A method and device for controlling the incidence of shoulder pain based on a laparoscopic ERAS pathway

Abstract

The present application relates to a kind of based on laparoscopic ERAS path incidence control method and device of shoulder pain, wherein the method comprises: collecting the history clinical data of laparoscopic surgery patient in perioperative period;Record the intervention selection of different perioperative management strategy set in postoperative shoulder pain risk scene of laparoscopy and corresponding estimated shoulder pain occurrence risk;According to IRL algorithm, the implicit reward function of doctor is deduced and learned from perioperative management strategy set;Shoulder pain risk prediction model is constructed;The real-time clinical data of current laparoscopic surgery patient in perioperative period is input into shoulder pain risk prediction model, when searching the perioperative intervention scheme of current laparoscopic surgery patient;According to the perioperative intervention scheme of current laparoscopic surgery patient, the perioperative management strategy in ERAS path is dynamically adjusted.The present application dynamically adjusts the perioperative management strategy in ERAS path according to real-time clinical data and intervention scheme, overcomes the defects of traditional ERAS path rigidity and lack of individual response.

Description

Technical Field

[0001] This invention relates to the field of medical artificial intelligence and clinical decision support technology, specifically to a method and device for controlling the incidence of shoulder pain based on the laparoscopic ERAS approach, and computer equipment. Background Technology

[0002] Laparoscopic surgery is widely used in clinical practice due to its advantages of minimal trauma and rapid recovery. However, postoperative complications, especially shoulder pain, are a common concern for patients undergoing laparoscopic surgery. Their high incidence significantly impacts postoperative comfort and recovery, prolongs hospital stays, and increases medical costs. The occurrence of postoperative shoulder pain after laparoscopy is related to various factors, including pneumoperitoneum pressure, the Trendelenburg position, anesthesia regimen, muscle relaxant use, postoperative analgesia, and individual differences.

[0003] To improve the quality of postoperative recovery, Enhanced Recovery After Surgery (ERAS) pathways are widely used in perioperative management. ERAS pathways aim to reduce surgical stress, accelerate recovery, and decrease complications by optimizing preoperative, intraoperative, and postoperative management measures. However, current ERAS pathways are often standardized and lack the ability to be finely adjusted for individual patients, especially when dealing with complications influenced by multiple factors, such as postoperative shoulder pain. Although some studies have attempted to identify risk factors for postoperative shoulder pain after laparoscopy, there is currently a lack of predictive models that can integrate patients' perioperative history and real-time clinical data in real time and output quantitative risk assessments of shoulder pain, making it difficult for clinicians to accurately assess patients' shoulder pain risk preoperatively or early intraoperatively. Furthermore, physicians accumulate a large amount of tacit knowledge and decision preferences over long-term practice, and existing systems struggle to effectively extract and learn implicit reward functions or preferences, failing to align recommended strategies with physicians' actual decision-making logic.

[0004] To address the aforementioned issues, this invention proposes a method for controlling the incidence of shoulder pain based on the laparoscopic ERAS pathway. This method dynamically adjusts the perioperative management strategy within the ERAS pathway based on real-time clinical data and recommended intervention protocols, thereby overcoming the rigidity and lack of individual responsiveness inherent in traditional ERAS pathways. Summary of the Invention

[0005] In view of the above problems, the present invention provides a method and device for controlling the incidence of shoulder pain based on the laparoscopic ERAS approach, as well as a computer device.

[0006] According to one aspect of the present invention, a method for controlling the incidence of shoulder pain based on laparoscopic ERAS approach is provided, comprising:

[0007] Historical perioperative clinical data of patients undergoing laparoscopic surgery were collected. The historical perioperative clinical data included basic patient information, surgical parameters, postoperative recovery data, and execution information of each step of the ERAS pathway. Among them, the surgical parameters included surgical type, duration, approach, pneumoperitoneum pressure setting, Trendelenburg position angle, surgical site, anesthesia plan, type and dosage of muscle relaxant, analgesics, and intraoperative complications.

[0008] This study records the intervention choices and corresponding predicted risk of shoulder pain in laparoscopic surgery patients and anesthesiologists under different perioperative management strategy sets for different shoulder pain risk scenarios after laparoscopic surgery. The perioperative management strategy set includes pneumoperitoneum pressure selection, anesthesia depth adjustment, muscle relaxant strategy, body position fine-tuning, postoperative analgesia plan, and rehabilitation guidance. Based on the IRL algorithm, with minimizing the risk of shoulder pain as the objective function, the implicit reward function of the physician is derived and learned from the perioperative management strategy set in reverse.

[0009] Using the historical perioperative clinical data as training samples and the implicit reward function as an auxiliary training signal, a shoulder pain risk prediction model is constructed. Real-time perioperative clinical data of current laparoscopic surgery patients are input into the shoulder pain risk prediction model, and a Monte Carlo tree search algorithm is used to search for perioperative intervention plans for these patients. The perioperative intervention plans include the range of intraoperative pneumoperitoneum pressure settings, depth of anesthesia and dosage of muscle relaxants, intraoperative positional fine-tuning, early postoperative analgesia, and early rehabilitation guidance.

[0010] The perioperative management strategy in the ERAS pathway is dynamically adjusted based on the current perioperative intervention plan for laparoscopic surgery patients.

[0011] In one alternative approach, collecting perioperative historical clinical data from patients undergoing laparoscopic surgery further includes:

[0012] Establish a medical data platform that includes HIS, electronic medical record (EMR), anesthesia information management (AIMS), and operating room management (ORM);

[0013] Historical perioperative clinical data of laparoscopic surgery patients can be extracted from the medical data platform via API interface;

[0014] The perioperative historical clinical data were integrated into a structured time series dataset.

[0015] In an alternative approach, the process of inversely deriving and learning the physician's implicit reward function from the perioperative management strategy set, based on the IRL algorithm with the objective function of minimizing the risk of shoulder pain, further includes:

[0016] Based on the historical clinical data of the perioperative period and the set of perioperative management strategies, a Markov decision process model is constructed. The state space of the Markov decision process model consists of patient basic information, surgical parameters, postoperative recovery data, and execution information of each link in the ERAS pathway. The action space of the Markov decision process model consists of pneumoperitoneum pressure selection, anesthesia depth adjustment, muscle relaxant strategy, body position fine-tuning, postoperative analgesia plan, and rehabilitation guidance. The state transition probabilities of the Markov decision process model are estimated from historical data through a neural network model.

[0017] Based on the intervention choices made by doctors under the perioperative management strategy set and the corresponding estimated risk of shoulder pain, an expert trajectory dataset was constructed; where each trajectory This indicates the doctor's state of mind during a complete surgical procedure. Select action The sequence was correlated with the corresponding estimated risk value for shoulder pain;

[0018] Based on the maximum entropy inverse reinforcement learning (IRL) algorithm, the implicit reward function parameters of the Markov decision process model are optimized with the objective function of minimizing the difference between the expert trajectory data and the expected risk of shoulder pain under the learning strategy, until the optimal implicit reward function is obtained.

[0019] In one alternative approach, optimizing the implicit reward function parameters of the Markov decision process model, based on the maximum entropy inverse reinforcement learning (IRL) algorithm and with the objective function of minimizing the difference between the expert trajectory data and the expected risk of shoulder pain under the learning strategy, further includes:

[0020] The state-action pairs in the expert trajectory dataset are mapped to a multidimensional feature space, and the expected vector of expert features is calculated.

[0021] Based on the current implicit reward function parameters and the state transition probability, the state access frequency distribution under the current policy is calculated using the forward and backward propagation algorithm, thereby obtaining the policy feature expectation vector.

[0022] Construct a maximum entropy objective function based on the expert feature expectation vector and the strategy feature expectation vector;

[0023] The current implicit reward function parameters are iteratively updated using the stochastic gradient descent algorithm. After each parameter update, the relative difference between the expected risk of shoulder pain under the current strategy and the doctor's expected risk is estimated using the importance sampling method based on the maximum entropy objective function. When the change in the current implicit reward function parameters meets a preset convergence threshold, the optimal implicit reward function parameters are output.

[0024] In one alternative approach, the shoulder pain risk prediction model includes:

[0025] The input layer is used to receive the perioperative historical clinical data;

[0026] The time series feature encoding layer consists of multiple time-aware graph attention networks, in which continuous time series data points are constructed as time graph nodes, and the graph attention layer is used to capture the dynamic correlation between different time steps and related indicators.

[0027] The data feature encoding layer consists of a multi-head self-attention layer and a multi-layer perceptron.

[0028] The text feature encoding layer includes multiple bidirectional masked autoencoders for extracting semantic representations from unstructured text data;

[0029] Relationship graph convolutional network layers are used to learn latent representations of entities and relationships from a medical knowledge graph;

[0030] The cross-modal fusion and reward guidance layer includes multiple cross-modal attention network layers, which are used to interact and aggregate among time series embeddings, structured feature embeddings, text feature embeddings, and knowledge graph enhanced features, dynamically calculating the contribution of different modalities to the final prediction; and receiving implicit reward functions as auxiliary training signals to guide the model to pay more attention to feature combinations that are strongly correlated with the doctor's implicit decision-making strategy and the goal of minimizing shoulder pain risk.

[0031] The risk prediction and uncertainty quantification output layer is used to predict the probability distribution of shoulder pain in current laparoscopic surgery patients. During training, the parameter distribution is optimized through variational inference to minimize the KL divergence between the predicted probability distribution and the actual occurrence of shoulder pain.

[0032] In one alternative approach, the perioperative intervention options for the current laparoscopic surgery patient are further included using a Monte Carlo tree search algorithm:

[0033] Initialize a Monte Carlo tree, where the root node is the current patient state, which is represented by a state feature vector calculated by the shoulder pain risk prediction model based on real-time clinical data;

[0034] Starting from the root node, child nodes are recursively selected using the UCT algorithm until a leaf node is reached. The selection criteria include the number of times a node is visited and its reward value. When a leaf node is not fully explored, untried intervention actions are selected from the perioperative management strategy set to expand the node. On the expanded new node, a rollout simulation is performed using a fast policy simulator to predict the shoulder pain risk status after executing the sequence of intervention actions. Based on the shoulder pain risk score obtained from the simulation results, the number of times all nodes are visited and the cumulative reward value are updated in reverse along the search path.

[0035] Construct a multi-objective reward function that includes immediate rewards and long-term rewards, wherein the immediate rewards are calculated based on the degree of risk reduction output by the shoulder pain risk prediction model;

[0036] The search terminates when the confidence level of the optimal solution of the multi-objective reward function exceeds a preset threshold, and a Pareto optimal set of intervention schemes is output. The final intervention scheme that balances risk minimization and clinical feasibility is selected from the Pareto optimal set through a multi-criteria decision analysis method.

[0037] In one alternative approach, the composite loss function of the shoulder pain risk prediction model is:

[0038] ;

[0039] in,

[0040] ;

[0041] ; These are wavelet basis functions; Fourier transform; It is the Sobolev norm; These are latent variables; The curvature regularization coefficient; It is an n-order Euler characteristic; It is a parametric manifold; For geodesic curvature; Gaussian curvature; This is the initial time. The current moment; For all trainable parameters of the shoulder pain risk prediction model; It is a time-varying weighting function; This is the decay rate function.

[0042] In one alternative approach, the multi-objective reward function is:

[0043]

[0044] in, ; Let be the surgical state field at time t; The intervention action to be taken at time t; To carry out intervention actions Postoperative state field The expected instantaneous rate of change of reward; The probability of a jump event; It is a Poisson random measure; The jump magnitude that results in a reward when a jump of size z occurs; This refers to the patient's potential pathophysiological state. For different potential physiological states Personal preferences; For the observed surgical state field and intervention actions Influence and drive the underlying physiological field Coupling terms; for Observable measurement.

[0045] According to another aspect of the present invention, a shoulder pain incidence control device based on laparoscopic ERAS approach is provided, comprising:

[0046] The clinical data acquisition module is used to collect perioperative historical clinical data of patients undergoing laparoscopic surgery. The perioperative historical clinical data includes basic patient information, surgical parameters, postoperative recovery data, and execution information of each step of the ERAS pathway. Among them, the surgical parameters include surgical type, duration, approach, pneumoperitoneum pressure setting, Trendelenburg position angle, surgical site, anesthesia plan, type and dosage of muscle relaxant, analgesics, and intraoperative complications.

[0047] The implicit reward function learning module is used to record the intervention choices and corresponding estimated risk of shoulder pain in different perioperative management strategy sets for laparoscopic surgeons and anesthesiologists in the context of postoperative shoulder pain risk scenarios after laparoscopy. The perioperative management strategy set includes pneumoperitoneum pressure selection, anesthesia depth adjustment, muscle relaxant strategy, body position fine-tuning, postoperative analgesia plan, and rehabilitation guidance. Based on the IRL algorithm with the objective function of minimizing the risk of shoulder pain, the implicit reward function of the doctors is derived and learned in reverse from the perioperative management strategy set.

[0048] The shoulder pain risk prediction and intervention program search module is used to construct a shoulder pain risk prediction model using the perioperative historical clinical data as training samples and the implicit reward function as an auxiliary training signal. Real-time perioperative clinical data of current laparoscopic surgery patients are input into the shoulder pain risk prediction model, and a Monte Carlo tree search algorithm is used to search for perioperative intervention programs for the current laparoscopic surgery patients. The perioperative intervention programs include the range of intraoperative pneumoperitoneum pressure settings, depth of anesthesia and dosage of muscle relaxants, range of intraoperative position adjustments, early postoperative analgesia programs, and early rehabilitation guidance programs.

[0049] The ERAS pathway dynamic adjustment module is used to dynamically adjust the perioperative management strategy in the ERAS pathway according to the current perioperative intervention plan for the laparoscopic surgery patient.

[0050] According to another aspect of the present invention, a computer device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus;

[0051] The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-mentioned shoulder pain incidence control method based on the laparoscopic ERAS approach.

[0052] According to the solution provided by the present invention, a method and related device for controlling the incidence of shoulder pain based on the laparoscopic ERAS pathway are disclosed. The method includes: collecting perioperative historical clinical data of laparoscopic surgery patients, wherein the perioperative historical clinical data includes patient basic information, surgical parameters, postoperative recovery data, and execution information of each step of the ERAS pathway; wherein the surgical parameters include surgical type, duration, approach, pneumoperitoneum pressure setting, Trendelenburg position angle, surgical site, anesthesia regimen, type and dosage of muscle relaxants, analgesics, and intraoperative complications; recording the intervention selection of laparoscopic surgeons and anesthesiologists under different perioperative management strategy sets for postoperative shoulder pain risk scenarios after laparoscopy and the corresponding estimated risk of shoulder pain; wherein the perioperative management strategy set includes pneumoperitoneum pressure selection, anesthesia depth adjustment, muscle relaxant strategy, body... This invention provides perioperative management strategies, including position fine-tuning, postoperative analgesia, and rehabilitation guidance. Based on the IRL algorithm with the objective function of minimizing shoulder pain risk, it inversely derives and learns the physician's implicit reward function from the set of perioperative management strategies. Using historical perioperative clinical data as training samples and the implicit reward function as auxiliary training signals, a shoulder pain risk prediction model is constructed. Real-time perioperative clinical data of current laparoscopic surgery patients is input into the shoulder pain risk prediction model, and a Monte Carlo tree search algorithm is used to search for perioperative intervention plans for the current laparoscopic surgery patients. These intervention plans include intraoperative pneumoperitoneum pressure settings, anesthesia depth and muscle relaxant dosage, intraoperative position fine-tuning, early postoperative analgesia, and early rehabilitation guidance. The perioperative management strategies in the ERAS pathway are dynamically adjusted based on the current perioperative intervention plans for laparoscopic surgery patients. This invention overcomes the shortcomings of traditional ERAS pathways, such as rigidity and lack of individual response, by dynamically adjusting perioperative management strategies in the ERAS pathway based on real-time clinical data and recommended intervention plans.

[0053] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0054] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0055] Figure 1 A flowchart illustrating the shoulder pain incidence control method based on the laparoscopic ERAS approach according to an embodiment of the present invention is shown.

[0056] Figure 2 A schematic diagram of the frame of the shoulder pain incidence control device based on the laparoscopic ERAS approach according to an embodiment of the present invention is shown.

[0057] Figure 3 A schematic diagram of the structure of a computer device according to an embodiment of the present invention is shown. Detailed Implementation

[0058] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0059] Figure 1 A flowchart illustrating a method for controlling the incidence of shoulder pain based on laparoscopic ERAS (Extended Recovery After Surgery) according to an embodiment of the present invention is shown. Specifically, as... Figure 1 As shown, it includes the following steps:

[0060] Step S101: Collect perioperative historical clinical data of patients undergoing laparoscopic surgery. The perioperative historical clinical data includes basic patient information, surgical parameters, postoperative recovery data, and execution information of each step of the ERAS pathway. Among them, the surgical parameters include surgical type, duration, approach, pneumoperitoneum pressure setting, Trendelenburg position angle, surgical site, anesthesia plan, type and dosage of muscle relaxant, analgesics, and intraoperative complications.

[0061] This embodiment includes not only intraoperative data, but also basic patient information (preoperative), postoperative recovery data (postoperative), and execution information of each stage of the ERAS pathway. By recording the execution information of each stage of ERAS, the impact of different levels of ERAS execution on shoulder pain can be assessed, and future ERAS pathway optimization can be guided.

[0062] Specifically, the hospital information system (HIS) is used to obtain basic patient information (age, gender, BMI, medical history, comorbidities, etc.), admission and discharge records, and diagnostic information. The electronic medical record system (EMR) is used to obtain progress notes, medical orders, examination and test results, nursing records, and preoperative assessment reports (such as ASA classification, Mallampati score, etc.). The anesthesia information management system (AIMS) automatically or semi-automatically records physiological parameters during anesthesia (heart rate, blood pressure, blood oxygen, end-tidal CO2, etc.), anesthetic drug use (type, dosage, and timing), muscle relaxant use (type, dosage, and monitoring values), pneumoperitoneum pressure settings, Trendelenburg position angle, operation duration, and intraoperative complications. The operating room management system (ORM) is used to obtain information such as surgical type, surgical site, surgical approach, start and end times of surgery, and the principal surgeon. The ERAS management module records the execution status of each step in the ERAS pathway, such as: preoperative oral carbohydrates, early postoperative analgesia regimen (PCA settings, types and dosages of analgesics), early postoperative mobilization, time of first ambulation, time of extubation, dietary recovery, and patient adherence to the ERAS pathway.

[0063] For example, the data of patient A who underwent laparoscopic cholecystectomy are shown below:

[0064] Age: 55

[0065] Gender: Female

[0066] BMI: 28.5 kg / m²

[0067] ASA classification: Grade II (mild systemic disease)

[0068] Past medical history: Hypertension (well controlled with oral antihypertensive medication), diabetes (controlled by diet).

[0069] Allergy history: Penicillin

[0070] Preoperative anxiety assessment: Moderate anxiety (HADS scale score)

[0071] Surgical type: Laparoscopic cholecystectomy

[0072] Surgery duration: 75 minutes

[0073] Access method: Standard four-hole method

[0074] (Pneumoperitoneum pressure setting) Starting: 12 mmHg

[0075] Maintain: 10-12 mmHg (real-time recording, average per minute)

[0076] Maximum instantaneous temperature: 14 mmHg

[0077] (Trendelenburg position angle) Starting point: 0 degrees

[0078] Intraoperative adjustment: 15-degree tilt (maintained for 60 minutes)

[0079] Surgical site: Right upper abdomen

[0080] Anesthesia protocol: Total intravenous anesthesia (TIVA)

[0081] Anesthetic drugs: Propofol (induction 150 mg, maintenance 5 mg / kg / h), remifentanil (induction 0.5 μg / kg / min, maintenance 0.2 μg / kg / min)

[0082] Anesthesia depth monitoring: BIS (mean 45-55)

[0083] Types and dosages of muscle relaxants: Rocuronium bromide (induction 50 mg, maintenance 20 mg), muscle relaxation monitoring (TOF-Guard, maintenance T1 / T4 > 90%)

[0084] Analgesics (intraoperative): Fentanyl (50 μg added intraoperatively)

[0085] Intraoperative complications: None

[0086] Liquid management: 1000ml crystalloid solution, 250ml colloid solution

[0087] (Postoperative recovery data) PACU stay time: 60 minutes

[0088] PACU upon exiting the room: VAS 3 points

[0089] 6 hours post-surgery: VAS score 2

[0090] 24 hours post-surgery: VAS score 5 (patient reports dull pain in shoulder)

[0091] 48 hours post-surgery: VAS score 3

[0092] First time getting out of bed: 8 hours post-surgery

[0093] First gas expulsion time: 12 hours post-surgery

[0094] Incidence of postoperative nausea and vomiting: None

[0095] Length of hospital stay: 2 days

[0096] (Execution information for each step of the ERAS pathway) Preoperative fasting and abstaining from fluids: Fast for 6 hours, abstain from fluids for 2 hours, and take 500ml of carbohydrates orally 2 hours before surgery.

[0097] (Early Postoperative Analgesia Protocol) Patient-Controlled Intravenous Analgesia (PCIA) Pump Settings: Morphine 1 mg / ml, background infusion 0.5 ml / h, additional 0.5 ml / time as needed, lockout time 10 minutes.

[0098] Oral pain reliever: Ibuprofen 200mg, three times a day

[0099] Postoperative early mobilization guidance: Nurses will assist patients to begin limb movement in bed 6 hours after surgery, and to try sitting up 8 hours after surgery.

[0100] Dietary recovery: Drink water 6 hours after surgery, liquid diet 8 hours after surgery, and regular diet 12 hours after surgery.

[0101] ERAS path compliance: High (all steps are executed as planned).

[0102] In one alternative approach, collecting perioperative historical clinical data from patients undergoing laparoscopic surgery further includes:

[0103] Establish a medical data platform that includes HIS, electronic medical record (EMR), anesthesia information management (AIMS), and operating room management (ORM);

[0104] Historical perioperative clinical data of laparoscopic surgery patients can be extracted from the medical data platform via API interface;

[0105] The perioperative historical clinical data were integrated into a structured time series dataset.

[0106] In this embodiment, for example, the HIS records patient A's age, gender, height, weight, past medical history (such as diabetes, hypertension), allergy history, and hospitalization date. EMR includes patient A's diagnosis (gallbladder stones), preoperative examination results (complete blood count, liver function, electrocardiogram, etc.), preoperative medications (such as antibiotics), informed consent signing, and the physician's preoperative assessment record (such as ASA classification). AIMS records patient A's physiological data in real time during anesthesia and surgery, such as heart rate, blood pressure, blood oxygen saturation, body temperature, and end-tidal carbon dioxide partial pressure (ETCO2) per minute. It also records the type, dosage, and administration time of anesthetic drugs (such as propofol, fentanyl, and cisatracurium), the pneumoperitoneum pressure setpoint (increasing from 12 mmHg to 15 mmHg over 30 minutes), real-time readings of the depth of anesthesia monitoring (BIS value), and any special intraoperative events recorded by the anesthesiologist (such as a single instance of hypotension, which was treated and resolved). ORM records the start and end times of the surgery, the type of surgery (laparoscopic cholecystectomy), key time points of major surgical steps, surgical instruments used, and the Trendelenburg position (adjusted from 0 degrees to 20 degrees over 2 hours) and the adjustment time. The perioperative historical clinical data were integrated into a structured time series as shown in Table 1.

[0107] Table 1

[0108]

[0109] Step S102: Record the intervention choices of laparoscopic surgeons and anesthesiologists under different perioperative management strategy sets for shoulder pain risk scenarios after laparoscopic surgery, and the corresponding estimated risk of shoulder pain occurrence; the perioperative management strategy set includes pneumoperitoneum pressure selection, anesthesia depth adjustment, muscle relaxant strategy, body position fine-tuning, postoperative analgesia plan, and rehabilitation guidance; based on the IRL algorithm with minimizing the risk of shoulder pain as the objective function, derive and learn the doctor's implicit reward function from the perioperative management strategy set.

[0110] In this embodiment, by learning the doctor's implicit reward function, the intervention is no longer just based on preset rules, but can understand how doctors weigh the impact of various perioperative management strategies (such as pneumoperitoneum pressure, depth of anesthesia, body position, etc.) on shoulder pain risk in different shoulder pain risk scenarios. This can generate personalized intervention plans that are closer to the doctor's experience and more in line with the individual patient's situation. For example, while ensuring the surgical field of view, pneumoperitoneum pressure can be reduced as much as possible to reduce the risk of shoulder pain.

[0111] For example, a patient is about to undergo laparoscopic cholecystectomy. Preoperative assessment shows poor tolerance to CO2 pneumoperitoneum, posing a higher risk of postoperative shoulder pain. A large number of historical surgical data for similar patients were recorded. Some surgeons tend to use lower pneumoperitoneum pressure (e.g., 8-10 mmHg) while maintaining a clear surgical field, combined with light anesthesia and moderate muscle relaxation, while emphasizing early shoulder movement and nonsteroidal anti-inflammatory drugs (NSAIDs) for analgesia. These patients have a lower incidence of shoulder pain. Other surgeons habitually use higher pneumoperitoneum pressure (e.g., 12-14 mmHg), deeper anesthesia, more complete muscle relaxation, and rely more on opioids for postoperative analgesia. These patients have a relatively higher incidence of shoulder pain. Surgeon A, observing a patient with stable vital signs but potential shoulder pain risk under these high-risk conditions, selected actions a1 (reducing pneumoperitoneum pressure to 10 mmHg) and a3 (lowering the Trendelenburg position) and predicted that the risk of shoulder pain would be reduced. Under the same conditions, Doctor B chooses actions a2 (maintaining pneumoperitoneum pressure of 12 mmHg) and a4 (adjusting the depth of anesthesia to BIS 50), focusing more on the surgical field and muscle relaxation. The IRL algorithm, by observing the aforementioned expert behavior, infers that Doctor A's reward function has a higher positive reward for low pneumoperitoneum pressure and reducing the Trendelenburg angle (i.e., Doctor A believes these actions are beneficial for reducing shoulder pain), while Doctor B's reward function has a higher positive reward for maintaining the surgical field and ensuring adequate muscle relaxation. The parameters of the implicit reward function are iteratively adjusted to ensure that the action sequences chosen by Doctor A and Doctor B under this reward function maximize their respective cumulative rewards. In this process, the goal of IRL is to minimize the difference between the reward derived from the doctor's behavior and the actual risk of shoulder pain; that is, the doctor's decision is to minimize the risk of shoulder pain. For example, if Doctor A reduces pneumoperitoneum pressure and the patient's shoulder pain incidence is low, IRL assigns a higher negative reward to the action of reducing pneumoperitoneum pressure (because the goal is to minimize shoulder pain, and a negative reward represents a reduction in risk). The IRL algorithm outputs a parameterized reward function, which can quantify the contribution or preference of different intervention actions in minimizing the risk of shoulder pain in a specific patient state. For example, in patients with high BMI and those undergoing prolonged surgery, reducing pneumoperitoneum pressure yields a higher positive reward. At specific surgical sites (such as upper abdominal surgery), the reward for postural adjustments is higher than that for lower abdominal surgery. The key is not only to learn how to identify high-risk scenarios for shoulder pain, but more importantly, to understand how expert physicians weigh and select various intervention strategies to minimize the risk of shoulder pain.

[0112] In an alternative approach, the process of inversely deriving and learning the physician's implicit reward function from the perioperative management strategy set, based on the IRL algorithm with the objective function of minimizing the risk of shoulder pain, further includes:

[0113] Based on the historical clinical data of the perioperative period and the set of perioperative management strategies, a Markov decision process model is constructed. The state space of the Markov decision process model consists of patient basic information, surgical parameters, postoperative recovery data, and execution information of each link in the ERAS pathway. The action space of the Markov decision process model consists of pneumoperitoneum pressure selection, anesthesia depth adjustment, muscle relaxant strategy, body position fine-tuning, postoperative analgesia plan, and rehabilitation guidance. The state transition probabilities of the Markov decision process model are estimated from historical data through a neural network model.

[0114] Based on the intervention choices made by doctors under the perioperative management strategy set and the corresponding estimated risk of shoulder pain, an expert trajectory dataset was constructed; where each trajectory This indicates the doctor's state of mind during a complete surgical procedure. Select action The sequence was correlated with the corresponding estimated risk value for shoulder pain;

[0115] Based on the maximum entropy inverse reinforcement learning (IRL) algorithm, the implicit reward function parameters of the Markov decision process model are optimized with the objective function of minimizing the difference between the expert trajectory data and the expected risk of shoulder pain under the learning strategy, until the optimal implicit reward function is obtained.

[0116] In this embodiment, quantifying the quality of physician decisions is often very complex, making it difficult to directly design precise reward functions to guide model learning. IRL (Internal Restriction Learning) observes physician (expert) behavior and infers their underlying reward mechanisms, thus avoiding the difficulty of manually designing reward functions. This allows the model to learn the implicit wisdom of physicians in making decisions after weighing various factors (such as shoulder pain risk, surgical outcome, patient comfort, resource consumption, etc.). Through the learned implicit reward function, it is possible to better understand the emphases of different physicians (or different hospital practices) in shoulder pain management. When processing expert trajectories, Maximum Entropy IRL does not require expert behavior to be perfect or deterministic; instead, it assumes that expert behavior exhibits the greatest uncertainty under the reward function, and can still learn the reward function even with some noise or suboptimal expert data.

[0117] In one alternative approach, optimizing the implicit reward function parameters of the Markov decision process model, based on the maximum entropy inverse reinforcement learning (IRL) algorithm and with the objective function of minimizing the difference between the expert trajectory data and the expected risk of shoulder pain under the learning strategy, further includes:

[0118] The state-action pairs in the expert trajectory dataset are mapped to a multidimensional feature space, and the expected vector of expert features is calculated.

[0119] Based on the current implicit reward function parameters and the state transition probability, the state access frequency distribution under the current policy is calculated using the forward and backward propagation algorithm, thereby obtaining the policy feature expectation vector.

[0120] Construct a maximum entropy objective function based on the expert feature expectation vector and the strategy feature expectation vector;

[0121] The current implicit reward function parameters are iteratively updated using the stochastic gradient descent algorithm. After each parameter update, the relative difference between the expected risk of shoulder pain under the current strategy and the doctor's expected risk is estimated using the importance sampling method based on the maximum entropy objective function. When the change in the current implicit reward function parameters meets a preset convergence threshold, the optimal implicit reward function parameters are output.

[0122] In this embodiment, for example, by traversing all expert trajectories and calculating the average of the eigenvectors of all (s,a) pairs, it is found that the average pneumoperitoneal pressure of the experts is 13 mmHg and the average body position is 25 degrees. Based on the current reward function parameters and state transition probabilities, it is calculated which action in each state under the current reward function will yield the greatest long-term reward. After each update, the expected shoulder pain risk value of different intervention options under the current strategy is evaluated by sampling and compared with the expected risk of the doctor's actual operation, to help adjust the weights until the reward function converges. The high weights for low pneumoperitoneal pressure and rapid surgery indicate that the doctor is making a trade-off between the lowest possible shoulder pain risk and the shortest possible surgical time.

[0123] Step S103: Using the historical perioperative clinical data as training samples and the implicit reward function as an auxiliary training signal, a shoulder pain risk prediction model is constructed; the real-time perioperative clinical data of the current laparoscopic surgery patient is input into the shoulder pain risk prediction model, and the perioperative intervention plan for the current laparoscopic surgery patient is searched using the Monte Carlo tree search algorithm; wherein, the perioperative intervention plan includes the range of intraoperative pneumoperitoneum pressure settings, depth of anesthesia and dosage of muscle relaxants, range of intraoperative position fine-tuning, early postoperative analgesia plan, and early rehabilitation guidance plan.

[0124] In this embodiment, real-time perioperative clinical data of patients undergoing laparoscopic surgery is received. The assessment of shoulder pain risk can be dynamically adjusted based on the patient's dynamic physiological changes, surgical progress, and real-time information on anesthesia effectiveness. The Monte Carlo Tree Search (MCTS) algorithm can assess the long-term impact of different intervention options in complex and uncertain environments by simulating possible future action paths and outcomes. The recommended intervention is not based on minimizing risk at a single moment, but rather considers the continuous medical process and potential future states to find the intervention path with the optimal long-term effect and lowest risk.

[0125] For example, after two hours of surgery, the patient's vital signs were stable, but the surgeon reported slight difficulties operating under low pneumoperitoneal pressure. Therefore, surgical time and surgeon convenience were reinstated as additional considerations in the MCTS reward function.

[0126] Optional actions:

[0127] B1: Maintain a pneumoperitoneum pressure of 10 mmHg.

[0128] B2: Increase the pneumoperitoneum pressure to 11 mmHg, and recommend that the nurse make a small positional adjustment every 15 minutes (e.g., tilt slightly to the left or right by 5 degrees and hold for 1-2 minutes to redistribute diaphragmatic stimulation).

[0129] B3: Increase pneumoperitoneum pressure to 12 mmHg and recommend that the anesthesiologist administer additional local infiltration anesthesia during the operation, and pre-place long-acting local anesthetic at the incision site.

[0130] Rollout simulation: The simulation found that B2 improved the ease of operation for physicians while maintaining a low risk of shoulder pain, and the increase in operation time was not significant.

[0131] MCTS recommends adjusting the pneumoperitoneum pressure to 11 mmHg, making minor adjustments to the patient's position every 15 minutes, and reminding the anesthesiologist to monitor the patient's diaphragmatic movement.

[0132] In one alternative approach, the shoulder pain risk prediction model includes:

[0133] The input layer is used to receive the perioperative historical clinical data;

[0134] The time series feature encoding layer consists of multiple time-aware graph attention networks, in which continuous time series data points are constructed as time graph nodes, and the graph attention layer is used to capture the dynamic correlation between different time steps and related indicators.

[0135] The data feature encoding layer consists of a multi-head self-attention layer and a multi-layer perceptron.

[0136] The text feature encoding layer includes multiple bidirectional masked autoencoders for extracting semantic representations from unstructured text data;

[0137] Relationship graph convolutional network layers are used to learn latent representations of entities and relationships from a medical knowledge graph;

[0138] The cross-modal fusion and reward guidance layer includes multiple cross-modal attention network layers, which are used to interact and aggregate among time series embeddings, structured feature embeddings, text feature embeddings, and knowledge graph enhanced features, dynamically calculating the contribution of different modalities to the final prediction; and receiving implicit reward functions as auxiliary training signals to guide the model to pay more attention to feature combinations that are strongly correlated with the doctor's implicit decision-making strategy and the goal of minimizing shoulder pain risk.

[0139] The risk prediction and uncertainty quantification output layer is used to predict the probability distribution of shoulder pain in current laparoscopic surgery patients. During training, the parameter distribution is optimized through variational inference to minimize the KL divergence between the predicted probability distribution and the actual occurrence of shoulder pain.

[0140] In this embodiment, by constructing time-series data points as time graph nodes, the dynamic correlations in continuous time-series data can be captured, such as the changing trend of pneumoperitoneum pressure at different time points and its interaction with other indicators such as heart rate and blood pressure. Integrating medical knowledge graphs through relational graph convolutional network layers incorporates expert domain knowledge (such as disease associations, drug mechanisms of action, and surgical complications) into feature learning, enhancing the understanding of complex medical concepts and providing additional contextual information in cases of sparse or incomplete training data. For example, knowledge graphs help the model understand the potential association between a certain anesthetic and specific complications. The cross-modal fusion and reward guidance layer receives the implicit reward function learned by inverse reinforcement learning (IRL) as an auxiliary training signal. When learning to predict shoulder pain risk, it not only focuses on the actual occurrence of shoulder pain but also learns how doctors weigh various factors in actual clinical decision-making (e.g., minimizing shoulder pain risk while considering surgical efficiency and patient safety). This is a semi-supervised or weakly supervised learning paradigm that effectively utilizes the implicit information in physician behavior. By using variational inference to optimize the parameter distribution during training, the uncertainty of prediction can be quantified.

[0141] In one alternative approach, the composite loss function of the shoulder pain risk prediction model is:

[0142]

[0143] in, ; ; ; These are wavelet basis functions; Fourier transform; It is the Sobolev norm; These are latent variables; The curvature regularization coefficient; It is an n-order Euler characteristic; It is a parametric manifold; For geodesic curvature; Gaussian curvature; This is the initial time. The current moment; For all trainable parameters of the shoulder pain risk prediction model; It is a time-varying weighting function; This is the decay rate function.

[0144] In this embodiment, the patient's condition and the effectiveness of medical interventions evolve over time. Through time-varying weighting and decay, the model can prioritize recent data while retaining historical information and dynamically adjust the weights of different loss terms according to time. Using Fourier transform and wavelet basis functions for data fitting in the Sobolev space, the model not only focuses on the point-to-point error of the original data but also on the smoothness of the data at different frequencies and scales, as well as the approximation quality of the function space, enabling it to capture potential trend changes and abrupt changes. Euler characteristic captures the topological structure of the parametric manifold, ensuring that the geometric properties of the model's parameter space remain stable during training and avoiding ill-conditioned structures. Geodesic curvature and Gaussian curvature are geometric measures of the manifold boundary and interior, respectively, making the parametric manifold flatter and ensuring that the path explored by the model in the parameter space is the shortest or most reasonable.

[0145] In one alternative approach, the perioperative intervention options for the current laparoscopic surgery patient are further included using a Monte Carlo tree search algorithm:

[0146] Initialize a Monte Carlo tree, where the root node is the current patient state, which is represented by a state feature vector calculated by the shoulder pain risk prediction model based on real-time clinical data;

[0147] Starting from the root node, child nodes are recursively selected using the UCT algorithm until a leaf node is reached. The selection criteria include the number of times a node is visited and its reward value. When a leaf node is not fully explored, untried intervention actions are selected from the perioperative management strategy set to expand the node. On the expanded new node, a rollout simulation is performed using a fast policy simulator to predict the shoulder pain risk status after executing the sequence of intervention actions. Based on the shoulder pain risk score obtained from the simulation results, the number of times all nodes are visited and the cumulative reward value are updated in reverse along the search path.

[0148] Construct a multi-objective reward function that includes immediate rewards and long-term rewards, wherein the immediate rewards are calculated based on the degree of risk reduction output by the shoulder pain risk prediction model;

[0149] The search terminates when the confidence level of the optimal solution of the multi-objective reward function exceeds a preset threshold, and a Pareto optimal set of intervention schemes is output. The final intervention scheme that balances risk minimization and clinical feasibility is selected from the Pareto optimal set through a multi-criteria decision analysis method.

[0150] In this embodiment, for example, patient A has a BMI of 28 and is about to undergo laparoscopic cholecystectomy. MCTS begins by selecting pathways downwards from the root node, prioritizing pathways that reduce pneumoperitoneum pressure, as historical data shows this is associated with a reduced risk of shoulder pain. At a certain leaf node, MCTS attempts to reduce pneumoperitoneum pressure from the standard 12 mmHg to 10 mmHg. Simulations are performed at a pneumoperitoneum pressure of 10 mmHg, incorporating different depths of anesthesia, muscle relaxant dosages, postural adjustments, and postoperative analgesia regimens (randomized or heuristically selected) to simulate the patient's entire surgical procedure and postoperative recovery. Each rollout inputs the simulated sequence into the shoulder pain risk prediction model to predict the final shoulder pain risk. Based on the simulation results (average risk reduction of 10%), the number of visits and cumulative reward value of that node and its parent nodes are updated. MCTS continues to explore other pathways according to UCT guidelines, such as postoperative abdominal wall block with ropivacaine or minor adjustments to the Trendelenburg position during surgery. During the simulation, in addition to the reduction in shoulder pain risk (immediate reward), other indicators were assessed, such as operation time (which may increase and affect long-term reward), muscle relaxant dosage (side effect risk, affecting long-term reward), and anesthetic drug cost. After thousands of simulation iterations, the MCTS reached its termination criteria. The following Pareto-optimal intervention sets were identified:

[0151] Option A: Pneumoperitoneum pressure 10 mmHg, moderate depth of anesthesia, standard muscle relaxant, one minor intraoperative position adjustment, postoperative ropivacaine abdominal wall block. Predicted risk of shoulder pain: 12%, operative time increased by 5 minutes.

[0152] Option B: Pneumoperitoneum pressure 11 mmHg, light anesthesia depth, reduced muscle relaxant dosage, no postural adjustments, postoperative intravenous analgesia pump. Predicted shoulder pain risk: 15%, no significant change in operation time.

[0153] Option C: Pneumoperitoneum pressure 10 mmHg, moderate anesthesia depth, reduced muscle relaxant dosage, two minor intraoperative position adjustments, and postoperative oral nonsteroidal anti-inflammatory drugs. Predicted risk of shoulder pain: 10%, with an additional 10 minutes to the surgical duration.

[0154] The physician can then choose based on the patient's specific circumstances (such as history of drug allergies and tolerance), hospital resources (whether abdominal wall block is available), and the Pareto optimal protocol. For example, if the patient is allergic to ropivacaine, option A is not an option; if operating room time is limited, option C, which significantly increases the operation time, may not be prioritized. Ultimately, the physician may choose option B because it achieves a better balance between risk reduction and surgical efficiency. During the operation, if the patient's real-time data changes (such as fluctuations in vital signs), the system can rerun the MCTS and recommend dynamically adjusted interventions based on the new patient status (e.g., suggesting fine-tuning the Trendelenburg position angle during surgery).

[0155] In one alternative approach, the multi-objective reward function is:

[0156]

[0157] in, ; Let be the surgical state field at time t; The intervention action to be taken at time t; To carry out intervention actions Postoperative state field The expected instantaneous rate of change of reward; The probability of a jump event; It is a Poisson random measure; The jump magnitude that results in a reward when a jump of size z occurs; This refers to the patient's potential pathophysiological state. For different potential physiological states Personal preferences; For the observed surgical state field and intervention actions Influence and drive the underlying physiological field Coupling terms; for Observable measurement.

[0158] In this embodiment, continuous reward considers the ongoing impact of the intervention on the patient's state (e.g., the immediate physiological response to pneumoperitoneum pressure adjustment) and the expected rate of change of immediate reward over a continuous time dimension. This is more refined than traditional walk-long reward and can capture more subtle physiological dynamics. Jump event reward introduces Poisson stochastic measures to model sudden, unpredictable jump events that may occur during surgery, such as intraoperative complications or sudden fluctuations in the patient's heart rate. This allows the reward function to assess the intervention's ability to respond to sudden events and the resulting rewards or penalties (e.g., avoiding serious consequences through timely adjustments).

[0159] Step S104: Dynamically adjust the perioperative management strategy in the ERAS pathway according to the current perioperative intervention plan for laparoscopic surgery patients.

[0160] In this embodiment, for example, patient A has a BMI of 28 and plans to undergo laparoscopic cholecystectomy. Based on the patient's initial data (age, BMI, type of surgery, etc.), the shoulder pain risk prediction model assesses the patient's shoulder pain risk as moderate to high. The Monte Carlo tree search algorithm generates an initial personalized ERAS perioperative management strategy based on this risk and the learned implicit reward function for physicians.

[0161] Pneumoperitoneum pressure: It is recommended to set it at 12 mmHg at the beginning of the operation, but it is suggested to try to gradually reduce it to 10 mmHg during the operation while maintaining a good surgical field.

[0162] Anesthesia depth and muscle relaxants: It is recommended to maintain the anesthesia depth at around BIS 45 and use muscle relaxants to maintain deep muscle relaxation (PTC < 5).

[0163] Posture adjustment: It is recommended to make a small adjustment (1 degree) to the Trendelenburg position every 45 minutes and ensure that the head-down-to-feet-up angle does not exceed 20 degrees.

[0164] Postoperative analgesia: It is recommended to administer a multimodal analgesia regimen immediately after surgery, including intravenous PCA (patient-controlled analgesia), dezocine, and oral ibuprofen. The VAS score should be assessed in the post-anesthesia recovery room. If the score is higher than 4, additional analgesics should be considered.

[0165] Early rehabilitation: Students can start sitting up 6 hours after surgery and get out of bed with assistance 12 hours after surgery to perform simple shoulder exercises (such as shrugs and shoulder rotations) and deep breathing exercises.

[0166] Intraoperative situation: 60 minutes into the operation, the patient's pneumoperitoneum pressure had decreased from 12 mmHg to 10 mmHg, but at this time the anesthesiologist noticed a slight increase in the patient's muscle tone (TOF 2 / 4) and the surgical team reported that the field of vision was slightly limited.

[0167] Real-time monitoring suggests the muscle relaxant effect may be insufficient. Based on visual field feedback, a reassessment of patient risk is recommended.

[0168] Muscle relaxants: It is recommended to administer additional muscle relaxants (such as rocuronium bromide 5mg) to restore deep muscle relaxation, maintain surgical field of vision, and reduce visceral traction.

[0169] Pneumoperitoneum pressure: Considering the issue of surgical field, it is recommended to temporarily restore the pneumoperitoneum pressure to 11 mmHg and observe the improvement in the surgical field before deciding whether to continue to reduce it.

[0170] The anesthesiologist accepted the suggestion to administer additional muscle relaxants and agreed to temporarily adjust the pneumoperitoneum pressure back to 11 mmHg.

[0171] Postoperative condition: 8 hours after the operation, the nurse assessed the patient's shoulder pain VAS score at 6, which was higher than expected (target VAS < 4).

[0172] A high shoulder pain score may necessitate adjustments to the analgesia regimen. Recommendation:

[0173] Postoperative analgesia: It is recommended to administer a single intravenous injection of parecoxib (or etoricoxib) in addition to the existing PCA and assess the effect. At the same time, reinforce the patient's instructions on shoulder exercises and deep breathing to promote carbon dioxide absorption and pain relief.

[0174] After the nurse reported to the doctor, the doctor adopted the suggestion of combined medication and instructed the nurse to strengthen rehabilitation guidance.

[0175] Through these dynamic adjustments, the ERAS pathway is no longer a static guideline, but a management system that can intelligently optimize based on individual patient conditions and real-time changes, thereby more effectively controlling the incidence of shoulder pain and improving patient satisfaction and surgical quality.

[0176] According to the solution provided by the present invention, a method and related device for controlling the incidence of shoulder pain based on the laparoscopic ERAS pathway are disclosed. The method includes: collecting perioperative historical clinical data of laparoscopic surgery patients, wherein the perioperative historical clinical data includes patient basic information, surgical parameters, postoperative recovery data, and execution information of each step of the ERAS pathway; wherein the surgical parameters include surgical type, duration, approach, pneumoperitoneum pressure setting, Trendelenburg position angle, surgical site, anesthesia regimen, type and dosage of muscle relaxants, analgesics, and intraoperative complications; recording the intervention selection of laparoscopic surgeons and anesthesiologists under different perioperative management strategy sets for postoperative shoulder pain risk scenarios after laparoscopy and the corresponding estimated risk of shoulder pain; wherein the perioperative management strategy set includes pneumoperitoneum pressure selection, anesthesia depth adjustment, muscle relaxant strategy, body... This invention provides perioperative management strategies, including position fine-tuning, postoperative analgesia, and rehabilitation guidance. Based on the IRL algorithm with the objective function of minimizing shoulder pain risk, it inversely derives and learns the physician's implicit reward function from the set of perioperative management strategies. Using historical perioperative clinical data as training samples and the implicit reward function as auxiliary training signals, a shoulder pain risk prediction model is constructed. Real-time perioperative clinical data of current laparoscopic surgery patients is input into the shoulder pain risk prediction model, and a Monte Carlo tree search algorithm is used to search for perioperative intervention plans for the current laparoscopic surgery patients. These intervention plans include intraoperative pneumoperitoneum pressure settings, anesthesia depth and muscle relaxant dosage, intraoperative position fine-tuning, early postoperative analgesia, and early rehabilitation guidance. The perioperative management strategies in the ERAS pathway are dynamically adjusted based on the current perioperative intervention plans for laparoscopic surgery patients. This invention overcomes the shortcomings of traditional ERAS pathways, such as rigidity and lack of individual response, by dynamically adjusting perioperative management strategies in the ERAS pathway based on real-time clinical data and recommended intervention plans.

[0177] Figure 2 This diagram illustrates a frame schematic of a shoulder pain incidence control device based on the laparoscopic ERAS approach according to an embodiment of the present invention. The shoulder pain incidence control device based on the laparoscopic ERAS approach includes:

[0178] The clinical data acquisition module 210 is used to collect perioperative historical clinical data of patients undergoing laparoscopic surgery. The perioperative historical clinical data includes basic patient information, surgical parameters, postoperative recovery data, and execution information of each step of the ERAS pathway. Among them, the surgical parameters include surgical type, duration, approach, pneumoperitoneum pressure setting, Trendelenburg position angle, surgical site, anesthesia plan, type and dosage of muscle relaxant, analgesics, and intraoperative complications.

[0179] The implicit reward function learning module 220 is used to record the intervention choices of laparoscopic surgeons and anesthesiologists under different perioperative management strategy sets for shoulder pain risk scenarios after laparoscopic surgery, as well as the corresponding estimated risk of shoulder pain occurrence. The perioperative management strategy set includes pneumoperitoneum pressure selection, anesthesia depth adjustment, muscle relaxant strategy, body position fine-tuning, postoperative analgesia plan, and rehabilitation guidance. Based on the IRL algorithm with the objective function of minimizing the risk of shoulder pain occurrence, the implicit reward function of the doctors is derived and learned in reverse from the perioperative management strategy set.

[0180] The shoulder pain risk prediction and intervention program search module 230 is used to construct a shoulder pain risk prediction model using the perioperative historical clinical data as training samples and the implicit reward function as an auxiliary training signal; inputting the real-time perioperative clinical data of the current laparoscopic surgery patient into the shoulder pain risk prediction model, and searching for perioperative intervention programs for the current laparoscopic surgery patient using the Monte Carlo tree search algorithm; wherein, the perioperative intervention program includes the range of intraoperative pneumoperitoneum pressure settings, anesthesia depth and muscle relaxant dosage, intraoperative position fine-tuning range, postoperative early analgesia program, and early rehabilitation guidance program;

[0181] The ERAS pathway dynamic adjustment module 240 is used to dynamically adjust the perioperative management strategy in the ERAS pathway according to the current perioperative intervention plan for the laparoscopic surgery patient.

[0182] Figure 3 The diagram shows a structural schematic of an embodiment of the computer device of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computer device.

[0183] like Figure 3 As shown, the computer device may include: a processor 302, a communications interface 304, a memory 306, and a communications bus 308.

[0184] The processor 302, communication interface 304, and memory 306 communicate with each other via communication bus 308. Communication interface 304 is used to communicate with other network elements such as clients or other servers. The processor 302 executes program 310, specifically performing the relevant steps in the above embodiment of the shoulder pain incidence control method based on the laparoscopic ERAS approach.

[0185] Specifically, program 310 may include program code that includes computer operation instructions.

[0186] Processor 302 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computer device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0187] Memory 306 is used to store program 310. Memory 306 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0188] According to the solution provided by the present invention, a method and related device for controlling the incidence of shoulder pain based on the laparoscopic ERAS pathway are disclosed. The method includes: collecting perioperative historical clinical data of laparoscopic surgery patients, wherein the perioperative historical clinical data includes patient basic information, surgical parameters, postoperative recovery data, and execution information of each step of the ERAS pathway; wherein the surgical parameters include surgical type, duration, approach, pneumoperitoneum pressure setting, Trendelenburg position angle, surgical site, anesthesia regimen, type and dosage of muscle relaxants, analgesics, and intraoperative complications; recording the intervention selection of laparoscopic surgeons and anesthesiologists under different perioperative management strategy sets for postoperative shoulder pain risk scenarios after laparoscopy and the corresponding estimated risk of shoulder pain; wherein the perioperative management strategy set includes pneumoperitoneum pressure selection, anesthesia depth adjustment, muscle relaxant strategy, body... This invention provides perioperative management strategies, including position fine-tuning, postoperative analgesia, and rehabilitation guidance. Based on the IRL algorithm with the objective function of minimizing shoulder pain risk, it inversely derives and learns the physician's implicit reward function from the set of perioperative management strategies. Using historical perioperative clinical data as training samples and the implicit reward function as auxiliary training signals, a shoulder pain risk prediction model is constructed. Real-time perioperative clinical data of current laparoscopic surgery patients is input into the shoulder pain risk prediction model, and a Monte Carlo tree search algorithm is used to search for perioperative intervention plans for the current laparoscopic surgery patients. These intervention plans include intraoperative pneumoperitoneum pressure settings, anesthesia depth and muscle relaxant dosage, intraoperative position fine-tuning, early postoperative analgesia, and early rehabilitation guidance. The perioperative management strategies in the ERAS pathway are dynamically adjusted based on the current perioperative intervention plans for laparoscopic surgery patients. This invention overcomes the shortcomings of traditional ERAS pathways, such as rigidity and lack of individual response, by dynamically adjusting perioperative management strategies in the ERAS pathway based on real-time clinical data and recommended intervention plans.

[0189] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination of all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed can be employed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose. Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims listing several devices, several of these devices may be embodied by the same hardware item. Unless otherwise specified, the steps in the above embodiments should not be construed as limiting the order of execution.

Claims

1. A method for controlling the incidence of shoulder pain based on laparoscopic ERAS approach, characterized in that, include: Historical perioperative clinical data of patients undergoing laparoscopic surgery were collected. The historical perioperative clinical data included basic patient information, surgical parameters, postoperative recovery data, and execution information of each step of the ERAS pathway. Among them, the surgical parameters included surgical type, duration, approach, pneumoperitoneum pressure setting, Trendelenburg position angle, surgical site, anesthesia plan, type and dosage of muscle relaxant, analgesics, and intraoperative complications. This study records the intervention choices and corresponding predicted risk of shoulder pain in laparoscopic surgery patients and anesthesiologists under different perioperative management strategy sets for different shoulder pain risk scenarios after laparoscopic surgery. The perioperative management strategy set includes pneumoperitoneum pressure selection, anesthesia depth adjustment, muscle relaxant strategy, body position fine-tuning, postoperative analgesia plan, and rehabilitation guidance. Based on the IRL algorithm, with minimizing the risk of shoulder pain as the objective function, the implicit reward function of the physician is derived and learned from the perioperative management strategy set in reverse. Using the historical perioperative clinical data as training samples and the implicit reward function as an auxiliary training signal, a shoulder pain risk prediction model is constructed. Real-time perioperative clinical data of current laparoscopic surgery patients are input into the shoulder pain risk prediction model, and a Monte Carlo tree search algorithm is used to search for perioperative intervention plans for these patients. The perioperative intervention plans include the range of intraoperative pneumoperitoneum pressure settings, depth of anesthesia and dosage of muscle relaxants, intraoperative positional fine-tuning, early postoperative analgesia, and early rehabilitation guidance. The perioperative management strategy in the ERAS pathway is dynamically adjusted based on the current perioperative intervention plan for laparoscopic surgery patients.

2. The method for controlling the incidence of shoulder pain based on the laparoscopic ERAS approach according to claim 1, characterized in that, The collection of perioperative historical clinical data from patients undergoing laparoscopic surgery further includes: Establish a medical data platform that includes HIS, electronic medical record (EMR), anesthesia information management (AIMS), and operating room management (ORM); Historical perioperative clinical data of laparoscopic surgery patients can be extracted from the medical data platform via API interface; The perioperative historical clinical data were integrated into a structured time series dataset.

3. The method for controlling the incidence of shoulder pain based on the laparoscopic ERAS approach according to claim 1, characterized in that, Based on the IRL algorithm with the objective function of minimizing the risk of shoulder pain, the implicit reward function for physicians is further derived and learned by inverse derivation from the perioperative management strategy set, including: Based on the historical clinical data of the perioperative period and the set of perioperative management strategies, a Markov decision process model is constructed. The state space of the Markov decision process model consists of patient basic information, surgical parameters, postoperative recovery data, and execution information of each link in the ERAS pathway. The action space of the Markov decision process model consists of pneumoperitoneum pressure selection, anesthesia depth adjustment, muscle relaxant strategy, body position fine-tuning, postoperative analgesia plan, and rehabilitation guidance. The state transition probabilities of the Markov decision process model are estimated from historical data through a neural network model. Based on the intervention choices made by doctors under the perioperative management strategy set and the corresponding estimated risk of shoulder pain, an expert trajectory dataset was constructed; where each trajectory This indicates the doctor's state of mind during a complete surgical procedure. Select action The sequence was correlated with the corresponding estimated risk value for shoulder pain; Based on the maximum entropy inverse reinforcement learning (IRL) algorithm, the implicit reward function parameters of the Markov decision process model are optimized with the objective function of minimizing the difference between the expert trajectory data and the expected risk of shoulder pain under the learning strategy, until the optimal implicit reward function is obtained.

4. The method for controlling the incidence of shoulder pain based on the laparoscopic ERAS approach according to claim 3, characterized in that, Based on the maximum entropy inverse reinforcement learning (IRL) algorithm, and with the objective function being minimizing the difference between the expert trajectory data and the expected risk of shoulder pain under the learning strategy, the optimization of the implicit reward function parameters of the Markov decision process model further includes: The state-action pairs in the expert trajectory dataset are mapped to a multidimensional feature space, and the expected vector of expert features is calculated. Based on the current implicit reward function parameters and the state transition probability, the state access frequency distribution under the current policy is calculated using the forward and backward propagation algorithm, thereby obtaining the policy feature expectation vector. Construct a maximum entropy objective function based on the expert feature expectation vector and the strategy feature expectation vector; The current implicit reward function parameters are iteratively updated using the stochastic gradient descent algorithm. After each parameter update, the relative difference between the expected risk of shoulder pain under the current strategy and the doctor's expected risk is estimated using the importance sampling method based on the maximum entropy objective function. When the change in the current implicit reward function parameters meets a preset convergence threshold, the optimal implicit reward function parameters are output.

5. The method for controlling the incidence of shoulder pain based on the laparoscopic ERAS approach according to claim 1, characterized in that, The shoulder pain risk prediction model includes: The input layer is used to receive the perioperative historical clinical data; The time series feature encoding layer consists of multiple time-aware graph attention networks, in which continuous time series data points are constructed as time graph nodes, and the graph attention layer is used to capture the dynamic correlation between different time steps and related indicators. The data feature encoding layer consists of a multi-head self-attention layer and a multi-layer perceptron. The text feature encoding layer includes multiple bidirectional masked autoencoders for extracting semantic representations from unstructured text data; Relationship graph convolutional network layers are used to learn latent representations of entities and relationships from a medical knowledge graph; The cross-modal fusion and reward guidance layer includes multiple cross-modal attention network layers, which are used to interact and aggregate among time series embeddings, structured feature embeddings, text feature embeddings, and knowledge graph enhanced features, dynamically calculating the contribution of different modalities to the final prediction; and receiving implicit reward functions as auxiliary training signals to guide the model to pay more attention to feature combinations that are strongly correlated with the doctor's implicit decision-making strategy and the goal of minimizing shoulder pain risk. The risk prediction and uncertainty quantification output layer is used to predict the probability distribution of shoulder pain in current laparoscopic surgery patients. During training, the parameter distribution is optimized through variational inference to minimize the KL divergence between the predicted probability distribution and the actual occurrence of shoulder pain.

6. The method for controlling the incidence of shoulder pain based on the laparoscopic ERAS approach according to claim 1, characterized in that, The Monte Carlo tree search algorithm is used to further search for perioperative intervention plans for patients undergoing laparoscopic surgery, including: Initialize a Monte Carlo tree, where the root node is the current patient state, which is represented by a state feature vector calculated by the shoulder pain risk prediction model based on real-time clinical data; Starting from the root node, child nodes are recursively selected using the UCT algorithm until a leaf node is reached. The selection criteria include the number of times a node is visited and its reward value. When a leaf node is not fully explored, untried intervention actions are selected from the perioperative management strategy set to expand the node. On the expanded new node, a rollout simulation is performed using a fast policy simulator to predict the shoulder pain risk status after executing the sequence of intervention actions. Based on the shoulder pain risk score obtained from the simulation results, the number of times all nodes are visited and the cumulative reward value are updated in reverse along the search path. Construct a multi-objective reward function that includes immediate rewards and long-term rewards, wherein the immediate rewards are calculated based on the degree of risk reduction output by the shoulder pain risk prediction model; The search terminates when the confidence level of the optimal solution of the multi-objective reward function exceeds a preset threshold, and a Pareto optimal set of intervention schemes is output. The final intervention scheme that balances risk minimization and clinical feasibility is selected from the Pareto optimal set through a multi-criteria decision analysis method.

7. The method for controlling the incidence of shoulder pain based on the laparoscopic ERAS approach according to claim 1, characterized in that, The composite loss function of the shoulder pain risk prediction model is: ； in, ； ； ; These are wavelet basis functions; Fourier transform; It is the Sobolev norm; These are latent variables; The curvature regularization coefficient; It is an n-order Euler characteristic; It is a parametric manifold; For geodesic curvature; Gaussian curvature; This is the initial time. The current moment; For all trainable parameters of the shoulder pain risk prediction model; It is a time-varying weighting function; This is the decay rate function.

8. The method for controlling the incidence of shoulder pain based on the laparoscopic ERAS approach according to claim 6, characterized in that, The multi-objective reward function is: ； in, ; Let be the surgical state field at time t; The intervention action to be taken at time t; To carry out intervention actions Postoperative state field The expected instantaneous rate of change of reward; The probability of a jump event; It is a Poisson random measure; The jump magnitude that results in a reward when a jump of size z occurs; This refers to the patient's potential pathophysiological state. For different potential physiological states Personal preferences; For the observed surgical state field and intervention actions Influence and drive the underlying physiological field Coupling terms; for Observable measurement.

9. A device for controlling the incidence of shoulder pain based on laparoscopic ERAS approach, characterized in that, include: The clinical data acquisition module is used to collect perioperative historical clinical data of patients undergoing laparoscopic surgery. The perioperative historical clinical data includes basic patient information, surgical parameters, postoperative recovery data, and execution information of each step of the ERAS pathway. Among them, the surgical parameters include surgical type, duration, approach, pneumoperitoneum pressure setting, Trendelenburg position angle, surgical site, anesthesia plan, type and dosage of muscle relaxant, analgesics, and intraoperative complications. The implicit reward function learning module is used to record the intervention choices and corresponding estimated risk of shoulder pain in different perioperative management strategy sets for laparoscopic surgeons and anesthesiologists in the context of postoperative shoulder pain risk scenarios after laparoscopy. The perioperative management strategy set includes pneumoperitoneum pressure selection, anesthesia depth adjustment, muscle relaxant strategy, body position fine-tuning, postoperative analgesia plan, and rehabilitation guidance. Based on the IRL algorithm with the objective function of minimizing the risk of shoulder pain, the implicit reward function of the doctors is derived and learned in reverse from the perioperative management strategy set. The shoulder pain risk prediction and intervention program search module is used to construct a shoulder pain risk prediction model using the perioperative historical clinical data as training samples and the implicit reward function as an auxiliary training signal. Real-time perioperative clinical data of current laparoscopic surgery patients are input into the shoulder pain risk prediction model, and a Monte Carlo tree search algorithm is used to search for perioperative intervention programs for the current laparoscopic surgery patients. The perioperative intervention programs include the range of intraoperative pneumoperitoneum pressure settings, depth of anesthesia and dosage of muscle relaxants, range of intraoperative position adjustments, early postoperative analgesia programs, and early rehabilitation guidance programs. The ERAS pathway dynamic adjustment module is used to dynamically adjust the perioperative management strategy in the ERAS pathway according to the current perioperative intervention plan for the laparoscopic surgery patient.

10. A computer device, comprising: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction that causes the processor to perform the operation corresponding to the shoulder pain incidence control method based on the laparoscopic ERAS approach as described in any one of claims 1-8.

Images