Bronchoscopic surgery sedation anesthetic dosage prediction method
Through the integrated learning model combined with linear fitting model and neural network model, the patient's preoperative, intraoperative and postoperative sign information is used to solve the problem that traditional anesthetic dose prediction depends on clinical experience, achieving a more accurate and stable anesthetic dose Predicted that the safety of anesthesia in bronchoscopy is ensured.
Patent Information
- Application Number
- CN202411435763.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-15
- Publication Date
- 2025-02-14
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The determination of traditional anesthetic dose depends on the clinical experience of the doctor and is easily subjectively affected, resulting in inaccurate dose control and poor anesthesia effect in patients. Especially in bronchoscopy, insufficient anesthetic dose can easily cause choking and airway stress responses. Excessiveness will lead to the risk of hypoxia and adverse reactions.
Using an integrated learning model, including a linear fit model, a first neural network model and a second neural network model, a strong learner is constructed to predict and optimize anesthesia dose by obtaining the patient's preoperative, intraoperative and postoperative sign information. .
Through the use of integrated learning models, the accuracy of anesthetic dose prediction can be improved, errors in individual models can be reduced, the generalization ability and robustness of the model can be enhanced, and the stability and safety of the anesthetic effect can be ensured.
Smart Images
Figure CN119446401A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of surgical anesthesia, and in particular to a method for predicting sedation anesthesia dosage for bronchoscopic surgery. Background Art
[0002] The precise dosage of anesthetic drugs is the decisive factor in ensuring anesthesia safety and is also the focus of the medical community during the implementation of clinical anesthesia. In order to do a good job in clinical anesthesia, medical staff must carry out pre-anesthesia preparation, work during anesthesia, and post-anesthesia work in sequence when performing each anesthesia.
[0003] The main work before anesthesia is to understand the condition, make a correct estimate, determine the anesthesia plan based on the condition, and choose the most appropriate anesthetic dose. The work during anesthesia is to perform anesthesia according to the anesthesia operating procedures to achieve the best anesthesia effect, so that the patient can complete the operation without pain, quietness, memory, or adverse reactions. At the same time, create good conditions for the operation and try to meet the special requirements of certain operations, such as muscle relaxation, hypothermia, and hypotension. Post-anesthesia work refers to sending the patient back to the ward after surgery, and doing various post-anesthesia treatments according to different conditions and operations, including continuous monitoring of the patient, preventing complications, and handling accidents in a timely manner. Sometimes the patient still feels pain and discomfort after surgery and needs additional anesthesia.
[0004] The determination of traditional anesthetic doses relies on the doctor's clinical experience. This manual method of adjusting the anesthetic dose is easily affected by the doctor's subjective influence, and often results in inaccurate dose control and poor patient anesthesia effects, which can easily lead to a series of problems before, during, and after surgery. In the field of painless endoscopy, especially compared with other painless endoscopy examinations, bronchoscopic diagnosis and treatment can easily cause severe choking and airway stress reactions due to excessive airway stimulation, while excessive anesthetic doses can cause the risk of hypoxia and adverse reactions after surgery. Therefore, sedation anesthesia for bronchoscopy is obviously unique, and there are many factors that affect its sedation effect, making it difficult to predict doses through traditional clinical experience-based methods. Summary of the invention
[0005] The technical problem to be solved by the present invention is to provide a method for predicting the sedation and anesthesia dosage for bronchoscopic surgery, which can effectively predict the sedation and anesthesia dosage for bronchoscopic surgery and assist anesthesiologists in performing anesthesia.
[0006] The present invention provides a method for predicting sedation and anesthesia dosage for bronchoscopic surgery, comprising the following steps:
[0007] S1, obtaining the preoperative physical sign information of each patient in previous bronchoscopic surgery and dividing it into a first training set and a first validation set, obtaining the intraoperative physical sign information of each patient and dividing it into a second training set and a second validation set, obtaining the postoperative physical sign information of each patient and dividing it into a third training set and a third validation set;
[0008] S2. construct an integrated learning model, wherein the weak learners in the integrated learning model include a linear fitting model, a first neural network model, and a second neural network model;
[0009] S3, training the linear fitting model through the first training set, and obtaining the preoperative sedation and anesthetic prediction measurement training set through the trained linear fitting model; training the first neural network model through the second training set and the preoperative sedation and anesthetic prediction measurement training set, and obtaining the intraoperative sedation and anesthetic prediction measurement training set through the trained first neural network model; training the second neural network model through the third training set and the intraoperative sedation and anesthetic prediction measurement training set, and obtaining the postoperative sedation and anesthetic prediction measurement training set through the trained second neural network model;
[0010] S4, constructing a strong learner based on the preoperative sedation and anesthetic prediction amount training set, the intraoperative sedation and anesthetic prediction amount training set, and the postoperative sedation and anesthetic prediction amount training set according to the integration strategy;
[0011] S5, respectively optimizing the linear fitting model, the first neural network model, and the second neural network model through the first validation set, the second validation set, and the third validation set, thereby obtaining an optimized strong learner;
[0012] S6. Predict the sedation and anesthesia dose for bronchoscopic surgery of the current patient through the optimized strong learner.
[0013] Compared with the prior art, the present application has the following advantages: the results of multiple weak learners can be integrated through the integrated learning model, which can reduce the errors of individual weak learners and thus improve the overall prediction accuracy. Each weak learner may perform well in different aspects. Through integration, the advantages of each weak learner can be fully utilized to improve the overall performance, reduce the risk of overfitting, and improve the generalization ability of the model. Even in the case of limited training data, the prediction ability of new data can be improved by integrating different weak learners, and the robustness to noise and abnormal data can be improved. When a weak learner is affected by abnormal data, other weak learners can compensate for it, thereby reducing the overall error and being more stable, so that the sedation and anesthesia dose for bronchoscopic surgery can be effectively predicted to assist anesthesiologists in anesthesia.
[0014] In a possible implementation, the preoperative physical sign information includes age, gender, height, weight, underlying diseases, and preoperative adverse reactions.
[0015] Compared with the prior art, the above-mentioned preoperative physical sign information is used as the necessary physical sign information for preoperative anesthesia, which can effectively reflect the patient's preoperative condition.
[0016] In a possible implementation, the intraoperative vital sign information includes systolic blood pressure, diastolic blood pressure, heart rate, pulse oxygen saturation, choking VAS score, sedation score, and hypoxia risk in various time periods.
[0017] Compared with the prior art, the above intraoperative vital sign information is used as the necessary vital sign information for intraoperative anesthesia, which can effectively reflect the intraoperative condition of the patient.
[0018] In a possible implementation, the postoperative physical sign information includes a comfort score and postoperative adverse reactions.
[0019] Compared with the prior art, the above-mentioned postoperative vital sign information is used as necessary information for postoperative anesthesia, which can effectively reflect the patient's postoperative condition.
[0020] In a possible implementation, the linear fitting model includes the following formula:
[0021]
[0022] Among them, L(y i ,γ) represents the loss function; y i represents the predicted amount of preoperative sedative anesthetics; γ represents the actual amount of preoperative sedative anesthetics.
[0023] Compared with the prior art, the above technical solution can fit the preoperative vital sign information so as to predict the anesthetic dose.
[0024] In a possible implementation, the hidden layers in the first neural network model and the second neural network model use the following activation function:
[0025]
[0026] Among them, σ(x) represents the sigmoid function, and x represents the function input value.
[0027] Compared with the prior art, the activation function using the above technical solution is continuously differentiable over the entire real number domain, which enables better gradient calculation during back-propagation, making network training more stable and alleviating gradient vanishing. The shape of the response curve can be automatically adjusted within different input ranges to adapt to different data distributions and model requirements without significantly increasing the computational burden.
[0028] In a possible implementation, the optimization of step S5 includes the following steps:
[0029] S51, calculating the residual between the first validation set and the predicted value of the linear fitting model, calculating the residual between the second validation set and the first neural network model, and calculating the residual between the third validation set and the second neural network model, the residual is calculated by the following formula:
[0030]
[0031] S52, fit the residual by the following formula:
[0032]
[0033] when
[0034] Among them, h(x i ) represents the i-th regression tree model;
[0035] S53, the optimization is completed by the following formula:
[0036]
[0037] Among them, v represents the learning rate.
[0038] Compared with the prior art, the above technical solution can effectively optimize the weak learner by analyzing the residuals, and ultimately make the strong learner more accurate. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION
[0040] First, those skilled in the art should understand that these implementations are only used to explain the technical principles of the embodiments of the present application, and are not intended to limit the protection scope of the embodiments of the present application. Those skilled in the art can make adjustments to them as needed to adapt to specific application scenarios.
[0041] The present application embodiment discloses a method for predicting sedation and anesthesia dosage for bronchoscopic surgery, comprising the following steps:
[0042] S1, obtaining the preoperative physical sign information of each patient in previous bronchoscopic surgery and dividing it into a first training set and a first validation set, obtaining the intraoperative physical sign information of each patient and dividing it into a second training set and a second validation set, obtaining the postoperative physical sign information of each patient and dividing it into a third training set and a third validation set;
[0043] S2. construct an integrated learning model, wherein the weak learners in the integrated learning model include a linear fitting model, a first neural network model, and a second neural network model;
[0044] S3, training the linear fitting model through the first training set, and obtaining the preoperative sedation and anesthetic prediction measurement training set through the trained linear fitting model; training the first neural network model through the second training set and the preoperative sedation and anesthetic prediction measurement training set, and obtaining the intraoperative sedation and anesthetic prediction measurement training set through the trained first neural network model; training the second neural network model through the third training set and the intraoperative sedation and anesthetic prediction measurement training set, and obtaining the postoperative sedation and anesthetic prediction measurement training set through the trained second neural network model;
[0045] S4, constructing a strong learner based on the preoperative sedation and anesthetic prediction amount training set, the intraoperative sedation and anesthetic prediction amount training set, and the postoperative sedation and anesthetic prediction amount training set according to the integration strategy;
[0046] S5, respectively optimizing the linear fitting model, the first neural network model, and the second neural network model through the first validation set, the second validation set, and the third validation set, thereby obtaining an optimized strong learner;
[0047] S6. Predict the sedation and anesthesia dose for bronchoscopic surgery of the current patient through the optimized strong learner.
[0048] Specifically, the integrated learning model includes three weak learners, which are a linear fitting model, a first neural network model, and a second neural network model. Preoperative physical sign information includes age, gender, height, weight, underlying diseases, and preoperative adverse reactions. Among them, underlying diseases include hypertension, diabetes, coronary heart disease, chronic lung disease, Parkinson's disease, depression, heart failure, liver and kidney failure, history of cerebral infarction, history of pulmonary infarction, history of respiratory failure, and history of heart failure. Preoperative adverse reactions include severe insomnia, tension, anxiety, dizziness, and allergies. These are the main conditions that affect the anesthesia dose for bronchoscopic surgery. Those skilled in the art can flexibly adjust the addition of underlying diseases and preoperative adverse reactions according to actual conditions to effectively reflect the patient's preoperative conditions so that the model is more accurate.
[0049] Intraoperative vital signs include systolic blood pressure, diastolic blood pressure, heart rate, pulse oxygen saturation, cough VAS score, sedation score, and hypoxia risk in each time period. Hypoxia risk is the incidence of pulse oxygen saturation below 90%. The above intraoperative vital signs are collected every five minutes during the operation to effectively reflect the patient's intraoperative condition.
[0050] Postoperative physical sign information includes comfort scores and postoperative adverse reactions. Postoperative adverse reactions include painful memories, drowsiness, fatigue, headache and dizziness, nausea and vomiting, and fatigue. These are the main conditions that affect postoperative adverse reactions. Those skilled in the art can flexibly adjust and add postoperative adverse reactions according to actual conditions to effectively reflect the patient's postoperative conditions, so as to make the model more accurate.
[0051] The linear fitting model includes the following formula:
[0052]
[0053] Among them, L(y i ,γ) represents the loss function; y i represents the predicted amount of preoperative sedative anesthetics; γ represents the actual amount of preoperative sedative anesthetics.
[0054] The above loss function can use mean square error, including the following formula:
[0055]
[0056] The mean square error loss function can fit the preoperative vital signs information to predict the anesthetic dose.
[0057] The hidden layers in the first neural network model and the second neural network model use the following activation function:
[0058]
[0059] Among them, σ(x) represents the sigmoid function, and x represents the function input value.
[0060] The first neural network model and the second neural network model both include an input layer, an output layer, two hidden layers, and corresponding fully connected layers. The swish activation function is continuously differentiable over the entire real number domain, which enables better gradient calculation during back propagation, making network training more stable, alleviating gradient vanishing, and automatically adjusting the shape of the response curve within different input ranges to adapt to different data distributions and model requirements without significantly increasing the computational burden.
[0061] The optimization of step S5 includes the following steps:
[0062] S51, calculating the residual between the first validation set and the predicted value of the linear fitting model, calculating the residual between the second validation set and the first neural network model, and calculating the residual between the third validation set and the second neural network model, the residual is calculated by the following formula:
[0063]
[0064] S52, fit the residual by the following formula:
[0065]
[0066] when
[0067] Among them, h(x i ) represents the i-th regression tree model;
[0068] S53, the optimization is completed by the following formula:
[0069]
[0070] Where v represents the learning rate. The above can effectively optimize the weak learner by analyzing the residual, and ultimately make the strong learner more accurate. The above is also the iterative optimization using the gradient boosting algorithm.
[0071] This application can integrate the results of multiple weak learners through an integrated learning model, which can reduce the errors of individual weak learners and thus improve the overall prediction accuracy. Each weak learner may perform well in different aspects. Through integration, the advantages of each weak learner can be fully utilized to improve the overall performance, reduce the risk of overfitting, and improve the generalization ability of the model. Even in the case of limited training data, different weak learners can be integrated to improve the prediction ability of new data and improve the robustness to noise and abnormal data. When a weak learner is affected by abnormal data, other weak learners can compensate for it, thereby reducing the overall error and being more stable, so that the sedation and anesthesia dose for bronchoscopic surgery can be effectively predicted to assist anesthesiologists in anesthesia.
[0072] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "in the present embodiment", "specific example", or "some examples" etc. means that the specific features, mechanisms, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, mechanisms, materials or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.
[0073] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.
Claims
1. A method for predicting sedation and anesthesia dose for bronchoscopic surgery, characterized in that: The following steps are involved: S1, obtaining the preoperative physical sign information of each patient in previous bronchoscopic surgery and dividing it into a first training set and a first validation set, obtaining the intraoperative physical sign information of each patient and dividing it into a second training set and a second validation set, obtaining the postoperative physical sign information of each patient and dividing it into a third training set and a third validation set; S2. Constructing an integrated learning model, wherein the weak learners in the integrated learning model include a linear fitting model, a first neural network model, and a second neural network model; S3, training the linear fitting model through the first training set, and obtaining a preoperative sedation and anesthetic prediction measurement training set through the trained linear fitting model; training the first neural network model through the second training set and the preoperative sedation and anesthetic prediction measurement training set, and obtaining an intraoperative sedation and anesthetic prediction measurement training set through the trained first neural network model; training the second neural network model through the third training set and the intraoperative sedation and anesthetic prediction measurement training set, and obtaining a postoperative sedation and anesthetic prediction measurement training set through the trained second neural network model; S4, constructing a strong learner based on the preoperative sedation and anesthetic prediction amount training set, the intraoperative sedation and anesthetic prediction amount training set, and the postoperative sedation and anesthetic prediction amount training set according to an integrated strategy; S5, respectively optimizing the linear fitting model, the first neural network model and the second neural network model through the first validation set, the second validation set and the third validation set, thereby obtaining the optimized strong learner; S6. Predicting the bronchoscopic surgery sedation and anesthesia dosage of the current patient by using the optimized strong learner.
2. The method for predicting sedation and anesthesia dosage for bronchoscopic surgery according to claim 1, characterized in that: The preoperative physical sign information includes age, gender, height, weight, underlying diseases and preoperative adverse reactions.
3. The method for predicting sedation and anesthesia dosage for bronchoscopic surgery according to claim 2, characterized in that: The intraoperative vital signs information includes systolic blood pressure, diastolic blood pressure, heart rate, pulse oxygen saturation, choking VAS score, sedation score and hypoxia risk in various time periods.
4. The method for predicting sedation and anesthesia dosage for bronchoscopic surgery according to claim 3, characterized in that: The postoperative physical sign information includes comfort score and postoperative adverse reactions.
5. The method for predicting sedation and anesthesia dosage for bronchoscopic surgery according to claim 4, characterized in that: The linear fitting model includes the following formula: Among them, L(y i ,γ) represents the loss function; y i represents the predicted amount of preoperative sedative anesthetics; γ represents the actual amount of preoperative sedative anesthetics.
6. The method for predicting sedation and anesthesia dosage for bronchoscopic surgery according to claim 5, characterized in that: The hidden layers in the first neural network model and the second neural network model use the following activation function: Among them, σ(x) represents the sigmoid function, and x represents the function input value.
7. The method for predicting sedation and anesthesia dosage for bronchoscopic surgery according to claim 6, characterized in that: The optimization of step S5 includes the following steps: S51, calculating the residual between the first validation set and the predicted value of the linear fitting model, calculating the residual between the second validation set and the first neural network model, and calculating the residual between the third validation set and the second neural network model, wherein the residual is calculated by the following formula: Wherein, m = 1, 2, 3...M; S52, fitting the residual by the following formula: Among them, h(x i ) represents the i-th regression tree model; S53, the optimization is completed by the following formula: Among them, v represents the learning rate.