Intelligent prediction model and method for postoperative complications of anesthetized patient

Abstract

The invention relates to the technical field of medical information, in particular to an intelligent prediction model and method for postoperative complications of anesthetized patients, and the method comprises the steps: collecting preoperative to postoperative complete-cycle clinical data of a patient through a medical data interface; analyzing operation codes to generate risk features, extracting vital sign dynamic features, and establishing a complication probability mapping relation through a multi-modal fusion network; combining the complication probability and pharmacokinetic parameters to construct an optimization model, and solving an individualized anesthetic dosage interval by using a gradient descent algorithm; vital signs are dynamically monitored in the operation, a dose re-optimization mechanism is triggered, the infusion rate is adjusted, and a closed-loop control link is formed; and generating a visual decision report. According to the method, through deep integration of complete-cycle clinical data and multi-modal feature modeling, preoperative physiological parameters, operation coding semantic information and intraoperative vital sign dynamic modes are subjected to fusion analysis, a nonlinear mapping relation between dosage and complication probability is constructed, and the risk prediction precision and individualized adaptability are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to an intelligent prediction model and method for postoperative complications in anesthetized patients. Background Art

[0002] In surgical operations, anesthesia is a key link to ensure the smooth progress of the operation. However, the occurrence of postoperative complications in anesthetized patients will significantly affect the patient's prognosis and medical quality. Therefore, it is of great clinical significance to accurately predict postoperative complications and formulate individualized anesthesia plans. In existing technologies, the prediction of postoperative complications after anesthesia mostly relies on single-dimensional clinical data or empirical models, and it is difficult to fully integrate multimodal information such as preoperative physiological parameters, surgical risk characteristics, and dynamic changes in intraoperative vital signs, resulting in insufficient prediction accuracy and inability to provide reliable support for clinical decision-making.

[0003] With the development of medical information technology, although some studies have attempted to construct complication prediction models through machine learning methods, traditional models often ignore the semantic parsing of surgical type codes and the time series characteristics of vital signs data, have limited ability to model the nonlinear relationship between anesthetic drug dosage and complications, and lack a dynamic adjustment mechanism, making it difficult to adapt to real-time changes in the patient's status during surgery. Therefore, they have great limitations in practical applications.

[0004] In addition, existing anesthesia dose optimization schemes are mostly based on population pharmacokinetic data, and do not fully incorporate individual patient differences and real-time monitoring of vital signs information. They cannot effectively minimize the risk of complications while ensuring the target anesthesia depth. There is an urgent need for an intelligent system that can integrate multi-source data, dynamically predict complication risks, and achieve personalized dose optimization to improve the accuracy and safety of anesthesia management. Summary of the Invention

[0005] In response to the above-mentioned shortcomings of the prior art, the present invention provides an intelligent prediction model and method for postoperative complications in anesthetized patients, which can effectively solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0007] The present invention provides an intelligent prediction method for postoperative complications in anesthetized patients, comprising the following steps:

[0008] S1. Collect the patient's full-cycle clinical data through the medical data integration interface, including preoperative physiological parameters, surgery type codes, medical images, intraoperative anesthetic drug infusion records, vital sign time-series monitoring data, and postoperative complication types and occurrence time labels. Use a time-series alignment strategy to construct complete medical record units for each patient from 24 hours before surgery to 72 hours after surgery.

[0009] S2. Construct an anesthetic dose-complication association prediction model, specifically including:

[0010] Performing multi-dimensional feature analysis on the surgery type code to generate a comprehensive surgery risk feature reflecting the risk level of the surgery operation;

[0011] Extract the dynamic change pattern of intraoperative vital signs data and generate dynamic features of intraoperative vital signs;

[0012] Preoperative physiological parameters, comprehensive surgical risk characteristics, and intraoperative vital sign dynamic characteristics are input into a multimodal fusion network to establish a nonlinear mapping relationship with the types and probabilities of postoperative complications.

[0013] S3. Establish a dose optimization decision model: Specifically including:

[0014] Based on the complication probability distribution output by the prediction model and combined with the patient's individual pharmacokinetic parameters, a constrained optimization problem is constructed with the goal of minimizing the risk of complications;

[0015] The optimal dosage interval that meets the target anesthesia depth is solved by the gradient descent algorithm, and the recommended parameters of the individualized anesthesia plan are generated;

[0016] S4. Deploy a real-time feedback adjustment system: Specifically including:

[0017] Dynamically monitor vital signs and anesthesia depth during surgery. When the actual monitored values ​​deviate from the safety thresholds preset by the prediction model, the dose parameter reoptimization mechanism is triggered.

[0018] Dynamically adjust the anesthetic drug infusion rate based on the updated complication risk prediction value to form a closed-loop control link;

[0019] S5. Generate a visual decision report, including:

[0020] Output clinical decision reports including expected complication types, risk probability cloud maps, and recommended dose safety margins;

[0021] A counterfactual deduction interface is provided to display the quantitative impact curve of different dosage strategies on the risk of complications.

[0022] Furthermore, the method for implementing the multi-dimensional feature analysis in step S2 includes the following steps:

[0023] S200, using word embedding technology to convert the surgery type code into a low-dimensional vector feature, wherein the word embedding technology uses the Word2Vec algorithm to map the surgery type code into a vector of fixed length;

[0024] S201. Extracting semantic features from the surgery-related text description using a pre-trained language model, wherein the pre-trained language model uses a BERT model to perform feature extraction on the surgery-related text description;

[0025] S202: Concatenate the encoding vector features and the semantic features to obtain a comprehensive surgical risk feature.

[0026] Furthermore, the method for extracting dynamic change patterns of intraoperative vital sign data includes:

[0027] S300: Construct a multi-scale sliding window for vital sign time series monitoring data, with window sizes ranging from 5 minutes to 60 minutes. Use a temporal convolutional network to extract local fluctuation features at different times, including heart rate variability, blood pressure fluctuation amplitude, and blood oxygen saturation periodic change rate.

[0028] S301. Modeling vital sign sequences spanning more than 30 minutes using a long short-term memory network to capture the continuous changing trends of vital signs under the influence of anesthetic drugs and identify the long-range dependence characteristics of the slope of anesthesia depth decay and the decay rate of circulatory system stability.

[0029] S302: Preset a dynamic threshold based on historical baseline data, where the historical baseline data is the normal reference data of the patient's vital signs within 24 hours before surgery. When the monitoring values ​​of five consecutive sampling points exceed three standard deviations of the baseline, an abnormal fluctuation flag is triggered, and an abnormal enhancement feature is obtained;

[0030] S303 , fusing the above-mentioned local fluctuation features, long-range dependency features, and abnormal enhancement features to generate a dynamic feature vector of intraoperative vital signs including change trend, fluctuation amplitude, periodicity, and abnormal event markers.

[0031] Furthermore, the establishment of a nonlinear mapping relationship with the postoperative complication type and probability of occurrence includes the following steps:

[0032] S400, normalizing the preoperative physiological parameters, comprehensive surgical risk characteristics, and intraoperative vital sign dynamic characteristics, and mapping the data to the [0, 1] interval; if there are missing values, filling them with mean values;

[0033] S401: Construct a multimodal fusion network, the network comprising:

[0034] The input layer sets corresponding input channels based on the three feature dimensions of preoperative physiological parameters, comprehensive surgical risk characteristics, and intraoperative vital signs dynamic characteristics, and inputs the preprocessed feature data;

[0035] The feature fusion layer uses a series connection method to fuse the input layer features into a high-dimensional vector;

[0036] The middle layer, consisting of convolutional neural network layers and recurrent neural network layers, extracts complex patterns and deep semantic information of fused features through multi-layer nonlinear transformations;

[0037] The output layer uses a fully connected layer with the same number of neurons as the number of postoperative complication types. The softmax activation function is used to output the probability distribution of each complication type.

[0038] S402, using the cross entropy loss function to define the model loss, and updating the network parameters through the back propagation algorithm to minimize the loss function. The formula is:

[0039]

[0040] in, is the total number of samples, is the true label of the i-th sample belonging to the j-th class. When sample i belongs to the j-th class, ,otherwise ; is the probability that sample i belongs to the jth class, is the number of complication types.

[0041] Furthermore, the step S3 of establishing a dose optimization decision model includes:

[0042] S500, based on the complication probability distribution output by the anesthetic dose-complication association prediction model, combined with the patient's individual pharmacokinetic parameters, the pharmacokinetic parameters including drug clearance rate and volume of distribution , constructing a constrained optimization problem with the goal of minimizing the risk of complications, including:

[0043] Construct the objective function, the formula is:

[0044]

[0045] in, is the probability of the jth complication occurring, is the weight of the jth complication, is the number of complication types;

[0046] S501. Set constraints, including: the relationship between the depth of anesthesia and the dosage u must satisfy ,in is the actual depth of anesthesia, is the target anesthesia depth, is the allowable depth error range; the dosage u must meet ,in The minimum allowable dose is The maximum dose allowed;

[0047] S502: Calculate the optimal dosage interval that satisfies the target anesthesia depth through the gradient descent algorithm and generate the recommended parameters for the individualized anesthesia plan, including: initializing the dosage u, setting the learning rate of the gradient descent algorithm , maximum number of iterations T and convergence threshold , calculate the objective function The gradient with respect to u ;

[0048] The dosage is updated according to the gradient descent algorithm, and the formula is:

[0049]

[0050] in, For the updated dosage, For the dosage before the update, check whether the dosage after the update meets the constraint conditions. If not, adjust it to the boundary value;

[0051] Determine whether the convergence condition is met. If Or the maximum number of iterations T, then stop the iteration; otherwise, return to the gradient calculation step to continue the iteration;

[0052] S503. When the algorithm converges, the optimal dosage interval is determined with the obtained dosage u as the center. ,in , , is a safety margin;

[0053] Output the optimal dosage range and related anesthesia plan recommended parameters to form individualized anesthesia plan recommendations.

[0054] Furthermore, the deployment of the real-time feedback adjustment system is specifically as follows:

[0055] Real-time monitoring: During the operation, the patient's vital signs and anesthesia depth indicators are continuously collected;

[0056] Threshold judgment: The data obtained from real-time monitoring is compared with the safety threshold pre-set by the prediction model. If the actual monitoring value exceeds the safety threshold range, it is considered that the monitoring value deviates from the preset safety threshold;

[0057] Re-optimization trigger: If the monitoring value deviates from the preset safety threshold, the dose parameter re-optimization mechanism is activated, the patient's current vital signs, anesthesia depth and complication risk prediction are reconsidered, and the constrained optimization problem with the goal of minimizing the complication risk is re-constructed. The relevant steps in the dose optimization decision model are repeated to solve the new optimal administration dose or dose range.

[0058] Rate adjustment: Based on the complication risk prediction value obtained through re-optimization, the infusion rate of anesthetic drugs is adjusted to form a closed-loop control chain from monitoring, judgment, optimization to adjustment.

[0059] An intelligent prediction system for postoperative complications in anesthesia patients, comprising:

[0060] The medical data integration module is used to collect and integrate patients' full-cycle clinical data, including preoperative physiological parameters, surgery type codes, intraoperative anesthetic drug records, vital sign time-series monitoring data, and postoperative complication labels, and construct personalized medical record units through time-series alignment strategies;

[0061] The correlation prediction module is used to integrate preoperative parameters, surgical risk characteristics, and dynamic characteristics of intraoperative vital signs to establish a prediction model for the type and probability of postoperative complications;

[0062] The dose optimization decision module uses a constrained optimization algorithm to solve the optimal anesthetic dose interval based on the probability distribution of complications and the patient's pharmacokinetic parameters, and generates an individualized dosing plan;

[0063] Real-time feedback regulation module, used to dynamically monitor vital signs and anesthesia depth during surgery, trigger dose re-optimization and adjust the infusion rate to form a closed-loop control link;

[0064] The visual report generation module is used to output clinical decision reports including complication risk probability cloud maps, recommended dose safety margins, and counterfactual deduction interfaces.

[0065] Furthermore, the association prediction module includes:

[0066] Multi-dimensional feature parsing unit, used to parse surgery type codes to generate comprehensive surgical risk features;

[0067] A dynamic feature extraction unit for extracting dynamic change patterns of intraoperative vital sign data;

[0068] The multimodal fusion unit is used to fuse preoperative parameters, comprehensive risk characteristics and dynamic characteristics to establish a nonlinear mapping relationship between postoperative complication types and probabilities.

[0069] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects:

[0070] 1. The present invention integrates full-cycle clinical data and multimodal feature modeling to fuse and analyze preoperative physiological parameters, surgical coding semantic information, and dynamic patterns of intraoperative vital signs, and constructs a nonlinear mapping relationship between dose and complication probability. Compared with traditional experience-based or single data source models, the present invention significantly improves the accuracy and individual adaptability of risk prediction.

[0071] 2. Continuously monitor vital signs and anesthesia depth indicators during surgery. When actual data deviate from the preset safety threshold, the system immediately triggers the dose re-optimization mechanism and dynamically adjusts the infusion rate based on the updated risk prediction value to restore blood pressure to a safe range. This "monitoring-prediction-adjustment" closed-loop control chain effectively avoids the lag of static models, proactively prevents and controls sudden risks during surgery, and ensures that the anesthesia depth is always within a safe range.

[0072] 3. The present invention uses data visualization technology to transform the complex anesthetic dose-risk relationship into an intuitive interactive interface, enabling doctors to quickly grasp key decision-making information. The system can generate a risk probability distribution map in real time, clearly showing the probability of complications corresponding to different anesthetic doses, while providing personalized dose recommendation ranges and predicting risk changes of different dose adjustment plans through simulation and deduction functions. This intelligent decision support significantly improves the efficiency and scientific nature of clinical plan formulation, enabling doctors to make optimal decisions based on a full grasp of risk information. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.

[0074] Figure 1 Schematic diagram of the method flow of the present invention;

[0075] Figure 2 Schematic diagram of the system structure of the present invention;

[0076] Figure 3 This is a heat map of the anesthetic dose-surgical risk bivariate complication probability of the present invention;

[0077] Figure 4 This is a multi-patient anesthetic dose-complication risk decision diagram of the present invention. DETAILED DESCRIPTION

[0078] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0079] The present invention will be further described below with reference to the embodiments.

[0080] Example 1: Reference Figure 1 In a specific embodiment, the present invention systematically improves the accuracy and safety of anesthesia management through full-cycle data collection, multimodal feature modeling, real-time closed-loop optimization and visual decision support. First, the medical data integration interface collects the patient's full-cycle clinical data from 24 hours before surgery to 72 hours after surgery, including preoperative physiological parameters such as blood pressure and heart rate, surgery type codes such as ICD-10, medical images, intraoperative anesthetic drug infusion records and vital sign time-series monitoring data. Through the time-series alignment strategy, multi-source heterogeneous data are aligned to a unified time axis according to individual patients to form a complete medical record unit. For example, the preoperative baseline blood pressure, intraoperative propofol infusion rate and postoperative complication labels of patients undergoing laparoscopic cholecystectomy are constructed into a structured data matrix after timestamp alignment to ensure data completeness and time-series consistency, providing reliable input for subsequent analysis.

[0081] In the feature modeling stage, the surgery type code is mapped to a low-dimensional vector through the Word2Vec algorithm to capture the semantic association between the codes, and the BERT model is combined to parse the deep semantic information of the text description related to the surgery. For example, the text description of "laparoscopic cholecystectomy, no adhesion" is extracted by BERT to generate a feature vector reflecting the low risk of the surgery. After fusing the coding vector and semantic features, a comprehensive surgical risk score is formed to quantify the impact of surgical operations on complications. Intraoperative vital signs data are dynamically extracted through multi-scale analysis: the temporal convolutional network captures the heart rate variability and blood pressure fluctuation amplitude within a 5-minute window, the long short-term memory network models the anesthesia depth decline trend for more than 30 minutes, and the dynamic threshold marks abnormal events where 5 consecutive sampling points exceed 3 times the baseline standard deviation. For example, the continuous downward trend of a patient's blood pressure during surgery is identified by LSTM as the anesthesia depth decline slope, and abnormal fluctuations trigger real-time warnings, such as Figure 3 As shown in the red area, the ability to respond to sudden risks is enhanced.

[0082] The multimodal fusion network normalizes preoperative parameters, surgical risk characteristics, and dynamic characteristics of vital signs, extracts complex patterns through convolutional neural networks and recurrent neural networks, and finally outputs the probability distribution of complications through Softmax. For example, after inputting the characteristics of a patient, the model predicts that the probability of respiratory depression is 15% and the probability of hypotension is 10%, corresponding to Figure 3 A medium dose of 1.5 mg / kg is associated with a low probability distribution for medium-risk surgical areas. This process is optimized using a cross-entropy loss function, achieving end-to-end automation from raw data to risk prediction, reducing manual intervention and improving prediction accuracy.

[0083] Individualized anesthesia dose optimization is combined with patient pharmacokinetic parameters to construct a constrained optimization problem with the goal of minimizing the weighted complication risk. The objective function integrates the complication weight and the probability of occurrence, and the constraint conditions ensure the depth of anesthesia and the safe dose range. The optimal dosage interval is solved iteratively by the gradient descent algorithm. For example, when a patient's drug clearance is 2.1L / h and the distribution volume is 25L, the optimized dosage interval is 1.2–1.8mg / kg. Figure 4 The green shaded area shows the robustness of the dynamic safety margin guarantee scheme.

[0084] The intraoperative real-time feedback system continuously monitors vital signs and anesthesia depth indicators. When the monitored values ​​deviate from the preset safety threshold, the dose parameter re-optimization mechanism is triggered. For example, when the patient's blood pressure drops sharply to 85mmHg, the system recalculates the risk and adjusts the infusion rate to 1.2mg / kg. After 10 minutes, the blood pressure returns to the safe range, forming a closed-loop control link to actively prevent and control intraoperative risks. The visual decision report generates a complication risk probability cloud map and a dose-risk decision map, such as Figure 3 The horizontal axis is the anesthetic dose, the vertical axis is the surgical risk score, and the color map shows the low-risk area; Figure 4 A scatter plot displays dose and risk values ​​for multiple patients, with green shading marking the recommended safety interval, intuitively assisting clinical decision-making. A counterfactual analysis interface quantifies the impact of different dosing strategies, such as a 5% increase in the probability of respiratory depression due to a 0.2 mg / kg dose increase, helping physicians assess potential risks.

[0085] Reference Figure 3 Through the collaborative analysis of dose and surgical risk, the low probability of complications in the medium-risk surgical area was verified at 1.5 mg / kg; Figure 4 , showing that the risk value within the safe interval is significantly lower than the threshold, supporting the effectiveness of the individualized plan. Through data fusion, dynamic optimization, and visual interaction, this invention significantly improves the accuracy of postoperative complication prediction, providing clinically scientific, adaptive, and intelligent decision-making tools.

[0086] The intelligent prediction method for postoperative complications in anesthesia patients comprises the following steps:

[0087] S1. Collect the patient's full-cycle clinical data through the medical data integration interface, including preoperative physiological parameters, surgery type codes, medical images, intraoperative anesthetic drug infusion records, vital sign time-series monitoring data, and postoperative complication types and occurrence time labels. Use a time-series alignment strategy to construct complete medical record units for each patient from 24 hours before surgery to 72 hours after surgery.

[0088] S2. Construct an anesthetic dose-complication association prediction model, specifically including:

[0089] Performing multi-dimensional feature analysis on the surgery type code to generate a comprehensive surgery risk feature reflecting the risk level of the surgery operation;

[0090] Extract the dynamic change pattern of intraoperative vital signs data and generate dynamic features of intraoperative vital signs;

[0091] Preoperative physiological parameters, comprehensive surgical risk characteristics, and intraoperative vital sign dynamic characteristics are input into a multimodal fusion network to establish a nonlinear mapping relationship with the types and probabilities of postoperative complications.

[0092] S3. Establish a dose optimization decision model: Specifically including:

[0093] Based on the complication probability distribution output by the prediction model and combined with the patient's individual pharmacokinetic parameters, a constrained optimization problem is constructed with the goal of minimizing the risk of complications;

[0094] The optimal dosage interval that meets the target anesthesia depth is solved by the gradient descent algorithm, and the recommended parameters of the individualized anesthesia plan are generated;

[0095] S4. Deploy a real-time feedback adjustment system: Specifically including:

[0096] Dynamically monitor vital signs and anesthesia depth during surgery. When the actual monitored values ​​deviate from the safety thresholds preset by the prediction model, the dose parameter reoptimization mechanism is triggered.

[0097] Dynamically adjust the anesthetic drug infusion rate based on the updated complication risk prediction value to form a closed-loop control link;

[0098] S5. Generate a visual decision report, including:

[0099] Output clinical decision reports including expected complication types, risk probability cloud maps, and recommended dose safety margins;

[0100] A counterfactual deduction interface is provided to display the quantitative impact curve of different dosage strategies on the risk of complications.

[0101] Furthermore, the method for implementing the multi-dimensional feature analysis in step S2 includes the following steps:

[0102] S200, using word embedding technology to convert the surgery type code into a low-dimensional vector feature, wherein the word embedding technology uses the Word2Vec algorithm to map the surgery type code into a vector of fixed length;

[0103] S201. Extracting semantic features from the surgery-related text description using a pre-trained language model, wherein the pre-trained language model uses a BERT model to perform feature extraction on the surgery-related text description;

[0104] S202: Concatenate the encoding vector features and the semantic features to obtain a comprehensive surgical risk feature.

[0105] Furthermore, the method for extracting dynamic change patterns of intraoperative vital sign data includes:

[0106] S300: Construct a multi-scale sliding window for vital sign time series monitoring data, with window sizes ranging from 5 minutes to 60 minutes. Use a temporal convolutional network to extract local fluctuation features at different times, including heart rate variability, blood pressure fluctuation amplitude, and blood oxygen saturation periodic change rate.

[0107] S301. Modeling vital sign sequences spanning more than 30 minutes using a long short-term memory network to capture the continuous changing trends of vital signs under the influence of anesthetic drugs and identify the long-range dependence characteristics of the slope of anesthesia depth decay and the decay rate of circulatory system stability.

[0108] S302: Preset a dynamic threshold based on historical baseline data, where the historical baseline data is the normal reference data of the patient's vital signs within 24 hours before surgery. When the monitoring values ​​of five consecutive sampling points exceed three standard deviations of the baseline, an abnormal fluctuation flag is triggered, and an abnormal enhancement feature is obtained;

[0109] S303 , fusing the above-mentioned local fluctuation features, long-range dependency features, and abnormal enhancement features to generate a dynamic feature vector of intraoperative vital signs including change trend, fluctuation amplitude, periodicity, and abnormal event markers.

[0110] Furthermore, the establishment of a nonlinear mapping relationship with the postoperative complication type and probability of occurrence includes the following steps:

[0111] S400, normalizing the preoperative physiological parameters, comprehensive surgical risk characteristics, and intraoperative vital sign dynamic characteristics, and mapping the data to the [0, 1] interval; if there are missing values, filling them with mean values;

[0112] S401: Construct a multimodal fusion network, the network comprising:

[0113] The input layer sets corresponding input channels based on the three feature dimensions of preoperative physiological parameters, comprehensive surgical risk characteristics, and intraoperative vital signs dynamic characteristics, and inputs the preprocessed feature data;

[0114] The feature fusion layer uses a series connection method to fuse the input layer features into a high-dimensional vector;

[0115] The middle layer, consisting of convolutional neural network layers and recurrent neural network layers, extracts complex patterns and deep semantic information of fused features through multi-layer nonlinear transformations;

[0116] The output layer uses a fully connected layer with the same number of neurons as the number of postoperative complication types. The softmax activation function is used to output the probability distribution of each complication type.

[0117] S402, using the cross entropy loss function to define the model loss, and updating the network parameters through the back propagation algorithm to minimize the loss function. The formula is:

[0118]

[0119] in, is the total number of samples, is the true label of the i-th sample belonging to the j-th class. When sample i belongs to the j-th class, ,otherwise ; is the probability that sample i belongs to the jth class, is the number of complication types.

[0120] Furthermore, the step S3 of establishing a dose optimization decision model includes:

[0121] S500, based on the complication probability distribution output by the anesthetic dose-complication association prediction model, combined with the patient's individual pharmacokinetic parameters, the pharmacokinetic parameters including drug clearance rate and volume of distribution , constructing a constrained optimization problem with the goal of minimizing the risk of complications, including:

[0122] Construct the objective function, the formula is:

[0123]

[0124] in, is the probability of the jth complication occurring, is the weight of the jth complication, is the number of complication types;

[0125] S501. Set constraints, including: the relationship between the depth of anesthesia and the dosage u must satisfy ,in is the actual depth of anesthesia, is the target anesthesia depth, is the allowable depth error range; the dosage u must meet ,in The minimum allowable dose is The maximum dose allowed;

[0126] S502: Calculate the optimal dosage interval that satisfies the target anesthesia depth through the gradient descent algorithm and generate the recommended parameters for the individualized anesthesia plan, including: initializing the dosage u, setting the learning rate of the gradient descent algorithm , maximum number of iterations T and convergence threshold , calculate the objective function The gradient with respect to u ;

[0127] The dosage is updated according to the gradient descent algorithm, and the formula is:

[0128]

[0129] in, For the updated dosage, For the dosage before the update, check whether the dosage after the update meets the constraint conditions. If not, adjust it to the boundary value;

[0130] Determine whether the convergence condition is met. If Or the maximum number of iterations T, then stop the iteration; otherwise, return to the gradient calculation step to continue the iteration;

[0131] S503. When the algorithm converges, the optimal dosage interval is determined with the obtained dosage u as the center. ,in , , is a safety margin;

[0132] Output the optimal dosage range and related anesthesia plan recommended parameters to form individualized anesthesia plan recommendations.

[0133] Furthermore, the deployment of the real-time feedback adjustment system is specifically as follows:

[0134] Real-time monitoring: During the operation, the patient's vital signs and anesthesia depth indicators are continuously collected;

[0135] Threshold judgment: The data obtained from real-time monitoring is compared with the safety threshold pre-set by the prediction model. If the actual monitoring value exceeds the safety threshold range, it is considered that the monitoring value deviates from the preset safety threshold;

[0136] Re-optimization trigger: If the monitoring value deviates from the preset safety threshold, the dose parameter re-optimization mechanism is activated, the patient's current vital signs, anesthesia depth and complication risk prediction are reconsidered, and the constrained optimization problem with the goal of minimizing the complication risk is re-constructed. The relevant steps in the dose optimization decision model are repeated to solve the new optimal administration dose or dose range.

[0137] Rate adjustment: Based on the complication risk prediction value obtained through re-optimization, the infusion rate of anesthetic drugs is adjusted to form a closed-loop control chain from monitoring, judgment, optimization to adjustment.

[0138] Example 2: Reference Figure 2 , an intelligent prediction system for postoperative complications in anesthesia patients, comprising:

[0139] The medical data integration module is used to collect and integrate patients' full-cycle clinical data, including preoperative physiological parameters, surgery type codes, intraoperative anesthetic drug records, vital sign time-series monitoring data, and postoperative complication labels, and construct personalized medical record units through time-series alignment strategies;

[0140] The correlation prediction module is used to integrate preoperative parameters, surgical risk characteristics, and dynamic characteristics of intraoperative vital signs to establish a prediction model for the type and probability of postoperative complications;

[0141] The dose optimization decision module uses a constrained optimization algorithm to solve the optimal anesthetic dose interval based on the probability distribution of complications and the patient's pharmacokinetic parameters, and generates an individualized dosing plan;

[0142] Real-time feedback regulation module, used to dynamically monitor vital signs and anesthesia depth during surgery, trigger dose re-optimization and adjust the infusion rate to form a closed-loop control link;

[0143] The visual report generation module is used to output clinical decision reports including complication risk probability cloud maps, recommended dose safety margins, and counterfactual deduction interfaces.

[0144] Furthermore, the association prediction module includes:

[0145] Multi-dimensional feature parsing unit, used to parse surgery type codes to generate comprehensive surgical risk features;

[0146] A dynamic feature extraction unit for extracting dynamic change patterns of intraoperative vital sign data;

[0147] The multimodal fusion unit is used to fuse preoperative parameters, comprehensive risk characteristics and dynamic characteristics to establish a nonlinear mapping relationship between postoperative complication types and probabilities.

[0148] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An intelligent prediction method for postoperative complications in anesthesia patients, characterized by: The following steps are involved: S1. Collect the patient's full-cycle clinical data through the medical data integration interface, including preoperative physiological parameters, surgery type codes, medical images, intraoperative anesthetic drug infusion records, vital sign time-series monitoring data, and postoperative complication types and occurrence time labels. Use a time-series alignment strategy to construct complete medical record units for each patient from 24 hours before surgery to 72 hours after surgery. S2. Construct an anesthetic dose-complication association prediction model, specifically including: Performing multi-dimensional feature analysis on the surgery type code to generate a comprehensive surgery risk feature reflecting the risk level of the surgery operation; Extract the dynamic change pattern of intraoperative vital signs data and generate dynamic features of intraoperative vital signs; Preoperative physiological parameters, comprehensive surgical risk characteristics, and intraoperative vital sign dynamic characteristics are input into a multimodal fusion network to establish a nonlinear mapping relationship with the types and probabilities of postoperative complications. S3. Establish a dose optimization decision model: Specifically including: Based on the complication probability distribution output by the prediction model and combined with the patient's individual pharmacokinetic parameters, a constrained optimization problem is constructed with the goal of minimizing the risk of complications; The optimal dosage interval that meets the target anesthesia depth is solved by the gradient descent algorithm, and the recommended parameters of the individualized anesthesia plan are generated; S4. Deploy a real-time feedback adjustment system: Specifically including: Dynamically monitor vital signs and anesthesia depth during surgery. When the actual monitored values ​​deviate from the safety thresholds preset by the prediction model, the dose parameter re-optimization mechanism is triggered. Dynamically adjust the anesthetic drug infusion rate based on the updated complication risk prediction value to form a closed-loop control link; S5. Generate a visual decision report, including: Output clinical decision reports including expected complication types, risk probability cloud maps, and recommended dose safety margins; A counterfactual deduction interface is provided to display the quantitative impact curve of different dosage strategies on the risk of complications.

2. The intelligent prediction method for postoperative complications in anesthetized patients according to claim 1, characterized in that: The method for implementing the multi-dimensional feature analysis in step S2 includes the following steps: S200, using word embedding technology to convert the surgery type code into a low-dimensional vector feature, wherein the word embedding technology uses the Word2Vec algorithm to map the surgery type code into a vector of fixed length; S201. Extracting semantic features from the surgery-related text description using a pre-trained language model, wherein the pre-trained language model uses a BERT model to perform feature extraction on the surgery-related text description; S202: Concatenate the encoding vector features and the semantic features to obtain comprehensive surgical risk features.

3. The intelligent prediction method for postoperative complications in anesthetized patients according to claim 2, characterized in that: The method for extracting dynamic change patterns of intraoperative vital sign data includes: S300: Construct a multi-scale sliding window for vital sign time series monitoring data, with window sizes ranging from 5 minutes to 60 minutes. Use a temporal convolutional network to extract local fluctuation features at different times, including heart rate variability, blood pressure fluctuation amplitude, and blood oxygen saturation periodic change rate. S301. Modeling vital sign sequences spanning more than 30 minutes using a long short-term memory network to capture the continuous changing trends of vital signs under the influence of anesthetic drugs and identify the long-range dependence characteristics of the slope of anesthesia depth decay and the decay rate of circulatory system stability. S302: Preset a dynamic threshold based on historical baseline data, where the historical baseline data is the normal reference data of the patient's vital signs within 24 hours before surgery. When the monitoring values ​​of five consecutive sampling points exceed three standard deviations of the baseline, an abnormal fluctuation flag is triggered, and an abnormal enhancement feature is obtained; S303 , fusing the above-mentioned local fluctuation features, long-range dependency features, and abnormal enhancement features to generate a dynamic feature vector of intraoperative vital signs including change trend, fluctuation amplitude, periodicity, and abnormal event markers.

4. The intelligent prediction method for postoperative complications in anesthetized patients according to claim 1, characterized in that: The establishment of a nonlinear mapping relationship with the postoperative complication type and occurrence probability comprises the following steps: S400, normalizing the preoperative physiological parameters, comprehensive surgical risk characteristics, and intraoperative vital sign dynamic characteristics, and mapping the data to the [0, 1] interval; if there are missing values, filling them with mean values; S401: Construct a multimodal fusion network, the network comprising: The input layer sets corresponding input channels based on the three feature dimensions of preoperative physiological parameters, comprehensive surgical risk characteristics, and intraoperative vital signs dynamic characteristics, and inputs the preprocessed feature data; The feature fusion layer uses a series connection method to fuse the input layer features into a high-dimensional vector; The middle layer, consisting of a convolutional neural network layer and a recurrent neural network layer, extracts complex patterns and deep semantic information of fused features through multi-layer nonlinear transformations; The output layer uses a fully connected layer with the same number of neurons as the number of postoperative complication types. The softmax activation function is used to output the probability distribution of each complication type. S402, using the cross entropy loss function to define the model loss, and updating the network parameters through the back propagation algorithm to minimize the loss function. The formula is: in, is the total number of samples, is the true label of the i-th sample belonging to the j-th class. When sample i belongs to the j-th class, ,otherwise ; is the probability that sample i belongs to the jth class, is the number of complication types.

5. The intelligent prediction method for postoperative complications in anesthetized patients according to claim 1, characterized in that: The step S3 of establishing a dose optimization decision model includes: S500, based on the complication probability distribution output by the anesthetic dose-complication association prediction model, combined with the patient's individual pharmacokinetic parameters, the pharmacokinetic parameters including drug clearance rate and volume of distribution , constructing a constrained optimization problem with the goal of minimizing the risk of complications, including: Construct the objective function, the formula is: in, is the probability of the jth complication occurring, is the weight of the jth complication, is the number of complication types; S501. Set constraints, including: the relationship between the depth of anesthesia and the dosage u must satisfy ,in is the actual depth of anesthesia, is the target anesthesia depth, is the allowable depth error range; the dosage u must meet ,in The minimum allowable dose is The maximum dose allowed; S502: Calculate the optimal dosage interval that satisfies the target anesthesia depth through the gradient descent algorithm and generate the recommended parameters for the individualized anesthesia plan, including: initializing the dosage u, setting the learning rate of the gradient descent algorithm , maximum number of iterations T and convergence threshold , calculate the objective function The gradient with respect to u ; The dosage is updated according to the gradient descent algorithm, and the formula is: in, For the updated dosage, For the dosage before the update, check whether the dosage after the update meets the constraint conditions. If not, adjust it to the boundary value; Determine whether the convergence condition is met. If Or the maximum number of iterations T, then stop the iteration; otherwise, return to the gradient calculation step to continue the iteration; S503. When the algorithm converges, the optimal dosage interval is determined with the obtained dosage u as the center. ,in , , is a safety margin; Output the optimal dosage range and related anesthesia plan recommended parameters to form individualized anesthesia plan recommendations.

6. The intelligent prediction method for postoperative complications in anesthetized patients according to claim 1, characterized in that: The deployment of the real-time feedback adjustment system is specifically as follows: Real-time monitoring: During the operation, the patient's vital signs and anesthesia depth indicators are continuously collected; Threshold judgment: The data obtained from real-time monitoring is compared with the safety threshold pre-set by the prediction model. If the actual monitoring value exceeds the safety threshold range, it is considered that the monitoring value deviates from the preset safety threshold; Re-optimization trigger: If the monitoring value deviates from the preset safety threshold, the dose parameter re-optimization mechanism is activated, the patient's current vital signs, anesthesia depth and complication risk prediction are reconsidered, and the constrained optimization problem with the goal of minimizing the complication risk is re-constructed. The relevant steps in the dose optimization decision model are repeated to solve the new optimal administration dose or dose range. Rate adjustment: Based on the complication risk prediction value obtained through re-optimization, the infusion rate of anesthetic drugs is adjusted to form a closed-loop control chain from monitoring, judgment, optimization to adjustment.

7. An intelligent prediction system for postoperative complications in anesthesia patients, characterized by: For implementing the method according to any one of claims 1 to 6, comprising: The medical data integration module is used to collect and integrate patients' full-cycle clinical data, including preoperative physiological parameters, surgery type codes, intraoperative anesthetic drug records, vital sign time-series monitoring data, and postoperative complication labels, and construct personalized medical record units through time-series alignment strategies; The correlation prediction module is used to integrate preoperative parameters, surgical risk characteristics, and dynamic characteristics of intraoperative vital signs to establish a prediction model for the type and probability of postoperative complications; The dose optimization decision module uses a constrained optimization algorithm to solve the optimal anesthetic dose interval based on the probability distribution of complications and the patient's pharmacokinetic parameters, and generates an individualized dosing plan; Real-time feedback regulation module, used to dynamically monitor vital signs and anesthesia depth during surgery, trigger dose re-optimization and adjust the infusion rate to form a closed-loop control link; The visual report generation module is used to output clinical decision reports including complication risk probability cloud maps, recommended dose safety margins, and counterfactual deduction interfaces.

8. The intelligent prediction system for postoperative complications of anesthesia patients according to claim 7, characterized in that: The association prediction module includes: Multi-dimensional feature parsing unit, used to parse surgery type codes to generate comprehensive surgical risk features; A dynamic feature extraction unit for extracting dynamic change patterns of intraoperative vital sign data; The multimodal fusion unit is used to fuse preoperative parameters, comprehensive risk characteristics and dynamic characteristics to establish a nonlinear mapping relationship between postoperative complication types and probabilities.

Images