of the low-frequency band and the representative fused signals of the high-frequency band, the modulation index of the frequency band combination is calculated using the phase binning method as the PAC feature of the frequency band combination, including: Obtain the representative fused signal of the low-frequency band L in any of the frequency band combinations at time t. Representative fused signal of high-frequency band H at time t ; right Perform Hilbert transform to extract the low-frequency instantaneous phase sequence. and to Perform Hilbert transform to extract the high-frequency instantaneous amplitude envelope sequence. ; The modulation index of any frequency band combination is calculated using the phase binning method. ,include: ; In the phase binning method, only the low-frequency instantaneous phase sequence is considered. Interval division, high-frequency instantaneous amplitude envelope sequence Without dividing the data into intervals, it is assigned to the corresponding phase bin according to the time synchronization principle, i.e., the time interval at time t. Belongs to The determined phase box , This indicates that the m-th element is divided into boxes. It is exist After dividing the phase range into M intervals, the corresponding intervals are... The probability distribution formed by the average value, It is its entropy value, where M is a positive integer greater than 1; Iterate through each frequency band combination to obtain the PAC features of 10 frequency band combinations, denoted as... , The modulation index corresponding to the Theta-Gamma band combination.
7. The sevoflurane anesthesia status monitoring system according to claim 1, characterized in that, Based on the specific mechanism of action of sevoflurane, the PAC features of each frequency band combination are weighted to obtain a weighted PAC feature vector, including: Based on the specific mechanism of action of sevoflurane, a weight vector w corresponding to the feature dimension is defined. , This represents the weighting coefficient of the i-th frequency band combination; Based on the weight vector, the PAC features of the 10 frequency band combinations are weighted, resulting in a weighted PAC feature vector, which is expressed as follows: ; Where ⊙ denotes element-wise multiplication. This represents the weighted PAC eigenvector. This represents the PAC characteristic sequence of 10 frequency band combinations.
8. The sevoflurane anesthesia status monitoring system according to claim 1, characterized in that, The hierarchical clustering algorithm is used to classify the weighted PAC feature vector in real time to obtain a preliminary classification result of the current anesthesia state, including: The system acquires multi-channel full-band EEG signals from multiple subjects within multiple sliding analysis windows. Each sliding analysis window's multi-channel full-band EEG signal is treated as a sample. The weighted PAC feature vector of each sample is extracted, and the anesthesia status category label of each sample is obtained. Based on the MOAA / S score, multiple samples are clustered into three anesthesia statuses: awake, moderately sedated, and deeply sedated. Multiple samples corresponding to each category of anesthesia status are randomly selected. Based on the weighted PAC feature vector of multiple samples, the template center of each category of anesthesia status is calculated to obtain three category template centers. Calculate the weighted Euclidean distance between the weighted PAC eigenvector of the current sliding analysis window and the center of each template: ; in, This represents the weight coefficient of the i-th dimension of the PAC feature in the weighted PAC feature vector. This represents the i-th dimension of the PAC feature in the weighted PAC feature vector of the current sliding analysis window. This represents the i-th dimension PAC feature in the center of the k-th template. This represents the weighted Euclidean distance between the weighted PAC eigenvector of the current sliding analysis window and the center of the k-th template. The anesthesia state category with the smallest weighted Euclidean distance to the weighted PAC feature vector of the current sliding analysis window is taken as the preliminary classification result of the anesthesia state of the current sliding analysis window.
9. The sevoflurane anesthesia status monitoring system according to claim 1, characterized in that, The anesthesia status monitoring process also includes status lock-in control, which includes: When the final classification result of the anesthesia status changes, the new final classification result of the anesthesia status will be maintained for at least the specified time. During the specified time, the current final classification result of the anesthesia status will remain unchanged.
10. The sevoflurane anesthesia status monitoring system according to claim 1, characterized in that, The automatic triggering weight adjustment feedback mechanism adjusts the weight vector w of each frequency band combination, including: The weighted PAC eigenvectors of all sliding analysis windows within a preset time period before the weight adjustment feedback mechanism is triggered are recorded as the feature matrix. Where g is the number of recent sliding analysis windows, D=10 is the PAC feature dimension, and the preliminary classification results of the anesthesia status are obtained for each sliding analysis window; Based on the preliminary classification results of the anesthesia status in each sliding analysis window, the samples from the g sliding analysis windows are divided into multiple anesthesia status groups; For the i-th PAC feature within each anesthesia state group, calculate its inter-class variance. Within-class variance : ; ; Where K is the number of current state categories, i.e., the number of anesthesia state groups. Let k be the number of samples in the k-th class. Let be the mean of the k-th class on the i-th PAC feature. The population mean of all samples on the i-th PAC feature; The current discriminative power score of the i-th PAC feature is: ; in, To prevent tiny positive numbers with a denominator of zero; The weighting coefficients of each frequency band combination are updated using the following smoothing update formula: ; in, Let be the current discriminative power score of the PAC features for the i-th frequency band combination. Let be the current discriminative power score of the PAC features for the i-th frequency band combination. The learning rate is small, and D is the total number of features, i.e., the number of frequency band combinations, D=10; The updated weighting coefficients for each frequency band combination Limited to a preset range Internally, and then normalized; Using the fine-tuned The weighted PAC feature vector of the current sliding analysis window is recalculated and fed into the hierarchical clustering real-time discrimination model for re-discrimination to obtain the final classification result of the anesthesia state corresponding to the current time window.
