A method and system for controlling high frequency ventilation anesthesia

Abstract

This invention discloses a high-frequency ventilation anesthesia control method and system, relating to the field of medical anesthesia control. The method includes acquiring clinical case information for high-frequency ventilation anesthesia; analyzing and extracting clinical case data for each case; preprocessing the clinical case data to obtain sample data; using an improved LSTM to filter and train the sample data to obtain initial expressions for ventilation parameters; acquiring intraoperative monitoring data from the sample data; establishing a monitoring curve with time as the horizontal axis and intraoperative monitoring data as the vertical axis; extracting features from the monitoring curve and establishing a correlation with intraoperative control data; configuring correlation factors based on the correlation to generate a control expression for the initial expression of ventilation parameters; and controlling the high-frequency ventilation anesthesia process based on the initial expression and the control expression. This invention achieves precise and coordinated control of the anesthesia and ventilation processes using the above method.

Description

Technical Field

[0001] This invention relates to the field of medical anesthesia control, and in particular to a high-frequency ventilation anesthesia control method and system. Background Technology

[0002] High-frequency ventilation anesthesia is a medical technique that combines high-frequency ventilation technology with clinical anesthesia depth control. It ensures oxygen supply and carbon dioxide removal for patients during surgery through a high-frequency, low-tidal-volume ventilation mode, while simultaneously controlling the infusion rate of anesthetic drugs to maintain a stable depth of anesthesia during surgery, thus achieving coordinated operation of ventilation support and anesthesia management.

[0003] Its main applications are concentrated in various surgeries such as thoracic surgery, general surgery, and pediatric surgery. It is especially suitable for minimally invasive thoracoscopic surgery, surgery for patients with lung diseases, and surgery for newborns and infants, where high precision in ventilation is required, or where patients have weak lung function and require strict control of airway pressure. In addition, in long and complex surgeries, it can also reduce the risk of intraoperative complications through stable ventilation and anesthesia.

[0004] Currently, in the field of high-frequency ventilation anesthesia, high-frequency ventilation systems and anesthesia control systems mostly operate independently, lacking a coordinated linkage mechanism. The adjustment of ventilation parameters and anesthesia parameters is fragmented and cannot be coordinated and adapted according to the patient's physiological feedback, which easily leads to regulatory imbalance.

[0005] The initial setting and intraoperative adjustment of control parameters (respiratory rate and tidal volume in high-frequency ventilation, and infusion rate of anesthesia drugs such as remifentanil) rely excessively on the clinical experience of anesthesiologists, which is highly subjective and difficult to accurately adapt to individual differences in age, weight, and lung function (such as FEV1) among different patients. Furthermore, no quantitative correlation has been established between individual signs, surgical type, intraoperative physiological feedback parameters (such as BIS index and end-tidal CO2), and control parameters. Intraoperative adjustments lack the support of precise mathematical models, making it difficult to achieve personalized and precise control.

[0006] Existing control models have fixed parameters and cannot dynamically adapt to fluctuations in the patient's physiological state and changes in the intensity of surgical stimulation during surgery. This results in insufficient model adaptability and control stability, increasing the risks of intraoperative respiratory depression, excessively shallow or deep anesthesia, and other problems.

[0007] To solve these problems, there is an urgent need for a high-frequency ventilation anesthesia control method and system. Summary of the Invention

[0008] To address the aforementioned issues, this application proposes a high-frequency ventilation anesthesia control method and system. The aim is to construct personalized initial control parameter expressions based on machine learning models, and then dynamically correct these expressions by combining real-time feedback of intraoperative physiological parameters. This ultimately forms a closed-loop high-frequency ventilation anesthesia collaborative control system, ensuring the quantification, accuracy, and personalization of the control process. The specific details are as follows: A method for controlling anesthesia with high-frequency ventilation includes the following steps: S1. Obtain clinical case information of high-frequency ventilation anesthesia, analyze and extract clinical case data for each case, and preprocess the clinical case data to obtain sample data. S2. An improved LSTM is used to filter and train the sample data to obtain the initial expression for the ventilation parameters; S4. Obtain intraoperative monitoring data from the sample data, establish a monitoring curve with time as the horizontal axis and intraoperative monitoring data as the vertical axis, extract features from the monitoring curve and establish a correlation with the intraoperative control data; S5. Based on the correlation relationship, configure the correlation factor for the initial expression of the ventilation parameter to generate the control expression; S6. Control the high-frequency ventilation anesthesia process based on the initial expression and the regulation expression.

[0009] Preferably, the specific content of the sample data obtained by preprocessing clinical case data in S1 includes: The clinical case data includes initial preoperative characteristic data, initial ventilation parameters, intraoperative monitoring data, intraoperative adjustment data, and effect evaluation; Multivariate correlation analysis was performed on the initial preoperative characteristic data and initial ventilation parameters, and relevant preoperative characteristic data were obtained by screening. Relevant preoperative characteristic data include: Individual trait parameters: age F1, weight F2, forced expiratory volume in one second F3; Surgical type parameters: Surgical site F4, Stimulation intensity F5; First BIS index: F6; The remaining data in the relevant preoperative characteristic data and clinical case data constitute pure clinical case data; Based on the surgical type parameter, the pure clinical case data were divided into surgical category groups; The effectiveness evaluation is coded and converted to obtain the effectiveness evaluation value. Individual cases in the surgical category group are sorted from high to low according to the effectiveness evaluation value. The top 80% of cases were selected as sample data. The sample data includes relevant preoperative characteristic data, initial ventilation parameters, intraoperative monitoring data, and intraoperative control data under different surgical category groups.

[0010] Preferably, in step S2, the improved LSTM filters the sample data to obtain relevant preoperative feature data and initial ventilation parameters, and uses the relevant preoperative feature data and initial ventilation parameters as input data and output data, respectively. Initial ventilation parameters include high-frequency ventilation respiratory rate, tidal volume, and remifentanil infusion rate; The improved LSTM includes an input layer, a feature enhancement layer, a gated LSTM hidden layer, a multi-target attention fully connected layer, and an output layer. Initial ventilation parameters include high-frequency ventilation respiratory rate Z1, tidal volume Z2, and remifentanil infusion rate Z3; The newly added clinical safety constraint branch in the LSTM hidden layer based on the gating mechanism ensures that the output temporal features conform to the physiological safety range. The expression is as follows: ; ; in, The improved input gate output value controls the input weights of the current feature. It is the sigmoid activation function. ; The input gate weight matrix, The time-series features are the output of the LSTM at the previous time step. This involves concatenating the output from the previous time step with the current enhanced features. For input gate bias, For element-wise multiplication, Clinical constraint weights; , To constrain branch weights and biases, To strengthen the safety constraint, a constraint strength coefficient is used. The clinical safety range vector is the core feature.

[0011] Preferably, the multi-target attention fully connected layer designs a dynamic attention weight allocation mechanism to address the collaborative needs of the three outputs, thereby strengthening the influence of key features on the core control target; ; ; in, The attention weights for the k-th control objective, i.e., the output. Using the cosine similarity function, we calculate the correlation strength between the fully connected layer features and the control target reference vector. The basic feature vector output by the fully connected layer. Let be the reference vector for the k-th control target. Let be the attention enhancement coefficient for the k-th control target. This is the weighted feature vector of the fully connected layer, used for prediction in the subsequent output layer.

[0012] Preferably, the initial expression for the ventilation parameters is: Initial respiratory rate: ; Initial tidal volume: ; Initial remifentanil rate: .

[0013] Preferably, in step S4, the intraoperative monitoring data from the sample data is obtained, a monitoring curve is established with time as the horizontal axis and intraoperative monitoring data as the vertical axis, features are extracted from the monitoring curve, and a correlation is established with the intraoperative control data. The intraoperative monitoring data includes the second BIS index. End-tidal carbon dioxide level ; Intraoperative control data include respiratory rate control values, tidal volume control values, and remifentanil rate control values; A monitoring curve was established with time as the horizontal axis and the second BIS index and end-tidal carbon dioxide value as the vertical axes. Several abnormal time periods were obtained by marking the monitoring curves based on the time periods of intraoperative control data; A pre-set warning period is provided, which is a stable period and is set before the abnormal period. The remaining period is defined as a stable period. A standard monitoring curve is obtained by migrating and fitting data during the stable periods under the same curve. Early warning features are obtained by extracting features from the early warning period, and abnormal features are obtained by extracting features from the abnormal period. The correlation between abnormal features and intraoperative control data is obtained by training the data, and then a correlation factor is generated.

[0014] Preferably, the initial expression for ventilation parameters is configured with correlation factors based on the correlation relationship, thereby generating the regulation expression: ; ; ; in, This represents the fitted real-time respiratory rate. To improve the initial respiratory rate output by the LSTM model, for right The positive correlation coefficient of the correlation factors The power-order fitting factor. This refers to end-tidal CO2 deviation. The sign function ensures that the adjustment direction is consistent with the deviation direction. This represents the clinically safe upper limit for respiratory rate. This represents the real-time tidal volume after fitting. To improve the initial tidal volume output of the LSTM model, These are the fitting coefficients for the deviation interval. These are the fitting coefficients for the large deviation interval; This is the deviation threshold; The fitted real-time remifentanil infusion rate; To improve the initial remifentanil rate output by the LSTM model, for right The negative correlation fitting coefficient, BIS bias, It is an exponential function, enabling rapid adjustment of the anesthetic drug rate; , The upper / lower limits of clinical safety for remifentanil rate; , This is a safety constraint.

[0015] Preferably, the specific content of controlling the high-frequency ventilation anesthesia process based on the initial expression and the regulation expression in S6 includes: Initial stage: High-frequency ventilation anesthesia input is directly based on the initial expression; Process stage: Preset fine-tuning values ​​are used to generate detection curves in real time during the detection process and match them during stable periods, warning periods, and abnormal periods; No adjustments are made during periods of stable testing. When the detection occurs within the warning period, fine-tuning is performed using the correlation between the fine-tuning value and the parameter. If the subsequent curve matches the stable period, the fine-tuning is stopped. If the subsequent curve does not match the stable period, continue to fine-tune by multiples of the fine-tuning value and repeat the matching process. If an abnormal time period is matched, adjustments are made based on the control expression; If a stable period is found, monitoring will continue.

[0016] A high-frequency ventilation anesthesia control system, comprising: Data processing unit: acquires clinical case information of high-frequency ventilation anesthesia, analyzes and extracts clinical case data for each case, and preprocesses the clinical case data to obtain sample data; Initial parameter determination unit: The sample data is filtered and trained using an improved LSTM to obtain the initial expression for the ventilation parameters; Intraoperative Relationship Determination Unit: Acquire intraoperative monitoring data from the sample data, establish a monitoring curve with time as the horizontal axis and intraoperative monitoring data as the vertical axis, extract features from the monitoring curve and establish a correlation with the intraoperative control data; Intraoperative control determination unit: Based on the initial expression of ventilation parameters with correlation, the control factors are configured and adjusted to generate the control expression; Control execution unit: controls the high-frequency ventilation anesthesia process based on the initial expression and the control expression.

[0017] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor invokes the computer program in the memory to implement the content of a high-frequency ventilation anesthesia control method.

[0018] A storage medium storing computer-executable instructions, which, when loaded and executed by a processor, implement the content of a high-frequency ventilation anesthesia control method.

[0019] In summary, the high-frequency ventilation anesthesia control method and system of the present invention have the following advantages compared with traditional technologies: 1. The collaborative system of this application breaks down the independent barriers between the ventilation and anesthesia systems, realizes parameter linkage and adaptation, and solves the problem of imbalance in traditional fragmented regulation; 2. This application relies on intraoperative BIS index, real-time monitoring of end-tidal CO2 and deviation analysis to construct a dynamic correction model, provide quantitative mathematical support, realize precise adjustment of intraoperative parameters, and reduce the risks of excessively shallow or deep anesthesia.

[0020] The technical method of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the steps of a high-frequency ventilation anesthesia control method according to the present invention. Figure 2 This is a unit diagram of a high-frequency ventilation anesthesia control system according to the present invention. Detailed Implementation

[0022] The technical method of the present invention will be further described below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application.

[0023] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the scope of this application and its application or use.

[0024] Techniques, systems, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, they should be considered part of the instruction manual.

[0025] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0026] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0027] Example 1 A high-frequency ventilation anesthesia control method, such as Figure 1 As shown, it includes the following steps: S1. Obtain clinical case information for high-frequency ventilation anesthesia, analyze and extract clinical case data for each case, and preprocess the clinical case data to obtain sample data.

[0028] Furthermore, the specific content of the sample data obtained by preprocessing clinical case data in S1 includes: The clinical case data includes initial preoperative characteristic data, initial ventilation parameters, intraoperative monitoring data, intraoperative adjustment data, and efficacy evaluation.

[0029] Multivariate correlation analysis was performed on the initial preoperative characteristic data and initial ventilation parameters, and relevant preoperative characteristic data were obtained by screening.

[0030] Relevant preoperative characteristic data include: Individual trait parameters: age F1, weight F2, forced expiratory volume in one second F3.

[0031] Surgical type parameters: Surgical site F4, Stimulation intensity F5.

[0032] The first BIS index is F6.

[0033] The remaining data in the relevant preoperative characteristic data and clinical case data constitute pure clinical case data.

[0034] Based on the surgical type parameter, the pure clinical case data were divided into surgical category groups.

[0035] The effectiveness evaluation is coded and converted to obtain the effectiveness evaluation value. Individual cases in the surgical category group are sorted from high to low according to the effectiveness evaluation value.

[0036] The top 80% of cases were selected as sample data.

[0037] The sample data includes relevant preoperative characteristic data, initial ventilation parameters, intraoperative monitoring data, and intraoperative control data under different surgical category groups.

[0038] S2. An improved LSTM is used to filter the sample data and train it to obtain the initial expression for the ventilation parameters.

[0039] Furthermore, the improved LSTM described in S2 filters the sample data to obtain relevant preoperative feature data and initial ventilation parameters, and uses the relevant preoperative feature data and initial ventilation parameters as input data and output data, respectively.

[0040] Initial ventilation parameters include high-frequency ventilation respiratory rate, tidal volume, and remifentanil infusion rate.

[0041] The improved LSTM includes an input layer, a feature enhancement layer, a gated LSTM hidden layer, a multi-target attention fully connected layer, and an output layer.

[0042] Initial ventilation parameters include high-frequency ventilation respiratory rate Z1, tidal volume Z2, and remifentanil infusion rate Z3.

[0043] The newly added clinical safety constraint branch in the LSTM hidden layer based on the gating mechanism ensures that the output temporal features conform to the physiological safety range. The expression is as follows: .

[0044] .

[0045] in, The improved input gate output value, ranging from [0,1], controls the input weights of the current feature. It is the sigmoid activation function. .

[0046] The input gate weight matrix (dimension = 32 × (32 + 12)) is learned during training. The temporal features output by the LSTM at the previous time step (dimension=32). This involves concatenating the output from the previous time step with the current enhanced features. The input gate bias (dimension=32) is trained and learned. For element-wise multiplication, This represents the clinical constraint weight, with a value range of [0,1]. It approaches 0 when the feature deviates from the safe range.

[0047] , To constrain branch weights and biases, the training process learns... To strengthen the safety constraints, a constraint strength coefficient (fixed = 0.8) is set. The core feature is the vector of clinical safety ranges (e.g., the safe weight range of 40~150kg).

[0048] Furthermore, the multi-target attention fully connected layer designs a dynamic attention weight allocation mechanism to address the collaborative needs of the three outputs, thereby strengthening the influence of key features on the core control objective.

[0049] .

[0050] .

[0051] in, The attention weights for the k-th control target (output) range from [0,1], and their sum is 1. Using the cosine similarity function, we calculate the correlation strength between the fully connected layer features and the control target reference vector. This is the base feature vector (dimension=8) output by the fully connected layer. The reference vector (dimension=8) for the k-th control objective is learned through training based on historical best control data. Attention enhancement coefficient for the k-th control target (set according to clinical priority): =1.0, =1.0, =1.2), This is the weighted feature vector of the fully connected layer, used for prediction in the subsequent output layer.

[0052] Furthermore, the initial expressions for the ventilation parameters are: Initial respiratory rate: .

[0053] Initial tidal volume: .

[0054] Initial remifentanil rate: .

[0055] S4. Obtain intraoperative monitoring data from the sample data, establish a monitoring curve with time as the horizontal axis and intraoperative monitoring data as the vertical axis, extract features from the monitoring curve and establish a correlation with the intraoperative control data.

[0056] Furthermore, in S4, intraoperative monitoring data from the sample data is obtained, and a monitoring curve is established with time as the horizontal axis and intraoperative monitoring data as the vertical axis. The specific details of feature extraction from the monitoring curve and establishing its correlation with intraoperative control data are as follows: The intraoperative monitoring data includes the second BIS index. End-tidal carbon dioxide level .

[0057] Intraoperative control data include respiratory rate control values, tidal volume control values, and remifentanil rate control values.

[0058] A monitoring curve was established with time as the horizontal axis and the second BIS index and end-tidal carbon dioxide value as the vertical axes.

[0059] Several abnormal periods were obtained by marking the monitoring curves based on the time periods of intraoperative control data.

[0060] A pre-set warning period is provided, which is a stable period and is set before the abnormal period. The remaining period is defined as a stable period.

[0061] Data migration and fitting were performed on the stable periods under the same curve to obtain a standard monitoring curve.

[0062] Early warning features are obtained by extracting features from the early warning period, and abnormal features are obtained by extracting features from the abnormal period.

[0063] The correlation between abnormal features and intraoperative control data is obtained by training the data, and then a correlation factor is generated.

[0064] S5. Based on the correlation relationship, configure the correlation factor for the initial expression of the ventilation parameter to generate the control expression.

[0065] Furthermore, based on the correlation relationships, correlation factors are configured for the initial expressions of ventilation parameters to generate control expressions: .

[0066] .

[0067] .

[0068] in, The fitted real-time respiratory rate (breaths / min).

[0069] To improve the initial respiratory rate (breaths / min) output of the LSTM model.

[0070] for right The positive correlation coefficient of the correlation factor (learned during training, with a value range of 0.5-1.2).

[0071] The power-law fitting factor (learned during training, fixed at 1.3, reflecting the nonlinear acceleration adjustment characteristic of respiratory rate as the deviation increases).

[0072] End-tidal CO2 deviation (mmHg, ΔY2=Y) 2,real -Y 2,target ).

[0073] The sign function ensures that the adjustment direction is consistent with the deviation direction (ΔY2 increases when positive). When negative, decrease ).

[0074] This is the clinically safe upper limit for respiratory rate (fixed at 35 breaths / min).

[0075] As a safety constraint, the closer Z1 is to the upper limit, the smaller the adjustment range should be to avoid exceeding the safety range.

[0076] The fitted real-time tidal volume (ml / kg) is given.

[0077] To improve the initial tidal volume (ml / kg) output of the LSTM model.

[0078] The fitting coefficient for the deviation interval (learned during training, with a value range of 0.02-0.05, suitable for smooth adjustment under small deviations).

[0079] The fitting coefficient for the large deviation interval (learned during training, with a value range of 0.01-0.02, adapting to conservative adjustments under large deviation to avoid over-ventilation).

[0080] The deviation threshold is fixed at 3 mmHg to distinguish between small and large deviation ranges.

[0081] The fitted real-time remifentanil infusion rate is denoted as .

[0082] To improve the initial remifentanil rate output by the LSTM model.

[0083] for right The negative correlation fitting coefficient (learned during training, with a value range of 0.03-0.08, reflecting the exponential response characteristics of drug rate with deviation).

[0084] This is a BIS bias.

[0085] It is an exponential function, enabling rapid adjustment of the anesthetic drug rate (when ΔY1 is positive). Decrease When ΔY1 increases and becomes negative, (This decreases, which aligns with the logic of negative correlation).

[0086] , Clinical safety limits for remifentanil rate (upper / lower limits).

[0087] ( , ): Safety constraints The closer to the upper limit, the smaller the adjustment range; the closer to the lower limit, the larger the adjustment range, thus balancing the depth of anesthesia and safety.

[0088] S6. Control the high-frequency ventilation anesthesia process based on the initial expression and the regulation expression.

[0089] Furthermore, the specific details of controlling the high-frequency ventilation anesthesia process based on the initial expression and the regulation expression in S6 include: Initial stage: High-frequency ventilation anesthesia input is directly based on the initial expression.

[0090] Process stage: Preset fine-tuning values ​​are used to generate detection curves in real time during the detection process and match them during stable periods, warning periods, and abnormal periods.

[0091] No adjustments are made during periods of stable activity.

[0092] When the detection occurs within the warning period, fine-tuning is performed using the correlation between the fine-tuning value and the parameter. If the subsequent curve matches the stable period, the fine-tuning is stopped.

[0093] If the subsequent curve does not match the stable period, continue to fine-tune by multiples of the fine-tuning value and repeat the matching process.

[0094] If an abnormal time period is matched, adjustments are made based on the control expression.

[0095] If a stable period is found, monitoring will continue.

[0096] Example 2 A high-frequency ventilation anesthesia control system, such as Figure 2 As shown, it includes: Data processing unit: Acquires clinical case information of high-frequency ventilation anesthesia, analyzes and extracts clinical case data for each case, and preprocesses the clinical case data to obtain sample data.

[0097] Initial parameter determination unit: The sample data is filtered and trained using an improved LSTM to obtain the initial expression of the ventilation parameters.

[0098] Intraoperative Relationship Determination Unit: Acquire intraoperative monitoring data from the sample data, establish a monitoring curve with time as the horizontal axis and intraoperative monitoring data as the vertical axis, extract features from the monitoring curve, and establish a correlation with the intraoperative control data.

[0099] Intraoperative control determination unit: Based on the initial expression of ventilation parameters with correlation, the control factors are configured and adjusted to generate the control expression.

[0100] Control execution unit: controls the high-frequency ventilation anesthesia process based on the initial expression and the control expression.

[0101] Example 3 An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor invokes the computer program in the memory to implement the content of a high-frequency ventilation anesthesia control method.

[0102] Example 4 A storage medium storing computer-executable instructions, which, when loaded and executed by a processor, implement the content of a high-frequency ventilation anesthesia control method.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical methods of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical methods of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical methods to deviate from the spirit and scope of the technical methods of the present invention.

Claims

1. A method of controlling high frequency ventilation anesthesia, characterized by, Includes the following steps: S1. Obtain clinical case information of high-frequency ventilation anesthesia, analyze and extract clinical case data for each case, and preprocess the clinical case data to obtain sample data. S2. An improved LSTM is used to filter and train the sample data to obtain the initial expression for the ventilation parameters; S3. Obtain intraoperative monitoring data from the sample data, establish a monitoring curve with time as the horizontal axis and intraoperative monitoring data as the vertical axis, extract features from the monitoring curve and establish a correlation with the intraoperative control data; S4. Based on the correlation relationship, configure the correlation factor for the initial expression of the ventilation parameter to generate the control expression; S5. The high-frequency ventilation anesthesia process is controlled based on the initial expression and the regulation expression.

2. The method of claim 1, wherein, The specific content of the sample data obtained by preprocessing clinical case data in S1 includes: The clinical case data includes initial preoperative characteristic data, initial ventilation parameters, intraoperative monitoring data, intraoperative adjustment data, and effect evaluation; Multivariate correlation analysis was performed on the initial preoperative characteristic data and initial ventilation parameters, and relevant preoperative characteristic data were obtained by screening. Relevant preoperative characteristic data include: Individual trait parameters: age F1, weight F2, forced expiratory volume in one second F3; Surgical type parameters: Surgical site F4, Stimulation intensity F5; First BIS index: F6; The remaining data in the relevant preoperative characteristic data and clinical case data constitute pure clinical case data; Based on the surgical type parameter, the pure clinical case data were divided into surgical category groups; The effectiveness evaluation is coded and converted to obtain the effectiveness evaluation value. Individual cases in the surgical category group are sorted from high to low according to the effectiveness evaluation value. The top 80% of cases were selected as sample data. The sample data includes relevant preoperative characteristic data, initial ventilation parameters, intraoperative monitoring data, and intraoperative control data under different surgical category groups.

3. The method of claim 2, wherein, The improved LSTM described in S2 filters the sample data to obtain relevant preoperative feature data and initial ventilation parameters, and uses the relevant preoperative feature data and initial ventilation parameters as input data and output data, respectively. Initial ventilation parameters include high-frequency ventilation respiratory rate, tidal volume, and remifentanil infusion rate; The improved LSTM includes an input layer, a feature enhancement layer, a gated LSTM hidden layer, a multi-target attention fully connected layer, and an output layer. Initial ventilation parameters include high-frequency ventilation respiratory rate Z1, tidal volume Z2, and remifentanil infusion rate Z3; The newly added clinical safety constraint branch in the LSTM hidden layer based on the gating mechanism ensures that the output temporal features conform to the physiological safety range. The expression is as follows: ； ； wherein, is the improved input gate output value, controlling the input weight of the current feature, is the sigmoid activation function, ; is an input gate weight matrix, is a time sequence feature output by the last time step LSTM, is a splicing of the output of the last time step and the current enhanced feature, is an input gate bias, is an element-wise multiplication, is a clinical constraint weight; , For constraint branch weight and bias, For constraint strength coefficient, strengthen security constraint strength, For the clinical safety range vector of core features.

4. The high-frequency ventilation anesthesia control method according to claim 3, characterized in that, To address the collaborative needs of the three outputs, the multi-target attention fully connected layer designs a dynamic attention weight allocation mechanism to enhance the influence of key features on the core control objective. ； ； in, The attention weights for the k-th control objective, i.e., the output. Using the cosine similarity function, we calculate the correlation strength between the fully connected layer features and the control target reference vector. The basic feature vector output by the fully connected layer. Let be the reference vector for the k-th control target. Let be the attention enhancement coefficient for the k-th control target. This is the weighted feature vector of the fully connected layer, used for prediction in the subsequent output layer.

5. The high-frequency ventilation anesthesia control method according to claim 3, characterized in that, The initial expressions for the ventilation parameters are: Initial respiratory rate: ; Initial tidal volume: ; Initial remifentanil rate: .

6. The high-frequency ventilation anesthesia control method according to claim 4, characterized in that, In S4, intraoperative monitoring data from the sample data is obtained. A monitoring curve is constructed with time as the horizontal axis and intraoperative monitoring data as the vertical axis. The specific details of feature extraction from the monitoring curve and establishing its correlation with intraoperative control data are as follows: The intraoperative monitoring data includes the second BIS index. End-tidal carbon dioxide level ; Intraoperative control data include respiratory rate control values, tidal volume control values, and remifentanil rate control values; A monitoring curve was established with time as the horizontal axis and the second BIS index and end-tidal carbon dioxide value as the vertical axes. Several abnormal time periods were obtained by marking the monitoring curves based on the time periods of intraoperative control data; A pre-set warning period is provided, which is a stable period and is set before the abnormal period. The remaining period is defined as a stable period. A standard monitoring curve is obtained by migrating and fitting data during the stable periods under the same curve. Early warning features are obtained by extracting features from the early warning period, and abnormal features are obtained by extracting features from the abnormal period. The correlation between abnormal features and intraoperative control data is obtained by training the data, and then a correlation factor is generated.

7. The high-frequency ventilation anesthesia control method according to claim 4, characterized in that, Based on the correlation relationships, the initial expression for ventilation parameters is configured with correlation factors, thereby generating the regulation expression: ； ； ； in, This represents the fitted real-time respiratory rate. To improve the initial respiratory rate output by the LSTM model, for right The positive correlation coefficient of the correlation factors The power-order fitting factor. This refers to end-tidal CO2 deviation. The sign function ensures that the adjustment direction is consistent with the deviation direction. This represents the clinically safe upper limit for respiratory rate. This represents the real-time tidal volume after fitting. To improve the initial tidal volume output of the LSTM model, These are the fitting coefficients for the deviation interval. These are the fitting coefficients for the large deviation interval; This is the deviation threshold; The fitted real-time remifentanil infusion rate; To improve the initial remifentanil rate output by the LSTM model, for right The negative correlation fitting coefficient, BIS bias, It is an exponential function, enabling rapid adjustment of the anesthetic drug rate; , The upper / lower limits of clinical safety for remifentanil rate; , This is a safety constraint.

8. The high-frequency ventilation anesthesia control method according to claim 7, characterized in that, The specific details of controlling the high-frequency ventilation anesthesia process based on the initial expression and the regulation expression in S6 include: Initial stage: High-frequency ventilation anesthesia input is directly based on the initial expression; Process stage: Preset fine-tuning values ​​are used to generate detection curves in real time during the detection process and match them during stable periods, warning periods, and abnormal periods; No adjustments are made during periods of stable testing. When the detection occurs within the warning period, fine-tuning is performed using the correlation between the fine-tuning value and the parameter. If the subsequent curve matches the stable period, the fine-tuning is stopped. If the subsequent curve does not match the stable period, continue to fine-tune by multiples of the fine-tuning value and repeat the matching process. If an abnormal time period is matched, adjustments are made based on the control expression; If a stable period is found, monitoring will continue.

9. A high-frequency ventilation anesthesia control system, characterized in that, include: Data processing unit: acquires clinical case information of high-frequency ventilation anesthesia, analyzes and extracts clinical case data for each case, and preprocesses the clinical case data to obtain sample data; Initial parameter determination unit: The sample data is filtered and trained using an improved LSTM to obtain the initial expression for the ventilation parameters; Intraoperative Relationship Determination Unit: Acquire intraoperative monitoring data from the sample data, establish a monitoring curve with time as the horizontal axis and intraoperative monitoring data as the vertical axis, extract features from the monitoring curve and establish a correlation with the intraoperative control data; Intraoperative control determination unit: Based on the initial expression of ventilation parameters with correlation, the control factors are configured and adjusted to generate the control expression; Control execution unit: controls the high-frequency ventilation anesthesia process based on the initial expression and the control expression.

10. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor, when calling the computer program in the memory, implements the content of the high-frequency ventilation anesthesia control method as described in any one of claims 1 to 8.

Images