Anesthesia monitoring method and system

Abstract

The invention provides an anesthesia monitoring method and system, and belongs to the technical field of anesthesia monitoring. The method comprises the following steps: acquiring multiple groups of normal state data of each physiological index of the anesthesia monitoring equipment, performing time sequence smoothing preprocessing by taking a physiological index monitoring range as a constraint, constructing a feature library and a discrimination rule of a time sequence correlation analysis model, acquiring real-time physiological index data, and determining whether the real-time physiological index data is normal or not. And searching an optimal feature combination in the feature space in combination with a sliding window statistical algorithm, calculating an abnormal score, and when the abnormal score is greater than a preset threshold, determining an abnormal physiological index and outputting an abnormal category. The method can accurately monitor abnormal physiological indexes in the anesthesia process, reliability and accuracy of anesthesia monitoring are improved, the system comprises a processor and a memory, and the method is achieved when programs or instructions stored in the memory are executed by the processor.

Description

Technical Field

[0001] This application belongs to the field of anesthesia monitoring technology, and more specifically, relates to anesthesia monitoring methods and systems. Background Technology

[0002] In modern medicine, anesthesia monitoring is a crucial part of the surgical process, directly affecting patient safety and the success of the operation. With the continuous development of medical technology, higher demands are being placed on the accuracy and timeliness of anesthesia monitoring.

[0003] Existing anesthesia monitoring methods have many problems in practical applications. These methods often analyze only single physiological indicators, neglecting the temporal correlations and mutual influences between these indicators. However, during anesthesia, the patient's physiological state is a complex system with close relationships between various indicators. For example, changes in heart rate may affect blood pressure, and fluctuations in blood pressure may affect blood oxygen saturation. Without comprehensively considering the relationships between these indicators, it is difficult to accurately determine the patient's true physiological condition, easily leading to misdiagnosis or missed diagnosis.

[0004] Existing anesthesia monitoring systems lack effective data preprocessing and feature extraction methods when processing real-time data, making it difficult to quickly and accurately identify abnormalities from large amounts of real-time data. Furthermore, in terms of abnormality detection and localization, existing methods often fail to accurately determine the specific physiological indicators causing the abnormality, hindering doctors from taking timely and effective intervention measures.

[0005] In the data processing process, existing methods do not make full use of the monitoring range of physiological indicators and do not incorporate them as constraints into the data preprocessing and model building process. As a result, the processed data may exceed the normal physiological range, affecting subsequent analysis and judgment. Summary of the Invention

[0006] The purpose of this application is to provide anesthesia monitoring methods and systems to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides an anesthesia monitoring method, the method comprising: S1: Acquire multiple sets of normal state data for various physiological indicators of the anesthesia monitoring equipment; S2: Using the physiological index monitoring range of the anesthesia monitoring device as a constraint, perform time-series smoothing-based data preprocessing on the normal state data; S3: Construct a feature library and discrimination rules for a time-series correlation analysis model based on the preprocessed normal state data; S4: Obtain real-time physiological indicator data; S5: Combining the sliding window statistical algorithm, the optimal feature combination that minimizes the matching error of the real-time physiological index data is searched in the input dimension of the feature library, i.e., the feature space. S6: Calculate the anomaly score of the real-time physiological indicator data based on the optimal feature combination; S7: When the abnormal score is greater than the preset abnormal threshold, calculate the influence value of the real-time physiological index data corresponding to each of the physiological indicators on the abnormal score, and determine the physiological index corresponding to the maximum influence value as the abnormal physiological index. S8: Output the abnormal category based on the abnormal physiological indicators to complete the anesthesia monitoring of the anesthesia monitoring device.

[0008] In one embodiment of this application, the physiological indicators include heart rate, blood pressure, blood oxygen saturation, respiratory rate, electroencephalogram (EEG) signal, electromyogram (EMG) signal, carbon dioxide partial pressure, and body temperature.

[0009] In one embodiment of this application, S2 specifically includes: Time series fitting is performed on normal state data belonging to the same physiological index category in each group of normal state data to obtain a smoothed sequence representing trend changes, wherein each of the smoothed sequences constitutes a multidimensional time series. Calculate the first difference of each smoothed sequence with respect to the time parameter; Calculate the correlation coefficient between pairwise first-order differences; Calculate the principal component directions of the multidimensional time series; Smoothing parameters for data preprocessing are set by combining the monitoring range of the physiological indicators and the correlation coefficient; Based on the monitoring range of the physiological indicators, and taking the principal component direction as the optimization direction, the smoothing parameter is applied to the multidimensional time series to obtain preprocessed normal state data.

[0010] In one embodiment of this application, S3 specifically includes: The preprocessed normal state data is used as a training set and input into the time series correlation analysis model; The time-series correlation analysis model is constructed by alternately updating the feature library and the discrimination rules, calibrating the unupdated feature library and discrimination rules respectively.

[0011] In one embodiment of this application, the matching error includes feature library matching error and discrimination rule verification error.

[0012] In one embodiment of this application, the search process for the optimal feature combination specifically includes: A combination of basic features is randomly selected from the feature space, and a standard sample representing the physiological state is generated based on the feature data extracted from the feature library using a specific algorithm. Based on the combination of the basic features, the matching probability between the standard sample and the real-time physiological index data is calculated using the discrimination rule. The cumulative absolute difference between the real-time physiological index data and the standard sample is calculated to obtain the feature library matching error. The cumulative absolute difference between the matching probabilities of the standard sample and the real-time physiological index data is calculated to obtain the verification error of the discrimination rule. Calculate the first weighted sum between the feature library matching error and the discrimination rule verification error; In the feature space, the basic feature combination is optimized by combining the sliding window statistical algorithm to obtain the optimized feature combination; The basic feature combination is updated using the optimized feature combination, and the matching probability of the standard sample and the real-time physiological index data is calculated based on the basic feature combination and the discrimination rule, until the number of iterations is greater than the preset number of iterations. The optimal feature combination is the basic feature combination that minimizes the first weighted sum between the feature library matching error and the discrimination rule verification error.

[0013] In one embodiment of this application, optimizing the basic feature combination in the feature space using the sliding window statistical algorithm to obtain an optimized feature combination specifically includes: The feature space is divided into sliding windows of fixed length to obtain local feature subsets; Based on the assumption of temporal continuity of the feature library, the local feature subset is initially screened; The optimized feature combination is obtained by secondary filtering of the local feature subset using the residual of the discrimination rule.

[0014] In one embodiment of this application, S6 specifically includes: Calculate the matching error and verification error under the optimal feature combination, respectively; Calculate a second weighted sum between the matching error and the verification error; The second weighted sum is used as the anomaly score output.

[0015] In one embodiment of this application, S7 specifically includes: Calculate the partial difference value of the abnormal score on various physiological indicators in the real-time physiological indicator data, and use the partial difference value as the influence value; The physiological index corresponding to the maximum impact value is output as the abnormal physiological index.

[0016] In one embodiment of this application, the present invention also includes an anesthesia monitoring system, the system comprising: a processor and a memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the anesthesia monitoring method described above.

[0017] The beneficial effects of the anesthesia monitoring method and system provided by this invention are as follows: The anesthesia monitoring method provided by this invention has several significant effects during anesthesia monitoring. Firstly, by acquiring multiple sets of normal-state data for various physiological indicators from the anesthesia monitoring equipment, and performing time-series smoothing-based data preprocessing constrained by the monitoring range of these physiological indicators, noise and fluctuations in the data can be effectively removed, making the data smoother and more stable, thereby improving data quality and reliability. This preprocessing method fully considers the normal monitoring range of physiological indicators, ensuring that the processed data conforms to physiological laws, laying a solid foundation for subsequent analysis.

[0018] When constructing the feature library and discrimination rules for the time-series correlation analysis model, preprocessed normal-state data is used as the training set input to the model. Calibration is performed by alternately updating the feature library and discrimination rules, enabling the model to fully learn the time-series correlations between physiological indicators and the characteristic patterns under normal conditions. This construction method accurately describes the changing patterns of physiological indicators under normal conditions, providing a reliable basis for subsequent anomaly detection.

[0019] After acquiring real-time physiological indicator data, a sliding window statistical algorithm is used to search for the optimal feature combination in the feature space. This method can fully utilize the information in the feature space to find the feature combination with the smallest matching error to the real-time physiological indicator data. By considering the feature library matching error and the discrimination rule verification error, the effectiveness of the feature combination can be comprehensively evaluated, thereby improving the accuracy of anomaly detection.

[0020] Anomaly scores for real-time physiological indicators are calculated based on the optimal feature combination. When the anomaly score exceeds a preset anomaly threshold, the impact of each real-time physiological indicator's corresponding data on the anomaly score is calculated, and the physiological indicator corresponding to the largest impact value is identified as the abnormal physiological indicator. This method can accurately locate abnormal physiological indicators.

[0021] This method comprehensively considers multiple physiological indicators, including heart rate, blood pressure, blood oxygen saturation, respiratory rate, electroencephalogram (EEG) signals, electromyogram (EMG) signals, carbon dioxide partial pressure, and body temperature, enabling comprehensive monitoring of the patient's physiological state. By analyzing the temporal correlations between these physiological indicators, it can more accurately determine the patient's true physiological condition, avoiding the limitations of single-indicator analysis.

[0022] During data preprocessing, time series fitting is performed on normal state data of the same physiological indicator category to obtain a smoothed sequence representing trend changes. The first difference, correlation coefficient, and principal component direction are calculated. Smoothing parameters are set in combination with the monitoring range of physiological indicators, which can fully consider the temporal characteristics and physiological constraints of the data, making the preprocessed data more consistent with the actual physiological situation.

[0023] In the search process for the optimal feature combination, dividing the feature space into a sliding window of fixed length for initial and secondary screening can improve search efficiency and accuracy, quickly find the optimal feature combination, and thus improve the real-time performance of anomaly detection.

[0024] By calculating the partial difference values ​​of abnormal scores on various physiological indicators in real-time physiological indicator data as the influence values, the contribution of each physiological indicator to the abnormal score can be accurately reflected, thereby accurately identifying abnormal physiological indicators.

[0025] The anesthesia monitoring system provided by this invention includes a processor and a memory. When the program or instructions stored in the memory are executed by the processor, they implement the steps of the above-mentioned anesthesia monitoring method. It has high reliability and stability, and can provide doctors with accurate anesthesia monitoring information in actual clinical applications, improve the quality and efficiency of anesthesia monitoring, and protect the life safety of patients. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a schematic diagram illustrating the working principle of the anesthesia monitoring method described in this invention. Figure 2 A flowchart for data preprocessing; Figure 3 Flowchart for constructing a time-series correlation analysis model; Figure 4 A flowchart for searching for the optimal feature combination; Figure 5 A flowchart for optimizing feature combinations for a sliding window. Detailed Implementation

[0028] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0029] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0030] Please see Figures 1-5 This invention provides an anesthesia monitoring method and system, the specific implementation steps of which are as follows: S1: Acquire multiple sets of normal-state data for various physiological indicators from anesthesia monitoring equipment. These physiological indicators include heart rate, blood pressure, blood oxygen saturation, respiratory rate, electroencephalogram (EEG) signals, electromyogram (EMG) signals, carbon dioxide partial pressure, and body temperature. Real-time monitoring of the patient's physiological indicators under normal physiological conditions is achieved using sensors and other devices, thereby collecting multiple sets of corresponding normal-state data.

[0031] S2: Using the physiological indicator monitoring range of the anesthesia monitoring equipment as a constraint, time-series smoothing-based data preprocessing is performed on the normal state data. Data belonging to the same physiological indicator category in each group of normal state data are fitted with time series data to obtain smoothed sequences representing trend changes. These smoothed sequences together form a multidimensional time series. Next, the first-order difference of each smoothed sequence relative to the time parameter is calculated, followed by the correlation coefficient between each pair of first-order differences. Simultaneously, the principal component direction of the multidimensional time series is calculated. Then, smoothing parameters for data preprocessing are set based on the physiological indicator monitoring range and the correlation coefficient. Finally, considering the physiological indicator monitoring range and using the principal component direction as the optimization direction, the smoothing parameters are applied to the multidimensional time series to obtain the preprocessed normal state data.

[0032] S3: Construct the feature library and discrimination rules for the time-series correlation analysis model based on the preprocessed normal state data. Input the preprocessed normal state data as the training set into the time-series correlation analysis model, and calibrate the unupdated feature library and discrimination rules by alternately updating them, thereby completing the construction of the time-series correlation analysis model.

[0033] S4: Acquire real-time physiological data. The monitoring module of the anesthesia monitoring equipment collects real-time data on various physiological indicators of the patient, such as heart rate and blood pressure.

[0034] S5: Combining the sliding window statistical algorithm, the optimal feature combination that minimizes the matching error of real-time physiological indicator data is searched in the feature space (input dimension of the feature library). This matching error includes the feature library matching error and the discrimination rule verification error. Basic feature combinations are randomly selected from the feature space. Based on the feature data extracted from the feature library, standard samples representing physiological states are generated using a specific algorithm. Based on the basic feature combinations, the matching probability between the standard samples and real-time physiological indicator data is calculated using discrimination rules. The cumulative absolute difference between the real-time physiological indicator data and the standard samples is calculated to obtain the feature library matching error; the cumulative absolute difference between the matching probabilities of the standard samples and the real-time physiological indicator data is also calculated to obtain the discrimination rule verification error. The first weighted sum of the feature library matching error and the discrimination rule verification error is calculated. In the feature space, the feature space is divided into fixed-length sliding windows to obtain local feature subsets. Based on the assumption of temporal continuity of the feature library, the local feature subsets are initially screened, and then further screened using the discrimination rule residuals to obtain the optimized feature combination. The basic feature combination is updated by optimizing the feature combination. The above steps of calculating the matching probability are repeated until the number of iterations is greater than the preset number of iterations. The basic feature combination that minimizes the first weighted sum value is output as the optimal feature combination.

[0035] S6: Calculate the anomaly score for real-time physiological indicator data based on the optimal feature combination. Calculate the matching error and validation error under the optimal feature combination, calculate the second weighted sum between the two, and output the second weighted sum as the anomaly score.

[0036] S7: When the abnormal score exceeds a preset abnormal threshold, calculate the impact of real-time physiological indicator data on the abnormal score for each physiological indicator, and determine the physiological indicator corresponding to the maximum impact value as the abnormal physiological indicator. Specifically, calculate the partial difference value of the abnormal score on various physiological indicators in the real-time physiological indicator data, use this partial difference value as the impact value, and output the physiological indicator corresponding to the maximum impact value as the abnormal physiological indicator.

[0037] S8: Outputs the abnormal category based on abnormal physiological indicators, and completes the anesthesia monitoring of the anesthesia monitoring equipment.

[0038] Example 1: In the implementation of the anesthesia monitoring method, this example mainly focuses on the detailed implementation of the time-series smoothing-based data preprocessing process in step S2. Specifically, it is as follows: After acquiring multiple sets of normal-state data for various physiological indicators from the anesthesia monitoring equipment, these data need to be preprocessed. These physiological indicators include heart rate, blood pressure, blood oxygen saturation, respiratory rate, electroencephalogram (EEG) signals, electromyogram (EMG) signals, partial pressure of carbon dioxide, and body temperature. For each set of normal-state data, data belonging to the same physiological indicator category need to be grouped together. For example, for the heart rate indicator, there may be normal-state data collected at multiple time points; these data should be aggregated.

[0039] Time series fitting is performed on normal state data of the same physiological indicator category. The purpose of time series fitting is to obtain a smooth sequence that represents the trend of change of the physiological indicator. When performing fitting, it is necessary to choose an appropriate fitting method. Different physiological indicators may have different characteristics of change, so it is necessary to choose an appropriate fitting algorithm according to the specific situation. For example, for physiological indicators with relatively gentle changes, such as body temperature, a simple linear fitting can be used; while for physiological indicators with more complex changes, such as heart rate, polynomial fitting or spline fitting methods may be needed. Through fitting, the original discrete data points are transformed into a continuous smooth sequence, thereby better reflecting the changing trend of the physiological indicator.

[0040] After time series fitting of normal state data for each physiological indicator category is completed, these smoothed sequences together form a multidimensional time series. The multidimensional time series contains smoothed change information of multiple physiological indicators, providing a more comprehensive reflection of the patient's physiological state.

[0041] It is necessary to calculate the first-order difference of each smoothed sequence relative to the time parameter. The first-order difference reflects the rate of change of the smoothed sequence over time. For each smoothed sequence, the differences between adjacent time points are calculated sequentially to obtain the first-order difference sequence. Through the first-order difference, the rate of change of physiological indicators can be understood more clearly.

[0042] Calculate the correlation coefficients between each pair of first-order differences. Correlation coefficients measure the correlation between the first-order differences of different physiological indicators. For example, calculate the correlation coefficient between the first-order differences of heart rate and blood pressure to understand the degree of correlation between the rates of change of heart rate and blood pressure. By calculating the correlation coefficients between all pairwise first-order differences, a correlation coefficient matrix can be obtained, which reflects the interrelationships between the rates of change of various physiological indicators.

[0043] It is also necessary to calculate the principal component directions of the multidimensional time series. Principal component analysis is a dimensionality reduction method. By calculating the principal component directions, the main directions of change in the data can be found, thus simplifying the data analysis. The principal component directions can reflect the most important trends of change in the multidimensional time series.

[0044] After obtaining the correlation coefficient and principal component orientation, smoothing parameters for data preprocessing are set based on the physiological indicator monitoring range. The physiological indicator monitoring range refers to the pre-set normal ranges for various physiological indicators on the anesthesia monitoring equipment. For example, the normal range for heart rate might be 60-100 beats / minute, and the normal range for blood pressure might be 90-140 mmHg systolic and 60-90 mmHg diastolic. When setting smoothing parameters, both the physiological indicator monitoring range and the correlation coefficient need to be considered. The purpose of the smoothing parameters is to control the degree of data smoothing, removing noise while preserving the main characteristics of the data. If the smoothing parameters are set too high, the data may be over-smoothed, losing important variation information; if they are set too low, noise removal may be ineffective. Therefore, reasonable settings based on the physiological indicator monitoring range and the correlation coefficient are necessary.

[0045] By combining the monitoring range of physiological indicators and using the principal component direction as the optimization direction, smoothing parameters are applied to the multidimensional time series to obtain preprocessed normal state data. Using the principal component direction as the optimization direction when applying smoothing parameters means that the data's trend along the principal component direction should be maintained as much as possible during the smoothing process. In this way, noise can be removed while retaining the most important information in the data, making the preprocessed normal state data more accurate and reliable, and providing high-quality data support for subsequent steps such as building time series correlation analysis models.

[0046] Example 2: In the implementation of the anesthesia monitoring method, this example mainly focuses on the detailed implementation method of constructing the time-series correlation analysis model in step S3. Specifically, it is as follows: After preprocessing the normal state data, preprocessed normal state data is obtained, which will serve as the basis for constructing the time-series correlation analysis model. At this point, the preprocessed normal state data needs to be input into the time-series correlation analysis model as a training set. The time-series correlation analysis model is an important model used to analyze the correlation between physiological indicator data and changes over time, and to make anomaly detection based on this correlation.

[0047] The core of building this model lies in simultaneously constructing a feature library and discrimination rules, and employing an alternating update approach. Specifically, the feature library is updated first. During this update, features are filtered and optimized based on existing discrimination rules. These rules may be initially set or updated through previous iterations. The features in the feature library are evaluated using these discrimination rules to determine the importance and effectiveness of each feature in representing physiological states. Features that perform poorly under the discrimination rules—those that cannot adequately reflect differences in physiological states—are filtered out and removed; while those that perform well are retained or further optimized.

[0048] After updating the feature library, the next step is to calibrate the discrimination rules based on the updated feature library. The purpose of the discrimination rules is to determine whether a physiological state is normal and the type of abnormality based on the features in the feature library. Since the feature library has been updated, the features within it may have changed, such as the number of features or their weights. Therefore, the discrimination rules need to be adjusted accordingly to adapt to the new feature library. When calibrating the discrimination rules, the parameters and conditions of the discrimination rules need to be redefined based on the features in the updated feature library. For example, it may be necessary to adjust the discrimination threshold or modify the combination of discrimination conditions to ensure that the discrimination rules can accurately discriminate physiological states based on the new feature library.

[0049] After calibrating the discrimination rules, the feature library is updated again. At this point, based on the calibrated discrimination rules, the features in the feature library are re-evaluated and optimized. Because the discrimination rules have been adjusted, their evaluation criteria for features may have changed, so features need to be screened again to further optimize the feature library. This process may uncover previously overlooked features or exclude features that are no longer important under the new discrimination rules.

[0050] Then, based on the updated feature library, the discrimination rules are calibrated again. This process of updating and calibrating the feature library and discrimination rules is repeated until a preset number of updates is reached. Each alternating update is an optimization of the model, allowing the feature library and discrimination rules to continuously adapt to and improve each other. In this process, the feature library gradually includes features that can more accurately represent physiological states, and the discrimination rules can gradually make more accurate judgments based on these features.

[0051] During the alternating update process, it is crucial to ensure the accuracy and appropriateness of each update and calibration. For feature library updates, it is essential to ensure that the process of selecting and optimizing features is based on the correct evaluation of the discrimination rules; features should not be arbitrarily deleted or added, as this could affect the model's accuracy. For discrimination rule calibration, adjustments should be made reasonably according to changes in the feature library, avoiding excessive adjustments that could cause the discrimination rules to deviate from actual requirements.

[0052] By repeatedly updating the feature library and discrimination rules, the model is continuously calibrated for outdated features and rules, gradually improving the temporal correlation analysis model. As the number of updates increases, the model's ability to analyze physiological data improves, enabling it to more accurately capture the temporal correlations between physiological indicators. This provides reliable model support for subsequent real-time physiological data processing and anomaly detection.

[0053] Throughout the model building process, each alternating update must be strictly performed according to the procedure to ensure that every operation is based on evidence and can effectively improve the model's performance. The preprocessed normal state data, used as the training set, plays a crucial role in the model building process. It provides real-world data support for updating the feature library and calibrating the discrimination rules, enabling the constructed model to better adapt to actual anesthesia monitoring scenarios.

[0054] Example 3: In the implementation of the anesthesia monitoring method, this example mainly focuses on the detailed implementation of step S5, which involves searching for the optimal feature combination using a sliding window statistical algorithm. Specifically, it is as follows: After obtaining real-time physiological indicator data in step S4, it is necessary to search for the optimal feature combination that minimizes the matching error of the real-time physiological indicator data in the input dimension of the feature library, i.e., the feature space. In this process, the matching error includes the feature library matching error and the discrimination rule verification error. The specific search process is as follows: Basic feature combinations are randomly selected from the feature space. The feature space is a set of all possible physiological indicator features, each corresponding to a specific dimension of the physiological indicator data, such as the rate of change of heart rate or the amplitude of blood pressure fluctuations. The basic feature combination is a set of features selected from the feature space, used to initially construct an analytical model of the physiological state.

[0055] Based on feature data extracted from a feature library, standard samples representing physiological states are generated using specific algorithms. The feature library stores a set of features from preprocessed normal state data; these features have been filtered and optimized to characterize patterns of normal physiological states. Based on basic feature combinations, corresponding feature data are extracted from the feature library, and standard samples are generated using specific algorithms (such as mean calculation and principal component analysis). These standard samples represent normal physiological state patterns under the basic feature combinations.

[0056] Based on the combination of fundamental features, the matching probability between standard samples and real-time physiological indicator data is calculated using discriminant rules. The discriminant rules are a set of rules used in the time-series correlation analysis model to determine physiological states, containing a series of judgment conditions and thresholds. For standard samples and real-time physiological indicator data, their features are input into the discriminant rules. Through the rule's judgment logic, the degree of matching between the two on various feature dimensions is obtained, and then the matching probability is comprehensively calculated. This probability reflects the similarity between real-time physiological indicator data and normal physiological state patterns.

[0057] The cumulative absolute difference between real-time physiological indicator data and standard samples is calculated to obtain the feature library matching error. For each feature in the basic feature combination, the absolute difference between the real-time physiological indicator data value and the standard sample value for that feature is calculated. Then, the absolute differences of all features are accumulated and summed to obtain the feature library matching error. This error reflects the overall deviation between real-time data and normal patterns in feature values.

[0058] The cumulative absolute difference between the matching probabilities of standard samples and real-time physiological indicator data is calculated to obtain the discrimination rule verification error. Specifically, the absolute difference between the matching probability of standard samples obtained through the discrimination rule and the matching probability of real-time physiological indicator data is calculated. If multiple matching probabilities exist (such as probabilities of different discrimination dimensions), these absolute differences are accumulated and summed to obtain the discrimination rule verification error. This error reflects the degree of deviation between real-time data and normal patterns in the discrimination rule judgment results.

[0059] Calculate the first weighted sum between the feature library matching error and the discrimination rule verification error, denoted as The calculation formula is as follows:

[0060] in, This indicates the feature library matching error. This indicates the verification error of the discrimination rule. and The preset weighting coefficients, and . Used to measure the importance of feature library matching error in the overall error. The weighting coefficient is used to measure the importance of the verification error of the discrimination rule. It can be set according to the actual anesthesia monitoring needs and historical data experience.

[0061] After obtaining the first weighted sum, the basic feature combination needs to be optimized in the feature space using a sliding window statistical algorithm to obtain an optimized feature combination. The specific steps are as follows: The feature space is divided into sliding windows of fixed length to obtain local feature subsets. The length of the sliding window is set according to the dimension of the feature space and computational efficiency requirements; for example, the feature space is divided sequentially into... Each feature is divided into a window, and each window corresponds to a local feature subset. The windows slide sequentially across the feature space to cover the entire feature space.

[0062] Based on the temporal continuity assumption of the feature library, an initial screening of local feature subsets is performed. The temporal continuity assumption of the feature library assumes that the changes in physiological indicators are continuous over time, and that features at adjacent time points have strong correlations. Therefore, within the local feature subset, features that are continuous over time and conform to normal physiological changes are selected, while features that are abrupt in time and do not conform to the continuity assumption are excluded.

[0063] A secondary screening of local feature subsets is performed using the discriminant rule residuals to obtain an optimized feature combination. The discriminant rule residual refers to the difference between the results obtained from real-time physiological indicator data using the discriminant rule and the results obtained from standard samples using the same rule. For the initially screened local feature subsets, the discriminant rule residuals for each feature are calculated. Features with smaller residuals are retained, while those with larger residuals are removed, resulting in an optimized feature combination. Features in this combination better conform to the judgment logic of the discriminant rule, thus reducing the matching error between real-time data and standard samples.

[0064] The algorithm updates the basic feature combination using optimized feature combinations, then returns to the step of "calculating the matching probability between the standard sample and the real-time physiological indicator data based on the basic feature combination and the discrimination rule." This process of calculating matching probability, error, weighted summation, and feature combination optimization is repeated until the number of iterations exceeds the preset number of iterations. The preset number of iterations is set according to the computational accuracy and efficiency requirements, for example, 100 or 200 times.

[0065] When the number of iterations reaches a preset value, the iteration stops, and the basic feature combination that minimizes the first weighted sum between the feature library matching error and the discrimination rule verification error is output as the optimal feature combination. This optimal feature combination can minimize the matching error of real-time physiological indicator data in the feature space, thereby more accurately reflecting the deviation between real-time physiological state and normal state, and providing reliable feature combination support for subsequent anomaly score calculation and abnormal physiological indicator determination.

[0066] Example 4: In the implementation of the anesthesia monitoring method, this example mainly focuses on the detailed implementation of step S6, which involves calculating the abnormal score of real-time physiological index data based on the optimal feature combination. A specific example is provided below for illustration.

[0067] Suppose that in a certain anesthesia monitoring scenario, the optimal feature combination has been found through step S5. This combination includes three features: the rate of change of heart rate, the amplitude of blood pressure fluctuations, and the downward trend of blood oxygen saturation. At this point, it is necessary to calculate the anomaly score of the real-time physiological data based on this optimal feature combination.

[0068] Acquire current real-time physiological data. For example, a patient's real-time heart rate changes at a rate of 5 beats per minute, blood pressure fluctuations are 15 mmHg of systolic blood pressure compared to baseline, and blood oxygen saturation decreases by 2% every 10 minutes. Simultaneously, using a feature library and discrimination rules, a standard sample representing a normal physiological state is calculated based on the optimal feature combination. It is assumed that the normal range for the standard sample's heart rate change rate is -2 to +3 beats per minute, the normal range for blood pressure fluctuations is no more than 10 mmHg of systolic blood pressure compared to baseline, and the normal range for the decrease in blood oxygen saturation is no more than 0.5% every 10 minutes.

[0069] Calculate the matching error and validation error under the optimal feature combination. For the matching error, calculate the cumulative absolute difference between the real-time physiological index data and the standard sample. In this example, the absolute difference of the heart rate change rate is |5-3|=2, the absolute difference of the blood pressure fluctuation amplitude is |15-10|=5, and the absolute difference of the blood oxygen saturation decreasing trend is |2-0.5|=1.5. Summing these absolute differences, the matching error is 2+5+1.5=8.5.

[0070] The validation error is calculated as the cumulative absolute difference between the matching probabilities of the standard sample and the real-time physiological indicator data. Assuming the matching probability of the standard sample is 0.9 and the matching probability of the real-time physiological indicator data is 0.3, the absolute difference between the two is |0.9 - 0.3| = 0.6, which is the validation error.

[0071] After obtaining the matching error and validation error, a second weighted sum needs to be calculated and used as the anomaly score output. This weighted sum calculation needs to consider the importance of the matching error and validation error in the anomaly score. For example, if the weight of the matching error is preset to 0.6 and the weight of the validation error to 0.4, then the second weighted sum is 8.5 × 0.6 + 0.6 × 0.4 = 5.1 + 0.24 = 5.34, which is the anomaly score for this real-time physiological indicator data.

[0072] For another example, the optimal feature combination might include three features: the rate of change of respiratory rate, the power spectral density of the EEG signal, and the amplitude of the EMG signal. A patient's real-time respiratory rate increases by 8 breaths per minute, the power spectral density of the EEG signal is 15 μV² / Hz higher than normal in a certain frequency band, and the EMG signal amplitude is 50 μV. In the standard sample, the normal range for the rate of change of respiratory rate is -3 to +5 breaths per minute, the normal range for the power spectral density of the EEG signal is no more than 10 μV² / Hz in that frequency band, and the normal range for the amplitude of the EMG signal is no more than 30 μV.

[0073] When calculating the matching error, the absolute difference in the rate of change of respiratory rate is |8-5|=3, the absolute difference in the power spectral density of EEG signals is |15-10|=5, and the absolute difference in the amplitude of EMG signals is |50-30|=20. The cumulative summation yields a matching error of 3+5+20=28. The matching probability of the standard sample is calculated to be 0.85 using the discrimination rule, and the matching probability of the real-time physiological indicator data is 0.2. The verification error is |0.85-0.2|=0.65. If the preset matching error weight is 0.7 and the verification error weight is 0.3, the second weighted sum is 28×0.7+0.65×0.3=19.6+0.195=19.795, meaning the anomaly score is 19.795.

[0074] In practical applications, different optimal feature combinations may include different physiological indicators, and the normal range setting for each feature needs to be based on a large amount of normal state data and medical standards. For example, for the feature of carbon dioxide partial pressure, its normal range may be set to 35-45 mmHg. When calculating the matching error, the absolute difference between the real-time carbon dioxide partial pressure value and the median or boundary value of this range will be included in the cumulative sum.

[0075] When calculating the matching error, it's important to note that the units and dimensions of each feature may differ. Therefore, whether standardization is necessary before summing depends on the specific algorithm design. For example, the unit for heart rate change rate is beats per minute, while the unit for blood pressure fluctuation amplitude is mmHg. Since these two units are different, direct summing might over-amplify or under-amplify the influence of some features. Therefore, it may be necessary to standardize the feature data first, such as converting it into standard scores, to make different features comparable before calculating the cumulative absolute difference.

[0076] The calculation of the verification error is based on the matching probability assessment of standard samples and real-time data by the discrimination rule. The discrimination rule may contain multiple sub-rules, each corresponding to different features or combinations of features. Therefore, the calculation of the matching probability may be a combination of the judgment results of multiple sub-rules. For example, the discrimination rule may first independently judge each feature to obtain the matching probability of each feature, and then obtain the final matching probability through weighted averaging or other combinations.

[0077] The weighting in the second weighted sum needs to comprehensively consider the actual needs of anesthesia monitoring. If more attention is paid to the numerical deviation of the feature itself, a higher weight may be assigned to the matching error; if more attention is paid to the comprehensive judgment result of the discrimination rules, a higher weight may be assigned to the validation error. The weighting can be adjusted jointly by medical experts and algorithm engineers based on clinical experience and historical data.

[0078] By calculating anomaly scores according to the above steps for different optimal feature combinations and real-time physiological indicator data, the degree of deviation between real-time physiological indicator data and the normal state can be objectively reflected. A higher anomaly score indicates a greater deviation of the real-time data from the normal state, suggesting a higher likelihood of anomalies requiring further attention and analysis.

[0079] Example 5: In the implementation of the anesthesia monitoring method, this example mainly focuses on step S7, where, when the abnormal score exceeds a preset abnormal threshold, the influence of real-time physiological indicator data corresponding to each physiological indicator on the abnormal score is calculated, and the physiological indicator corresponding to the maximum influence value is determined as the abnormal physiological indicator. The detailed implementation method is as follows: After calculating the abnormality score of the real-time physiological indicator data in step S6, the abnormality score needs to be compared with a preset abnormality threshold. The preset abnormality threshold is a critical value pre-set based on a large amount of clinical data and medical experience, used to determine whether there is an abnormality in the physiological indicators. For example, the preset abnormality threshold may be set to 5. When the abnormality score is greater than 5, it is considered that there is an abnormality, and further determination of the specific abnormal physiological indicator is required.

[0080] Suppose that in a certain anesthesia monitoring scenario, the calculated abnormality score is 6, which is greater than the preset abnormality threshold of 5. In this case, step S7 needs to be initiated to determine the abnormal physiological indicators. First, the deviation values ​​of the abnormality score on various physiological indicators in the real-time physiological indicator data need to be calculated, and these deviation values ​​are used as the influence values. These deviation values ​​reflect the degree to which the change in each physiological indicator contributes to the abnormality score, that is, the magnitude of the impact of the degree to which the real-time data of that physiological indicator deviates from the normal state data on the abnormality score.

[0081] When calculating the partial differential value, it is necessary to consider the weight and influence of each physiological indicator in the abnormality score calculation process. Different physiological indicators may have different importance in the calculation of the abnormality score, which may be due to their different degrees of importance in reflecting the patient's physiological state, or different weights in the feature combination. For example, heart rate and blood pressure may have a larger weight in the abnormality score, while body temperature may have a smaller weight.

[0082] Taking specific physiological indicators as an example, let's assume that real-time physiological indicator data includes three indicators: heart rate, blood pressure, and blood oxygen saturation. First, let's analyze the impact of heart rate on abnormal scores. Real-time heart rate data may deviate from normal data. For example, the normal heart rate range is 60-100 beats per minute, while the real-time heart rate is 110 beats per minute, deviating from the normal range. We need to calculate the deviation of this heart rate deviation from the abnormal score, that is, how much the abnormal score increases when this deviation occurs in heart rate.

[0083] Similarly, for blood pressure indicators, the normal systolic blood pressure range is 90-140 mmHg, and the diastolic blood pressure range is 60-90 mmHg. If the real-time systolic blood pressure is 150 mmHg and the diastolic blood pressure is 95 mmHg, the partial variance of this blood pressure deviation on the abnormality score needs to be calculated. For blood oxygen saturation, the normal range is generally above 95%. If the real-time blood oxygen saturation is 92%, its partial variance on the abnormality score also needs to be calculated.

[0084] The calculation of partial variance values ​​requires consideration of the outlier scoring model and the role of each physiological indicator within that model. For example, in the calculation of outlier scores, the deviation of each physiological indicator might be multiplied by a corresponding weighting coefficient and then accumulated into the outlier score. Therefore, the calculation of partial variance values ​​can be understood as multiplying the deviation of each physiological indicator by its corresponding weighting coefficient.

[0085] Assuming that in the anomaly scoring calculation model, the weighting coefficients for heart rate, blood pressure, and blood oxygen saturation are all 0.3, and the real-time heart rate is 10 beats / minute higher than normal (each deviation of 1 beat / minute corresponds to an impact value of 0.2), then the partial difference for heart rate is 10 × 0.2 × 0.4 = 0.8. Similarly, if the real-time systolic blood pressure is 10 mmHg higher than normal (each deviation of 1 mmHg corresponds to an impact value of 0.15), and the diastolic blood pressure is 5 mmHg higher than normal (each deviation of 1 mmHg corresponds to an impact value of 0.1), then the partial difference for blood pressure is (10 × 0.15 + 5 × 0.1) × 0.3 = (1.5 + 0.5) × 0.3 = 2 × 0.3 = 0.6. Finally, if the real-time blood oxygen saturation is 3% lower than normal (each deviation of 1% corresponds to an impact value of 0.1), then the partial difference for blood oxygen saturation is 3 × 0.1 × 0.3 = 0.09.

[0086] After calculating the partial differences for each physiological indicator, it is necessary to compare the magnitudes of these values ​​and identify the largest influencing value. In the example above, the influence value of heart rate is 0.8, the influence value of blood pressure is 0.6, and the influence value of blood oxygen saturation is 0.09. The largest influence value is 0.8, which corresponds to the physiological indicator heart rate. Therefore, heart rate is identified as an abnormal physiological indicator.

[0087] For example, in another scenario, real-time physiological indicators include respiratory rate, EEG signals, EMG signals, and body temperature. Assuming the abnormal score exceeds a preset threshold, the impact value of each indicator needs to be calculated. The real-time respiratory rate is 8 breaths / minute higher than normal, a certain characteristic value of the EEG signal is 15 μV higher than normal, the amplitude of the EMG signal is 20 μV higher than normal, and the body temperature is 0.5℃ higher than normal. In the abnormal score model, the weight of respiratory rate is 0.3, the weight of EEG signal is 0.35, the weight of EMG signal is 0.25, and the weight of body temperature is 0.1. The impact value for each deviation of 1 breath / minute in respiratory rate is 0.15, the impact value for each 1 μV deviation in the EEG signal characteristic value is 0.08, the impact value for each 1 μV deviation in the EMG signal amplitude is 0.05, and the impact value for each 0.1℃ deviation in body temperature is 0.2.

[0088] The deviation values ​​for respiratory rate are 8 × 0.15 × 0.3 = 0.36, for EEG signal deviations are 15 × 0.08 × 0.35 = 0.42, for EMG signal deviations are 20 × 0.05 × 0.25 = 0.25, and for body temperature deviations are 0.5 ÷ 0.1 × 0.2 × 0.1 = 5 × 0.2 × 0.1 = 0.1. Comparing these values, 0.42 is the largest, and the corresponding EEG signal is the abnormal physiological indicator.

[0089] In practical applications, there may be more types of physiological indicators, including those mentioned above such as heart rate, blood pressure, blood oxygen saturation, respiratory rate, electroencephalogram (EEG) signals, electromyogram (EMG) signals, partial pressure of carbon dioxide, and body temperature. For each physiological indicator, its bias variability in abnormality scores needs to be calculated using the method described above.

[0090] Calculating the partial differential value requires accurately obtaining the deviation between real-time data and normal state data for each physiological indicator, as well as the weight coefficient of that physiological indicator in the abnormality scoring model and the impact value corresponding to each unit deviation. The setting of these parameters needs to be based on a large amount of clinical data and medical research to ensure that the calculated partial differential value can truly reflect the degree of influence of each physiological indicator on the abnormality score.

[0091] After determining the physiological indicator corresponding to the maximum impact value, it is output as an abnormal physiological indicator. This process can accurately locate the specific abnormal physiological indicator, providing a clear direction for subsequent abnormality classification and corresponding clinical treatment. For example, if the abnormal physiological indicator is determined to be heart rate, the doctor can focus on the patient's cardiac function and conduct further examinations and treatments; if it is an abnormal electroencephalogram (EEG) signal, it may be necessary to pay attention to the patient's depth of anesthesia or neurological status.

[0092] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for anesthesia monitoring, characterized in that, include: S1: Acquire multiple sets of normal state data for various physiological indicators of the anesthesia monitoring equipment; S2: Using the physiological index monitoring range of the anesthesia monitoring device as a constraint, perform time-series smoothing-based data preprocessing on the normal state data; S3: Construct a feature library and discrimination rules for a time-series correlation analysis model based on the preprocessed normal state data; S4: Obtain real-time physiological indicator data; S5: Combining the sliding window statistical algorithm, the optimal feature combination that minimizes the matching error of the real-time physiological index data is searched in the input dimension of the feature library, i.e., the feature space. S6: Calculate the anomaly score of the real-time physiological indicator data based on the optimal feature combination; S7: When the abnormal score is greater than the preset abnormal threshold, calculate the influence value of the real-time physiological index data corresponding to each physiological index on the abnormal score, and determine the physiological index corresponding to the maximum influence value as the abnormal physiological index. S8: Output the abnormal category based on the abnormal physiological indicators to complete the anesthesia monitoring of the anesthesia monitoring device.

2. The anesthesia monitoring method according to claim 1, characterized in that, The physiological indicators include heart rate, blood pressure, blood oxygen saturation, respiratory rate, electroencephalogram (EEG) signal, electromyogram (EMG) signal, carbon dioxide partial pressure, or body temperature.

3. The anesthesia monitoring method according to claim 1, characterized in that, S2 specifically includes: Time series fitting is performed on normal state data belonging to the same physiological index category in each group of normal state data to obtain a smoothed sequence representing trend changes, wherein each of the smoothed sequences constitutes a multidimensional time series. Calculate the first difference of each smoothed sequence with respect to the time parameter; Calculate the correlation coefficient between pairwise first-order differences; Calculate the principal component directions of the multidimensional time series; Smoothing parameters for data preprocessing are set by combining the monitoring range of the physiological indicators and the correlation coefficient; Based on the monitoring range of the physiological indicators, and taking the principal component direction as the optimization direction, the smoothing parameter is applied to the multidimensional time series to obtain preprocessed normal state data.

4. The anesthesia monitoring method according to claim 1, characterized in that, S3 specifically includes: The preprocessed normal state data is used as a training set and input into the time series correlation analysis model; The time-series correlation analysis model is constructed by alternately updating the feature library and the discrimination rules, calibrating the unupdated feature library and discrimination rules respectively.

5. The anesthesia monitoring method according to claim 1, characterized in that, The matching error includes feature library matching error and discrimination rule verification error.

6. The anesthesia monitoring method according to claim 5, characterized in that, The search process for the optimal feature combination specifically includes: A combination of basic features is randomly selected from the feature space, and a standard sample representing the physiological state is generated based on the feature data extracted from the feature library using a specific algorithm. Based on the combination of the basic features, the matching probability between the standard sample and the real-time physiological index data is calculated using the discrimination rule. The cumulative absolute difference between the real-time physiological index data and the standard sample is calculated to obtain the feature library matching error. The cumulative absolute difference between the matching probabilities of the standard sample and the real-time physiological index data is calculated to obtain the verification error of the discrimination rule. Calculate the first weighted sum between the feature library matching error and the discrimination rule verification error; In the feature space, the basic feature combination is optimized by combining the sliding window statistical algorithm to obtain the optimized feature combination; The basic feature combination is updated using the optimized feature combination, and the matching probability of the standard sample and the real-time physiological index data is calculated based on the basic feature combination and the discrimination rule, until the number of iterations is greater than the preset number of iterations. The optimal feature combination is the basic feature combination that minimizes the first weighted sum between the feature library matching error and the discrimination rule verification error.

7. The anesthesia monitoring method according to claim 6, characterized in that, In the feature space, the basic feature combination is optimized using the sliding window statistical algorithm to obtain an optimized feature combination, specifically including: The feature space is divided into sliding windows of fixed length to obtain local feature subsets; Based on the assumption of temporal continuity of the feature library, the local feature subset is initially screened; The optimized feature combination is obtained by secondary filtering of the local feature subset using the residual of the discrimination rule.

8. The anesthesia monitoring method according to claim 1, characterized in that, S6 specifically includes: Calculate the matching error and verification error under the optimal feature combination, respectively; Calculate a second weighted sum between the matching error and the verification error; The second weighted sum is used as the anomaly score output.

9. The anesthesia monitoring method according to claim 8, characterized in that, Specifically, S7 includes: Calculate the partial difference value of the abnormal score on various physiological indicators in the real-time physiological indicator data, and use the partial difference value as the influence value; The physiological index corresponding to the maximum impact value is output as the abnormal physiological index.

10. An anesthesia monitoring system, characterized in that, include: Processor and memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the anesthesia monitoring method as described in any one of claims 1 to 9.

Images