Anesthesia depth detection method based on electroencephalogram signals

Abstract

The invention relates to the technical field of medical signal processing and detection, in particular to an electroencephalogram signal-based anesthesia depth detection method, which comprises the following steps of: setting a dynamic weighting coefficient according to a power amplitude fluctuation standard deviation and a frequency drift speed slope, and adjusting a frequency axis deviation direction and a power peak absolute difference to obtain an electroencephalogram signal; re-calculating the weighted dynamic matching degree in combination with harmonic peak width full width at half maximum measurement and alpha frequency band amplitude abrupt change point counting, and determining the adjusted matching level; and matching the anesthesia state transition trend sequence with the updated template in the standard mode library by adopting a time sequence comparison method, and determining an anesthetic-induced electroencephalogram mode state recognition result according to the continuous appearing frame number statistics of the resonance mode, the mode interruption interval duration accumulation and the transition trend direction continuity. According to the method, the instantaneous frequency change rate is dynamically associated with the dosage, the precision and stability of anesthesia depth monitoring are remarkably improved, and reliable technical support is provided for clinical anesthesia management.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals. Background Technology

[0002] Electroencephalography (EEG) monitoring plays a central role in modern anesthesiology because it directly reflects the brain's real-time response to anesthetic drugs. This helps doctors accurately assess the patient's level of consciousness and depth of anesthesia, avoiding the serious risks associated with intraoperative awareness or excessively deep anesthesia. Research in this field is directly related to surgical safety and patient prognosis, and therefore has received widespread attention.

[0003] Most current EEG monitoring methods rely on the power spectrum distribution of signals or relative intensity changes in certain fixed frequency bands to assess anesthetic status. However, these methods often suffer from drift or delays when dealing with different individuals, different drugs, and continuous changes in drug dosage. Especially when anesthetic drugs take effect rapidly or their concentration fluctuates, EEG signals exhibit brief but intense oscillations at specific frequencies. These oscillations are not continuous and stable but rather have obvious fluctuations and intermittent characteristics. Existing analytical methods struggle to accurately capture and utilize these dynamic changes, making the assessment of anesthetic depth unreliable at critical moments.

[0004] One of the most significant manifestations of EEG signals under the influence of anesthetic drugs such as propofol is the enhanced resonance phenomenon in the alpha band. This resonance is not only characterized by an increased amplitude at a specific frequency, but more importantly, it is accompanied by a clear pattern of oscillation duration and intensity fluctuations. The duration of this resonance varies depending on the dosage and administration stage, sometimes lasting only a few seconds, and sometimes extending to more than ten seconds. The actual transition in the anesthetic state often occurs during the appearance, maintenance, and disappearance of these resonant segments. Simply extracting the frequency and amplitude at a single moment as features fails to reflect the complete dynamic characteristics of the resonance process. Conversely, attempting to directly compare all instantaneous oscillations with a fixed template results in numerous mismatches due to duration differences, making it difficult for the system to determine which typical drug-induced pattern the current signal corresponds to. This contradiction between the instantaneity of the resonance and the stability required for pattern matching simultaneously limits the sensitivity and specificity of real-time monitoring for transitions in the anesthetic state.

[0005] How to accurately identify specific resonant segments induced by drugs in real-world monitoring scenarios where EEG signals change rapidly, and effectively correlate them with standard resonant modes at different doses, thereby achieving precise tracking of anesthesia depth, has become a key issue facing current EEG-based methods for detecting anesthesia depth. Summary of the Invention

[0006] This invention provides a method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals, mainly comprising: Raw EEG waveform data under the influence of anesthetic drugs were acquired. Resonant frequency and power amplitude features in the alpha band of each channel signal were extracted to form a frequency-amplitude feature vector. At the same time, the frequency axis offset direction, the absolute difference of power peaks and the full width at half maximum (FWHM) measurement of the resonant peak width were recorded to obtain a preliminary description of the resonant mode. For the frequency-amplitude feature vector, the cosine similarity algorithm is used to compare it with the template in the pre-established standard pattern library. The cosine value of the vector angle is calculated to quantify the similarity score. The preliminary matching level is determined based on the frequency axis offset direction and the absolute difference of the power peak. At the same time, the multi-peak identification results and amplitude spectrum asymmetry detection features within the frequency band are marked to obtain the preliminary matching level results. If the cosine of the vector angle is lower than the preset threshold, the number of consecutive frames of the resonant mode is counted on the signal time axis, the standard deviation of power amplitude fluctuation and the slope of frequency drift velocity are calculated, and the time stability is analyzed by combining the instantaneous resonant frequency drift direction judgment and the instantaneous difference sequence of frequency change rate generation, and the stability evaluation result including instantaneous frequency change rate monitoring is obtained. The dynamic weighting coefficients are set according to the standard deviation of power amplitude fluctuation and the slope of frequency drift velocity. The frequency axis offset direction and the absolute difference of power peak are adjusted. The dynamic matching degree after weighting is recalculated by combining the full width at half maximum (FWHM) measurement of the resonant peak width and the count of amplitude mutation points in the alpha band, and the adjusted matching level is determined. The system acquires real-time infusion rate data of anesthetic drugs, updates the offset direction record and amplitude difference parameter of the corresponding template in the standard mode library according to the adjusted matching level, frequency axis offset direction and power peak absolute difference, and integrates the instantaneous resonant frequency and dose corresponding interval positioning and amplitude change slope to continuously calculate and generate dose-resonant mode mapping relationship. By real-time monitoring of EEG signals to extract instantaneous resonant frequency and alpha band amplitude changes, the dose interval of the current signal is located based on the dose-resonance mode mapping relationship. The direction of the transition trend of the anesthesia state is determined by combining the amplitude peak and baseline deviation statistics and the instantaneous difference sequence of frequency change rate, and the transition trend sequence of the anesthesia state is obtained. The transition trend sequence of the anesthesia state was matched with the updated template in the standard pattern library using a time series comparison method. The results of the EEG pattern state recognition induced by anesthetic drugs were determined based on the statistics of the number of consecutive resonant patterns, the cumulative duration of pattern interruption intervals, and the continuity of the transition trend direction.

[0007] The aforementioned method for detecting anesthesia depth based on electroencephalogram (EEG) signals involves acquiring raw EEG waveform data under anesthetic drug influence, processing the signals from each channel using Fast Fourier Transform (FFT), extracting resonant frequencies and power amplitude features within the alpha band to form a frequency-amplitude feature vector, and simultaneously recording the frequency axis offset direction, the absolute difference in power peak values, and the full width at half maximum (FWHM) measurement of the resonant peak width to obtain a preliminary description of the resonant mode, including: Raw EEG waveform data is acquired using a multi-channel EEG acquisition device; The raw EEG waveform data was processed channel by channel using Fast Fourier Transform to obtain the spectral data for each channel; The resonant frequency and power amplitude in the alpha band are extracted from the spectral data of each channel to generate a frequency-amplitude feature vector. Calculate the frequency axis offset direction and the absolute difference between the power peak value based on the frequency-amplitude eigenvector to obtain the offset direction sequence and the difference sequence; For each resonant peak position in the frequency-amplitude eigenvector, the full width at half maximum (FWHM) of the resonant peak is measured to obtain the FWHM sequence. If the same direction appears consecutively in the offset direction sequence, it is determined that the frequency drift trend is valid, and the resonant frequency sequence and full width at half maximum (FWHM) sequence of the corresponding channel are extracted. Based on the resonant frequency sequence, power amplitude sequence, and full width at half maximum (FWHM) sequence when the frequency drift trend is established, the type of EEG resonant mode under the current anesthetic effect is determined.

[0008] The aforementioned method for detecting anesthesia depth based on electroencephalogram (EEG) signals involves comparing the frequency-amplitude feature vector with templates in a pre-established standard pattern library using a cosine similarity algorithm. The cosine value of the vector angle is calculated to quantify the similarity score. A preliminary matching level is determined based on the frequency axis offset direction and the absolute difference in power peak values. Simultaneously, multi-peak recognition results within the frequency band and amplitude spectrum asymmetry detection features are labeled to obtain the preliminary matching level result, including: Obtain the frequency-amplitude feature vector; The cosine similarity algorithm is used to calculate the cosine value of the angle between the frequency-amplitude feature vector and the vector of each template in the standard pattern library to obtain the similarity score; The initial matching level is determined based on the similarity score; Determine whether the offset is positive or negative based on the direction of the frequency axis offset, and obtain the offset direction mark; Based on the absolute difference in power peak values, determine whether the difference belongs to the small, medium, or large level, and obtain the difference level label; The initial matching level is adjusted based on the combination of offset direction markers and gap level markers to obtain the adjusted matching level. For the amplitude spectrum within the frequency band, the multi-peak distribution is identified and the amplitude spectrum asymmetry is detected, yielding multi-peak identification results and asymmetry detection results; The adjusted matching level is associated with the multi-peak recognition results and asymmetric detection results to obtain the final preliminary matching level result.

[0009] The aforementioned method for detecting anesthesia depth based on electroencephalogram (EEG) signals, wherein if the cosine value of the vector angle is lower than a preset threshold, the number of consecutive frames of resonant modes are counted on the signal time axis, the standard deviation of power amplitude fluctuation and the slope of frequency drift velocity are calculated, and the temporal stability is analyzed by combining the instantaneous resonant frequency drift direction judgment and the instantaneous difference sequence generation of frequency change rate, to obtain a stability assessment result including instantaneous frequency change rate monitoring, including: Obtain the cosine of the angle between the current signal frame vector and the reference vector; If the cosine value of the included angle is lower than the preset threshold, the subsequent processing will proceed; otherwise, it will be determined as a non-resonant state and the evaluation will end. The number of consecutive frames that satisfy the resonance condition is counted on the signal time axis to obtain the number of consecutively occurring frames; For the power sequence corresponding to consecutive frames, calculate the standard deviation of power amplitude fluctuation to obtain the standard deviation of power amplitude fluctuation. Linear fitting is performed on the instantaneous resonant frequency sequence within a consecutive number of frames to obtain the frequency drift velocity slope; The instantaneous frequency drift direction is determined by the sign of the slope of the frequency drift velocity. The frequency difference between adjacent frames is calculated from the instantaneous resonant frequency sequence within a consecutive number of frames to obtain the instantaneous difference sequence of frequency change rate; The mean and variance of the instantaneous difference sequence of the rate of change of frequency are calculated to obtain the statistic of the rate of change of frequency; The stability determination logic incorporates the standard deviation of power amplitude fluctuation, the slope of frequency drift velocity, the instantaneous frequency drift direction, and the statistics of frequency change rate. If the standard deviation of power amplitude fluctuation is less than the preset fluctuation limit and the absolute value of frequency drift velocity slope is less than the preset drift limit, then the time dimension stability is judged to be high; otherwise, it is judged to be low. By combining the instantaneous frequency drift direction and the frequency change rate statistics, the stability judgment result is corrected, and the final stability assessment result including instantaneous frequency change rate monitoring is obtained.

[0010] The aforementioned method for detecting anesthesia depth based on electroencephalogram (EEG) signals, wherein a dynamic weighting coefficient is set according to the standard deviation of power amplitude fluctuation and the slope of frequency drift velocity, the frequency axis offset direction and the absolute difference of power peak value are adjusted, and the weighted dynamic matching degree is recalculated by combining the measurement of the full width at half maximum (FWHM) of the resonance peak and the counting of amplitude mutation points in the alpha band, and the adjusted matching level is determined, including: By collecting data on power amplitude and fluctuation standards, an initial signal feature dataset is obtained. Based on the collected signal features, a preset standardization processing method is used to obtain standardized basic signal data. Based on standardized signal baseline data, a joint analysis of frequency drift and velocity slope is performed to obtain the trend parameters between the two and determine the dynamic adjustment benchmark for frequency drift. By dynamically adjusting the frequency drift reference and combining the data of frequency offset and power peak, the correlation deviation between the two is calculated, and it is determined whether the deviation exceeds the preset threshold range. If it does, the frequency offset is corrected to obtain the corrected offset data. Based on the corrected offset data and the difference correction processing logic, the power peak correction result after difference correction is obtained, and the power adjustment direction of the signal is determined. By analyzing the direction of power adjustment and combining the measurement data of the resonance peak width and full width at half maximum (FWHM), the resonance characteristic parameters of the signal are calculated, and the stability index of the resonance peak is obtained. Based on the stability index of the resonance peak and combined with the statistical data of amplitude mutation, the mutation frequency of the alpha band is analyzed to determine whether the mutation frequency meets the preset stability conditions. If it does not meet the conditions, the mutation point is smoothed to obtain the smoothed mutation distribution data. By using smoothed mutation distribution data and combining it with the matching degree calculation rules, the support vector machine algorithm is used to evaluate the overall matching degree of the signal and determine the final matching level classification result.

[0011] The aforementioned method for detecting anesthesia depth based on electroencephalogram (EEG) signals involves dynamically adjusting a reference for frequency drift, combining data on frequency offset and power peak values, and calculating the correlation deviation between the two using the following formula:

[0012] in, Indicates the correlation deviation value, Δ f Indicates frequency offset. Indicates the frequency offset mean. Indicates the standard deviation of frequency offset. Indicates peak power. Indicates the average peak power. Indicates the standard deviation of peak power; The process involves obtaining the power peak correction result after gap correction based on the corrected offset data and the gap correction processing logic. The calculation formula used is as follows:

[0013] in, λ represents the final peak power after gap correction, γ represents the overall amplitude compensation coefficient, and λ represents the exponentially decaying frequency offset penalty parameter. This indicates the frequency offset value after the calibration process.

[0014] The aforementioned method for detecting anesthesia depth based on electroencephalogram (EEG) signals involves acquiring real-time anesthetic infusion rate data, updating the offset direction record and amplitude difference parameters of the corresponding template in the standard pattern library according to the adjusted matching level, frequency axis offset direction, and absolute difference in power peak values, and simultaneously incorporating the instantaneous resonant frequency and dose-corresponding interval location and amplitude change slope continuous calculation to generate a dose-resonant mode mapping relationship, including: The infusion rate sequence is obtained by collecting real-time data of the current anesthetic drug infusion rate through sensors. Calculate the instantaneous resonant frequency for the infusion rate sequence and determine the current instantaneous resonant frequency value; The instantaneous resonant frequency is compared with the pre-divided dose-corresponding interval to obtain the current dose-corresponding interval location result; Based on the current dose-corresponding interval positioning results, extract the reference offset direction record and reference amplitude difference parameter of the corresponding template from the standard pattern library; The peak power of the current infusion rate sequence is calculated using spectral analysis. The peak power value is then subtracted from the baseline amplitude difference parameter to obtain the absolute difference in peak power. If the absolute difference between the peak power values ​​exceeds the preset threshold, the offset direction is determined to be positive or negative based on the offset direction of the frequency axis, and the offset direction determination result is obtained. Based on the offset direction judgment result and the absolute difference of the power peak, the offset direction record and amplitude difference parameter of the corresponding template in the standard mode library are overwritten and updated to obtain the updated offset direction record and updated amplitude difference parameter. The updated offset direction record and the updated amplitude difference parameter are combined with the current dose-corresponding interval positioning result and the continuous calculation result of amplitude change slope to update the mapping entries of the corresponding interval in the dose-resonance mode mapping relationship, and the latest dose-resonance mode mapping relationship is obtained.

[0015] The aforementioned method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals involves extracting instantaneous resonant frequencies and alpha band amplitude changes from real-time monitoring of EEG signals, locating the dose interval of the current signal based on the dose-resonance mode mapping relationship, and determining the direction of the transition trend of the anesthesia state by combining the statistical analysis of the amplitude peak and baseline deviation and the instantaneous difference sequence of the frequency change rate, thereby obtaining the anesthesia state transition trend sequence, including: Raw EEG data sequences are obtained by real-time acquisition of EEG signals; The instantaneous resonant frequency and alpha band amplitude changes were extracted from the raw EEG data sequence to obtain the frequency sequence and amplitude sequence; The instantaneous resonant frequency is input into the dose-resonant mode mapping to obtain the current dose range identifier; The amplitude peak value is calculated from the amplitude sequence and the deviation is statistically analyzed with a pre-established baseline to obtain the deviation statistical sequence; The frequency rate of change sequence is obtained by calculating the instantaneous difference sequence from the frequency sequence; If the peak deviation of the amplitude in the deviation statistics sequence continues to increase and the frequency change rate sequence shows a positive acceleration, then the transition trend direction is judged to be deepening; if the peak deviation of the amplitude in the deviation statistics sequence continues to decrease and the frequency change rate sequence shows a negative deceleration, then the transition trend direction is judged to be shallowing, thus obtaining the single transition trend direction judgment result. The transition trend sequence of anesthesia state is obtained by connecting the results of multiple consecutive single transition trend direction judgments in chronological order.

[0016] The aforementioned method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals, wherein a time-series comparison method is used to match the transition trend sequence of the anesthesia state with an updated template in a standard pattern library, and the results of identifying the EEG pattern state induced by anesthetic drugs are determined based on the statistical analysis of the number of consecutive occurrences of the resonant pattern, the cumulative duration of the pattern interruption interval, and the continuity of the transition trend direction. Time series analysis was used to extract transition trend features under anesthesia from the collected electroencephalogram (EEG) signal data, and preliminary trend sequence data were obtained. By comparing the extracted transition trend sequences with the updated templates in the standard pattern library one by one, the similarity distribution between each template and the trend sequence is determined. If a template in the similarity distribution has a matching value higher than the preset threshold, the resonant mode corresponding to that template is taken as the preliminary identification result, and its consecutive frame count and interruption interval duration are obtained. Based on the statistical results of the number of consecutive frames and the duration of the interruption interval, the persistence and stability index of the resonance mode are calculated to determine whether it meets the preset stability conditions. If the stability index meets the conditions, the direction continuity of the transition trend is further analyzed, the smoothness parameter of the direction change is obtained, and the final EEG pattern state recognition result is determined. By comparing the recognition results with the standard patterns, the template data in the pattern library is updated to optimize the accuracy of subsequent comparisons and obtain an updated template set. For the updated template set, the above matching and statistical process is executed repeatedly to continuously track changes in EEG patterns under anesthesia and determine the dynamic consistency of state recognition.

[0017] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: 1. Enhanced ability to capture and analyze transient resonant modes: By introducing dynamic time-domain features such as the standard deviation of power amplitude fluctuation and the slope of frequency drift velocity, and combining the statistics of the number of consecutive occurrences of resonant modes and the analysis of the interruption interval duration, the system can accurately identify and track the unsteady oscillations in the alpha band under anesthesia, significantly enhancing the response sensitivity to the transient process of anesthesia depth.

[0018] 2. Enhance the system's individualized adaptation and dynamic calibration capabilities: The system uses dynamic weighting coefficients to adjust the frequency axis offset direction and the absolute difference of power peak in real time, and continuously updates the template parameters in the standard pattern library based on the current monitoring data. This enables the system to adapt to the physiological differences of different patients and the EEG response characteristics of different anesthesia stages, effectively reducing misjudgments caused by individual variations and dose fluctuations.

[0019] 3. Achieve dynamic correlation and trend prediction of dose-resonance mode: By establishing and maintaining the dose-resonance mode mapping relationship in real time, the infusion rate of anesthetic drugs is dynamically bound to the EEG resonance characteristics, which supports the real-time positioning of the current anesthesia depth range and the accurate judgment of the state transition trend, providing an immediate and reliable decision-making basis for the fine control of clinical drug dosage.

[0020] 4. Improve the continuity, stability and anti-interference of state recognition. By adopting time series comparison method, and combining multi-dimensional stability indicators such as resonance mode persistence, directional continuity and interruption sparsity, the system can achieve smooth tracking and robust recognition of the transition trend of anesthesia state, which greatly improves the specificity, consistency and reliability of the monitoring system in complex environments.

[0021] In summary, this invention integrates dynamic feature extraction, adaptive template matching, real-time dose mapping, and multi-dimensional trend analysis to construct a highly responsive, adaptable, and accurate intelligent detection system for anesthesia depth, providing solid technical support for the safety management and individualized precision medication in clinical anesthesia. Attached Figure Description

[0022] Figure 1This is a flowchart of an anesthesia depth detection method based on electroencephalogram (EEG) signals according to the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] like Figure 1 This embodiment of a method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals may specifically include: S101. Obtain raw EEG waveform data under the action of anesthetic drugs through a multi-channel EEG signal acquisition device. Perform fast Fourier transform processing on the signals of each channel to extract the resonant frequency and power amplitude features in the alpha band, forming a frequency-amplitude feature vector. At the same time, record the frequency axis offset direction, the absolute difference of the power peak, and the full width at half maximum (FWHM) measurement of the resonant peak width to obtain a preliminary description of the resonant mode.

[0025] Raw EEG waveform data was acquired using a multi-channel EEG acquisition device. A Fast Fourier Transform (FFT) was performed on each channel of the raw EEG waveform data to obtain the spectral data for each channel. The resonant frequency and power amplitude within the alpha band were extracted from the spectral data of each channel, generating a frequency-amplitude feature vector. The frequency axis offset direction and the absolute difference between the peak power and the peak power were calculated based on the frequency-amplitude feature vector, resulting in an offset direction sequence and a difference sequence. For each resonant peak position in the frequency-amplitude feature vector, the full width at half maximum (FWHM) of the resonant peak was measured, obtaining a FWHM sequence. If the same direction appears consecutively in the offset direction sequence, a frequency drift trend is identified, and the resonant frequency sequence and FWHM sequence for the corresponding channel are extracted. Based on the resonant frequency sequence, power amplitude sequence, and FWHM sequence when the frequency drift trend is established, the EEG resonant mode category under the current anesthetic effect is determined.

[0026] After acquiring raw EEG waveform data using a multi-channel EEG acquisition device, a fast Fourier transform is performed independently on each channel to obtain the corresponding spectral data.

[0027] For example, in the signal acquired by the 8-channel device, the spectrum of the 3rd channel shows a significant peak in the alpha band of 8–13 Hz.

[0028] Specifically, the resonant frequency and power amplitude of the alpha band are extracted from the spectrum of each channel to form a frequency-amplitude feature vector.

[0029] In one embodiment, when a subject is in a quiet state with their eyes closed, channel Cz extracts a resonant frequency of 9.8 Hz, corresponding to a power amplitude of 12.4 μV² / Hz, while channel O1 extracts 10.2 Hz with an amplitude of 14.7 μV² / Hz. These values ​​together constitute the feature vector.

[0030] Preferably, the frequency axis offset direction and the absolute difference of the power peak are calculated based on the frequency-amplitude eigenvector.

[0031] For example, within a continuous 5-second period, the resonant frequency of channel Cz changes from 9.8Hz to 9.5Hz and then to 9.2Hz, with the shift direction always being negative, indicating a drift towards lower frequencies; simultaneously, the peak power of channel O1 decreases from 14.7μV² / Hz to 11.2μV² / Hz, with a difference sequence recorded as -3.5μV² / Hz. The continuous occurrence of negative shift directions in the sequence confirms a frequency drift trend.

[0032] In one possible implementation, the full width at half maximum (FWHM) is measured for each resonance peak location.

[0033] For example, the initial full width at half maximum (FWHM) of the Cz channel resonance peak is 1.6 Hz. As the propofol concentration increases, the FWHM of the same peak gradually increases to 2.4 Hz and 2.9 Hz, forming a FWHM sequence. This reflects the diffusion characteristics of EEG rhythms.

[0034] It should be noted that when the same direction appears more than three times consecutively in the offset direction sequence, the resonant frequency sequence, power amplitude sequence, and full width at half maximum (FWHM) sequence of the corresponding channel are extracted.

[0035] For example, during propofol induction, the channel Fz resonant frequency sequence is 10.1Hz→9.7Hz→9.3Hz→8.9Hz, the power amplitude sequence is 15.2→13.8→11.5→9.4μV² / Hz, and the full width at half maximum (FWHM) sequence is 1.4Hz→1.8Hz→2.3Hz→2.7Hz.

[0036] For example, during the deepening phase of propofol anesthesia, these sequences collectively characterize a pattern of alpha resonance shifting to lower frequencies, significant power attenuation, and broadened spectral peaks, which can be categorized as a "low-frequency-dominated diffuse alpha suppression mode." In contrast, during the light anesthesia phase, the frequency shift is not significant, the power remains relatively high, and the full width at half maximum (FWHM) is narrow, which may be classified as a "stable alpha-dominated mode."

[0037] Understandably, this multi-dimensional sequence-based resonant mode classification can more accurately distinguish changes in EEG state at different depths of propofol action, providing a reliable objective indicator for monitoring anesthesia depth, which helps reduce intraoperative awareness risks and optimize drug dosage.

[0038] S102. The cosine similarity algorithm is used to compare the frequency-amplitude feature vector with the template in the pre-established standard pattern library. The cosine value of the vector angle is calculated to quantify the similarity score. The preliminary matching level is determined based on the frequency axis offset direction and the absolute difference of the power peak. At the same time, the multi-peak identification results and amplitude spectrum asymmetry detection features within the frequency band are marked to obtain the preliminary matching level results.

[0039] Obtain the frequency-amplitude feature vector. Calculate the cosine of the angle between the frequency-amplitude feature vector and the vector of each template in the standard pattern library using a cosine similarity algorithm to obtain a similarity score. Determine the initial matching level based on the similarity score. Determine whether the offset is positive or negative based on the frequency axis offset direction to obtain an offset direction label. Determine whether the difference is small, medium, or large based on the absolute difference in power peak values ​​to obtain a difference level label. Adjust the initial matching level based on the offset direction label and the difference level label to obtain an adjusted matching level. For the amplitude spectrum within the frequency band, identify multi-peak distributions and detect amplitude spectrum asymmetry to obtain multi-peak identification results and asymmetry detection results. Associate the adjusted matching level with the multi-peak identification results and asymmetry detection results to obtain the final initial matching level result.

[0040] The cosine similarity algorithm is used to calculate the cosine of the angle between the frequency-amplitude feature vector and the vector of each template in the standard pattern library, thus obtaining the similarity score; the calculation formula is:

[0041] Where v represents the frequency-amplitude feature vector to be measured. This represents the k-th template vector in the standard pattern library. Let represent the cosine similarity score of the k-th template. The numerator is the dot product of the two vectors, and the denominator is the product of the magnitudes of the two vectors.

[0042] This represents the similarity score of the k-th template obtained from the calculation.

[0043] For example, after obtaining the frequency-amplitude feature vector, the cosine similarity algorithm is used to calculate the cosine value of the angle between the vector and each template in the standard pattern library, thereby obtaining the similarity score.

[0044] Specifically, the alpha band feature vector generated by the fast Fourier transform of the 8-channel EEG signal collected from a subject under the influence of propofol was found to have a cosine similarity of 0.92 with the "stable alpha-dominant pattern" template in the database and a similarity of 0.41 with the "low-frequency dominant diffuse alpha-inhibition pattern" template. Based on this, the matching level was initially determined to be high similarity.

[0045] Preferably, the matching level is divided according to the similarity score. For example, a score higher than 0.85 is considered a high match, 0.60 to 0.85 is a medium match, and a score lower than 0.60 is a low match.

[0046] In one embodiment, when the similarity score is 0.78, the system marks it as a moderate match level, providing a basis for subsequent adjustments.

[0047] In one possible implementation, the offset direction is determined as positive or negative based on the frequency axis offset direction, and an offset direction mark is obtained.

[0048] For example, if the resonant frequency of the Cz channel changes from 10.0Hz to 9.7Hz, 9.4Hz, and 9.1Hz in successive observations, and the shift direction is always negative, it is marked as a negative drift; if the frequency increases from 9.5Hz to 9.8Hz and 10.1Hz, it is marked as a positive drift.

[0049] Specifically, the difference level is determined based on the absolute difference in peak power, and is divided into three levels: small, medium, and large.

[0050] For example, if the peak power of the O1 channel drops from 15.1 μV² / Hz to 14.0 μV² / Hz in a short period of time, with an absolute difference of 1.1 μV² / Hz, it is classified as a small level; if it drops to 10.5 μV² / Hz, with a difference of 4.6 μV² / Hz, it is classified as a medium level; and if the drop exceeds 7 μV² / Hz, it is classified as a large level.

[0051] It should be noted that the initial matching level is adjusted by combining the offset direction marker with the gap level marker.

[0052] For example, when the offset direction is negative and the gap level is large, the moderate matching level is downgraded to low matching; if the offset is negative but the gap is only small, the matching level is maintained or only slightly downgraded; when the offset is positive and the gap is small, the matching level can be appropriately upgraded. This adjustment mechanism makes the matching results more closely reflect the dynamic changes in the effects of anesthetics. For the amplitude spectrum within the frequency band, multi-peak distribution is identified and amplitude spectrum asymmetry is detected, yielding multi-peak identification results and asymmetric detection results.

[0053] For example, when propofol is applied in a shallow layer, the amplitude spectrum of channel P3 shows a single main peak at 10.3 Hz with no obvious secondary peaks, indicating a single-peak distribution with good symmetry. In the deepening stage, channel Fz shows a multi-peak distribution with a main peak of 9.2 Hz and a secondary peak of 8.1 Hz. At the same time, the amplitude on the left side of the main peak decays faster, showing right-skewed asymmetry.

[0054] It is understandable that the adjusted matching level is associated with the multi-peak recognition results and asymmetric detection results to ultimately form a comprehensive matching level result.

[0055] For example, when the adjusted matching level is moderate and multi-peak distribution and significant asymmetry are detected, the system outputs a preliminary match of "transition to low-frequency diffusion inhibition mode," thus more accurately depicting the evolution of EEG from a stable alpha state to an inhibited state. This multi-dimensional fusion matching method can significantly improve the sensitivity and specificity of changes in anesthesia depth, providing stronger objective support for real-time clinical monitoring, and helping to finely control the dosage of anesthetic drugs and reduce related risks.

[0056] S103. If the cosine value of the vector angle is lower than the preset threshold, the number of consecutive frames of the resonant mode is counted on the signal time axis, the standard deviation of power amplitude fluctuation and the slope of frequency drift velocity are calculated, and the time stability is analyzed by combining the instantaneous resonant frequency drift direction judgment and the instantaneous differential sequence of frequency change rate generation, and the stability evaluation result including instantaneous frequency change rate monitoring is obtained.

[0057] Obtain the cosine of the angle between the current signal frame vector and the reference vector. If the cosine is lower than a preset threshold, proceed to the next processing step; otherwise, determine it as a non-resonant state and end the evaluation. Count the number of consecutive frames that meet the resonance condition from the signal time axis to obtain the number of consecutively occurring frames. Calculate the standard deviation of power amplitude fluctuation for the power sequence corresponding to the number of consecutively occurring frames. Perform linear fitting on the instantaneous resonant frequency sequence within the number of consecutively occurring frames to obtain the frequency drift velocity slope. Determine the instantaneous frequency drift direction based on the sign of the frequency drift velocity slope. Calculate the frequency difference between adjacent frames from the instantaneous resonant frequency sequence within the number of consecutively occurring frames to obtain the instantaneous difference sequence of frequency change rate. Calculate the mean and variance of the instantaneous difference sequence of frequency change rate to obtain the frequency change rate statistic. Input the standard deviation of power amplitude fluctuation, frequency drift velocity slope, instantaneous frequency drift direction, and frequency change rate statistic into the stability comprehensive judgment logic. If the standard deviation of power amplitude fluctuation is less than the preset fluctuation limit and the absolute value of the frequency drift velocity slope is less than the preset drift limit, the time dimension stability is judged to be high; otherwise, it is judged to be low. The stability judgment result is corrected by combining the instantaneous frequency drift direction and frequency change rate statistics, resulting in a final stability assessment result that includes instantaneous frequency change rate monitoring.

[0058] For example, after obtaining the frequency-amplitude feature vector of the current signal frame, the cosine value of the angle between it and the reference vector is first calculated. If the cosine value is lower than the preset threshold of 0.75, it is determined that the current frame does not meet the resonance condition, and it directly enters the subsequent non-resonance processing flow; if the cosine value reaches 0.88, it is confirmed that it is in a resonance state, and the time dimension analysis continues.

[0059] Specifically, the number of frames that satisfy the resonance condition is counted on the continuous time axis of the signal.

[0060] For example, if, during a subject's continuous propofol infusion, the feature vectors of 12 consecutive frames all have a cosine similarity higher than 0.75 with the reference template, then the number of consecutive frames is recorded as 12. This number of consecutive frames becomes the core basis for subsequent stability assessment.

[0061] In one embodiment, for the power sequence corresponding to these 12 frames, the standard deviation of the fluctuation of the peak power amplitude in each frame is calculated. Assuming the power values ​​are 14.2 μV² / Hz, 14.5 μV² / Hz, 13.9 μV² / Hz, 14.3 μV² / Hz, etc., the standard deviation of the fluctuation is only 0.28 μV² / Hz, indicating that the power amplitude remains relatively stable.

[0062] Preferably, the instantaneous resonant frequency sequence within the same continuous frame is linearly fitted to obtain the frequency drift velocity slope.

[0063] For example, the frequency changes sequentially from 10.2Hz, 10.1Hz, 10.0Hz, and 9.9Hz, with a fitting slope of -0.1Hz / frame, indicating the presence of a slow negative drift.

[0064] It should be noted that the instantaneous frequency drift direction is directly determined by the sign of the slope. In this example, the slope is negative, which is marked as a negative drift; if the slope changes to +0.08 Hz / frame, it is determined to be a positive drift. This direction determination provides a crucial basis for subsequent stability correction. Furthermore, the frequency difference sequence between adjacent frames, such as -0.1 Hz, -0.1 Hz, -0.1 Hz, and -0.1 Hz, is calculated to obtain the instantaneous difference sequence of the frequency change rate. Its mean is -0.1 Hz / frame, and its variance is extremely small at 0.002 Hz² / frame², indicating that the frequency change is relatively smooth rather than abrupt.

[0065] Understandably, the stability assessment logic incorporates the standard deviation of power amplitude fluctuation, the slope of frequency drift velocity, the instantaneous frequency drift direction, and the statistics of frequency change rate. If the standard deviation of power fluctuation is less than the preset limit of 0.5 μV² / Hz, and the absolute value of the frequency drift velocity slope is less than 0.15 Hz / frame, then the time dimension stability is initially determined to be high.

[0066] In one possible implementation, the stability results are further refined by combining the characteristics of a small mean and low variance in the negative drift direction and frequency change rate. Even with slight negative drift, the final stability assessment remains high due to the gradual change and stable power. This result indicates that although the current EEG alpha rhythm shows a slow, low-frequency shift trend, the overall resonant state remains stable. This helps to determine that the anesthetic effect is in a relatively stable maintenance phase rather than a phase of rapid deepening or weakening, thus providing a reliable time stability reference for clinical dose fine-tuning.

[0067] S104. Set dynamic weighting coefficients based on the standard deviation of power amplitude fluctuation and the slope of frequency drift velocity, adjust the frequency axis offset direction and the absolute difference of power peak, and recalculate the weighted dynamic matching degree by combining the full width at half maximum (FWHM) measurement of the resonant peak width and the count of amplitude mutation points in the alpha band, and determine the adjusted matching level.

[0068] By collecting data on power amplitude and fluctuation standards, an initial signal characteristic dataset is obtained. For the collected signal characteristics, a pre-defined standardization method is used to obtain standardized signal baseline data. Based on the standardized signal baseline data, a joint analysis of frequency drift and velocity slope is performed to obtain the trend parameters between the two, determining the dynamic adjustment benchmark for frequency drift. Using the dynamic adjustment benchmark for frequency drift, combined with data on frequency offset and power peak value, the correlation deviation value between the two is calculated. It is determined whether the deviation value exceeds a pre-defined threshold range. If it does, the frequency offset is corrected, obtaining the corrected offset data. Based on the corrected offset data, combined with the gap correction processing logic, the power peak correction result after gap correction is obtained, determining the power adjustment direction of the signal. By analyzing the power adjustment direction, combined with the measurement data of resonance peak width and full width at half maximum (FWHM), the resonance characteristic parameters of the signal are calculated, obtaining the stability index of the resonance peak. Based on the stability index of the resonance peak, combined with the statistical data of amplitude abrupt changes, the abrupt frequency of the alpha band is analyzed to determine whether the abrupt frequency meets the pre-defined stability conditions. If not, the abrupt points are smoothed, obtaining smoothed abrupt distribution data. By using smoothed mutation distribution data and combining it with the matching degree calculation rules, the support vector machine algorithm is used to evaluate the overall matching degree of the signal and determine the final matching level classification result.

[0069] The method involves dynamically adjusting the frequency drift benchmark, combining data on frequency offset and power peak, and calculating the correlation deviation between the two using the following formula:

[0070] Where D represents the correlation deviation value, Δ f Indicates frequency offset. Indicates the frequency offset mean. Indicates the standard deviation of frequency offset. Indicates peak power. Indicates the average peak power. Indicates the standard deviation of peak power; The process involves obtaining the power peak correction result after gap correction based on the corrected offset data and the gap correction processing logic. The calculation formula used is as follows:

[0071] in, λ represents the final peak power after gap correction, γ represents the overall amplitude compensation coefficient, and λ represents the exponentially decaying frequency offset penalty parameter. This indicates the frequency offset value after the calibration process.

[0072] For example, after acquiring the power spectrum characteristics of the raw EEG signal, the acquired signal feature dataset is first normalized using a preset standardization processing method to obtain standardized basic signal data.

[0073] Specifically, the power amplitude and frequency values ​​of each frame are mapped to a uniform scale to avoid the impact of dimensional differences on subsequent analysis.

[0074] In one embodiment, for alpha band data acquired during propofol anesthesia, the raw power values ​​range from 8 μV² / Hz to 22 μV² / Hz. After standardization, these values ​​are uniformly distributed in the 0-1 interval for easy trend comparison. Based on the standardized signal baseline data, a joint analysis is performed on frequency drift and drift velocity slope to obtain the trend parameters between the two, thereby determining the dynamic adjustment benchmark for frequency drift.

[0075] For example, when the frequency of consecutive frames gradually decreases from 10.3Hz to 9.7Hz, and the slope shows a stable negative trend of -0.12Hz / frame, this slope can be used as a benchmark for subsequent deviation correction.

[0076] It should be noted that this dynamic benchmark can reflect the evolution direction of the current resonance state in real time, avoiding misjudgments caused by using a fixed threshold.

[0077] Preferably, the correlation deviation value D between the frequency drift and the power peak is calculated by combining the dynamic adjustment reference of the frequency drift with the data of frequency offset and power peak. If D exceeds a preset threshold range, the frequency offset is specifically corrected to obtain the corrected offset data.

[0078] In one possible implementation, when the average frequency offset of a certain signal deviates significantly from the reference, and the peak power increases abnormally, with a correlation deviation reaching 1.4 and exceeding the threshold of 1.2, the system automatically applies reverse compensation to the frequency offset, with a correction amplitude of approximately 0.15Hz, making subsequent analysis closer to the true resonance center. Based on the corrected offset data and combined with the gap correction processing logic, the power peak correction result after gap correction is obtained, and the direction of signal power adjustment is determined.

[0079] For example, an amplitude compensation coefficient γ is introduced to raise the overall power peak, while an exponentially decaying frequency offset penalty parameter λ is used to weaken the impact of frames with large offsets, ultimately resulting in a smooth power adjustment curve. This directional judgment helps identify whether the power tends to converge or diverge. Furthermore, by analyzing the power adjustment direction and combining the measurement data of the resonant peak width and full width at half maximum (FWHM), the resonant characteristic parameters of the signal are calculated, and the stability index of the resonant peak is obtained.

[0080] Specifically, when the peak width is maintained at around 1.8Hz and the full width at half maximum (FWHM) variation is less than 0.3Hz, the stability index is high, indicating a good resonance mode.

[0081] Understandably, this stability index directly reflects the concentration of the alpha rhythm. Based on the stability index of the resonance peak, combined with statistical data on amplitude abrupt changes, the frequency of abrupt changes in the alpha band is analyzed. If the frequency of abrupt changes does not meet the preset stability conditions, the abrupt change points are smoothed to obtain smoothed abrupt change distribution data.

[0082] For example, if a signal exhibits two amplitude jumps exceeding 30%, Gaussian smoothing makes the abrupt change distribution more continuous, avoiding interference from isolated anomalies in the overall judgment. Using the smoothed abrupt change distribution data, combined with the matching degree calculation rules, a support vector machine algorithm is employed to evaluate the overall matching degree of the signal and determine the final matching level classification result.

[0083] In one embodiment, after inputting multidimensional features into a support vector machine, the output is categorized into three levels: high matching, medium matching, and low matching. When the matching degree is at the high matching level, it indicates that the current EEG signal is highly consistent with the target resonant template, providing a reliable basis for maintaining a stable depth of anesthesia in clinical practice. This multi-step progressive processing method can significantly improve the accuracy of assessment of temporal stability and frequency-power consistency, thereby supporting more precise control of anesthetic drug dosage.

[0084] S105. Obtain real-time infusion rate data of anesthetic drugs, update the offset direction record and amplitude difference parameter of the corresponding template in the standard mode library according to the adjusted matching level, frequency axis offset direction and power peak absolute difference, and simultaneously integrate the instantaneous resonant frequency and dose corresponding interval positioning and amplitude change slope to continuously calculate and generate dose-resonant mode mapping relationship.

[0085] Real-time infusion rate data of the current anesthetic drug is acquired using sensors to obtain an infusion rate sequence. The instantaneous resonant frequency is calculated for the infusion rate sequence to determine the current instantaneous resonant frequency value. This instantaneous resonant frequency is compared with a pre-defined dose-corresponding interval to obtain the current dose-corresponding interval location result. Based on the current dose-corresponding interval location result, the baseline offset direction record and baseline amplitude difference parameter of the corresponding template are extracted from the standard pattern library. The power peak value of the current infusion rate sequence is calculated using spectral analysis, and this value is subtracted from the baseline amplitude difference parameter to obtain the absolute power peak difference. If the absolute power peak difference exceeds a preset threshold, the offset direction (positive or negative) is determined based on the frequency axis offset direction to obtain the offset direction determination result. Based on the offset direction determination result and the absolute power peak difference, the offset direction record and amplitude difference parameter of the corresponding template in the standard pattern library are updated to obtain the updated offset direction record and updated amplitude difference parameter. The updated offset direction record and the updated amplitude difference parameter are combined with the current dose-corresponding interval positioning result and the continuous calculation result of amplitude change slope to update the mapping entries of the corresponding interval in the dose-resonance mode mapping relationship, and the latest dose-resonance mode mapping relationship is obtained.

[0086] Specifically, the instantaneous resonant frequency is calculated for the infusion rate sequence to determine the current instantaneous resonant frequency value; the calculation method used is as follows: Step 1: Calculate the discrete differential:

[0087] in, For infusion rate sequence, for The first-order difference; Step 2: Calculate the discrete signal and Teager energy operator:

[0088] Step 3: Calculate the instantaneous frequency :

[0089] in, The sampling rate.

[0090] Among them, the power peak value of the current infusion rate sequence is calculated by using the spectrum analysis method, and the power peak value is obtained. Then, it is subtracted from the reference amplitude difference parameter to obtain the absolute difference of the power peak value. For example, after obtaining the real-time infusion rate sequence of propofol, the instantaneous resonant frequency is first calculated to reflect the dynamic changes in brain electrical activity under the current drug effect.

[0091] Specifically, when the infusion rate is slowly increased from 2.5 mg / kg / h to 3.1 mg / kg / h at 30-second intervals, the instantaneous frequency gradually increases from 9.8 Hz to 10.4 Hz, demonstrating a positive frequency drift caused by dose escalation. This real-time estimation method can promptly capture the impact of infusion changes on the resonant state.

[0092] In one possible implementation, the calculated instantaneous resonant frequency is compared with a pre-defined dose interval.

[0093] For example, 9.5Hz to 10.0Hz corresponds to the low-dose maintenance range, 10.1Hz to 10.6Hz corresponds to the standard anesthesia depth range, and above 10.7Hz corresponds to the high-dose range. If the current instantaneous frequency is 10.3Hz, then the system is positioned within the standard anesthesia depth range, and the baseline offset direction for this range is extracted from the standard pattern library as slightly positive, and the baseline amplitude difference parameter is at a moderate level.

[0094] Preferably, the power peak is extracted from the EEG signal corresponding to the current infusion rate sequence using a spectral analysis method.

[0095] It should be noted that when the signal displays a clear peak shape in the alpha band, with a peak amplitude of 15 μV² / Hz, subtracting this from the reference amplitude difference parameter of 12 μV² / Hz yields an absolute difference of 3 μV² / Hz in the power peak. If this difference exceeds the preset threshold of 2.5 μV² / Hz, it indicates that the current power characteristics have significantly deviated from the standard template.

[0096] It is understandable that targeted processing is carried out based on the results of the frequency axis offset direction judgment.

[0097] For example, if the instantaneous frequency increases positively for three consecutive frames, the offset direction is determined to be positive, and the absolute difference in power peak values ​​is significant, then the offset direction record of the corresponding template in the standard pattern library is updated from neutral to positive, and the amplitude difference parameter is adjusted to a higher value accordingly. This overlay update allows the template to gradually adapt to the individual patient's current EEG response characteristics. Furthermore, the updated offset direction record and amplitude difference parameter are combined with the localization results of the current dose-corresponding interval and the slope of amplitude change.

[0098] For example, if the amplitude change slope shows a slow upward trend of power (0.8 μV² / Hz per minute) over the past minute, then in the dose-resonance mode mapping relationship, the mapping entries for the standard anesthesia depth range will be updated to a positive offset-dominant, high-amplitude type. This updated mapping relationship more closely reflects the real-time physiological state and supports subsequent dose fine-tuning decisions.

[0099] In one embodiment, this dynamic update mechanism based on infusion rate sequence can significantly improve the adaptability of the pattern library to individual patient differences, avoid matching bias caused by the use of static templates, and thus provide a more accurate resonant mode reference for maintaining a stable depth of anesthesia.

[0100] S106. By real-time monitoring of EEG signals, the instantaneous resonant frequency and alpha band amplitude changes are extracted. Based on the dose-resonance mode mapping relationship, the dose interval of the current signal is located. The direction of the transition trend of the anesthesia state is determined by combining the amplitude peak and baseline deviation statistics and the instantaneous difference sequence of frequency change rate. The transition trend sequence of the anesthesia state is obtained.

[0101] Raw EEG data sequences are obtained by real-time acquisition of EEG signals. Instantaneous resonant frequencies and alpha band amplitude changes are extracted from the raw EEG data sequences to obtain frequency and amplitude sequences. The instantaneous resonant frequencies are input into a dose-resonance mode mapping to obtain the current dose interval identifier. The amplitude peak value is calculated from the amplitude sequence and its deviation is statistically analyzed against a pre-established baseline to obtain a deviation statistics sequence. The instantaneous difference sequence is calculated from the frequency sequence to obtain a frequency change rate sequence. If the amplitude peak deviation in the deviation statistics sequence continuously increases and the frequency change rate sequence shows a positive acceleration, the transition trend is determined to be deepening; if the amplitude peak deviation in the deviation statistics sequence continuously decreases and the frequency change rate sequence shows a negative deceleration, the transition trend is determined to be shallowing, thus obtaining the single transition trend direction judgment result. Multiple consecutive single transition trend direction judgment results are concatenated in chronological order to obtain the anesthesia state transition trend sequence.

[0102] For example, after obtaining the original sequence based on the real-time acquired EEG signals, the instantaneous resonant frequency sequence is first extracted using the spectral sliding window method, and the amplitude sequence is calculated from the power integral of the alpha band.

[0103] Specifically, when the patient was in the late stage of anesthesia induction, the instantaneous resonant frequency gradually transitioned from 8.7 Hz to 9.9 Hz, while the amplitude sequence showed a slow increase from a lower 8 μV² / Hz to about 12 μV² / Hz, reflecting the gradual evolution of brain electrical activity to a deeper state.

[0104] In one possible implementation, the current instantaneous resonant frequency is input into a dynamically updated dose-resonant mode mapping for interval localization. For example, if the current frequency is 9.8 Hz, the mapping determines that it falls within the edge interval of the transition from low dose to standard depth, thus identifying the current dose interval as the early transition period.

[0105] It should be noted that the peak position is continuously tracked from the amplitude sequence and the deviation is calculated point by point with the individualized baseline amplitude to form a deviation statistical sequence.

[0106] Preferably, the baseline amplitude is set to the average level of the alpha peak in a patient's awake and quiet state, which is 11 μV² / Hz. If the real-time amplitude peaks are 11.2 μV² / Hz, 12.1 μV² / Hz, and 13.4 μV² / Hz respectively, the deviation statistical sequence shows an increasing trend of +0.2, +1.1, and +2.4, indicating that the amplitude is continuously deviating from the baseline and increasing.

[0107] It is understandable that the frequency change rate sequence can be obtained by taking the first difference of the frequency sequence. For example, if the frequency sequence in the last 30 seconds is 9.6Hz, 9.7Hz, 9.9Hz, and 10.1Hz, the corresponding change rates are approximately +0.1Hz / s, +0.2Hz / s, and +0.2Hz / s, respectively, showing a positive trend and a slight acceleration.

[0108] In one embodiment, when the deviation statistical sequence shows that the peak deviation of the amplitude increases continuously for three consecutive points or more, while the frequency change rate sequence remains positive and shows signs of acceleration, the direction of the single transition trend is judged as deepening. Conversely, if the peak deviation of the amplitude continuously decreases from +2.8 to +1.5, +0.7, and -0.1, and the frequency change rate turns to a negative deceleration such as -0.15Hz / s and -0.22Hz / s, the single judgment result is shallowing. Furthermore, the five consecutive single judgment results are connected in chronological order, for example, deepening, deepening, deepening, remaining unchanged, deepening, forming an anesthesia state transition trend sequence of "deepening-deepening-deepening-remaining unchanged-deepening". This sequence can clearly depict the overall trend of the anesthesia depth continuously evolving towards deeper levels in the past few minutes.

[0109] Preferably, when the proportion of deepening judgments in the transition trend sequence exceeds 70% and the last two frames are both deepening, it can serve as a reliable basis for indicating that the subsequent dose needs to be appropriately reduced; when the proportion of shallowing is dominant and the frequency continues to decelerate negatively, it supports a slight increase in dose to avoid excessive shallowing. Through this multi-dimensional fusion judgment, the transition trend sequence provides clinicians with a continuous and dynamic reference for the direction of changes in anesthesia depth, which helps to intervene in advance to avoid abrupt changes in depth.

[0110] S107. Using time series comparison method, the transition trend sequence of the anesthesia state is matched with the updated template in the standard pattern library. Based on the statistics of the number of consecutive occurrences of the resonance mode, the cumulative duration of the mode interruption interval, and the continuity of the transition trend direction, the EEG pattern state recognition result induced by anesthetic drugs is determined.

[0111] Time series analysis was employed to extract transitional trend features under anesthesia from the acquired EEG signal data, yielding preliminary trend sequence data. The extracted transitional trend sequences were matched one-to-one with updated templates in a standard pattern library to determine the similarity distribution between each template and the trend sequence. If a template's matching value exceeded a preset threshold, its corresponding resonance mode was used as the preliminary identification result, and its consecutive frame count and interruption interval duration were obtained. Based on the statistical results of consecutive frame counts and interruption interval durations, the persistence and stability indices of the resonance mode were calculated to determine if it met preset stability conditions. If the stability indices met the conditions, the directional continuity of the transition trend was further analyzed to obtain the smoothness parameter of directional changes, determining the final EEG pattern state recognition result. By comparing the recognition result with the standard pattern, the template data in the pattern library was updated to optimize the accuracy of subsequent comparisons, resulting in an updated template set. For the updated template set, the above matching and statistical processes were repeatedly executed to continuously track changes in EEG patterns under anesthesia and assess the dynamic consistency of state recognition.

[0112] The persistence index of the resonant mode is calculated based on the statistical results of the number of consecutive frames and the duration of the interruption interval. The calculation formula used is as follows:

[0113] Where N represents the number of consecutive frame segments. denoted by , i represents the number of consecutive frames in the i-th segment, and PS represents the persistence index. This formula quantifies the persistence of the resonant mode by calculating the average length of consecutive frames.

[0114] The stability index of the resonant mode is calculated based on the statistical results of the number of consecutive frames and the duration of the interruption interval. The calculation formula used is as follows:

[0115] Where M represents the number of interrupt segments. represents the duration of the interrupt interval of segment j, and SS represents the stability index. This formula reflects the sparsity of interrupts by the reciprocal of the interrupt frequency. The larger the value, the higher the stability.

[0116] For example, when analyzing EEG signal data under anesthesia, transition trend features can be extracted using time series analysis to initially form trend sequence data.

[0117] Specifically, suppose a 5-minute trend data segment is extracted from a patient's electroencephalogram (EEG) signal. It is observed that certain characteristic values ​​of the signal exhibit periodic fluctuations within a specific time period. These fluctuations may indicate potential changes in the anesthesia state. To address this phenomenon, the frequency and amplitude of these fluctuations can be used as key features to initially construct a trend sequence for subsequent comparative analysis.

[0118] For example, when matching a trend sequence with templates in a standard pattern library, a point-by-point comparison can be used to calculate the similarity distribution of each frame of data. Assume the standard pattern library contains typical resonance pattern templates for various anesthesia states, such as superficial anesthesia and standard deep anesthesia. If the similarity between the current trend sequence and a certain template reaches 85%, which is higher than a preset 75% threshold, then the resonance pattern corresponding to that template can be used as a preliminary identification result. This method helps to quickly pinpoint possible anesthesia state ranges, providing a foundation for subsequent analysis.

[0119] For example, by statistically analyzing the number of consecutive frames and the duration of interruptions in the initial identification results, the persistence and stability of the resonant mode can be further evaluated. Suppose the identification results show that a certain mode appears continuously for 20 frames with only two brief interruptions, each lasting 3 seconds and 5 seconds. By analyzing the consecutive frame segments and interruption intervals, the persistence of this mode can be quantified as high, while the sparseness of interruptions also indicates good stability. This analysis method can effectively screen out reliable pattern recognition results, laying the foundation for dynamic tracking of anesthesia.

[0120] For example, when determining the continuity of a transition trend, the smoothness parameter of the directional change in the trend sequence can be considered. Suppose that during a 10-minute monitoring period, the trend direction gradually changes from a deepening state to a level state without significant jumps; the smoothness parameter would be 0.9, close to the ideal value of 1.0. This indicates a relatively natural transition in the anesthesia state, and the identification result has high reliability. This smoothness analysis helps avoid misjudgments and ensures the dynamic consistency of state identification.

[0121] For example, when updating template data in the standard pattern library, the templates can be dynamically adjusted based on the recognition results. Suppose a patient's EEG pattern during a specific anesthesia phase deviates from the existing templates, the system will integrate newly acquired feature data into the templates, forming an updated template set. This update mechanism continuously optimizes the adaptability of the pattern library, improves the accuracy of subsequent matching, and provides a more reliable reference for clinical monitoring.

[0122] For example, when performing cyclic matching and statistics on an updated template set, subtle changes in the anesthesia state can be continuously tracked. Suppose that during one hour of continuous monitoring, the system detects through cyclic comparison that the anesthesia state has slightly deviated from the standard depth, but remains generally stable. This dynamic consistency assessment provides continuous feedback to healthcare professionals, helping them to adjust intervention strategies promptly and ensure patient safety.

[0123] For example, the comprehensive evaluation of persistence and stability metrics is a crucial step in the entire analysis process. If a pattern consistently exhibits high persistence and stability metrics across multiple cyclic matching sessions—for instance, an average continuous frame length of 15 frames and a low interruption frequency—then the pattern recognition result can be considered highly reliable. This multi-dimensional analysis effectively improves the robustness of state recognition, providing strong support for clinical decision-making.

[0124] If the technical solution of this application involves the collection, storage, use, processing, transmission, provision, disclosure, or deletion of personal information, the products using this technical solution have clearly and understandably informed the users of the personal information processing rules before processing personal information, and have obtained the individuals' voluntary consent in accordance with the law. If the technical solution of this application involves sensitive personal information (such as biometrics, religious beliefs, specific identities, medical and health information, financial accounts, and location tracking), the products using this solution have obtained the individuals' separate consent before processing sensitive personal information, and have also met the requirement of "express consent," ensuring that individuals make authorization decisions voluntarily based on full knowledge.

[0125] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals, characterized in that, The method includes: Raw EEG waveform data under the influence of anesthetic drugs were acquired. Resonant frequency and power amplitude features in the alpha band of each channel signal were extracted to form a frequency-amplitude feature vector. At the same time, the frequency axis offset direction, the absolute difference of power peaks and the full width at half maximum (FWHM) measurement of the resonant peak width were recorded to obtain a preliminary description of the resonant mode. For the frequency-amplitude feature vector, the cosine similarity algorithm is used to compare it with the template in the pre-established standard pattern library. The cosine value of the vector angle is calculated to quantify the similarity score. The preliminary matching level is determined based on the frequency axis offset direction and the absolute difference of the power peak. At the same time, the multi-peak identification results and amplitude spectrum asymmetry detection features within the frequency band are marked to obtain the preliminary matching level results. If the cosine of the vector angle is lower than the preset threshold, the number of consecutive frames of the resonant mode is counted on the signal time axis, the standard deviation of power amplitude fluctuation and the slope of frequency drift velocity are calculated, and the time stability is analyzed by combining the instantaneous resonant frequency drift direction judgment and the instantaneous difference sequence of frequency change rate generation, and the stability evaluation result including instantaneous frequency change rate monitoring is obtained. The dynamic weighting coefficients are set according to the standard deviation of power amplitude fluctuation and the slope of frequency drift velocity. The frequency axis offset direction and the absolute difference of power peak are adjusted. The dynamic matching degree after weighting is recalculated by combining the full width at half maximum (FWHM) measurement of the resonant peak width and the count of amplitude mutation points in the alpha band, and the adjusted matching level is determined. The system acquires real-time infusion rate data of anesthetic drugs, updates the offset direction record and amplitude difference parameter of the corresponding template in the standard mode library according to the adjusted matching level, frequency axis offset direction and power peak absolute difference, and integrates the instantaneous resonant frequency and dose corresponding interval positioning and amplitude change slope to continuously calculate and generate dose-resonant mode mapping relationship. By real-time monitoring of EEG signals to extract instantaneous resonant frequency and alpha band amplitude changes, the dose interval of the current signal is located based on the dose-resonance mode mapping relationship. The direction of the transition trend of the anesthesia state is determined by combining the amplitude peak and baseline deviation statistics and the instantaneous difference sequence of frequency change rate, and the transition trend sequence of the anesthesia state is obtained. The transition trend sequence of the anesthesia state was matched with the updated template in the standard pattern library using a time series comparison method. The results of the EEG pattern state recognition induced by anesthetic drugs were determined based on the statistics of the number of consecutive resonant patterns, the cumulative duration of pattern interruption intervals, and the continuity of the transition trend direction.

2. The method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The process involves acquiring raw EEG waveform data under anesthetic effects, processing the signals from each channel using Fast Fourier Transform (FFT), extracting resonant frequencies and power amplitude features within the alpha band to form a frequency-amplitude feature vector, and simultaneously recording the frequency axis offset direction, the absolute difference in power peak values, and the full width at half maximum (FWHM) measurement of the resonant peak width to obtain a preliminary description of the resonant modes, including: Raw EEG waveform data is acquired using a multi-channel EEG acquisition device; The raw EEG waveform data was processed channel by channel using Fast Fourier Transform to obtain the spectral data for each channel; The resonant frequency and power amplitude in the alpha band are extracted from the spectral data of each channel to generate a frequency-amplitude feature vector. Calculate the frequency axis offset direction and the absolute difference between the power peak value based on the frequency-amplitude eigenvector to obtain the offset direction sequence and the difference sequence; For each resonant peak position in the frequency-amplitude eigenvector, the full width at half maximum (FWHM) of the resonant peak is measured to obtain the FWHM sequence. If the same direction appears consecutively in the offset direction sequence, it is determined that the frequency drift trend is valid, and the resonant frequency sequence and full width at half maximum (FWHM) sequence of the corresponding channel are extracted. Based on the resonant frequency sequence, power amplitude sequence, and full width at half maximum (FWHM) sequence when the frequency drift trend is established, the type of EEG resonant mode under the current anesthetic effect is determined.

3. The method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The frequency-amplitude feature vector is compared with templates in a pre-established standard pattern library using a cosine similarity algorithm. The cosine value of the vector angle is calculated to quantify the similarity score. A preliminary matching level is determined based on the frequency axis offset direction and the absolute difference in power peak values. Simultaneously, multi-peak identification results and amplitude spectrum asymmetry detection features within the frequency band are marked to obtain the preliminary matching level results, including: Obtain the frequency-amplitude feature vector; The cosine similarity algorithm is used to calculate the cosine value of the angle between the frequency-amplitude feature vector and the vector of each template in the standard pattern library to obtain the similarity score; The initial matching level is determined based on the similarity score; Determine whether the offset is positive or negative based on the direction of the frequency axis offset, and obtain the offset direction mark; Based on the absolute difference in power peak values, determine whether the difference belongs to the small, medium, or large level, and obtain the difference level label; The initial matching level is adjusted based on the combination of offset direction markers and gap level markers to obtain the adjusted matching level. For the amplitude spectrum within the frequency band, the multi-peak distribution is identified and the amplitude spectrum asymmetry is detected, yielding multi-peak identification results and asymmetry detection results; The adjusted matching level is associated with the multi-peak recognition results and asymmetric detection results to obtain the final preliminary matching level result.

4. The method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals according to claim 1, characterized in that, If the cosine of the vector angle is lower than a preset threshold, the number of consecutive frames of the resonant mode on the signal time axis is counted, the standard deviation of power amplitude fluctuation and the slope of frequency drift velocity are calculated, and the time stability is analyzed by combining the instantaneous resonant frequency drift direction judgment and the instantaneous difference sequence of frequency change rate generation, resulting in a stability assessment result including instantaneous frequency change rate monitoring, including: Obtain the cosine of the angle between the current signal frame vector and the reference vector; If the cosine value of the included angle is lower than the preset threshold, the subsequent processing will proceed; otherwise, it will be determined as a non-resonant state and the evaluation will end. The number of consecutive frames that satisfy the resonance condition is counted on the signal time axis to obtain the number of consecutively occurring frames; For the power sequence corresponding to consecutive frames, calculate the standard deviation of power amplitude fluctuation to obtain the standard deviation of power amplitude fluctuation. Linear fitting is performed on the instantaneous resonant frequency sequence within a consecutive number of frames to obtain the frequency drift velocity slope; The instantaneous frequency drift direction is determined by the sign of the slope of the frequency drift velocity. The frequency difference between adjacent frames is calculated from the instantaneous resonant frequency sequence within a consecutive number of frames to obtain the instantaneous difference sequence of frequency change rate; The mean and variance of the instantaneous difference sequence of the rate of change of frequency are calculated to obtain the statistic of the rate of change of frequency; The stability determination logic incorporates the standard deviation of power amplitude fluctuation, the slope of frequency drift velocity, the instantaneous frequency drift direction, and the statistics of frequency change rate. If the standard deviation of power amplitude fluctuation is less than the preset fluctuation limit and the absolute value of frequency drift velocity slope is less than the preset drift limit, then the time dimension stability is judged to be high; otherwise, it is judged to be low. By combining the instantaneous frequency drift direction and the frequency change rate statistics, the stability judgment result is corrected, and the final stability assessment result including instantaneous frequency change rate monitoring is obtained.

5. The method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The dynamic weighting coefficients are set based on the standard deviation of power amplitude fluctuation and the slope of frequency drift velocity. The frequency axis offset direction and the absolute difference between the power peak value are adjusted. The dynamic matching degree is recalculated by combining the full width at half maximum (FWHM) measurement of the resonant peak and the count of amplitude mutation points in the alpha band. The adjusted matching level is then determined, including: By collecting data on power amplitude and fluctuation standards, an initial signal feature dataset is obtained. Based on the collected signal features, a preset standardization processing method is used to obtain standardized basic signal data. Based on standardized signal baseline data, a joint analysis of frequency drift and velocity slope is performed to obtain the trend parameters between the two and determine the dynamic adjustment benchmark for frequency drift. By dynamically adjusting the frequency drift reference and combining the data of frequency offset and power peak, the correlation deviation between the two is calculated, and it is determined whether the deviation exceeds the preset threshold range. If it does, the frequency offset is corrected to obtain the corrected offset data. Based on the corrected offset data and the difference correction processing logic, the power peak correction result after difference correction is obtained, and the power adjustment direction of the signal is determined. By analyzing the direction of power adjustment and combining the measurement data of the resonance peak width and full width at half maximum (FWHM), the resonance characteristic parameters of the signal are calculated, and the stability index of the resonance peak is obtained. Based on the stability index of the resonance peak and combined with the statistical data of amplitude mutation, the mutation frequency of the alpha band is analyzed to determine whether the mutation frequency meets the preset stability conditions. If it does not meet the conditions, the mutation point is smoothed to obtain the smoothed mutation distribution data. By using smoothed mutation distribution data and combining it with the matching degree calculation rules, the support vector machine algorithm is used to evaluate the overall matching degree of the signal and determine the final matching level classification result.

6. The method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals according to claim 5, characterized in that, The method involves dynamically adjusting the frequency drift benchmark, combining data on frequency offset and power peak, and calculating the correlation deviation between the two using the following formula: in, Indicates the correlation deviation value, Δ f Indicates frequency offset. Indicates the frequency offset from the mean. Indicates the standard deviation of frequency offset. Indicates peak power. Indicates the average peak power. Indicates the standard deviation of peak power; The process involves obtaining the power peak correction result after gap correction based on the corrected offset data and the gap correction processing logic. The calculation formula used is as follows: in, λ represents the final peak power after gap correction, γ represents the overall amplitude compensation coefficient, and λ represents the exponentially decaying frequency offset penalty parameter. This indicates the frequency offset value after the calibration process.

7. The method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The process involves acquiring real-time anesthetic infusion rate data, updating the offset direction record and amplitude difference parameters of the corresponding template in the standard pattern library based on the adjusted matching level, frequency axis offset direction, and absolute difference in power peak values. Simultaneously, it incorporates the instantaneous resonant frequency and dose-corresponding interval location and amplitude change slope continuous calculation to generate a dose-resonant mode mapping relationship, including: The infusion rate sequence is obtained by collecting real-time data of the current anesthetic drug infusion rate through sensors. Calculate the instantaneous resonant frequency for the infusion rate sequence and determine the current instantaneous resonant frequency value; The instantaneous resonant frequency is compared with the pre-divided dose-corresponding interval to obtain the current dose-corresponding interval location result; Based on the current dose-corresponding interval positioning results, extract the reference offset direction record and reference amplitude difference parameter of the corresponding template from the standard pattern library; The peak power of the current infusion rate sequence is calculated using spectral analysis. The peak power value is then subtracted from the baseline amplitude difference parameter to obtain the absolute difference in peak power. If the absolute difference between the peak power values ​​exceeds the preset threshold, the offset direction is determined to be positive or negative based on the offset direction of the frequency axis, and the offset direction determination result is obtained. Based on the offset direction judgment result and the absolute difference of the power peak, the offset direction record and amplitude difference parameter of the corresponding template in the standard mode library are overwritten and updated to obtain the updated offset direction record and updated amplitude difference parameter. The updated offset direction record and the updated amplitude difference parameter are combined with the current dose-corresponding interval positioning result and the continuous calculation result of amplitude change slope to update the mapping entries of the corresponding interval in the dose-resonance mode mapping relationship, and the latest dose-resonance mode mapping relationship is obtained.

8. The method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The process involves extracting instantaneous resonant frequency and alpha band amplitude changes by real-time monitoring of EEG signals, locating the dose interval of the current signal based on the dose-resonance mode mapping relationship, and determining the direction of the anesthetic state transition trend by combining the amplitude peak and baseline deviation statistics and the instantaneous difference sequence of frequency change rate, thus obtaining the anesthetic state transition trend sequence, including: Raw EEG data sequences are obtained by real-time acquisition of EEG signals; The instantaneous resonant frequency and alpha band amplitude changes were extracted from the raw EEG data sequence to obtain the frequency sequence and amplitude sequence; The instantaneous resonant frequency is input into the dose-resonant mode mapping to obtain the current dose range identifier; The amplitude peak value is calculated from the amplitude sequence and the deviation is statistically analyzed with a pre-established baseline to obtain the deviation statistical sequence; The frequency rate of change sequence is obtained by calculating the instantaneous difference sequence from the frequency sequence; If the peak deviation of the amplitude in the deviation statistics sequence continues to increase and the frequency change rate sequence shows a positive acceleration, then the transition trend direction is judged to be deepening; if the peak deviation of the amplitude in the deviation statistics sequence continues to decrease and the frequency change rate sequence shows a negative deceleration, then the transition trend direction is judged to be shallowing, thus obtaining the single transition trend direction judgment result. The transition trend sequence of anesthesia state is obtained by connecting the results of multiple consecutive single transition trend direction judgments in chronological order.

9. The method for detecting the depth of anesthesia based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The method employs time-series comparison to match the transition trend sequence of the anesthesia state with the updated template in the standard pattern library. Based on the statistics of consecutive occurrences of the resonant mode, the cumulative duration of the mode interruption interval, and the continuity of the transition trend direction, the EEG pattern state recognition result induced by anesthetic drugs is determined, including: Time series analysis was used to extract transition trend features under anesthesia from the collected electroencephalogram (EEG) signal data, and preliminary trend sequence data were obtained. By comparing the extracted transition trend sequences with the updated templates in the standard pattern library one by one, the similarity distribution between each template and the trend sequence is determined. If a template in the similarity distribution has a matching value higher than the preset threshold, the resonant mode corresponding to that template is taken as the preliminary identification result, and its consecutive frame count and interruption interval duration are obtained. Based on the statistical results of the number of consecutive frames and the duration of the interruption interval, the persistence and stability index of the resonance mode are calculated to determine whether it meets the preset stability conditions. If the stability index meets the conditions, the direction continuity of the transition trend is further analyzed, the smoothness parameter of the direction change is obtained, and the final EEG pattern state recognition result is determined. By comparing the recognition results with the standard patterns, the template data in the pattern library is updated to optimize the accuracy of subsequent comparisons and obtain an updated template set. For the updated template set, the above matching and statistical process is executed repeatedly to continuously track changes in EEG patterns under anesthesia and determine the dynamic consistency of state recognition.

Images