Intraoperative patient anesthesia state evaluation system based on electroencephalogram signal and muscle relaxation monitoring

Abstract

The invention discloses an intra-operative patient anesthesia state evaluation system based on electroencephalogram signals and muscle relaxation monitoring, and the system comprises a weight generation module which is used for generating a first weight and a second weight according to a personalized vector, an electroencephalogram consciousness prediction index, a personalized vector and a muscle relaxation monitoring prediction index; wherein the first weight represents the contribution degree of the personalized vector to the electroencephalogram consciousness index, and the second weight represents the contribution degree of the personalized vector to the muscle relaxation monitoring index; the anesthesia state generation module is used for generating a corresponding anesthesia state according to the anesthesia state evaluation value; according to the anesthesia state evaluation method, the first weight and the second weight are introduced, the electroencephalogram consciousness index predicted by the model, the muscle relaxation monitoring index and the personalized vector obtained in real time are combined, the influence of the personalized features on the anesthesia state can be quantified, the anesthesia state evaluation considers the physiological feature difference of the patient, and the evaluation accuracy is improved. And the anesthesia state evaluation is more personalized and targeted.

Description

Technical Field

[0001] The present invention relates to anesthesia assessment, in particular to an intraoperative patient anesthesia state assessment system based on electroencephalogram (EEG) signals and muscle relaxation monitoring. Background Art

[0002] Currently, intraoperative patient anesthetic status assessment methods based on EEG signals and muscle relaxation monitoring primarily rely on trained models to predict indicators such as the EEG consciousness index and muscle relaxation monitoring index. These predictive indicators can effectively reflect the patient's anesthetic status and help anesthesiologists determine the depth of anesthesia and degree of muscle relaxation. However, traditional prediction methods typically use pre-trained models to directly predict these indicators and then assess the patient's anesthetic status based on these predicted values.

[0003] For example, the patent document with patent publication number CN116712084A discloses an anesthetic consciousness state assessment system and method based on information integration theory. It starts from the problem of anesthetic consciousness state assessment, takes integrated information theory as the basic support, uses the binning method to solve the problem that integrated information estimation cannot be applied to measured EEG signals, and assesses the anesthetic consciousness state from a theoretical perspective. However, the defect in the prior art is that the prediction results often rely on the direct output of some prediction models. In this case, the predicted anesthesia index cannot reflect individual differences well. In particular, the influence of personalized signs (such as age, gender, weight, etc.) on the anesthetic state has not been fully considered. Although the prediction results can be optimized through model training, due to the neglect of the weight calculation of the patient's personalized data, it may not be possible to accurately assess the anesthetic state of each patient, resulting in limitations in the prediction accuracy, which in turn affects the subsequent state parameter adjustment of the patient during surgery. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides an intraoperative patient anesthesia status assessment system based on EEG signals and muscle relaxation monitoring. It evaluates the current intraoperative patient anesthesia status through personalized vectors, first weights and second weights based on a prediction model, and real-time EEG consciousness index and muscle relaxation monitoring index, thereby solving the technical problems raised in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0006] A system for evaluating the anesthesia status of patients during surgery based on EEG signals and muscle relaxation monitoring, comprising:

[0007] Real-time data acquisition module, used to obtain the personalized vector, EEG consciousness index and muscle relaxation monitoring index of the patient during the current operation;

[0008] An index prediction module, configured to input the personalized vector of the current intraoperative patient into a pre-trained dual-index monitoring model, and output an EEG consciousness prediction index and a muscle relaxation monitoring prediction index;

[0009] A weight generation module, configured to generate a first weight and a second weight based on the personalized vector and the EEG consciousness prediction index, and based on the personalized vector and the muscle relaxation monitoring prediction index;

[0010] Among them, the first weight represents the contribution of the personalized vector to the EEG consciousness index, and the second weight represents the contribution of the personalized vector to the muscle relaxation monitoring index;

[0011] An evaluation module, configured to calculate an evaluation value of the anesthetic state of the patient currently undergoing surgery based on the first weight, the second weight, an EEG consciousness index, and a muscle relaxation monitoring index;

[0012] The anesthesia state generating module is used to generate a corresponding anesthesia state according to the anesthesia state evaluation value.

[0013] In some embodiments, obtaining a personalized vector of a patient currently undergoing surgery includes:

[0014] S1-1, collecting several pre-anesthetic vital sign parameters and real-time anesthesia dosage of the patient currently undergoing surgery;

[0015] S1-2, standardizing several pre-anesthetic vital sign parameters and real-time anesthetic dosage to generate several personalized features;

[0016] S1-3. Perform feature concatenation on a plurality of the personalized features to obtain the personalized vector.

[0017] In some embodiments, the device for acquiring the EEG consciousness index is a BIS monitor or a qCON consciousness monitor, and the device for acquiring the muscle relaxation monitoring index is a TOF monitor or a neuromuscular monitor.

[0018] In some embodiments, the pre-training step of the dual exponential monitoring model includes:

[0019] A1. Obtaining a training sample set for the dual-exponential monitoring model;

[0020] A2. Select the MLP model as the initial training model for the dual exponential monitoring model and initialize the first output node and the second output node in its output layer;

[0021] The first output node and the second output node are used to map the personalized vector into an EEG consciousness prediction index and a muscle relaxation monitoring prediction index, respectively;

[0022] A3. Extracting an initial batch of training samples from the training sample set and inputting them into the initialized MLP model for iterative training;

[0023] A4. If iterative training reaches convergence conditions;

[0024] A5. Derive the MLP model with the current model parameters and define it as the dual-exponential monitoring model.

[0025] In some embodiments, obtaining a training sample set for the dual-exponential monitoring model includes:

[0026] A1-1. Obtain historical personalized vectors, historical EEG consciousness index, and historical muscle relaxation monitoring index of several historical intraoperative patients;

[0027] A1-2. Define the historical personalized vector of each historical intraoperative patient as a feature variable, define the historical EEG consciousness index as the first target label, and define the historical muscle relaxation monitoring index as the second target label;

[0028] A1-3. Pair the feature variable with the first target label and the second target label respectively to generate a training sample with dual target labels;

[0029] A1-4. Summarize several historical intraoperative patient training samples to generate a training sample set for the dual-exponential monitoring model.

[0030] In some embodiments, an MLP model is selected as an initial training model for the dual exponential monitoring model, and a first output node and a second output node are initialized in an output layer thereof, including:

[0031] A2-1, linearly activate the first output node and the second output node respectively to output continuous numerical values of the EEG consciousness prediction index and the muscle relaxation monitoring prediction index;

[0032] A2-2. Define the loss functions of the first output node and the second output node respectively;

[0033] Among them, the loss function of the first output node is the first matching loss between the EEG consciousness prediction index and the EEG consciousness index in the training sample; the loss function of the second output node is the second matching loss between the muscle relaxation monitoring prediction index and the muscle relaxation monitoring index in the training sample;

[0034] The expression of the first matching loss is:

[0035]

[0036] The expression of the second matching loss is:

[0037]

[0038] Among them, Loss EEG Represents the first matching loss, Loss NM represents the second matching loss, N represents the total number of training samples, i represents the sample index of the training sample, EEG pre represents the EEG consciousness prediction index, EEG his Represents the historical EEG consciousness index corresponding to the first target label in the training sample, NM pre NM stands for muscle relaxation monitoring prediction index his Represents the historical muscle relaxation monitoring index corresponding to the second target label in the training sample;

[0039] A2-3. Define a total matching loss of the first output node and the second output node according to a weighted sum of the first matching loss and the second matching loss;

[0040] The expression of the total matching loss is:

[0041] Loss total =λ1·Loss EEG +λ2·Loss NM ;

[0042] Among them, Loss total represents the total matching loss, λ1 is the weight coefficient of the first matching loss, and λ2 is the weight coefficient of the second matching loss.

[0043] In some embodiments, extracting an initial batch of training samples from the training sample set and inputting them into the initialized MLP model for iterative training includes:

[0044] A3-1. Accept the initial batch of training samples as input vectors of the MLP model;

[0045] A3-2, forward propagating the input vector, and outputting the EEG consciousness prediction index and muscle relaxation monitoring prediction index through the first output node and the second output node respectively;

[0046] A3-3. Calculate the first matching loss of the EEG awareness index and the EEG awareness prediction index in the initial batch of training samples, and calculate the second matching loss of the muscle relaxation monitoring index and the muscle relaxation monitoring prediction index in the initial batch of training samples;

[0047] A3-4. Calculate the total matching loss based on the first matching loss and the second matching loss;

[0048] A3-5. If one of the first matching loss, the second matching loss, or the total matching loss is higher than a predefined loss threshold, updating the model parameters according to a gradient descent algorithm;

[0049] A3-6. Extract the next batch of training samples and input them into the real-time model after updating the model parameters for iterative training until the iterative training reaches the convergence condition; wherein, the convergence condition is: the first matching loss, the second matching loss and the total matching loss are all lower than the corresponding predefined loss threshold.

[0050] In some embodiments, generating a first weight and a second weight based on the personalized vector and the EEG awareness prediction index, and based on the personalized vector and the muscle relaxation monitoring prediction index, respectively, includes:

[0051] S3-1. Extracting characteristic values of several personalized features in the personalized vector;

[0052] S3-2, linearly summing the characteristic values of several personalized features to generate a physical sign index;

[0053] S3-3, defining the sigmoid ratios of the physical sign index and the EEG consciousness prediction index, and the sigmoid ratios of the physical sign index and the muscle relaxation monitoring prediction index as the first weight and the second weight, respectively;

[0054] The definition expression of the first weight is:

[0055]

[0056] The definition expression of the second weight is:

[0057]

[0058] Among them, w E represents the first weight, w N represents the second weight, f represents the physical sign index, and exp represents the exponential function.

[0059] In some embodiments, the expression for calculating the anesthesia status evaluation value of the current intraoperative patient is:

[0060] S status =tanh(w E EEG act )+tanh(w N ·NM act );

[0061] Among them, S status Indicates the anesthetic state assessment value, EEG act Indicates the EEG consciousness index collected in real time during the current operation, NM actIt represents the muscle relaxation monitoring index collected in real time from the patient during the current operation. Tanh represents the hyperbolic tangent function, which is used to smooth and nonlinearly transform the first weight and EEG consciousness index, as well as the anesthetic state assessment value of the second weight and muscle relaxation monitoring index.

[0062] The present invention provides an intraoperative patient anesthesia status assessment system based on EEG signals and muscle relaxation monitoring, which has the following features:

[0063] Beneficial effects:

[0064] The present invention introduces a first weight and a second weight in the anesthetic state assessment method, combines the EEG consciousness index, muscle relaxation monitoring index and the personalized vector obtained in real time predicted by the model, thereby introducing a personalized factor when assessing the anesthetic state. By combining the personalized vector (including vital sign parameters and anesthetic dosage, etc.) with the prediction index to calculate the weight, it is possible to quantify the impact of personalized characteristics on the anesthetic state. The anesthetic state assessment takes into account the differences in the physiological characteristics of the patient, making the anesthetic state assessment more personalized and targeted.

[0065] Moreover, the present invention can more accurately evaluate the current patient's anesthetic state through the anesthetic state evaluation value of hyperbolic tangent activation. The anesthetic state evaluation value can not only take into account the current EEG consciousness index and muscle relaxation monitoring index, but also comprehensively evaluate the anesthetic state in combination with the personalized vector composed of several personalized vital sign parameters of the patient during surgery. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a structural block diagram of a system for evaluating the anesthesia state of an intraoperative patient based on EEG signals and muscle relaxation monitoring according to the present invention;

[0067] Figure 2 This is a schematic diagram of an evaluation process of an intraoperative patient anesthesia status evaluation system based on EEG signals and muscle relaxation monitoring according to the present invention;

[0068] Figure 3 Schematic diagram of the pre-training process of the dual-exponential monitoring model of the present invention; DETAILED DESCRIPTION

[0069] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0070] MLP model: A general feedforward neural network that can achieve multi-target output tasks by defining multiple neurons (or output interfaces) in the output layer, such as simultaneously predicting the EEG consciousness index and the muscle relaxation monitoring index.

[0071] Example 1: Please refer to Figures 1 to 2 The present invention provides an intraoperative patient anesthesia status assessment system based on EEG signals and muscle relaxation monitoring, comprising:

[0072] Real-time data acquisition module, used to obtain the personalized vector, EEG consciousness index and muscle relaxation monitoring index of the patient during the current operation;

[0073] An index prediction module, configured to input the personalized vector of the current intraoperative patient into a pre-trained dual-index monitoring model, and output an EEG consciousness prediction index and a muscle relaxation monitoring prediction index;

[0074] A weight generation module, configured to generate a first weight and a second weight based on the personalized vector and the EEG consciousness prediction index, and based on the personalized vector and the muscle relaxation monitoring prediction index;

[0075] Among them, the first weight represents the contribution of the personalized vector to the EEG consciousness index, and the second weight represents the contribution of the personalized vector to the muscle relaxation monitoring index;

[0076] An evaluation module, configured to calculate an evaluation value of the anesthetic state of the patient currently undergoing surgery based on the first weight, the second weight, an EEG consciousness index, and a muscle relaxation monitoring index;

[0077] The anesthesia state generating module is used to generate a corresponding anesthesia state according to the anesthesia state evaluation value.

[0078] Specifically, the process of generating anesthesia states is based on the anesthesia state evaluation value calculated in real time, and mapping the evaluation value to different anesthesia states through certain rules. The specific steps are as follows:

[0079] Defining the level of anesthesia: The anesthesia state assessment value is a continuous value that can be divided into different anesthesia depth levels according to the set threshold. For example, the anesthesia state assessment value is mapped to the following states:

[0080] ① Deep anesthesia: When the evaluation value is lower than a certain threshold (such as 40), it means that the patient is in a deep anesthesia state and is suitable for surgical operation.

[0081] ② Appropriate anesthesia depth: When the assessment value is between 40 and 60, the patient's anesthesia depth is appropriate and the surgery can continue.

[0082] ③ Awakening state: When the assessment value is higher than 60, the patient may be in a state of awakening or clear consciousness, and the anesthetic dose needs to be adjusted appropriately.

[0083] Threshold Mapping: Thresholds are set based on clinical experience or historical data to more accurately classify and regulate anesthetic states. For example, multiple state levels can be set within an assessment range and adjusted based on real-time data.

[0084] Generate anesthesia state: The calculated anesthesia state assessment value is converted into a specific anesthesia state through threshold mapping, such as deep anesthesia, appropriate anesthesia, awakening, etc. Finally, the current anesthesia state of the patient during surgery is output.

[0085] In this embodiment, the patient's personalized characteristics, EEG signals and muscle relaxation monitoring signals are collected in real time, and the EEG consciousness prediction index and muscle relaxation monitoring prediction index are calculated based on these data through a pre-trained dual-exponential monitoring model, and a weight coefficient is further generated. The patient's current anesthetic state assessment value is then calculated based on the weight coefficient, thereby achieving the evaluation of different anesthetic states based on the patient's personalized data during surgery (such as age, weight, anesthetic dosage, etc.), and then the anesthetic dosage can be adjusted in real time according to the patient's physiological characteristics, avoiding the inadaptability of the unified anesthetic dose to different patients in the traditional method.

[0086] In this embodiment, the step of obtaining the personalized vector by the real-time data acquisition module includes:

[0087] S1-1, collecting several pre-anesthetic vital sign parameters and real-time anesthesia dosage of the patient currently undergoing surgery;

[0088] S1-2, standardizing several pre-anesthetic vital sign parameters and real-time anesthetic dosage to generate several personalized features;

[0089] S1-3. Perform feature concatenation on a plurality of the personalized features to obtain the personalized vector.

[0090] Specifically, several pre-anesthetic vital sign parameters include the age, gender, height, weight and BMI index of the current intraoperative patient; of course, they can further include: basal heart rate, basal blood pressure, basal respiratory rate and basal body temperature; that is, the above-mentioned pre-anesthetic vital sign parameters should be the vital sign parameters of the current intraoperative patient when not anesthetized, and are the physiological and clinical status of the intraoperative patient before anesthesia. Through pre-anesthetic vital sign parameters and real-time anesthesia dosage, a personalized vector of the current physical state of the current intraoperative patient is comprehensively expressed. The "personalized vector" is a multi-dimensional one-dimensional feature vector that contains vital sign parameters that may affect the anesthetic state (such as age, weight, anesthetic dosage, etc.). These parameters are determined before anesthesia begins, and they act together with the actual amount of anesthetic drugs used to affect the anesthetic state of the current intraoperative patient.

[0091] In this embodiment, the device for acquiring the EEG consciousness index is a BIS monitor or a qCON consciousness monitor, and the device for acquiring the muscle relaxation monitoring index is a TOF monitor or a neuromuscular monitor.

[0092] Specifically, the EEG consciousness index assesses the patient's state of consciousness and depth of anesthesia by analyzing the patient's electroencephalogram (EEG) signals. It generates a quantitative index using EEG signal processing algorithms (such as frequency analysis, power spectrum analysis, and entropy calculation). The commonly used range is 0 to 100, where:

[0093] 0: Indicates the complete absence of brain electrical activity (such as deep anesthesia or brain death).

[0094] 40-60: Indicates the appropriate depth of anesthesia.

[0095] 100: Indicates full consciousness.

[0096] The EEG consciousness index is monitored by attaching EEG electrode patches to the patient's scalp during surgery to collect EEG signals, which are processed by the above-mentioned acquisition device to generate the corresponding EEG consciousness index.

[0097] The muscle relaxation monitoring index is used to evaluate the effect of anesthetic drugs on muscle relaxation by monitoring the patient's neuromuscular conduction activity. Commonly used indices include:

[0098] TOF (Train-of-Four) ratio: The ratio of the four contraction strengths of the muscle response after nerve stimulation, ranging from 0% to 100%.

[0099] 0%: Complete muscle relaxation (no contraction response).

[0100] >90%: Usually indicates that the patient has regained normal muscle function.

[0101] PTC (Post-Tetanic Count): Observe the muscle contraction response after strong stimulation during deep muscle relaxation.

[0102] The muscle relaxation index is monitored during surgery by placing stimulating electrodes on the patient's peripheral nerves (such as the ulnar nerve and peroneal nerve) and sensors on the corresponding muscles (such as the thumb or toe). The device delivers electrical stimulation and records the intensity of the muscle contraction response, thereby calculating the time-of-flight (TOF) ratio or post-contraction (PTC) value.

[0103] Example 2: See Figure 3 The technical solution of this embodiment 2 differs from that of embodiment 1 in that it discloses a pre-training step of the double exponential monitoring model described in embodiment 1, and the pre-training step includes:

[0104] A1. Obtaining a training sample set for the dual-exponential monitoring model;

[0105] A2. Select the MLP model as the initial training model for the dual exponential monitoring model and initialize the first output node and the second output node in its output layer;

[0106] The first output node and the second output node are used to map the personalized vector into an EEG consciousness prediction index and a muscle relaxation monitoring prediction index, respectively;

[0107] A3. Extracting an initial batch of training samples from the training sample set and inputting them into the initialized MLP model for iterative training;

[0108] A4. If iterative training reaches convergence conditions;

[0109] A5. Derive the MLP model with the current model parameters and define it as the dual-exponential monitoring model.

[0110] Specifically, this embodiment uses an MLP (multi-layer perceptron) model as the initial training model. Through training, the model is able to predict two key anesthesia indices based on a personalized vector (including the patient's vital signs and anesthesia dosage, etc.): the EEG consciousness prediction index and the muscle relaxation monitoring prediction index. Among them, the MLP model is a classic neural network structure that can handle complex nonlinear relationships, so that the patient's personalized data can be effectively mapped to outputs related to the anesthesia state. In order to achieve this goal, the output layer of the MLP model is designed with dual output nodes, which are used to predict these two anesthesia monitoring indices respectively.

[0111] Furthermore, the step A1 specifically includes:

[0112] A1-1. Obtain historical personalized vectors, historical EEG consciousness index, and historical muscle relaxation monitoring index of several historical intraoperative patients;

[0113] A1-2. Define the historical personalized vector of each historical intraoperative patient as a feature variable, define the historical EEG consciousness index as the first target label, and define the historical muscle relaxation monitoring index as the second target label;

[0114] A1-3. Pair the feature variable with the first target label and the second target label respectively to generate a training sample with dual target labels;

[0115] A1-4. Summarize several historical intraoperative patient training samples to generate a training sample set for the dual-exponential monitoring model.

[0116] Specifically, this example constructs a dual-target label training sample set using historical patient-specific vital sign parameters and anesthesia dosages. The historical personalized vector serves as the feature variable, paired with the corresponding EEG awareness index and muscle relaxation monitoring index as the first and second target labels, respectively, to generate training samples. By aggregating data from multiple patients, these dual-target samples can construct a mapping from personalized vital signs to anesthesia status.

[0117] Furthermore, the step A2 specifically includes:

[0118] A2-1, linearly activate the first output node and the second output node respectively to output continuous numerical values of the EEG consciousness prediction index and the muscle relaxation monitoring prediction index;

[0119] A2-2. Define the loss functions of the first output node and the second output node respectively;

[0120] Among them, the loss function of the first output node is the first matching loss between the EEG consciousness prediction index and the EEG consciousness index in the training sample; the loss function of the second output node is the second matching loss between the muscle relaxation monitoring prediction index and the muscle relaxation monitoring index in the training sample;

[0121] The expression of the first matching loss is:

[0122]

[0123] The expression of the second matching loss is:

[0124]

[0125] Among them, Loss EEG Represents the first matching loss, Loss NM represents the second matching loss, N represents the total number of training samples, i represents the sample index of the training sample, EEG pre Represents the EEG consciousness prediction index, which is the predicted value of the EEG consciousness state output by the training model, reflecting the depth of anesthesia or the state of consciousness. his Represents the historical EEG consciousness index corresponding to the first target label in the training sample, NM pre Represents the muscle relaxation monitoring prediction index. The predicted value of the muscle relaxation monitoring status is output by the training model and is used to reflect the degree of muscle relaxation. NM his Represents the historical muscle relaxation monitoring index corresponding to the second target label in the training sample;

[0126] A2-3. Define a total matching loss of the first output node and the second output node according to a weighted sum of the first matching loss and the second matching loss;

[0127] The expression of the total matching loss is:

[0128] Loss total =λ1·Loss EEG +λ2·Loss NM ;

[0129] Among them, Loss total represents the total matching loss, λ1 is the weight coefficient of the first matching loss, and λ2 is the weight coefficient of the second matching loss.

[0130] Specifically, the total matching loss in the present embodiment combines the errors of the EEG consciousness index and the muscle relaxation monitoring index to ensure that the two prediction targets are optimized simultaneously during the training process. It is possible to consider two anesthesia monitoring indices (EEG consciousness index and muscle relaxation monitoring index) simultaneously by using dual output nodes and dual matching losses. The loss function of each output node optimizes the prediction task separately, but the total loss ensures that the errors of the two indices are minimized, thereby improving the overall prediction performance.

[0131] Furthermore, the step A3 specifically includes:

[0132] A3-1. Accept the initial batch of training samples as input vectors of the MLP model;

[0133] A3-2, forward propagating the input vector, and outputting the EEG consciousness prediction index and muscle relaxation monitoring prediction index through the first output node and the second output node respectively;

[0134] A3-3. Calculate the first matching loss of the EEG awareness index and the EEG awareness prediction index in the initial batch of training samples, and calculate the second matching loss of the muscle relaxation monitoring index and the muscle relaxation monitoring prediction index in the initial batch of training samples;

[0135] A3-4. Calculate the total matching loss based on the first matching loss and the second matching loss;

[0136] A3-5. If one of the first matching loss, the second matching loss, or the total matching loss is higher than a predefined loss threshold, updating the model parameters according to a gradient descent algorithm;

[0137] A3-6. Extract the next batch of training samples and input them into the real-time model after updating the model parameters for iterative training until the iterative training reaches the convergence condition; wherein, the convergence condition is: the first matching loss, the second matching loss and the total matching loss are all lower than the corresponding predefined loss threshold.

[0138] Specifically, the present embodiment training process is completed by multiple iterative training, and each round of training can optimize model parameters until loss function converges. The training process is divided into multiple steps, and the output value is first calculated by forward propagation, and then the error between the prediction and the true value is calculated by loss function, and then the model parameters are optimized by gradient descent algorithm. The goal of each training is to reduce the prediction error of the model so that the model can accurately predict the EEG consciousness index and muscle relaxation monitoring index. Through multiple iterations, the model can learn the relationship between personalized vector and anesthetic state from a large amount of training data. As training proceeds, the model is more accurate in understanding personalized vectors, and therefore more accurate predictions can be provided in real-time anesthetic state assessment, ensuring the accurate assessment of anesthesia depth and muscle relaxation degree.

[0139] Example 3: The technical solution of Example 3 is different from that of Example 1 and Example 2 in that the steps of generating the first weight and the second weight and the calculation process of the anesthesia state evaluation value are disclosed.

[0140] The steps of generating the first weight and the second weight include:

[0141] S3-1. Extracting characteristic values of several personalized features in the personalized vector;

[0142] S3-2, linearly summing the characteristic values of several personalized features to generate a physical sign index;

[0143] S3-3, defining the sigmoid ratios of the physical sign index and the EEG consciousness prediction index, and the sigmoid ratios of the physical sign index and the muscle relaxation monitoring prediction index as the first weight and the second weight, respectively;

[0144] The definition expression of the first weight is:

[0145]

[0146] The definition expression of the second weight is:

[0147]

[0148] Among them, w E represents the first weight, w N represents the second weight, f represents the physical sign index, which is obtained by summing the characteristic values of several physical signs in the personalized vector and represents the physiological and clinical characteristics of the individual patient. exp is an exponential function used to smooth the relationship between the physical sign index and the predicted index using a sigmoid function, ensuring that the ratio can adapt to a wider range of nonlinear changes.

[0149] Specifically, the first weight and the second weight are used to represent the contribution of the personalized vector to the EEG consciousness prediction index and the muscle relaxation monitoring prediction index, respectively. Among them, the first weight and the second weight smooth the relationship between the sign index and the prediction index by using the Sigmoid ratio method, and the first weight and the second weight reflect the contribution of the personalized vector to the anesthetic state. The introduction of the Sigmoid function enables the ratio between the sign index and the prediction index to not only adapt to a wide range of nonlinear changes, but also ensures a more accurate response to the input data. As a result, the assessment of the anesthetic state does not rely solely on a simple linear relationship, but introduces a nonlinear mapping, which helps to better capture changes in the depth of anesthesia and the degree of muscle relaxation.

[0150] Furthermore, the expression for calculating the anesthesia state evaluation value of the current intraoperative patient is:

[0151] S status =tanh(w E EEG act )+tanh(w N ·NM act );

[0152] Among them, S status Indicates the anesthetic state assessment value, EEG act Indicates the EEG consciousness index collected in real time during the current operation, NM act It represents the muscle relaxation monitoring index collected in real time from the patient during the current operation. Tanh represents the hyperbolic tangent function, which is used to smooth and nonlinearly transform the first weight and EEG consciousness index, as well as the anesthetic state assessment value of the second weight and muscle relaxation monitoring index.

[0153] When calculating the anesthesia state evaluation value in this embodiment, the hyperbolic tangent function is introduced to perform smoothing and nonlinear transformation on the relationship between the first weight and the EEG consciousness index and the relationship between the second weight and the muscle relaxation monitoring index.

[0154] The hyperbolic tangent function can smoothly map input values within the interval [-1, 1] and enhance adaptability to nonlinear changes. By combining the first weight with the EEG consciousness index and the second weight with the muscle relaxation monitoring index, and applying the hyperbolic tangent function, the unreasonable effects of oversmoothing can be effectively avoided.

[0155] Specifically, in the assessment of anesthesia status, the changes in the patient's EEG consciousness and muscle relaxation level are often not linear. The introduction of the hyperbolic tangent function can transform the linear relationship between two different weights and prediction indexes into a nonlinear relationship, making the model's response to input more flexible and accurate.

[0156] Furthermore, controlling the range of assessment values is crucial for adjusting anesthetic dosages during anesthetic state assessment. The hyperbolic tangent function maps input values to the range [-1, 1], avoiding extreme output values. The smoothing effect of the hyperbolic tangent function ensures that anesthetic state assessment values remain within a reasonable range, avoiding excessively high or low assessment values that could affect anesthetic decisions.

[0157] In summary, the anesthetic state assessment value based on hyperbolic tangent activation can more accurately assess the patient's current anesthetic state and provide more precise real-time feedback to the anesthesiologist. This anesthetic state assessment value not only considers the current EEG consciousness index and muscle relaxation monitoring index, but also combines a personalized vector composed of several personalized vital sign parameters to comprehensively evaluate the anesthetic state during surgery, thereby supporting the dosage of anesthetic drugs and ensuring the safety and accuracy of the anesthesia process.

[0158] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, computer, server or data center to another website, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.) means.

[0159] The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, or a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0160] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division of a waterway underwater terrain change analysis system and method. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0161] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A system for evaluating the anesthesia status of patients during surgery based on EEG signals and muscle relaxation monitoring, characterized in that: include: Real-time data acquisition module, used to obtain the personalized vector, EEG consciousness index and muscle relaxation monitoring index of the patient during the current operation; An index prediction module, configured to input the personalized vector of the current intraoperative patient into a pre-trained dual-index monitoring model, and output an EEG consciousness prediction index and a muscle relaxation monitoring prediction index; A weight generation module, configured to generate a first weight and a second weight based on the personalized vector and the EEG consciousness prediction index, and based on the personalized vector and the muscle relaxation monitoring prediction index; Among them, the first weight represents the contribution of the personalized vector to the EEG consciousness index, and the second weight represents the contribution of the personalized vector to the muscle relaxation monitoring index; An evaluation module, configured to calculate an evaluation value of the anesthetic state of the patient currently undergoing surgery based on the first weight, the second weight, an EEG consciousness index, and a muscle relaxation monitoring index; The anesthesia state generating module is used to generate a corresponding anesthesia state according to the anesthesia state evaluation value.

2. The intraoperative patient anesthesia status assessment system based on EEG signals and muscle relaxation monitoring according to claim 1, characterized in that: Get the personalized vector of the current intraoperative patient, including: S1-1, collecting several pre-anesthetic vital sign parameters and real-time anesthesia dosage of the patient currently undergoing surgery; S1-2, standardizing several pre-anesthetic vital sign parameters and real-time anesthetic dosage to generate several personalized features; S1-3. Perform feature concatenation on a plurality of the personalized features to obtain the personalized vector.

3. The intraoperative patient anesthesia status assessment system based on EEG signals and muscle relaxation monitoring according to claim 1, characterized in that: The device for acquiring the EEG consciousness index is a BIS monitor or a qCON consciousness monitor, and the device for acquiring the muscle relaxation monitoring index is a TOF monitor or a neuromuscular monitor.

4. The intraoperative patient anesthesia status assessment system based on EEG signals and muscle relaxation monitoring according to claim 1, characterized in that: The pre-training step of the dual exponential monitoring model includes: A1. Obtaining a training sample set for the dual-exponential monitoring model; A2. Select the MLP model as the initial training model for the dual exponential monitoring model and initialize the first output node and the second output node in its output layer; The first output node and the second output node are used to map the personalized vector into an EEG consciousness prediction index and a muscle relaxation monitoring prediction index, respectively; A3. Extracting an initial batch of training samples from the training sample set and inputting them into the initialized MLP model for iterative training; A4. If iterative training reaches convergence conditions; A5. Derive the MLP model with the current model parameters and define it as the dual-exponential monitoring model.

5. The intraoperative patient anesthesia status assessment system based on EEG signals and muscle relaxation monitoring according to claim 4, characterized in that: Obtaining a training sample set for the dual-exponential monitoring model includes: A1-1. Obtain historical personalized vectors, historical EEG consciousness index, and historical muscle relaxation monitoring index of several historical intraoperative patients; A1-2. Define the historical personalized vector of each historical intraoperative patient as a feature variable, define the historical EEG consciousness index as the first target label, and define the historical muscle relaxation monitoring index as the second target label; A1-3. Pair the feature variable with the first target label and the second target label respectively to generate a training sample with dual target labels; A1-4. Summarize several historical intraoperative patient training samples to generate a training sample set for the dual-exponential monitoring model.

6. The intraoperative patient anesthesia status assessment system based on EEG signals and muscle relaxation monitoring according to claim 4, characterized in that: The MLP model is selected as the initial training model of the dual exponential monitoring model, and the first output node and the second output node are initialized in its output layer, including: A2-1, linearly activate the first output node and the second output node respectively to output continuous numerical values of the EEG consciousness prediction index and the muscle relaxation monitoring prediction index; A2-2. Define the loss functions of the first output node and the second output node respectively; Among them, the loss function of the first output node is the first matching loss between the EEG consciousness prediction index and the EEG consciousness index in the training sample; the loss function of the second output node is the second matching loss between the muscle relaxation monitoring prediction index and the muscle relaxation monitoring index in the training sample; The expression of the first matching loss is: The expression of the second matching loss is: Among them, Loss EEG Represents the first matching loss, Loss NM represents the second matching loss, N represents the total number of training samples, i represents the sample index of the training sample, EEG pre represents the EEG consciousness prediction index, EEG his Represents the historical EEG consciousness index corresponding to the first target label in the training sample, NM pre NM stands for muscle relaxation monitoring prediction index his Represents the historical muscle relaxation monitoring index corresponding to the second target label in the training sample; A2-3. Define a total matching loss of the first output node and the second output node according to a weighted sum of the first matching loss and the second matching loss; The expression of the total matching loss is: Loss total =λ1·Loss EEG +λ2·Loss NM ; Among them, Loss total represents the total matching loss, λ1 is the weight coefficient of the first matching loss, and λ2 is the weight coefficient of the second matching loss.

7. The intraoperative patient anesthesia status assessment system based on EEG signals and muscle relaxation monitoring according to claim 4, characterized in that: Extracting an initial batch of training samples from the training sample set and inputting them into the initialized MLP model for iterative training, including: A3-1. Accept the initial batch of training samples as input vectors of the MLP model; A3-2, forward propagating the input vector, and outputting the EEG consciousness prediction index and muscle relaxation monitoring prediction index through the first output node and the second output node respectively; A3-3. Calculate the first matching loss of the EEG awareness index and the EEG awareness prediction index in the initial batch of training samples, and calculate the second matching loss of the muscle relaxation monitoring index and the muscle relaxation monitoring prediction index in the initial batch of training samples; A3-4. Calculate the total matching loss based on the first matching loss and the second matching loss; A3-5. If one of the first matching loss, the second matching loss, or the total matching loss is higher than a predefined loss threshold, updating the model parameters according to a gradient descent algorithm; A3-6. Extract the next batch of training samples and input them into the real-time model after updating the model parameters for iterative training until the iterative training reaches the convergence condition; wherein, the convergence condition is: the first matching loss, the second matching loss and the total matching loss are all lower than the corresponding predefined loss threshold.

8. The intraoperative patient anesthesia status assessment system based on EEG signals and muscle relaxation monitoring according to claim 1, characterized in that: According to the personalized vector and the EEG consciousness prediction index, as well as the personalized vector and the muscle relaxation monitoring prediction index, a first weight and a second weight are generated respectively, including: S3-1. Extracting characteristic values of several personalized features in the personalized vector; S3-2, linearly summing the characteristic values of several personalized features to generate a physical sign index; S3-3, defining the sigmoid ratios of the physical sign index and the EEG consciousness prediction index, and the sigmoid ratios of the physical sign index and the muscle relaxation monitoring prediction index as the first weight and the second weight, respectively; The definition expression of the first weight is: The definition expression of the second weight is: Among them, w E represents the first weight, w N represents the second weight, f represents the physical sign index, and exp represents the exponential function.

9. The intraoperative patient anesthesia status assessment system based on EEG signals and muscle relaxation monitoring according to claim 1, characterized in that: The expression for calculating the anesthesia status evaluation value of the patient currently undergoing surgery is: S status =tanh(w E ·EEG act )+tanh(w N ·NM act ); Among them, S status Indicates the anesthetic state assessment value, EEG act Indicates the EEG consciousness index collected in real time during the current operation, NM act It represents the muscle relaxation monitoring index collected in real time from the patient during the current operation. Tanh represents the hyperbolic tangent function, which is used to smooth and nonlinearly transform the first weight and EEG consciousness index, as well as the anesthetic state assessment value of the second weight and muscle relaxation monitoring index.

Images