Gastrointestinal endoscopy anesthesia medication model training method, device and equipment and storage medium

Abstract

The embodiment of the invention relates to the technical field of anesthesia complication prevention, and discloses a gastrointestinal endoscopy anesthesia medicine model training method, device and equipment and a storage medium. The trained gastrointestinal endoscopy anesthesia medication model can automatically obtain related operation data according to the vital sign data of the patient, so that an anesthesia operation can be performed on the patient, the labor cost of painless gastrointestinal endoscopy can be reduced, the risk of anesthesia operation errors caused by human errors can be reduced, and the patient experience is improved. And convenience is provided for patients performing painless gastrointestinal endoscopy.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of anesthesia technology, and specifically to a method, device, equipment and storage medium for training an anesthesia medication model for gastrointestinal endoscopy. Background Art

[0002] Painless gastroenteroscopy is a medical examination method that helps doctors diagnose digestive system diseases. The difference between it and traditional gastroenteroscopy is that it reduces the patient's discomfort and pain during the operation by using sedatives or anesthetics, thereby improving the acceptability of the examination. In order to ensure the patient's intraoperative anesthesia safety and reduce the potential risk of postoperative complications, it is crucial to conduct a preoperative anesthesia evaluation for patients who are about to undergo painless gastroenteroscopy.

[0003] In the common gastrointestinal endoscopy surgery nowadays, in order to reduce the side effects of anesthesia and improve the patient's comfort, patients often choose painless diagnosis and treatment when undergoing gastrointestinal endoscopy. Painless gastrointestinal endoscopy requires intravenous injection of propofol for general anesthesia. This method requires experienced anesthesiologists to monitor the operation throughout the process. The anesthesia effect is highly linked to the experience level of the anesthesiologist. However, the number of experienced anesthesiologists in each hospital is limited, which leads to higher labor costs for the implementation of painless gastrointestinal endoscopy, which limits the application of painless gastrointestinal endoscopy.

[0004] At the same time, painless gastroenteroscopy performed under the full monitoring of anesthesiologists is also prone to the risk of incorrect drug infusion rate due to human misjudgment, which is not conducive to the safety and stability of the operation. Summary of the Invention

[0005] In view of the above problems, the embodiments of the present application provide a gastrointestinal endoscopy anesthesia medication model training method, device, equipment and storage medium, which are used to solve the problem that when existing clinicians perform continuous anesthesia on patients, they need experienced doctors to monitor and control the drug injection throughout the process, resulting in high human resource costs.

[0006] According to one aspect of an embodiment of the present application, a method for training a gastrointestinal endoscopy anesthesia medication model is provided, including: acquiring the vital signs data and surgical data of a target patient in real time; acquiring standard vital signs data, wherein the standard vital signs data reflects the normal vital signs of humans during anesthesia; judging whether the vital signs data are consistent with the standard vital signs data; if so, inputting the vital signs data and the surgical data together as training data into a learning model to obtain a gastrointestinal endoscopy anesthesia medication model, wherein the gastrointestinal endoscopy anesthesia medication model is used to output surgical data that can maintain the vital signs data of clinical patients consistent with the standard vital signs data.

[0007] In an optional manner, the real-time acquisition of the target patient's vital signs data and surgical data includes: real-time acquisition of the target patient's heart rate, blood pressure, and blood oxygen data as the vital signs data; real-time acquisition of the target patient's drug infusion dose and endoscopic operation signal data as the surgical data.

[0008] In an optional manner, after obtaining the standard vital signs data, it also includes: obtaining the patient's anesthesia depth BIS value; constructing a reward model, comparing the patient's anesthesia depth BIS value with the preset anesthesia depth BIS value, and when the patient's anesthesia depth BIS value does not match the preset anesthesia depth BIS value, deducting a preset score; after inputting the vital signs data and the surgical data into the learning model as training data to obtain a gastrointestinal endoscopy anesthesia medication model, it also includes: adjusting the parameters of the gastrointestinal endoscopy anesthesia medication model based on the reward model and the score.

[0009] In an optional embodiment, the preset anesthesia depth BIS value is 60.

[0010] In an optional manner, after inputting the vital signs data and the surgical data together as training data into the learning model to obtain a gastrointestinal endoscopy anesthesia medication model, the method further includes: obtaining the vital signs data of a clinical patient; inputting the vital signs data of the clinical patient into the gastrointestinal endoscopy anesthesia medication model to obtain clinical surgical data; anesthetizing the clinical patient based on the clinical surgical data; obtaining the anesthesia depth BIS value of the clinical patient after anesthesia; determining whether the anesthesia depth BIS value of the clinical patient is equal to 60; if so, continuing to train the gastrointestinal endoscopy anesthesia medication model using the vital signs data and the clinical surgical data together as training data; if not, jumping to the step of obtaining the vital signs data of the clinical patient.

[0011] In one optional embodiment, the surgical data includes an anesthetic drug infusion rate.

[0012] According to another aspect of an embodiment of the present application, a gastrointestinal endoscopy anesthesia medication model training device is provided, comprising: a first acquisition module, a second acquisition module, a first judgment module, and an input module. The first acquisition module is used to acquire the target patient's vital signs data and surgical data in real time; the second acquisition module is used to acquire standard vital signs data, wherein the standard vital signs data reflects the normal vital signs of humans during anesthesia; the first judgment module is used to determine whether the vital signs data is consistent with the standard vital signs data; the input module is used to input the vital signs data and the surgical data as training data into a learning model to obtain a gastrointestinal endoscopy anesthesia medication model, and the gastrointestinal endoscopy anesthesia medication model is used to output surgical data that can maintain the clinical patient's vital signs data consistent with the standard vital signs data.

[0013] According to another aspect of an embodiment of the present application, a gastrointestinal endoscopy anesthesia medication model training device is provided, including: a processor, a memory, a communication interface and a communication bus, the processor, the memory and the communication interface communicate with each other through the communication bus; the memory is used to store at least one program, and the program enables the processor to execute the operation of any one of the above gastrointestinal endoscopy anesthesia medication model training methods.

[0014] According to another aspect of an embodiment of the present application, a computer-readable storage medium is provided, in which executable instructions are stored. The executable instructions enable a gastrointestinal endoscopy anesthesia medication model training device to perform the operations of any one of the above gastrointestinal endoscopy anesthesia medication model training methods.

[0015] By training the anesthesia medication model for gastrointestinal endoscopy, the trained anesthesia medication model for gastrointestinal endoscopy can automatically obtain relevant surgical data based on the patient's vital signs data, so as to perform anesthesia surgery on the patient. This can reduce the labor cost of painless gastrointestinal endoscopy and the risk of anesthesia surgery errors due to human error, providing convenience for patients undergoing painless gastrointestinal endoscopy.

[0016] The above description is only an overview of the technical solutions of the embodiments of the present application. In order to more clearly understand the technical means of the embodiments of the present application, they can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the embodiments of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A structural schematic diagram of an anesthesia administration data preprocessing device is provided for this application. DETAILED DESCRIPTION

[0018] The exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0019] With the continuous progress of society, people's demand for medical quality is getting higher and higher. In gastrointestinal endoscopic surgery, more and more people choose painless gastrointestinal endoscopy.

[0020] The implementation of painless gastroenteroscopy requires anesthesia of the patient, and different patients have different tolerances to anesthetics. The doctor who performs anesthesia needs to monitor the patient's various data at all times and adjust the injection dose of anesthetics in real time according to the patient's data to ensure that the patient's anesthesia can be carried out in real time and stably during the operation, avoiding anesthesia complications in the patient and allowing the gastroenteroscopy operation to proceed smoothly.

[0021] For painless gastroenteroscopy, commonly used drugs include propofol and fentanyl. These drugs need to be injected into the patient's body through intravenous injection under the supervision of a professional anesthesiologist to put the patient into an anesthetized state. However, the dosage of anesthetic drugs required for patients with different physical conditions to enter the optimal anesthesia state is also different. Doctors need to monitor the patient's vital signs data in real time during the infusion process to determine the drug injection speed in order to adapt to the physical conditions of different patients and provide the most appropriate anesthesia rate.

[0022] However, it is common now that doctors need to collect various aspects of patient data, analyze multiple data, and ultimately rely on experience to judge the possible anesthetic complications of the patient and impose corresponding preventive measures. The judgment process is relatively complicated and requires high experience and technical skills of the doctor. When encountering patients with special physical conditions, doctors often find it difficult to accurately judge the type of anesthetic complications that the patient may have, resulting in hidden dangers in the safety of anesthesia surgery.

[0023] The inventors of this application have come up with the idea that the model can be trained by combining the clinical medication data of anesthetic drugs for gastrointestinal endoscopy with the physiological data of the patient as training data to obtain a gastrointestinal endoscopy anesthetic medication model with high accuracy and good output results, thereby freeing up doctor resources and making it possible for patients to undergo painless gastrointestinal endoscopy without the need for experienced doctors to monitor the entire process. Instead, it can be performed by medical staff such as nurses who have mastered the basic operating methods, thereby reducing the difficulty of surgical anesthesia operations and improving patient comfort and safety.

[0024] Based on this, the inventor has designed a method for training an anesthesia medication model for gastrointestinal endoscopy after research. By training the anesthesia medication model for gastrointestinal endoscopy, the trained anesthesia medication model for gastrointestinal endoscopy can automatically obtain relevant surgical data based on the patient's vital signs data, so as to perform anesthesia surgery on the patient. This can reduce the labor cost of painless gastrointestinal endoscopy, and can also reduce the risk of anesthesia surgery errors due to human errors, providing convenience for patients undergoing painless gastrointestinal endoscopy.

[0025] In some embodiments, the present application provides a gastrointestinal endoscopy anesthesia medication model training device, the gastrointestinal endoscopy anesthesia medication model training device includes: a first acquisition module, a second acquisition module, a first judgment module and an input module. The first acquisition module is used to acquire the vital signs data and surgical data of the target patient in real time; the second acquisition module is used to acquire standard vital signs data, wherein the standard vital signs data reflects the normal vital signs of humans during anesthesia; the first judgment module is used to judge whether the vital signs data is consistent with the standard vital signs data; the input module is used to input the vital signs data and the surgical data as training data into the learning model to obtain a gastrointestinal endoscopy anesthesia medication model, and the gastrointestinal endoscopy anesthesia medication model is used to output surgical data that can maintain the vital signs data of clinical patients consistent with the standard vital signs data.

[0026] In some feasible embodiments, the anesthesiologist's clinical actions, corresponding patient feedback, and special events are stored in a database to ensure data integrity and traceability. The anesthesiologist's clinical actions include, but are not limited to, the drug infusion rate; patient feedback includes pulse rate, pulse oximetry, BIS values, etc.; and special events include patient movement and jaw thrust. In this embodiment, data storage uses SQL for structured data storage.

[0027] Furthermore, through the powerful statistical analysis function of R language, the SQL raw data is cleaned, normalized and feature extracted to eliminate noise and outliers, providing a high-quality data set for the training of the anesthesia medication model for gastrointestinal endoscopy.

[0028] In some embodiments, the gastrointestinal endoscopy anesthesia medication model training device further includes a judgment module and a processing module. The judgment module is configured to determine whether the target patient has a history of anesthesia complications based on the medical history information; and the processing module is configured to include the target patient's history of anesthesia complications in the history of anesthesia complications information if the target patient has a history of anesthesia complications.

[0029] In some embodiments, the first acquisition module further includes a first real-time acquisition unit and a second real-time acquisition unit, the first real-time acquisition unit being used to acquire the heart rate, blood pressure and blood oxygen data of the target patient in real time as the vital signs data; the second real-time acquisition unit being used to acquire the drug infusion dose and endoscopic operation signal data of the target patient in real time as the surgical data.

[0030] In some embodiments, the gastrointestinal endoscopy anesthesia medication model training device further includes a third acquisition module, a reward model module, and a parameter adjustment module. The third acquisition module is used to obtain the patient's anesthesia depth BIS value; the reward model module is used to construct a reward model, compare the patient's anesthesia depth BIS value with a preset anesthesia depth BIS value, and deduct a preset score when the patient's anesthesia depth BIS value does not match the preset anesthesia depth BIS value; and the parameter adjustment module is used to adjust the parameters of the gastrointestinal endoscopy anesthesia medication model based on the reward model and the score.

[0031] The reward model adjusts the drug infusion rate based on the patient's anesthesia depth BIS value, and compares the patient's anesthesia depth BIS value with the preset anesthesia depth BIS value, and calculates the deviation between the two for immediate feedback. The greater the deviation, the more points will be deducted. The reward value calculation formula of the reward model is: R t =-λ·|BIS t -BIS target ∣ k . Among them: R t is the reward value at time step t, and negative values ​​indicate penalties; λ is the penalty coefficient, which controls the penalty intensity and is usually set to 1 to 5; BIS t BIS is the patient's anesthesia depth. target is the preset BIS value of anesthesia depth; k is an exponential parameter, and k = 2 is often used to strengthen the penalty of large deviations. For example, if BIS target =60, BIS t =50,λ=2,k=2, then R t =-2×(50-60) 2 =-200.

[0032] When the BIS value of the patient's anesthesia depth does not match the preset BIS value of anesthesia depth, the preset score is deducted. The deduction of the preset score is based on a comprehensive consideration of clinical risk classification and time cumulative effect. Specifically, it includes three aspects: deviation degree graded penalty, time dimension penalty enhancement, and composite indicator linkage penalty. Among them, the deviation degree graded penalty includes mild deviation, moderate deviation, and severe deviation. Time dimension penalty enhancement is to increase the deduction value according to the duration of the patient's anesthesia depth BIS value range. Optionally, if the patient's anesthesia depth BIS value is abnormal and accompanied by other abnormal physiological indicators, the deduction value is increased.

[0033] In this embodiment, the BIS value range for a patient's depth of anesthesia is mildly deviated from 55 to 59 or 61 to 65, with a deduction of 10 points per session; the BIS value range for a patient's depth of anesthesia is moderately deviated from 50 to 54 or 66 to 70, with a deduction of 30 points per session; and the BIS value range for a patient's depth of anesthesia is severe: greater than 70 or less than 50, with a deduction of 60 points per session. For every additional minute of duration outside the range, an additional 20% of the deduction is deducted. If the patient's BIS value is abnormal and accompanied by abnormalities in other physiological indicators, the deduction weight is increased by 50%. For example, if a patient's BIS value for depth of anesthesia is 57, the duration is 2 minutes, and the MAP is <65 mmHg, the deduction of the preset score is calculated as: (10 + 2 × 10 × 20%) × 150% = 21. The values ​​and weights for the deviation level classification and time dimension penalty enhancement in this embodiment are merely examples and can be set according to actual needs.

[0034] The parameter adjustment module is used to adjust the parameters of the gastrointestinal endoscopy anesthesia medication model based on the reward model and the score. That is, the parameter adjustment module is triggered when the reward value of the reward model is negative. Specifically, by constructing a real-time sedation depth curve and a pump speed curve, the phase difference between the first-order derivative functions of the two curves is calculated to determine the maximum probability lag time. According to the lag time and historical data, the response multiple of the sedation / analgesia depth to the pump speed is fitted, and the new pump speed is calculated based on the response function and the real-time deviation. The formula for calculating the new pump speed is: new pump speed = current pump speed + adjustment parameter, and the calculation formula for the adjustment parameter is: Example: If BIS target =60, BIS t =70, response multiple = 50, lag time = 90 seconds, then BIS value deviation = 70-60 = 10, 5.56ml / min.

[0035] In some feasible embodiments, a gastrointestinal endoscopy anesthesia medication model is trained through structured data fusion, offline reinforcement learning dataset construction, deep reinforcement learning framework, simulation testing platform construction, multi-agent fusion, and model iteration. The following describes each method in detail.

[0036] Structured data fusion constructs a state space by integrating multi-source heterogeneous data to comprehensively characterize the depth of anesthesia. Specifically, the patient's physiological data, drug infusion data, and basic information of the patient are integrated, and signals of different sampling rates are aggregated through a sliding window to generate a synchronous feature vector to eliminate timing deviations. Dynamic indicators are constructed through ΔBIS / Δt (to capture the delayed effect of drugs) and HR variability (SDNN, reflecting autonomic nervous stability); and the original EEG signal is decomposed into the power spectral density of the δ (0.5–4Hz), θ (4–8Hz), and α (8–13Hz) frequency bands to enhance the characterization of anesthesia depth and form frequency domain features. The state space dimension is set to st = [basic information, physiological parameters t, drug data t, Δ physiological parameters t-1:t ]Among them, the patient's physiological parameters include deep anesthesia BIS value, blood pressure, and heart rate.

[0037] Construct an offline reinforcement learning dataset and autonomously optimize the anesthesia depth control target, for example, the anesthesia depth BIS value is stable in the target range and the vital signs are stable. Set the reward value of the reinforcement learning model and build a Markov reinforcement learning environment. Specifically, set the reward function according to the anesthesia depth BIS value, the stability of vital signs, and the drug economy: t =w1·I BIS -w2|MAP-MAP base ∣-w3·D drug -w4·I event , where w1, w2, w3, and w4 are weight coefficients, which are positive values; I BIS , I event is an exponential function that converts discrete events into reward signals; D drug is the standardized cumulative dose of the drug; MAP, MAP base are mean arterial pressure and patient baseline blood pressure, respectively. BIS Indicates that when the BIS value of deep anesthesia is within the target range, w1 is rewarded, otherwise there is no reward; I event "w4" indicates a deduction if an adverse event occurs; otherwise, no deduction is made. Adverse events include, but are not limited to, hypoxemia, motor reactions, and the risk of deep sedation. A Markov reinforcement learning environment is established. Specifically, the state space integrates vital sign time series data with the drug concentration vector output by the PK / PD model, and the action space is defined as the continuous drug infusion rate. The transfer function simulates delayed effects through the PK / PD model, and the reward function integrates multi-objective optimization and safety event penalties. The current interaction round terminates when the surgery ends, a vital sign exceeds a threshold, or the maximum time step is reached, triggering a federated learning-compatible safety log.

[0038] The deep reinforcement learning framework trains nearly 30 current mainstream reinforcement learning model frameworks, such as the deep Q-network (DQN) and the proximal policy optimization (PPO) algorithm. The training process includes pre-training, offline training, online fine-tuning, and federated deployment. Specifically, pre-training imitates expert medication records through behavioral cloning (BC) and uses a hierarchical loss function to initialize the policy network. Offline training uses conservative Q-learning to add regularization terms to constrain Q-value estimation and prevent distribution shift. Online fine-tuning integrates PK / PD drug efficacy delay models through dynamic fusion output of multiple agents and state transfer. Federated deployment jointly trains policy networks across hospitals, differential privacy protects patient data, and lifelong learning copes with model degradation.

[0039] Optionally, training can be performed using the Deep Q-Network (DQN) reinforcement learning model framework. Specifically, multiple models are integrated through a three-layer structure: a base layer, where DQN provides experience replay and a target network foundation to ensure training stability; an extension layer, where improved algorithms such as Double / Dueling are implemented as plug-ins to enhance performance; and a system layer, where federated learning enables cross-scenario generalization and evolutionary algorithms dynamically optimize model combinations.

[0040] Optionally, training can be performed using the Proximal Policy Optimization (PPO) algorithm to reinforce the learning model framework. Specifically, the Proximal Policy Optimization (PPO) algorithm implements reinforcement learning training through four mechanisms: a collaborative core mechanism, GAE advantage estimation, synchronously clipping the objective function, and constraining the policy update amplitude; a refined network architecture, an actor-critic shared feature layer, policy network output adaptation to discrete / continuous action spaces, and orthogonal initialization to improve stability; engineering-level optimization, MPI parallel sampling to accelerate data collection, dynamic adjustment ∈ and learning rate decay to balance convergence efficiency; domain-specific expansion, embedded action masks to ensure medical safety, and federated learning to address data privacy issues.

[0041] A simulation test platform was built to test the output results of anesthesia induction and maintenance doses of different models. Based on clinical experience, the output results of different reward frameworks were evaluated to see if they were consistent with clinical logic. Appropriate model groups were selected to further tune the model hyperparameters to optimize model performance. Specifically, a simulation test platform was built by simulating population parameters and testing key scenarios. The simulated population can be divided into an elderly group, an obese group, etc. The key scenarios of the test include the induction phase and the maintenance phase. In this embodiment, the metabolic characteristics of the elderly group were simulated, with a 20% decrease in liver metabolic rate, a decrease in clearance rate, and a 50% increase in BIS response delay. Within 2 minutes, the BIS value dropped from 98 to 60-80 to enter the induction phase, the body movement response rate was less than 5%, and the respiratory depression events were less than 10% to enter the maintenance phase. Optionally, the learning rate, discount factor, and target network update coefficient were systematically tuned through network search and adaptive strategies, and parameters such as the experience replay buffer and batch size were combined for joint optimization.

[0042] Multi-agent fusion assigns trained and optimized agents to different clinical logic conditions and adjusts weight distribution coefficients, ensuring that the output of each action is the optimal solution for multiple agents, thereby increasing the robustness of the model's overall output. Specifically, multi-agent fusion includes three specialized agents: a sedation control agent responsible for BIS stability, whose weights dynamically adjust with BIS deviation; a safety monitoring agent responsible for intervening in SpO2 / MAP abnormalities, with LSTM prediction veto power; and a metabolic prediction agent, which estimates drug clearance based on PK / PD models and guides medication reduction decisions during the recovery period. The final action is output through adaptive state space partitioning and nonlinear weighted fusion, while also embedding action masks and a federated learning update mechanism to achieve coordinated optimization of safety and robustness.

[0043] Model iteration, through a trial-and-error mechanism based on reinforcement learning, allows the model to dynamically adjust drug infusion strategies to accommodate individual patient differences and dynamic changes during examinations. Dynamic optimization is achieved through online fine-tuning and trial-and-error mechanisms. Specifically, online fine-tuning collects new data under safe supervision, incrementally updates the policy network, and, through multi-center joint training, addresses data fragmentation issues for specific populations. The trial-and-error mechanism adds Gaussian noise ∈~N(0,σ) to the action output, with σ decaying during training. If three consecutive actions result in adverse events, the weight w4 in the reward function is automatically increased.

[0044] In some embodiments, the gastrointestinal endoscopy anesthesia medication model training device further includes a fourth acquisition module, a model input module, an application module, a fifth acquisition module, a second judgment module, a first processing module, and a second processing module. The fourth acquisition module is used to acquire the vital signs data of clinical patients; the model input module is used to input the vital signs data of clinical patients into the gastrointestinal endoscopy anesthesia medication model to obtain clinical operation data; the application module is used to perform anesthesia on the clinical patients based on the clinical operation data; the fifth acquisition module is used to acquire the anesthesia depth BIS value of the clinical patients after anesthesia; the second judgment module is used to determine whether the anesthesia depth BIS value of the clinical patients is equal to 60; the first processing module is used to use the vital signs data of the clinical patients and the clinical operation data as training data to continue training the gastrointestinal endoscopy anesthesia medication model when the anesthesia depth BIS value of the clinical patients is equal to 60; the second processing module is used to jump to execute the steps of the fourth acquisition module when the anesthesia depth BIS value of the clinical patients is not equal to 60.

[0045] According to another aspect of an embodiment of the present application, a device for pre-processing anesthesia administration data is provided, which may include: a processor 502 , a memory 506 , a communication interface 504 , and a communication bus 508 .

[0046] The processor 502, memory 506 and communication interface 504 communicate with each other via a communication bus 508. The memory 506 is used to store at least one program 510, which enables the processor 502 to execute the relevant steps in the above-mentioned gastrointestinal endoscopy anesthesia medication model training method embodiment.

[0047] Specifically, the program 510 may include program code including computer-executable instructions.

[0048] The processor 502 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. The one or more processors included in the anesthesia administration data preprocessing device may be processors of the same type, such as one or more CPUs, or may be processors of different types, such as one or more CPUs and one or more ASICs.

[0049] The memory 506 is used to store the program 510. The memory 506 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0050] Program 510 can be specifically called by processor 502 to enable the anesthesia administration data preprocessing device to perform the following operations: obtain the target patient's vital signs data and surgical data in real time; obtain standard vital signs data, wherein the standard vital signs data reflects the normal vital signs of humans during anesthesia; determine whether the vital signs data are consistent with the standard vital signs data; if so, input the vital signs data and the surgical data together as training data into the learning model to obtain a gastrointestinal endoscopy anesthesia medication model, and the gastrointestinal endoscopy anesthesia medication model is used to output surgical data that can maintain the clinical patient's vital signs data consistent with the standard vital signs data.

[0051] An embodiment of the present application further provides a computer-readable storage medium storing executable instructions. When the executable instructions are executed on an anesthesia administration data preprocessing device, the anesthesia administration data preprocessing device executes the gastrointestinal endoscopy anesthesia medication model training method in any of the above embodiments.

[0052] The executable instructions can be specifically used to enable the anesthesia administration data preprocessing device to perform the following operations: obtain the vital signs data and surgical data of the target patient in real time; obtain standard vital signs data, wherein the standard vital signs data reflects the normal vital signs of humans during anesthesia; determine whether the vital signs data are consistent with the standard vital signs data; if so, input the vital signs data and the surgical data together as training data into the learning model to obtain a gastrointestinal endoscopy anesthesia medication model, and the gastrointestinal endoscopy anesthesia medication model is used to output surgical data that can maintain the clinical patient's vital signs data consistent with the standard vital signs data.

[0053] The algorithm or demonstration provided here are not inherently relevant to any particular computer, virtual system or other equipment. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing this type of system. In addition, the present application embodiment is not directed to any specific programming language yet. It should be understood that various programming languages ​​can be utilized to realize the content of the present application described here, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the present application.

[0054] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known methods, structures, and techniques are not shown in detail so as not to obscure the understanding of this description.

[0055] Similarly, it should be understood that in order to streamline the present application and assist in understanding one or more of the various aspects of the invention, in the above description of the exemplary embodiments of the present application, the various features of the embodiments of the present application are sometimes grouped together into a single embodiment, figure, or description thereof.

[0056] Those skilled in the art will appreciate that the modules in the devices in the embodiments can be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and can be divided into multiple submodules or subunits or subassemblies. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying abstract and drawings) and all processes or units of any method or device disclosed in this manner can be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying abstract and drawings) can be replaced by alternative features providing the same, equivalent or similar purpose.

[0057] The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present application may be implemented by means of hardware comprising a number of distinct elements and by means of a suitably programmed computer. The use of the words first, second, and third, etc., does not denote any order. These words may be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be construed as limiting the order in which they are performed.

Claims

1. A gastrointestinal endoscopy anesthesia medication model training method, characterized in that: include: Obtain target patients' vital signs and surgical data in real time; Acquiring standard vital sign data, wherein the standard vital sign data reflects normal vital signs of a human during anesthesia; Determining whether the vital sign data is consistent with the standard vital sign data; If so, the vital signs data and the surgical data are input into the learning model together as training data to obtain a gastrointestinal endoscopy anesthesia medication model, and the gastrointestinal endoscopy anesthesia medication model is used to output surgical data that can maintain the vital signs data of clinical patients consistent with the standard vital signs data.

2. The gastrointestinal endoscopy anesthesia medication model training method according to claim 1, characterized in that: The real-time acquisition of vital sign data and surgical data of the target patient includes: Acquiring the heart rate, blood pressure, and blood oxygen data of the target patient in real time as the vital sign data; The drug infusion dosage and endoscopic operation signal data of the target patient are acquired in real time as the surgical data.

3. The gastrointestinal endoscopy anesthesia medication model training method according to claim 1, characterized in that: After obtaining the standard vital sign data, the method further includes: Obtain the patient's anesthesia depth BIS value; Constructing a reward model, comparing the BIS value of the patient's anesthesia depth with a preset BIS value of the anesthesia depth, and deducting a preset score when the BIS value of the patient's anesthesia depth does not match the preset BIS value of the anesthesia depth; After inputting the vital sign data and the surgical data as training data into the learning model to obtain a gastrointestinal endoscopy anesthesia medication model, the method further includes: Parameters of the gastrointestinal endoscopy anesthesia medication model are adjusted based on the reward model and the score.

4. The gastrointestinal endoscopy anesthesia medication model training method according to claim 3, characterized in that: The preset anesthesia depth BIS value is 60.

5. The gastrointestinal endoscopy anesthesia medication model training method according to claim 1, characterized in that: After inputting the vital sign data and the surgical data as training data into the learning model to obtain a gastrointestinal endoscopy anesthesia medication model, the method further includes: Obtaining vital sign data of clinical patients; Inputting the vital sign data of the clinical patient into the gastrointestinal endoscopy anesthesia medication model to obtain clinical operation data; performing anesthesia on the clinical patient according to the clinical operation data; Obtaining a BIS value of anesthesia depth of the clinical patient after anesthesia; Determine whether the anesthesia depth BIS value of the clinical patient is equal to 60; If yes, the vital sign data of the clinical patient and the clinical operation data are used together as training data to continue training the gastroenteroscopic anesthesia medication model; If not, the process jumps to the step of obtaining the vital sign data of the clinical patient.

6. The gastrointestinal endoscopy anesthesia medication model training method according to claim 1, characterized in that: The surgical data includes anesthesia drug infusion rate.

7. A gastrointestinal endoscopy anesthesia medication model training device, characterized in that: include: The first acquisition module is used to obtain the vital signs data and surgical data of the target patient in real time; A second acquisition module is used to acquire standard vital signs data, wherein the standard vital signs data reflects normal vital signs of humans during anesthesia; A first judgment module is used to judge whether the vital sign data is consistent with the standard vital sign data; An input module is used to input the vital signs data and the surgical data as training data into a learning model to obtain a gastrointestinal endoscopy anesthesia medication model, and the gastrointestinal endoscopy anesthesia medication model is used to output surgical data that can maintain the vital signs data of clinical patients consistent with the standard vital signs data.

8. A gastrointestinal endoscopy anesthesia medication model training device, characterized in that: include: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one program, and the program enables the processor to perform the operations of the gastrointestinal endoscopy anesthesia medication model training method as described in any one of claims 1-6.

9. A storage medium, characterized in that: The storage medium stores executable instructions, which, when executed on the gastrointestinal endoscopy anesthesia medication model training device, enable the gastrointestinal endoscopy anesthesia medication model training device to perform the operation of the gastrointestinal endoscopy anesthesia medication model training method according to any one of claims 1 to 6.

Images