Anesthesia depth classification method and system based on artificial intelligence

By combining the long and short-term memory network of attention mechanism and the codec network that divides physiological signal components, the problems of poor correlation between anesthesia depth and anesthetic dose and shallow classification dimensions are solved, and the accurate prediction of anesthetic dose and depth are achieved, and the depth of anesthetic is improved. Automated process of classification.

CN118468166BActive Publication Date: 2024-11-01THE FIRST AFFILIATED HOSPITAL OF FUJIAN MEDICAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410934996.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-12
Publication Date
2024-11-01
Estimated Expiration
2044-07-12

AI Technical Summary

Technical Problem

Among the existing anesthesia depth classification methods, the correlation between anesthesia depth and specific anesthetic dose is poor. The traditional method can predict the classification dimensions of the single combination of EEG data and has limited application scope. Moreover, the prediction of anesthetic dose cannot provide complete anesthesia targets and duration, which is difficult to In combination with the anaesthetic dosage scheme prediction, the classification results are low in quantification.

Method used

A long-term and short-term memory artificial neural network combined with attention mechanism is used to predict anesthesia doses, and anesthesia depth quantization prediction is carried out through physiological signal component division and codec network that improves channel attention, enhancing timing feature processing and frequency domain signal processing, and improving Reliability of anesthetic dose prediction and classification dimensions and availability of anesthetic depth classification.

Benefits of technology

Improve the classification dimensions and the availability and generality of prediction results of anesthesia depth classification, enhance the reliability and accuracy of anesthetic dose prediction, provide specific anesthesia depth quantitative index prediction data, improve the availability of data, and make the anesthesia depth. The overall automation process of classification provides technical reference.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118468166B_ABST
    Figure CN118468166B_ABST
Patent Text Reader

Abstract

The present invention discloses an artificial intelligence-based anesthesia depth classification method and system, the method includes data collection, data preprocessing, anesthesia dose program prediction, anesthesia depth quantitative prediction, anesthesia depth classification. The present invention relates to the field of anesthesia depth classification technology, specifically refers to an artificial intelligence-based anesthesia depth classification method and system, the scheme first performs anesthesia dose program prediction, and uses the numerical value given in the personalized anesthesia scheme and the physiological signal data to perform anesthesia depth classification; the long short-term memory artificial neural network combined with the attention mechanism is used to perform anesthesia dose program prediction, which improves the reliability and accuracy of anesthesia dose prediction; the encoding and decoding network combined with physiological signal component division enhancement and improved channel attention is used to perform anesthesia depth quantitative prediction, which improves the availability of data and provides a technical reference for the overall automation process of anesthesia depth classification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of anesthesia depth classification, and in particular to an anesthesia depth classification method and system based on artificial intelligence. Background Art

[0002] The AI-based anesthesia depth classification method is a method that uses machine learning and data mining technology to evaluate and classify the depth of anesthesia of patients. Its main function is to help anesthesiologists monitor the depth of anesthesia of patients more accurately to ensure that patients are in a safe state of anesthesia during surgery. This method can monitor and evaluate the depth of anesthesia in real time by analyzing the patient's physiological signal data combined with advanced machine learning algorithms. By classifying the patient's anesthesia depth into different levels or states, anesthesiologists can better adjust the dosage of anesthetic drugs to achieve the ideal anesthesia effect, while minimizing the risk of anesthesia and the occurrence of complications, which can improve the accuracy and safety of anesthesia management and provide more reliable protection and care for patients during surgery.

[0003] However, in the existing anesthesia depth classification methods, there is a poor correlation between the existing technology for anesthesia depth classification and specific anesthetic doses, and the traditional anesthesia depth classification only combines EEG data for prediction, which has the technical problem of shallow classification dimension and limited applicability; in the existing anesthesia depth classification methods, there is a technical problem that the existing anesthetic dose prediction generally only performs preoperative prediction on the numerical value of fixed anesthetics, but cannot provide complete anesthesia goals, anesthesia duration and specific anesthesia plan, thereby affecting the overall effect of anesthesia dose prediction; in the existing anesthesia depth classification methods, there is a technical problem that the existing anesthesia depth classification only simply and generally divides the anesthesia depth into several fixed types, the degree of quantification of the classification results is low, and it is difficult to cooperate with the anesthesia dose plan prediction to further improve the classification dimension and interpretability. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an anesthesia depth classification method and system based on artificial intelligence. In view of the fact that the existing anesthesia depth classification method has a poor correlation between the anesthesia depth classification and the specific anesthetic dose, and the traditional anesthesia depth classification is predicted by combining EEG data alone, which has the technical problem of shallow classification dimension and limited applicability, the present solution creatively first performs a programmatic prediction of anesthesia dose, and classifies the anesthesia depth together with the numerical value given in the personalized anesthesia plan and the physiological signal data, thereby improving the classification dimension of the anesthesia depth classification and the availability and versatility of the prediction results; in view of the fact that the existing anesthesia depth classification method generally only performs preoperative prediction on the numerical value of the fixed anesthetic, but cannot provide a complete anesthesia goal, anesthesia duration and specific anesthesia plan, thereby affecting the overall effect of the anesthesia dose prediction, the present solution creatively uses a long short-term memory artificial neural network combined with an attention mechanism to perform anesthesia dose. Scheme-based prediction, through the long short-term memory network to enhance the temporal characteristics of anesthesia duration, and further add artificial neural hidden layer neurons to enhance the prediction effect of applicable schemes for various anesthetics, thereby improving the reliability and accuracy of anesthetic dose prediction; in the existing anesthesia depth classification methods, there is a technical problem that the existing anesthesia depth classification simply and generally divides the anesthesia depth into several fixed types, the classification results have a low degree of quantification, and it is difficult to cooperate with the anesthesia dose scheme prediction to further improve the classification dimension and interpretability. This scheme creatively uses an encoding and decoding network that combines physiological signal component division enhancement and improved channel attention to quantitatively predict anesthesia depth. Through physiological signal component division and enhancement, physiological signals in different frequency domains are processed separately, and by combining improved channel attention, anesthesia depth classification combined with anesthetic dose prediction values ​​and multi-channel integration is performed, and specific anesthesia depth quantitative index prediction data is obtained, which improves the data availability and provides a technical reference for the overall automation process of anesthesia depth classification.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides an artificial intelligence-based anesthesia depth classification method, the method comprising the following steps:

[0006] Step S1: data collection;

[0007] Step S2: data preprocessing;

[0008] Step S3: Anesthesia dosage planning prediction;

[0009] Step S4: quantitative prediction of anesthesia depth;

[0010] Step S5: Classification of anesthesia depth.

[0011] Furthermore, in step S1, the data collection is used to collect basic patient data required for anesthesia depth classification, specifically, to obtain original data of anesthesia depth classification from medical system and medical record data;

[0012] The raw data of anesthesia depth specifically includes physiological signal data, basic patient data and anesthesia target data.

[0013] Furthermore, in step S2, the data preprocessing is used to optimize and enhance the original data, specifically, data preprocessing is performed on the anesthesia depth classification original data to obtain optimized anesthesia depth classification data.

[0014] Further, in step S3, the anesthetic dose program prediction is used to predict the details of the anesthetic dose usage program, specifically, based on the patient optimization data and the anesthetic target optimization data in the optimized anesthesia depth classification data, a long short-term memory artificial neural network combined with an attention mechanism is used to perform anesthetic dose program prediction to obtain anesthetic dose program recommendation data;

[0015] The long short-term memory artificial neural network combined with the attention mechanism specifically includes a long short-term memory layer, an attention layer, an artificial neural hidden layer and an artificial neural output layer;

[0016] The long short-term memory layer is used to process the temporal characteristics in the anesthesia target data and serves as the artificial neural input layer of the entire model;

[0017] The attention layer is used to enhance the model's ability to discover and predict key information features;

[0018] The artificial neural hidden layer is used to process the complex characteristics of the anesthetic dose and perform a programmatic prediction of the anesthetic dose;

[0019] The artificial neural output layer is used to output a prediction scheme of anesthesia dosage;

[0020] The step of using a long short-term memory artificial neural network combined with an attention mechanism to perform anesthesia dosage scheme prediction to obtain anesthesia dosage scheme recommendation data includes:

[0021] Step S31: constructing a long short-term memory layer, specifically by sequentially constructing an input gate, a forget gate, and an output gate, performing time series feature processing, obtaining a long short-term memory prediction output, and using the long short-term memory prediction output as an input of the attention layer;

[0022] Step S32: constructing an attention layer, specifically, calculating the influence of the hidden state of the long short-term memory layer on the current output by calculating the normalized score, the attention weight and the context vector, obtaining a high-price feature of the anesthetic dose, and using the high-price feature of the anesthetic dose as the output data of the artificial neural hidden layer;

[0023] Step S33: constructing an artificial neural hidden layer, specifically constructing an artificial neural network hidden layer after the attention layer, and performing a weighted and biased calculation of the calculated predicted value for anesthesia dose prediction by constructing the hidden layer to obtain an artificial neural hidden layer predicted output;

[0024] Step S34: constructing an artificial neural output layer, specifically, performing numerical calculation of anesthesia dosage scheme by obtaining the predicted output of the artificial neural hidden layer, and obtaining anesthesia dosage predicted output;

[0025] Step S35: training the anesthesia dose prediction model, specifically, training the model by constructing the long short-term memory layer, the attention layer, the artificial neural hidden layer, and the artificial neural output layer, to obtain the anesthesia dose prediction model Model DP ;

[0026] Step S36: Anesthesia dose scheme prediction, specifically using the anesthesia dose prediction model Model DP , perform anesthetic dosage scheme prediction and obtain anesthetic dosage scheme recommendation data;

[0027] The anesthesia dosage regimen recommendation data specifically includes anesthetic drug classification predicted dosage data, maintenance injection rate predicted data, maintenance dosage adjustment data and anesthesia combination strategy data;

[0028] The anesthetic drug classification predicted dose data specifically includes induction drug data, anesthetic drug data, muscle relaxant data and maintenance drug data;

[0029] The anesthesia combination strategy data specifically refers to data for recommending specific drug combinations based on anesthesia target data.

[0030] Further, in step S4, the quantitative prediction of anesthesia depth is used to quantitatively predict anesthesia depth in combination with the use scheme details, specifically, based on the enhanced physiological signal data in the optimized anesthesia depth classification data and the anesthesia dosage scheme recommendation data, a coding and decoding network that combines physiological signal component division enhancement and improved channel attention is used to quantitatively predict anesthesia depth, to obtain anesthesia depth quantitative index prediction data;

[0031] The encoding and decoding network combined with physiological signal component division enhancement and improved channel attention specifically includes an input embedding layer, a signal enhancement layer, a frequency domain attention layer, an encoding and decoding layer, an improved channel attention layer and a linear output layer;

[0032] The input embedding layer is used to receive the model input and perform conversion processing and feature extraction on the input physiological signal;

[0033] The signal enhancement layer is used to decompose the original physiological signal data according to the frequency components and process them separately;

[0034] The frequency domain attention layer is used to introduce a multi-head attention mechanism for feature optimization;

[0035] The codec layer is used to construct a variable-voltage encoder and add layer normalization and residual links;

[0036] The improved channel attention layer is used to integrate the respective predicted outputs of the frequency components and obtain weighted features;

[0037] The linear output layer is used to linearly change the weighted features and obtain the final prediction output;

[0038] The step of using an encoding / decoding network that combines physiological signal component division enhancement and improved channel attention to perform quantitative prediction of anesthesia depth to obtain anesthesia depth quantitative index prediction data includes:

[0039] Step S41: constructing an input embedding layer, specifically taking the enhanced physiological signal data and the anesthesia dosage regimen recommendation data as model inputs, performing signal feature extraction on the enhanced physiological signal data, and performing anesthesia information basic feature extraction on the anesthesia dosage regimen recommendation data, to obtain signal feature data and numerical feature data;

[0040] Step S42: constructing a signal enhancement layer, specifically dividing the signal characteristic data according to the frequency to obtain frequency-divided signal data;

[0041] Step S43: construct a frequency domain attention layer, specifically introduce a multi-head self-attention mechanism, perform multi-head self-attention processing on each group of frequency feature signals in the frequency division signal data, obtain frequency domain self-attention grouped data, and perform inverse Fourier transform and padding operations on the frequency domain self-attention grouped data to convert the frequency domain self-attention grouped data into time domain feature data. The calculation formula is:

[0042] Z=-1(Pad(MultiHead(Q, K, V))+X+Z′);

[0043] Where Z is the time domain feature data, -1(·) is the inverse Fourier transform, Pad(·) is the zero padding function, MultiHead(·) is the multi-head self-attention function, Q is the query vector, K is the key vector, V is the value vector, Z′ is the frequency domain self-attention grouping data, and X is the numerical feature data;

[0044] Step S44: constructing a codec layer, specifically, after the frequency domain attention layer, constructing a transformer block for feature encoding, and constructing layer normalization and residual linking, constructing the codec layer, and performing hierarchical feature processing on the frequency division signal data to obtain hierarchical feature data;

[0045] Step S45: constructing an improved channel attention layer, specifically, constructing an improved channel attention layer through a squeeze block and an excitation block, and outputting and integrating hierarchical feature data by constructing the improved channel attention layer, specifically including the following steps:

[0046] Step S451: constructing an extrusion block, specifically performing a standard convolution operation, and performing weight calculation of the hierarchical feature data through global average pooling to obtain a hierarchical feature weight value;

[0047] Step S452: constructing an excitation block, specifically using two fully connected layers and a nonlinear activation function to construct an excitation function, transforming the output signal of the excitation function through an S-type activation function, and updating the weights of the hierarchical feature data and re-adjusting the activation through channel multiplication to obtain hierarchical feature channel attention weights;

[0048] The calculation formula for constructing the activation function is:

[0049] F S =Sig(W2ReLU(W1z));

[0050] In the formula, F S is the activation function, Sig(·) is the S-type activation function, W2 is the weight of the second fully connected layer, ReLU(·) is the nonlinear activation function, W1 is the weight of the first fully connected layer, and z is the hierarchical feature data;

[0051] Step S453: hierarchical feature integration, specifically, integrating the attention weights of hierarchical feature channels to obtain weighted feature data, and the calculation formula is:

[0052] F = Concat(z1,z2,z3,z4);

[0053] Where F is the weighted feature data, Concat(·) is the feature concatenation operation, z1 is the hierarchical feature channel attention weight corresponding to the first level frequency, z2 is the hierarchical feature channel attention weight corresponding to the second level frequency, z3 is the hierarchical feature channel attention weight corresponding to the third level frequency, and z4 is the hierarchical feature channel attention weight corresponding to the fourth level frequency;

[0054] Step S46: constructing a linear output layer, specifically performing a linear transformation on the weighted feature data to obtain linear quantification data of anesthesia depth index;

[0055] Step S47: training the anesthesia depth quantitative prediction model, specifically, training the model by constructing the input embedding layer, constructing the signal enhancement layer, constructing the frequency domain attention layer, constructing the encoding and decoding layer, constructing the improved channel attention layer and constructing the linear output layer, to obtain the anesthesia depth quantitative prediction model Model DA ;

[0056] Step S48: quantitative prediction of anesthesia depth, specifically using the anesthesia depth quantitative prediction model Model DA , based on the enhanced physiological signal data in the optimized anesthesia depth classification data and the anesthesia dosage scheme recommendation data, a quantitative prediction of anesthesia depth is performed to obtain anesthesia depth quantitative index prediction data.

[0057] Furthermore, in step S5, the anesthesia depth classification is used to integrate the prediction results of the anesthetic dose and the anesthetic depth to obtain a complete anesthesia depth classification result, specifically combining the anesthetic dose scheme recommendation data and the anesthetic depth quantitative index prediction data, and combining the anesthetic depth classification original data, to perform an integrated prediction of the anesthesia depth, and obtain the anesthetic dose and quantitative classification prediction result data.

[0058] The present invention provides an artificial intelligence-based anesthesia depth classification system, comprising a data collection module, a data preprocessing module, an anesthesia dose program prediction module, an anesthesia depth quantitative prediction module and an anesthesia depth classification module;

[0059] The data collection module is used for data collection, and obtains the original data of anesthesia depth classification through data collection, and sends the original data of anesthesia depth classification to the data preprocessing module and the anesthesia depth classification module;

[0060] The data preprocessing module is used for data preprocessing, and obtains optimized anesthesia depth classification data through data preprocessing, and sends the optimized anesthesia depth classification data to the anesthesia dose scheme prediction module and the anesthesia depth quantification prediction module;

[0061] The anesthesia dose program prediction module is used for anesthesia dose program prediction, obtains anesthesia dose program recommendation data through anesthesia dose program prediction, and sends the anesthesia dose program recommendation data to the anesthesia depth quantification prediction module and the anesthesia depth classification module;

[0062] The anesthesia depth quantification prediction module is used for the anesthesia depth quantification prediction module, obtains anesthesia depth quantification index prediction data through the anesthesia depth quantification prediction module, and sends the anesthesia depth quantification index prediction data to the anesthesia depth classification module;

[0063] The anesthesia depth classification module is used for anesthesia depth classification, and obtains anesthesia dose and quantitative classification prediction result data through anesthesia depth classification.

[0064] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0065] (1) In the existing anesthesia depth classification methods, there is a poor correlation between the existing technology for anesthesia depth classification and specific anesthesia dose, and the traditional anesthesia depth classification is only combined with EEG data for prediction, which has the technical problems of shallow classification dimension and limited applicability. This scheme creatively first performs a programmatic prediction of anesthesia dose, and classifies anesthesia depth through the numerical value given in the personalized anesthesia scheme and physiological signal data, thereby improving the classification dimension of anesthesia depth classification and the availability and versatility of the prediction results;

[0066] (2) In view of the technical problem that the existing anesthesia depth classification method generally only predicts the value of a fixed anesthetic before surgery, but cannot provide a complete anesthesia goal, anesthesia duration and specific anesthesia plan, thus affecting the overall effect of anesthesia dose prediction, this scheme creatively uses a long short-term memory artificial neural network combined with an attention mechanism to predict anesthesia dose schemes, enhances the temporal feature processing of anesthesia duration through a long short-term memory network, and further adds artificial neural hidden layer neurons to enhance the prediction effect of applicable plans for various anesthetics, thereby improving the reliability and accuracy of anesthesia dose prediction;

[0067] (3) In view of the technical problem that the existing anesthesia depth classification methods simply and generally classify the anesthesia depth into several fixed types, the quantization degree of the classification results is low, and it is difficult to cooperate with the anesthesia dose scheme prediction to further improve the classification dimension and interpretability, this scheme creatively uses an encoding and decoding network that combines physiological signal component division enhancement and improved channel attention to quantitatively predict anesthesia depth. Through physiological signal component division and enhancement, physiological signals in different frequency domains are processed separately, and by combining improved channel attention, anesthesia depth classification is performed by combining anesthetic dose prediction values ​​and multi-channel integration. Specific anesthesia depth quantitative index prediction data is obtained, which improves the data availability and provides a technical reference for the overall automation process of anesthesia depth classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 A schematic diagram of a flow chart of an anesthesia depth classification method based on artificial intelligence provided by the present invention;

[0069] Figure 2 A schematic diagram of an anesthesia depth classification system based on artificial intelligence provided by the present invention;

[0070] Figure 3 This is a schematic diagram of the process of data preprocessing in step S2;

[0071] Figure 4 A schematic diagram of the process of predicting the anesthesia dosage scheme in step S3;

[0072] Figure 5 Schematic diagram of the process of quantitative prediction of anesthesia depth in step S4.

[0073] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0074] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0075] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0076] Example 1, see Figure 1 The present invention provides an artificial intelligence-based anesthesia depth classification method, which comprises the following steps:

[0077] Step S1: data collection;

[0078] Step S2: data preprocessing;

[0079] Step S3: Anesthesia dosage planning prediction;

[0080] Step S4: quantitative prediction of anesthesia depth;

[0081] Step S5: Classification of anesthesia depth.

[0082] By performing the above operations, in the existing anesthesia depth classification method, there is a poor correlation between the existing technology for anesthesia depth classification and the specific anesthetic dose, and the traditional anesthesia depth classification only combines EEG data for prediction, which has the technical problems of shallow classification dimension and limited applicability. This scheme creatively first performs a schemed prediction of anesthesia dose, and classifies the anesthesia depth together with the numerical values ​​given in the personalized anesthesia scheme and the physiological signal data, thereby improving the classification dimension of anesthesia depth classification and the availability and versatility of the prediction results.

[0083] Example 2, see Figure 1 and Figure 2 In step S1, the data collection is used to collect basic patient data required for anesthesia depth classification, specifically, to obtain original data of anesthesia depth classification from medical system and medical record data;

[0084] The raw data of anesthesia depth specifically includes physiological signal data, basic patient data and anesthesia target data;

[0085] The physiological signal data specifically includes electroencephalogram signal data, electromyography signal data and blood pressure data;

[0086] The patient's basic data specifically includes basic information data, medical history data and health status data;

[0087] The anesthesia target data specifically includes anesthesia usage data and anesthesia duration data;

[0088] The EEG signal data is used to directly evaluate the bispectral index score value of the anesthesia depth;

[0089] The electromyographic signal data and blood pressure data are used to indirectly affect the bispectral index score value of the anesthesia depth;

[0090] The basic information data is used to personalize the model performance and optimize the individual differences in prediction, specifically including age, gender, height and weight information;

[0091] The medical history data and health status data are used to adjust the model performance and optimize the predicted classification accuracy through medical history records and the current health status of the patient, specifically including past medical history, allergy history, drug use history and anesthesia history and current health status information;

[0092] The anesthesia usage data is used to indicate specific scenarios where anesthesia is required;

[0093] The anesthesia duration data is used to indicate the duration of anesthesia, specifically including the expected duration, actual duration and average duration.

[0094] Example 3, see Figure 1 , Figure 2 and Figure 3 , this embodiment is based on the above embodiment, in step S2, the data preprocessing is used to optimize and enhance the original data, specifically, the data preprocessing is performed on the anesthesia depth classification original data to obtain optimized anesthesia depth classification data;

[0095] The data preprocessing specifically includes the following steps:

[0096] Step S21: enhancing physiological signal data, specifically, performing data filtering operations on the EEG signal data and the EMG signal data, and performing outlier detection, specifically, deleting data whose amplitude exceeds a set threshold in the signal data, and then performing data segmentation on the EEG signal data, the EMG signal data, and the blood pressure data to obtain enhanced physiological signal data;

[0097] Step S22: optimizing the basic data of the patient, specifically, performing a data filtering operation on the basic information data, medical history data and health status data, and using the K nearest neighbor method to fill in missing values, and then performing mathematical statistical analysis on the basic data of the patient to obtain optimized patient data;

[0098] Step S23: optimizing anesthesia target data, specifically performing a data filtering operation on the anesthesia target data and performing a mathematical statistical analysis on the anesthesia duration data to obtain anesthesia target optimization data;

[0099] Step S24: integrated preprocessing, specifically, performing data preprocessing through the physiological signal data enhancement, the patient basic data optimization and the anesthesia target data optimization to obtain optimized anesthesia depth classification data.

[0100] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the anesthetic dose scheme prediction is used to predict the details of the anesthetic dose usage scheme. Specifically, based on the patient optimization data and the anesthetic target optimization data in the optimized anesthesia depth classification data, a long short-term memory artificial neural network combined with an attention mechanism is used to perform anesthetic dose scheme prediction to obtain anesthetic dose scheme recommendation data;

[0101] The long short-term memory artificial neural network combined with the attention mechanism specifically includes a long short-term memory layer, an attention layer, an artificial neural hidden layer and an artificial neural output layer;

[0102] The long short-term memory layer is used to process the temporal characteristics in the anesthesia target data and serves as the artificial neural input layer of the entire model;

[0103] The attention layer is used to enhance the model's ability to discover and predict key information features;

[0104] The artificial neural hidden layer is used to process the complex characteristics of the anesthetic dose and perform a programmatic prediction of the anesthetic dose;

[0105] The artificial neural output layer is used to output a prediction scheme of anesthesia dosage;

[0106] The step of using a long short-term memory artificial neural network combined with an attention mechanism to perform anesthesia dosage scheme prediction to obtain anesthesia dosage scheme recommendation data includes:

[0107] Step S31: constructing a long short-term memory layer, specifically by sequentially constructing an input gate, a forget gate, and an output gate, performing time series feature processing, obtaining a long short-term memory prediction output, and using the long short-term memory prediction output as an input of the attention layer;

[0108] The calculation formula for constructing the long short-term memory layer and performing time series feature processing is:

[0109]

[0110] In the formula, i t is the input gate output, f t is the output of the forget gate, Sig(·) is the S-type activation function, W i is the input gate weight, h t-1is the hidden state vector at the previous time instant t-1, t is the time index, x t is the input data at time t, b i is the input gate bias term, W f is the forget gate weight, b f is the forget gate bias term, is the candidate cell state, tanh(·) is the tanh activation function, W C is the cell state weight, b C is the cell state bias term, C t is the cell state at the current time t, C t-1 is the cell state at the previous time t-1, o t is the output gate output, W o is the output gate weight, b o is the output gate bias term, h t is the hidden state vector at the current time t, [·,·] is the vector concatenation operator, is the element-wise multiplication operator;

[0111] Step S32: constructing an attention layer, specifically, calculating the influence of the hidden state of the long short-term memory layer on the current output by calculating the normalized score, the attention weight and the context vector, obtaining a high-price feature of the anesthetic dose, and using the high-price feature of the anesthetic dose as the output data of the artificial neural hidden layer;

[0112] The calculation formula for constructing the attention layer is:

[0113]

[0114] In the formula, e t is the unnormalized score value of the attention weight, u is the context learning vector, W is the slightly normalized weight matrix, t is the time index, and h t is the hidden state vector at the current time t, b is the unnormalized bias term, α t is the attention weight value at the current time t, exp(·) is the natural base function, j is the normalized score calculation time index, e j is the unnormalized score value corresponding to the jth moment, s t is the context weighted vector, T is the total time length of the time series data;

[0115] Step S33: constructing an artificial neural hidden layer, specifically constructing an artificial neural network hidden layer after the attention layer, and performing a weighted and biased calculation of the calculated predicted value for anesthesia dose prediction by constructing the hidden layer to obtain an artificial neural hidden layer predicted output;

[0116] Step S34: constructing an artificial neural output layer, specifically, performing numerical calculation of anesthesia dosage scheme by obtaining the predicted output of the artificial neural hidden layer, and obtaining anesthesia dosage predicted output;

[0117] Step S35: training the anesthesia dose prediction model, specifically, training the model by constructing the long short-term memory layer, the attention layer, the artificial neural hidden layer, and the artificial neural output layer, to obtain the anesthesia dose prediction model Model DP ;

[0118] The model training specifically includes the following steps:

[0119] Step S351: data set division, specifically dividing the optimized anesthesia depth classification data into 90% training samples and 10% test samples;

[0120] Step S352: Model parameter setting, specifically setting the anesthesia dose prediction model Model DP The input, output, optimal number of iterations parameter, batch size parameter and other hyperparameters of

[0121] Step S353: training and evaluation, specifically optimizing model weights, bias terms and hyperparameter values ​​according to the output of each iterative training;

[0122] Step S354: iterative training, specifically repeating the operation described in step S353 until the model performance reaches the optimal value or reaches the maximum number of iterations;

[0123] Step S355: a performance evaluation test, specifically using the test sample to perform root mean square error and mean absolute error analysis, perform a performance evaluation test, and evaluate the model performance through the performance evaluation test;

[0124] Step S36: Anesthesia dose scheme prediction, specifically using the anesthesia dose prediction model Model DP , perform anesthetic dosage scheme prediction and obtain anesthetic dosage scheme recommendation data;

[0125] The anesthesia dosage regimen recommendation data specifically includes anesthetic drug classification predicted dosage data, maintenance injection rate predicted data, maintenance dosage adjustment data and anesthesia combination strategy data;

[0126] The anesthetic drug classification predicted dose data specifically includes induction drug data, anesthetic drug data, muscle relaxant data and maintenance drug data;

[0127] The anesthesia combination strategy data specifically refers to data for recommending specific drug combinations based on anesthesia target data.

[0128] By executing the above operations, in order to address the technical problem that in the existing anesthesia depth classification method, the existing anesthetic dose prediction generally only performs preoperative prediction on the numerical value of a fixed anesthetic, but cannot provide a complete anesthesia goal, anesthesia duration and specific anesthesia plan, thereby affecting the overall effect of anesthesia dose prediction, this scheme creatively uses a long short-term memory artificial neural network combined with an attention mechanism to perform anesthesia dose scheme prediction, enhances the temporal feature processing of anesthesia duration through a long short-term memory network, and further adds artificial neural hidden layer neurons to enhance the prediction effect of applicable plans for various anesthetics, thereby improving the reliability and accuracy of anesthesia dose prediction.

[0129] Example 5, see Figure 1 , Figure 2 and Figure 5 This embodiment is based on the above embodiment. In step S4, the quantitative prediction of anesthesia depth is used to quantitatively predict the anesthesia depth in combination with the use scheme details. Specifically, based on the enhanced physiological signal data in the optimized anesthesia depth classification data and the anesthesia dose scheme recommendation data, the encoding and decoding network combined with the physiological signal component division enhancement and improved channel attention is used to quantitatively predict the anesthesia depth to obtain the anesthesia depth quantitative index prediction data;

[0130] The encoding and decoding network combined with physiological signal component division enhancement and improved channel attention specifically includes an input embedding layer, a signal enhancement layer, a frequency domain attention layer, an encoding and decoding layer, an improved channel attention layer and a linear output layer;

[0131] The input embedding layer is used to receive the model input and perform conversion processing and feature extraction on the input physiological signal;

[0132] The signal enhancement layer is used to decompose the original physiological signal data according to the frequency components and process them separately;

[0133] The frequency domain attention layer is used to introduce a multi-head attention mechanism for feature optimization;

[0134] The codec layer is used to construct a variable-voltage encoder and add layer normalization and residual links;

[0135] The improved channel attention layer is used to integrate the respective predicted outputs of the frequency components and obtain weighted features;

[0136] The linear output layer is used to linearly change the weighted features and obtain the final prediction output;

[0137] The step of using an encoding / decoding network that combines physiological signal component division enhancement and improved channel attention to perform quantitative prediction of anesthesia depth to obtain anesthesia depth quantitative index prediction data includes:

[0138] Step S41: constructing an input embedding layer, specifically taking the enhanced physiological signal data and the anesthesia dosage regimen recommendation data as model inputs, performing signal feature extraction on the enhanced physiological signal data, and performing anesthesia information basic feature extraction on the anesthesia dosage regimen recommendation data, to obtain signal feature data and numerical feature data;

[0139] Step S42: constructing a signal enhancement layer, specifically dividing the signal characteristic data according to the frequency to obtain frequency-divided signal data;

[0140] The frequency division signal data specifically includes a first-level frequency, a second-level frequency, a third-level frequency and a fourth-level frequency;

[0141] The value range of the first-level frequency is [0.5Hz, 7Hz];

[0142] The value range of the second-level frequency is [7Hz, 15Hz];

[0143] The value range of the third frequency is [15Hz, 31Hz];

[0144] The value range of the fourth frequency is [31Hz, 46Hz];

[0145] Step S43: construct a frequency domain attention layer, specifically introduce a multi-head self-attention mechanism, perform multi-head self-attention processing on each group of frequency feature signals in the frequency division signal data, obtain frequency domain self-attention grouped data, and perform inverse Fourier transform and padding operations on the frequency domain self-attention grouped data to convert the frequency domain self-attention grouped data into time domain feature data. The calculation formula is:

[0146] Z=-1(Pad(MultiHead(Q, K, V))+X+Z′);

[0147] Where Z is the time domain feature data, -1 (·) is the inverse Fourier transform, Pad(·) is the zero padding function, MultiHead(·) is the multi-head self-attention function, Q is the query vector, K is the key vector, V is the value vector, Z′ is the frequency domain self-attention grouping data, and X is the numerical feature data;

[0148] Step S44: constructing a codec layer, specifically, after the frequency domain attention layer, constructing a transformer block for feature encoding, and constructing layer normalization and residual linking, constructing the codec layer, and performing hierarchical feature processing on the frequency division signal data to obtain hierarchical feature data;

[0149] Step S45: constructing an improved channel attention layer, specifically, constructing an improved channel attention layer through a squeeze block and an excitation block, and outputting and integrating hierarchical feature data by constructing the improved channel attention layer, specifically including the following steps:

[0150] Step S451: constructing an extrusion block, specifically performing a standard convolution operation, and performing weight calculation of the hierarchical feature data through global average pooling to obtain a hierarchical feature weight value;

[0151] Step S452: constructing an excitation block, specifically using two fully connected layers and a nonlinear activation function to construct an excitation function, transforming the output signal of the excitation function through an S-type activation function, and updating the weights of the hierarchical feature data and re-adjusting the activation through channel multiplication to obtain hierarchical feature channel attention weights;

[0152] The calculation formula for constructing the activation function is:

[0153] F S =Sig(W2ReLU(W1z));

[0154] In the formula, F S is the activation function, Sig(·) is the S-type activation function, W2 is the weight of the second fully connected layer, ReLU(·) is the nonlinear activation function, W1 is the weight of the first fully connected layer, and z is the hierarchical feature data;

[0155] Step S453: hierarchical feature integration, specifically, integrating the attention weights of hierarchical feature channels to obtain weighted feature data, and the calculation formula is:

[0156] F=Concat(z1,z2,z3,z4);

[0157] Where F is the weighted feature data, Concat(·) is the feature concatenation operation, z1 is the hierarchical feature channel attention weight corresponding to the first level frequency, z2 is the hierarchical feature channel attention weight corresponding to the second level frequency, z3 is the hierarchical feature channel attention weight corresponding to the third level frequency, and z4 is the hierarchical feature channel attention weight corresponding to the fourth level frequency;

[0158] Step S46: constructing a linear output layer, specifically performing a linear transformation on the weighted feature data to obtain linear quantification data of anesthesia depth index;

[0159] Step S47: training the anesthesia depth quantitative prediction model, specifically, training the model by constructing the input embedding layer, constructing the signal enhancement layer, constructing the frequency domain attention layer, constructing the encoding and decoding layer, constructing the improved channel attention layer and constructing the linear output layer, to obtain the anesthesia depth quantitative prediction model ModelDA ;

[0160] Step S48: quantitative prediction of anesthesia depth, specifically using the anesthesia depth quantitative prediction model Model DA , performing quantitative prediction of anesthesia depth based on the enhanced physiological signal data in the optimized anesthesia depth classification data and the anesthesia dosage regimen recommendation data to obtain anesthesia depth quantitative index prediction data;

[0161] The predicted value of the quantitative index of anesthesia depth specifically refers to the bispectral index score value;

[0162] Table 1 is a table of specific classification results corresponding to the quantification and type of anesthesia depth in Example 5 of the present invention. As shown in the table, when the value of the bispectral index score is [0, 35), the anesthesia depth type is deep anesthesia;

[0163] When the bispectral index score is [35,70), the anesthesia depth type is moderate anesthesia;

[0164] When the bispectral index score is [70,85), the anesthesia depth type is mild anesthesia;

[0165] When the bispectral index score is [85,100], the anesthesia depth type is awake;

[0166] Table 1 Specific classification results corresponding to anesthesia depth quantification and type

[0167]

[0168] By performing the above operations, in view of the technical problem that in the existing anesthesia depth classification methods, the existing anesthesia depth classification simply and generally divides the anesthesia depth into several fixed types, the quantization degree of the classification results is low, and it is difficult to cooperate with the anesthesia dose scheme prediction to further improve the classification dimension and interpretability, this scheme creatively adopts an encoding and decoding network that combines physiological signal component division enhancement and improved channel attention to quantitatively predict anesthesia depth. Through physiological signal component division and enhancement, physiological signals in different frequency domains are processed separately, and by combining improved channel attention, anesthesia depth classification combined with anesthetic dose prediction values ​​and multi-channel integration is performed, and specific anesthesia depth quantitative index prediction data is obtained, which improves the data availability and provides a technical reference for the overall automation process of anesthesia depth classification.

[0169] Example 6, see Figure 1 and Figure 2This embodiment is based on the above embodiment. In step S5, the anesthesia depth classification is used to integrate the prediction results of the anesthetic dose and the anesthetic depth to obtain a complete anesthesia depth classification result. Specifically, the anesthesia dose scheme recommendation data and the anesthesia depth quantitative index prediction data are combined with the anesthesia depth classification original data to perform an integrated prediction of the anesthesia depth, and obtain the anesthetic dose and quantitative classification prediction result data.

[0170] Embodiment 7, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, and the present invention provides an artificial intelligence-based anesthesia depth classification system, including a data collection module, a data preprocessing module, an anesthesia dose scheme prediction module, an anesthesia depth quantification prediction module and an anesthesia depth classification module;

[0171] The data collection module is used for data collection, and obtains the original data of anesthesia depth classification through data collection, and sends the original data of anesthesia depth classification to the data preprocessing module and the anesthesia depth classification module;

[0172] The data preprocessing module is used for data preprocessing, and obtains optimized anesthesia depth classification data through data preprocessing, and sends the optimized anesthesia depth classification data to the anesthesia dose scheme prediction module and the anesthesia depth quantification prediction module;

[0173] The anesthesia dose program prediction module is used for anesthesia dose program prediction, obtains anesthesia dose program recommendation data through anesthesia dose program prediction, and sends the anesthesia dose program recommendation data to the anesthesia depth quantification prediction module and the anesthesia depth classification module;

[0174] The anesthesia depth quantification prediction module is used for the anesthesia depth quantification prediction module, obtains anesthesia depth quantification index prediction data through the anesthesia depth quantification prediction module, and sends the anesthesia depth quantification index prediction data to the anesthesia depth classification module;

[0175] The anesthesia depth classification module is used for anesthesia depth classification, and obtains anesthesia dose and quantitative classification prediction result data through anesthesia depth classification.

[0176] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0177] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0178] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An artificial intelligence-based anesthesia depth classification method, characterized by: The method comprises the following steps: Step S1: Data collection, obtaining the original data of anesthesia depth classification, including physiological signal data, patient basic data and anesthesia target data; Step S2: data preprocessing to obtain optimized anesthesia depth classification data; Step S3: Anesthesia dosage scheme prediction is used to predict the details of the anesthesia dosage scheme, specifically, based on the patient optimization data and the anesthesia target optimization data in the optimized anesthesia depth classification data, anesthesia dosage scheme prediction is performed using a long short-term memory artificial neural network combined with an attention mechanism to obtain anesthesia dosage scheme recommendation data; The anesthetic dosage regimen recommendation data specifically includes anesthetic drug classification predicted dosage data, maintenance injection rate predicted data, maintenance dosage adjustment data and anesthetic combination strategy data; Step S4: quantitative prediction of anesthesia depth, which is used to quantitatively predict the anesthesia depth in combination with the use scheme details, specifically, based on the enhanced physiological signal data in the optimized anesthesia depth classification data and the anesthesia dosage scheme recommendation data, a coding and decoding network that combines physiological signal component division enhancement and improved channel attention is used to quantitatively predict the anesthesia depth, and obtain anesthesia depth quantitative index prediction data; The encoding and decoding network combined with physiological signal component division enhancement and improved channel attention specifically includes an input embedding layer, a signal enhancement layer, a frequency domain attention layer, an encoding and decoding layer, an improved channel attention layer and a linear output layer; The step of using an encoding / decoding network that combines physiological signal component division enhancement and improved channel attention to perform quantitative prediction of anesthesia depth to obtain anesthesia depth quantitative index prediction data includes: Step S41: constructing an input embedding layer; Step S42: constructing a signal enhancement layer; Step S43: constructing a frequency domain attention layer; Step S44: constructing a codec layer; Step S45: constructing an improved channel attention layer; Step S46: constructing a linear output layer; Step S47: training anesthesia depth quantitative prediction model; Step S48: quantitative prediction of anesthesia depth; In step S41, the input embedding layer is constructed by taking the enhanced physiological signal data and the anesthesia dosage regimen recommendation data as model inputs, performing signal feature extraction on the enhanced physiological signal data, and performing anesthesia information basic feature extraction on the anesthesia dosage regimen recommendation data to obtain signal feature data and numerical feature data; In step S42, the signal enhancement layer is constructed by dividing the signal characteristic data according to the frequency to obtain frequency-divided signal data; In step S43, the frequency domain attention layer is constructed by introducing a multi-head self-attention mechanism, performing multi-head self-attention processing on each group of frequency feature signals in the frequency division signal data to obtain frequency domain self-attention grouped data, and performing inverse Fourier transform and padding operations on the frequency domain self-attention grouped data to convert the frequency domain self-attention grouped data into time domain feature data. The calculation formula is: Z= -1 (Pad(MultiHead(Q,K,V))+X+Z′); Where Z is the time domain feature data, -1 (·) is the inverse Fourier transform, Pad(·) is the zero padding function, MultiHead(·) is the multi-head self-attention function, Q is the query vector, K is the key vector, V is the value vector, Z′ is the frequency domain self-attention grouping data, and X is the numerical feature data; In step S44, the construction of the codec layer is specifically to construct a transformer block for feature encoding after the frequency domain attention layer, and to construct layer normalization and residual linking, to construct the codec layer, and to perform hierarchical feature processing on the frequency division signal data to obtain hierarchical feature data; In step S45, the construction of the improved channel attention layer specifically refers to constructing the improved channel attention layer through the squeeze block and the excitation block, and outputting and integrating the hierarchical feature data by constructing the improved channel attention layer, which specifically includes the following steps: Step S451: constructing an extrusion block, specifically performing a standard convolution operation, and performing weight calculation of the hierarchical feature data through global average pooling to obtain a hierarchical feature weight value; Step S452: constructing an excitation block, specifically using two fully connected layers and a nonlinear activation function to construct an excitation function, transforming the output signal of the excitation function through an S-type activation function, and updating the weights of the hierarchical feature data and re-adjusting the activation through channel multiplication to obtain hierarchical feature channel attention weights; The calculation formula for constructing the activation function is: F S =Sig(W2ReLU(W1z)); In the formula, F S is the activation function, Sig(·) is the S-type activation function, W2 is the weight of the second fully connected layer, ReLU(·) is the nonlinear activation function, W1 is the weight of the first fully connected layer, and z is the hierarchical feature data; Step S453: hierarchical feature integration, specifically, integrating the attention weights of hierarchical feature channels to obtain weighted feature data, and the calculation formula is: F = Concat(z1,z2,z3,z4); Where F is the weighted feature data, Concat(·) is the feature concatenation operation, z1 is the hierarchical feature channel attention weight corresponding to the first level frequency, z2 is the hierarchical feature channel attention weight corresponding to the second level frequency, z3 is the hierarchical feature channel attention weight corresponding to the third level frequency, and z4 is the hierarchical feature channel attention weight corresponding to the fourth level frequency; In step S46, the linear output layer is constructed by linearly transforming the weighted feature data to obtain linear quantification data of the anesthesia depth index; In step S47, the anesthesia depth quantitative prediction model training is specifically carried out by constructing the input embedding layer, the signal enhancement layer, the frequency domain attention layer, the encoding and decoding layer, the improved channel attention layer and the linear output layer to obtain the anesthesia depth quantitative prediction model Model DA ; In step S48, the quantification prediction of the depth of anesthesia is specifically performed using the quantification prediction model Model DA , performing quantitative prediction of anesthesia depth based on the enhanced physiological signal data in the optimized anesthesia depth classification data and the anesthesia dosage regimen recommendation data to obtain anesthesia depth quantitative index prediction data; Step S5: Classify the depth of anesthesia to obtain the anesthesia dose and quantitative classification prediction result data.

2. The method for classifying depth of anesthesia based on artificial intelligence according to claim 1, characterized in that: In step S3, the long short-term memory artificial neural network combined with the attention mechanism specifically includes a long short-term memory layer, an attention layer, an artificial neural hidden layer and an artificial neural output layer; The long short-term memory layer is used to process the temporal characteristics in the anesthesia target data and serves as the artificial neural input layer of the entire model; The attention layer is used to enhance the model's ability to discover and predict key information features; The artificial neural hidden layer is used to process the complex characteristics of the anesthetic dose and perform a programmatic prediction of the anesthetic dose; The artificial neural output layer is used to output a prediction plan for anesthesia dosage.

3. The method for classifying depth of anesthesia based on artificial intelligence according to claim 2, characterized in that: In step S3, the step of using a long short-term memory artificial neural network combined with an attention mechanism to perform anesthesia dosage scheme prediction to obtain anesthesia dosage scheme recommendation data includes: Step S31: constructing a long short-term memory layer, specifically by sequentially constructing an input gate, a forget gate, and an output gate, performing time series feature processing, obtaining a long short-term memory prediction output, and using the long short-term memory prediction output as an input of the attention layer; Step S32: constructing an attention layer, specifically, calculating the influence of the hidden state of the long short-term memory layer on the current output by calculating the normalized score, the attention weight and the context vector, obtaining a high-price feature of the anesthetic dose, and using the high-price feature of the anesthetic dose as the output data of the artificial neural hidden layer; Step S33: constructing an artificial neural hidden layer, specifically constructing an artificial neural network hidden layer after the attention layer, and performing a weighted and biased calculation of the calculated predicted value for anesthesia dose prediction by constructing the hidden layer to obtain an artificial neural hidden layer predicted output; Step S34: constructing an artificial neural output layer, specifically, performing numerical calculation of anesthesia dosage scheme by obtaining the predicted output of the artificial neural hidden layer, and obtaining anesthesia dosage predicted output; Step S35: training the anesthesia dose prediction model, specifically, training the model by constructing the long short-term memory layer, the attention layer, the artificial neural hidden layer, and the artificial neural output layer, to obtain the anesthesia dose prediction model Model DP ; Step S36: Anesthesia dose scheme prediction, specifically using the anesthesia dose prediction model Model DP , perform anesthetic dosage scheme prediction and obtain anesthetic dosage scheme recommendation data; The anesthetic dosage regimen recommendation data specifically includes anesthetic drug classification predicted dosage data, maintenance injection rate predicted data, maintenance dosage adjustment data and anesthetic combination strategy data; The anesthetic drug classification predicted dose data specifically includes induction drug data, anesthetic drug data, muscle relaxant data and maintenance drug data; The anesthesia combination strategy data specifically refers to data for recommending specific drug combinations based on anesthesia target data.

4. The method for classifying anesthesia depth based on artificial intelligence according to claim 3, characterized in that: In step S4, the encoding / decoding network for combining physiological signal component division enhancement and improved channel attention specifically includes an input embedding layer, a signal enhancement layer, a frequency domain attention layer, an encoding / decoding layer, an improved channel attention layer and a linear output layer; The input embedding layer is used to receive the model input and perform conversion processing and feature extraction on the input physiological signal; The signal enhancement layer is used to decompose the original physiological signal data according to the frequency components and process them separately; The frequency domain attention layer is used to introduce a multi-head attention mechanism for feature optimization; The codec layer is used to construct a variable-voltage encoder and add layer normalization and residual links; The improved channel attention layer is used to integrate the respective predicted outputs of the frequency components and obtain weighted features; The linear output layer is used to perform linear changes on the weighted features and obtain the final prediction output.

5. The method for classifying depth of anesthesia based on artificial intelligence according to claim 4, characterized in that: In step S5, the anesthesia depth classification is used to integrate the prediction results of the anesthetic dose and the anesthetic depth to obtain a complete anesthesia depth classification result. Specifically, the anesthesia dose scheme recommendation data and the anesthesia depth quantitative index prediction data are combined with the anesthesia depth classification original data to perform an integrated prediction of the anesthesia depth, and obtain the anesthetic dose and quantitative classification prediction result data.

6. An artificial intelligence-based anesthesia depth classification system, used to implement an artificial intelligence-based anesthesia depth classification method as claimed in any one of claims 1 to 5, characterized in that: It includes a data collection module, a data preprocessing module, an anesthesia dose planning prediction module, an anesthesia depth quantitative prediction module and an anesthesia depth classification module.

7. The artificial intelligence-based anesthesia depth classification system according to claim 6, characterized in that: The data collection module is used for data collection, and obtains the original data of anesthesia depth classification through data collection, and sends the original data of anesthesia depth classification to the data preprocessing module and the anesthesia depth classification module; The data preprocessing module is used for data preprocessing, and obtains optimized anesthesia depth classification data through data preprocessing, and sends the optimized anesthesia depth classification data to the anesthesia dose scheme prediction module and the anesthesia depth quantification prediction module; The anesthesia dose program prediction module is used for anesthesia dose program prediction, obtains anesthesia dose program recommendation data through anesthesia dose program prediction, and sends the anesthesia dose program recommendation data to the anesthesia depth quantification prediction module and the anesthesia depth classification module; The anesthesia depth quantification prediction module is used for the anesthesia depth quantification prediction module, obtains anesthesia depth quantification index prediction data through the anesthesia depth quantification prediction module, and sends the anesthesia depth quantification index prediction data to the anesthesia depth classification module; The anesthesia depth classification module is used for anesthesia depth classification, and obtains anesthesia dose and quantitative classification prediction result data through anesthesia depth classification.