Intelligent prediction and dose optimization system for postoperative anesthesia complications based on multi-modal data fusion
The intelligent prediction and dosage optimization system for post-anesthesia complications through multimodal data fusion solves the problem of insufficient data integration in existing anesthesia management systems, achieves high-precision complication prediction and dosage optimization, improves the transparency and safety of the system, and is suitable for medical institutions of different sizes.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-02
- Publication Date
- 2026-03-24
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing anesthesia management systems fail to effectively integrate multimodal data and lack real-time dynamic update capabilities, resulting in insufficient accuracy in complication prediction and inadequate transparency and operability in dosage optimization. They also struggle to address individual patient differences and changes during the surgical procedure.
The intelligent prediction and dosage optimization system for post-anesthesia complications using multimodal data fusion achieves precise coupling and real-time adjustment of multi-source heterogeneous data through a multimodal data acquisition module, a dynamic hierarchical fusion module, a multi-stage hybrid prediction model, a closed-loop dosage optimization module, and an edge-cloud collaborative decision-making module.
It significantly improves the accuracy of postoperative complication prediction, reduces the incidence of complications, ensures drug concentration within a safe range, reduces manual operations by doctors, improves the transparency and safety of the system, and supports federated learning and customized adaptation of multi-center data.
Smart Images
Figure CN121725977A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical surgery technology, specifically to an intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion. Background Technology
[0002] Anesthesia management is a core and crucial link in the entire surgical procedure. Its accuracy directly determines the safety of the surgery and the postoperative prognosis of the patient. Postoperative complications are the main risk point of anesthesia management. They not only prolong the patient's hospitalization and increase medical costs, but may also cause multiple organ dysfunction and even endanger life in severe cases. According to clinical data, the overall incidence of postoperative complications under the traditional anesthesia management model is as high as 15%-30%. Among high-risk patients, the elderly, patients with liver and kidney dysfunction, and patients undergoing complex surgeries have a complication rate of more than 40%. Therefore, it is urgent to improve the accuracy and individualization of anesthesia decision-making through intelligent technology.
[0003] Existing anesthesia management systems generally suffer from the following shortcomings. First, these systems rely excessively on traditional physiological signals, such as heart rate and blood pressure, and fail to effectively integrate multimodal data, including genomics, drug metabolism, and behavioral data. Consequently, they struggle to fully reflect individual patient differences and neglect the unique physiological and biological characteristics of each patient. Second, current complication predictions are mostly limited to preoperative static assessments, lacking real-time dynamic updates during surgery. This makes it impossible to respond promptly to changes in the patient's condition, and the prediction results are often not directly related to dosage optimization, still relying on physicians' experience for adjustments. In terms of multimodal data processing, shallow fusion methods, such as feature splicing or weighted averaging, are often used. These methods fail to fully consider the temporal characteristics of data at different stages and the deep-seated correlations between cross-domain data, resulting in limitations on the accuracy and real-time performance of the prediction model. The lack of in-depth data mining makes it difficult for physicians to trace the basis for prediction results and dosage adjustments, resulting in a lack of transparency and operability, further reducing the system's clinical acceptance. Summary of the Invention
[0004] The purpose of this invention is to provide an intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion, which has the advantage of improving the accuracy of prediction of the incidence of post-anesthesia complications and solves the problems mentioned in the background art.
[0005] To achieve the above objectives, the present invention provides the following technical solution: a multimodal data fusion-based intelligent prediction and dosage optimization system for post-anesthesia complications, comprising a system body, wherein the system body includes a multimodal data acquisition module, a dynamic hierarchical fusion module, a multi-stage hybrid prediction model, a closed-loop dosage optimization module, an edge-cloud collaborative decision-making module, and a clinical decision interpretation module.
[0006] The multimodal data acquisition module is used to acquire preoperative genomic data, intraoperative real-time physiological signals, drug metabolism dynamic data, postoperative behavioral indicators and environmental correlation data. The multimodal data acquisition module includes a minimally invasive metabolic sensor, a facial micro-expression capture unit and a cross-device data synchronization interface.
[0007] The dynamic hierarchical fusion module adopts a hierarchical strategy of preoperative feature weighted fusion, intraoperative incremental flow fusion, and postoperative cross-domain correlation fusion, combined with a data quality dynamic evaluation unit, to achieve precise coupling of multi-source heterogeneous data.
[0008] The multi-stage hybrid prediction model includes a preoperative static risk grading sub-model, an intraoperative dynamic risk prediction sub-model, and a postoperative multi-complication collaborative prediction sub-model. The model has adaptive update capabilities based on transfer learning.
[0009] The closed-loop dose optimization module is based on a reinforcement learning and pharmacokinetic and pharmacodynamic coupling model. It dynamically adjusts the infusion rate and dose safety boundary of anesthetic drugs according to real-time predicted risk values, drug metabolic concentrations, and individual genetic characteristics.
[0010] The edge node of the edge-cloud collaborative decision-making module enables real-time data preprocessing and emergency dosage adjustment, completes model training and big data analysis in the cloud, and outputs visualized decision suggestions and emergency response instructions at the terminal.
[0011] The clinical decision interpretation module outputs clinically interpretable evidence for prediction and dosage optimization through feature contribution visualization and decision path tracing functions.
[0012] Furthermore, as a preferred embodiment of the present invention, in the multimodal data acquisition module, the minimally invasive metabolic sensor monitors the blood concentration of anesthetic drugs and the concentration of metabolites in real time through a subcutaneous microneedle array, with a sampling frequency ≥1Hz and a detection error ≤5%. The facial micro-expression capture unit collects the characteristics of the patient's facial muscle activity during surgery through a high-definition camera and extracts behavioral indicators of eyelid opening and closing and mouth corner deviation.
[0013] Furthermore, as a preferred embodiment of the present invention, in the dynamic hierarchical fusion module, the preoperative fusion uses an attention mechanism to weight gene data and medical history features, thereby highlighting the high-risk factors of drug-metabolizing enzyme gene polymorphism and allergy history. The intraoperative fusion achieves incremental coupling of physiological signals and drug concentration data through a sliding time window, eliminating equipment noise and outliers. The postoperative fusion is based on graph neural networks to mine hidden correlations between physiological indicators, behavioral data, and complication types.
[0014] Furthermore, as a preferred embodiment of the present invention, in the multi-stage hybrid prediction model, the preoperative static risk grading sub-model adopts an improved random forest algorithm and combines Bayesian optimization for hyperparameter tuning; the intraoperative dynamic risk prediction sub-model is based on the temporal attention mechanism LSTM to capture the time-series correlation between physiological signals and drug concentrations; and the postoperative multi-complication collaborative prediction sub-model constructs a multi-complication correlation map based on graph neural networks, and simultaneously predicts the probability and associated risk of postoperative nausea and vomiting, cognitive dysfunction, and infectious complications.
[0015] Furthermore, as a preferred embodiment of the present invention, in the closed-loop dose optimization module, the reinforcement learning agent takes the lowest risk of complications and the best anesthetic effect as dual objective functions. The state space includes real-time drug concentration, physiological indicators, and predicted risk values, while the action space is the adjustment amount of the anesthetic drug infusion rate. The PK / PD coupling model dynamically corrects drug metabolism parameters based on individual genetic characteristics to achieve individualized adaptation of dose adjustment.
[0016] Furthermore, as a preferred embodiment of the present invention, in the edge-cloud collaborative decision-making module, the response latency of the edge node is ≤100ms, supporting emergency dose adjustment for sudden intraoperative situations. The cloud uses a federated learning mechanism to complete multi-center model training without sharing the original data, thereby improving the model's generalization ability. The terminal is equipped with a touch-screen visual interface, supporting doctors to manually intervene in the dosage plan and provide feedback to the system for model iteration.
[0017] Furthermore, as a preferred embodiment of the present invention, the clinical decision interpretation module further includes an emergency response submodule, which automatically triggers an audible and visual alarm and outputs graded treatment suggestions when the intraoperative complication risk value is greater than or equal to a preset threshold and an abnormal drug concentration is detected. These suggestions include: Level 1 risk prompts the infusion pump to suspend drug administration; Level 2 risk prompts a 20% reduction in the infusion rate; and Level 3 risk prompts close monitoring.
[0018] The method of using the intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion in this invention includes the following steps: S1: Collect patient genomic data, drug-metabolizing enzyme gene polymorphism, medical history data, and physical examination data before surgery. Input the data into the preoperative static risk classification sub-model through the preoperative weighted fusion strategy of the dynamic hierarchical fusion module to obtain the complication risk classification results. S2: Based on the risk stratification of S1, and combined with individual patient characteristics, initialize the anesthetic drug dosage regimen and infusion parameters; S3: During the operation, physiological signals, drug metabolism concentrations, and surgical operation data are acquired in real time through the multimodal data acquisition module. After being processed by intraoperative incremental flow fusion, the data is input into the intraoperative dynamic risk prediction sub-model, and the dynamic risk value of complications is output in real time. S4: The closed-loop dose optimization module dynamically adjusts the infusion rate and dose based on the risk value of S3, real-time drug concentration, and PK / PD coupling model parameters through a reinforcement learning agent, while updating the dose safety boundary. S5: Postoperatively, collect patient behavioral indicators such as facial micro-expressions, limb range of motion, pain scores, and inflammatory factor levels. After postoperative cross-domain correlation fusion, input them into the postoperative multi-complication collaborative prediction sub-model to predict the probability of short-term complications. S6: The clinical decision interpretation module outputs the basis for prediction and dosage adjustment records throughout the entire process, while the edge-cloud collaboration module enables data synchronization and adaptive model updates.
[0019] Furthermore, as a preferred embodiment of the present invention, in step S4, the dose adjustment adopts a closed-loop mechanism of prediction, feedback, and optimization. Real-time data is collected every 5 seconds, the risk value change rate is calculated, and when the risk value increase rate is ≥0.1 / minute, the reinforcement learning agent is triggered to make an emergency adjustment, and the individual metabolic parameters in the PK / PD model are corrected at the same time.
[0020] Furthermore, as a preferred embodiment of the present invention, in step S6, the adaptive update of the model is achieved through transfer learning. The full-cycle data of new patients is used as incremental samples, and fine-tuning is performed on the basis of the pre-trained model in the cloud. This allows the model to adapt to the distribution of patient data in different hospitals without retraining.
[0021] Beneficial Effects: The technical solution of this application possesses the following advantages: This invention has the advantage of improving the accuracy of predicting the incidence of post-anesthesia complications. In actual use, through the combined use of the system body, multimodal data acquisition module, dynamic hierarchical fusion module, multi-stage hybrid prediction model, closed-loop dose optimization module, edge-cloud collaborative decision-making module, and clinical decision interpretation module, the prediction accuracy of post-anesthesia complications can be significantly improved. Through the innovative design of multimodal data fusion and hybrid model, the prediction accuracy of post-anesthesia complications can reach or exceed 92%, improving the reliability of prediction. Through a precise anesthetic dose optimization mechanism, the proportion of drug concentration maintained within the safe range can exceed 95%, thereby significantly reducing the incidence of complications caused by improper dosage. It is expected to reduce the incidence of complications by 30%-40%. The system's closed-loop dose optimization mechanism adjusts the anesthetic dose in real time to ensure that the anesthetic dose is maintained within the safe range. Throughout the procedure, the drug concentration was maintained at an optimal level, avoiding overdose or underdose. Automated data acquisition, complication prediction, and dosage adjustment significantly reduced manual intervention by doctors, alleviating their workload. The system's rapid response time, with intraoperative dosage adjustments responding within 100 milliseconds, greatly shortened decision-making time and improved efficiency. Furthermore, the interpretability module provided clinicians with clear decision-making support, enabling them to understand the scientific basis behind each clinical decision, thus enhancing the system's transparency and trustworthiness. Combined with an emergency response mechanism, the system's clinical applicability and safety were further improved. It also exhibited excellent scalability, supporting federated learning of multi-center data and allowing for customized adaptation based on different hospital patient groups and surgical types, making it suitable for medical institutions of varying sizes and facilitating nationwide promotion and application.
[0022] It should be understood that all combinations of the foregoing concepts and the additional concepts described in more detail below can be considered part of the inventive subject matter of this disclosure, provided that such concepts do not contradict each other. Attached Figure Description
[0023] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the entire lifecycle workflow framework of the system of the present invention; Figure 2 This is a schematic diagram of the closed-loop dose optimization process framework of the present invention; Figure 3 This is a schematic diagram of the edge-cloud collaborative decision-making process framework of the present invention; Figure 4 This is a schematic diagram of the dynamic hierarchical data fusion process framework of the present invention.
[0024] The meanings of the labels in the figures are as follows: 1. System body; 2. Multimodal data acquisition module; 3. Dynamic hierarchical fusion module; 4. Multi-stage hybrid prediction model; 5. Closed-loop dose optimization module; 6. Edge-cloud collaborative decision-making module; 7. Clinical decision interpretation module. Detailed Implementation
[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. To better understand the technical content of the present invention, specific embodiments are provided and described in conjunction with the accompanying drawings. Various aspects of the present invention are described in this disclosure with reference to the accompanying drawings, which illustrate numerous illustrative embodiments. It should be understood that the various concepts and embodiments described above, as well as those described in more detail below, can be implemented in any of many ways. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.
[0026] As attached Figure 1 To be continued Figure 4 As shown: This embodiment provides an intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion, including system body 1. System body 1 includes a multimodal data acquisition module 2, a dynamic hierarchical fusion module 3, a multi-stage hybrid prediction model 4, a closed-loop dosage optimization module 5, an edge-cloud collaborative decision-making module 6, and a clinical decision interpretation module 7.
[0027] The multimodal data acquisition module 2 is used to acquire preoperative genomic data, intraoperative real-time physiological signals, drug metabolism dynamic data, postoperative behavioral indicators and environmental correlation data. The multimodal data acquisition module 2 includes a minimally invasive metabolic sensor, a facial micro-expression capture unit and a cross-device data synchronization interface.
[0028] The dynamic hierarchical fusion module 3 adopts a hierarchical strategy of preoperative feature weighted fusion, intraoperative incremental flow fusion, and postoperative cross-domain correlation fusion, combined with the data quality dynamic evaluation unit, to achieve precise coupling of multi-source heterogeneous data.
[0029] The multi-stage hybrid prediction model 4 includes a preoperative static risk grading sub-model, an intraoperative dynamic risk prediction sub-model, and a postoperative multi-complication collaborative prediction sub-model. The model has adaptive update capabilities based on transfer learning.
[0030] The closed-loop dose optimization module 5 is based on reinforcement learning and a coupled pharmacokinetic and pharmacodynamic model. It dynamically adjusts the infusion rate and dose safety boundary of anesthetic drugs according to real-time predicted risk values, drug metabolic concentrations, and individual genetic characteristics.
[0031] The edge-cloud collaborative decision-making module enables real-time data preprocessing and emergency dosage adjustment at 6 edge nodes, completes model training and big data analysis in the cloud, and outputs visualized decision suggestions and emergency response instructions at the terminal.
[0032] The Clinical Decision Interpretation Module 7 outputs clinically interpretable evidence for prediction and dosage optimization through feature contribution visualization and decision path tracing functions.
[0033] Specifically, in the multimodal data acquisition module 2, the minimally invasive metabolic sensor monitors the blood concentration of anesthetic drugs and the concentration of metabolites in real time through a subcutaneous microneedle array, with a sampling frequency of ≥1Hz and a detection error of ≤5%. The facial micro-expression capture unit collects the characteristics of the patient's facial muscle activity during surgery through a high-definition camera and extracts behavioral indicators of eyelid opening and closing and mouth corner deviation.
[0034] Specifically, in the dynamic hierarchical fusion module 3, the preoperative fusion uses an attention mechanism to weight gene data and medical history features, thereby highlighting the high-risk factors of drug metabolism enzyme gene polymorphism and allergy history. The intraoperative fusion uses a sliding time window to achieve incremental coupling of physiological signals and drug concentration data, eliminating equipment noise and outliers. The postoperative fusion is based on graph neural networks to mine hidden correlations between physiological indicators, behavioral data and complication types.
[0035] Specifically, in the multi-stage hybrid prediction model 4, the preoperative static risk grading sub-model adopts an improved random forest algorithm and combines Bayesian optimization for hyperparameter tuning. The intraoperative dynamic risk prediction sub-model is based on the temporal attention mechanism LSTM to capture the time series correlation between physiological signals and drug concentrations. The postoperative multi-complication collaborative prediction sub-model constructs a multi-complication correlation map based on graph neural networks, and simultaneously predicts the probability and associated risk of postoperative nausea and vomiting, cognitive dysfunction, and infectious complications.
[0036] Specifically, in the closed-loop dose optimization module 5, the reinforcement learning agent takes the lowest risk of complications and the best anesthetic effect as dual objective functions. The state space includes real-time drug concentration, physiological indicators, and predicted risk values, while the action space is the adjustment amount of the anesthetic drug infusion rate. The PK / PD coupling model dynamically corrects drug metabolism parameters based on individual genetic characteristics to achieve individualized adaptation of dose adjustment.
[0037] Specifically, in the edge-cloud collaborative decision-making module 6, the edge node response latency is ≤100ms, supporting emergency dose adjustments for sudden intraoperative situations. The cloud uses a federated learning mechanism to complete multi-center model training without sharing the original data, improving the model's generalization ability. The terminal is equipped with a touch-screen visual interface, supporting doctors to manually intervene in the dosage plan and provide feedback to the system for model iteration.
[0038] Specifically, the clinical decision interpretation module 7 also includes an emergency response submodule. When the risk value of intraoperative complications is greater than or equal to a preset threshold or when an abnormal drug concentration is detected, an audible and visual alarm is automatically triggered, and a graded treatment suggestion is output. This includes a first-level risk prompt to suspend drug administration via the infusion pump, a second-level risk prompt to reduce the infusion rate by 20%, and a third-level risk prompt for close monitoring.
[0039] The method of using the intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion in this invention includes the following steps: S1: Collect patient genomic data, drug-metabolizing enzyme gene polymorphisms, medical history data, and physical examination data before surgery. Input the preoperative weighted fusion strategy of the dynamic hierarchical fusion module 3 into the preoperative static risk grading sub-model to obtain the complication risk grading results. S2: Based on the risk stratification of S1, and combined with individual patient characteristics, initialize the anesthetic drug dosage regimen and infusion parameters; S3: During the operation, physiological signals, drug metabolism concentrations, and surgical operation data are acquired in real time through the multimodal data acquisition module 2. After being processed by intraoperative incremental flow fusion, the data is input into the intraoperative dynamic risk prediction sub-model, and the dynamic risk value of complications is output in real time. S4: The closed-loop dose optimization module 5 dynamically adjusts the infusion rate and dose based on the risk value, real-time drug concentration and PK / PD coupling model parameters in S3 through a reinforcement learning agent, while updating the dose safety boundary. S5: Postoperatively, collect patient behavioral indicators such as facial micro-expressions, limb range of motion, pain scores, and inflammatory factor levels. After postoperative cross-domain correlation fusion, input them into the postoperative multi-complication collaborative prediction sub-model to predict the probability of short-term complications. S6: The clinical decision interpretation module 7 outputs the full-process prediction basis and dosage adjustment records, and the edge-cloud collaboration module realizes data synchronization and model adaptive updates.
[0040] Specifically, in step S4, the dose adjustment adopts a closed-loop mechanism of prediction, feedback, and optimization. Real-time data is collected every 5 seconds to calculate the risk value change rate. When the risk value rise rate is ≥0.1 / minute, the reinforcement learning agent is triggered to make an emergency adjustment, and the individual metabolic parameters in the PK / PD model are corrected at the same time.
[0041] Specifically, in step S6, the adaptive model update is achieved through transfer learning. The full-cycle data of new patients is used as incremental samples, and fine-tuning is performed on the pre-trained model in the cloud. This allows the model to adapt to the distribution of patient data in different hospitals without retraining.
[0042] It should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.
[0043] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention shall be determined by the claims.
Claims
1. A multimodal data fusion-based intelligent prediction and dosage optimization system for post-anesthesia complications, comprising a system body (1), characterized in that: The system body (1) includes a multimodal data acquisition module (2), a dynamic hierarchical fusion module (3), a multi-stage hybrid prediction model (4), a closed-loop dose optimization module (5), an edge-cloud collaborative decision-making module (6), and a clinical decision interpretation module (7). The multimodal data acquisition module (2) is used to acquire preoperative genomic data, intraoperative real-time physiological signals, drug metabolism dynamic data, postoperative behavioral indicators and environmental correlation data. The multimodal data acquisition module (2) includes a minimally invasive metabolic sensor, a facial micro-expression capture unit and a cross-device data synchronization interface. The dynamic hierarchical fusion module (3) adopts a hierarchical strategy of preoperative feature weighted fusion, intraoperative incremental flow fusion, and postoperative cross-domain association fusion, combined with the data quality dynamic evaluation unit, to achieve precise coupling of multi-source heterogeneous data; The multi-stage hybrid prediction model (4) includes a preoperative static risk grading sub-model, an intraoperative dynamic risk prediction sub-model, and a postoperative multi-complication collaborative prediction sub-model. The model has adaptive update capability based on transfer learning. The closed-loop dose optimization module (5) is based on a reinforcement learning and pharmacokinetic and pharmacodynamic coupling model. It dynamically adjusts the infusion rate and dose safety boundary of anesthetic drugs according to real-time predicted risk value, drug metabolism concentration and individual gene characteristics. The edge-cloud collaborative decision-making module (6) enables real-time data preprocessing and emergency dose adjustment at the edge node, completes model training and big data analysis in the cloud, and outputs visualized decision suggestions and emergency response instructions at the terminal. The clinical decision interpretation module (7) outputs clinically interpretable evidence for prediction and dosage optimization through feature contribution visualization and decision path tracing functions.
2. The intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion according to claim 1, characterized in that: In the multimodal data acquisition module (2), the minimally invasive metabolic sensor monitors the blood concentration of anesthetic drugs and the concentration of metabolites in real time through the subcutaneous microneedle array, with a sampling frequency ≥1Hz and a detection error ≤5%. The facial micro-expression capture unit collects the characteristics of the patient's facial muscle activity during surgery through a high-definition camera and extracts behavioral indicators of eyelid opening and closing and mouth corner deviation.
3. The intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion according to claim 1, characterized in that: In the dynamic hierarchical fusion module (3), the preoperative fusion uses attention mechanism to weight gene data and medical history features, thereby highlighting the high-risk factors of drug metabolism enzyme gene polymorphism and allergy history. The intraoperative fusion achieves incremental coupling of physiological signals and drug concentration data through sliding time windows, eliminating equipment noise and outliers. The postoperative fusion is based on graph neural network to mine the hidden correlation between physiological indicators, behavioral data and complication types.
4. The intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion according to claim 1, characterized in that: In the multi-stage hybrid prediction model (4), the preoperative static risk grading sub-model adopts an improved random forest algorithm and combines Bayesian optimization for hyperparameter tuning. The intraoperative dynamic risk prediction sub-model is based on the temporal attention mechanism LSTM to capture the time series correlation between physiological signals and drug concentration. The postoperative multi-complication collaborative prediction sub-model constructs a multi-complication correlation map based on graph neural network and simultaneously predicts the probability and associated risk of postoperative nausea and vomiting, cognitive dysfunction, and infection complications.
5. The intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion according to claim 1, characterized in that: In the closed-loop dose optimization module (5), the reinforcement learning agent takes the lowest risk of complications and the best anesthetic effect as dual objective functions. The state space includes real-time drug concentration, physiological indicators and predicted risk value, and the action space is the adjustment amount of anesthetic drug infusion rate. The PK / PD coupling model dynamically corrects drug metabolism parameters according to individual gene characteristics to achieve individualized adaptation of dose adjustment.
6. The intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion according to claim 1, characterized in that: In the edge-cloud collaborative decision-making module (6), the edge node response latency is ≤100ms, which supports emergency dose adjustment in case of sudden intraoperative situations. The cloud completes multi-center model training without sharing the original data through the federated learning mechanism, thereby improving the model's generalization ability. The terminal is equipped with a touch-screen visualization interface, which supports doctors to manually intervene in the dosage plan and feed it back to the system for model iteration.
7. The intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion according to claim 1, characterized in that: The clinical decision interpretation module (7) also includes an emergency response submodule. When the risk value of intraoperative complications is greater than or equal to the preset threshold and an abnormal drug concentration is detected, an audible and visual alarm is automatically triggered, and graded treatment suggestions are output. These suggestions include: Level 1 risk: linking the infusion pump to suspend drug administration; Level 2 risk: suggesting a 20% reduction in the infusion rate; and Level 3 risk: prompting close monitoring.
8. The method of using the intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion according to claims 1-7, characterized in that, Includes the following steps: S1: Collect patient genomic data, drug metabolism enzyme gene polymorphism, medical history data, and physical examination data before surgery. Input the preoperative static risk classification sub-model through the preoperative weighted fusion strategy of the dynamic hierarchical fusion module (3) to obtain the complication risk classification result. S2: Based on the risk stratification of S1, and combined with individual patient characteristics, initialize the anesthetic drug dosage regimen and infusion parameters; S3: During the operation, physiological signals, drug metabolism concentration and surgical operation data are acquired in real time through the multimodal data acquisition module (2). After being processed by the intraoperative incremental flow fusion, the data are input into the intraoperative dynamic risk prediction sub-model and the dynamic risk value of complications is output in real time. S4: Closed-loop dose optimization module (5) Based on the risk value of S3, real-time drug concentration and PK / PD coupling model parameters, dynamically adjusts the infusion rate and dose through reinforcement learning agent, and updates the dose safety boundary at the same time; S5: Postoperatively, collect patient behavioral indicators such as facial micro-expressions, limb range of motion, pain scores, and inflammatory factor levels. After postoperative cross-domain correlation fusion, input them into the postoperative multi-complication collaborative prediction sub-model to predict the probability of short-term complications. S6: The clinical decision interpretation module (7) outputs the full-process prediction basis and dose adjustment record, and the edge-cloud collaboration module realizes data synchronization and model adaptive update.
9. The method of using the intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion according to claim 8, characterized in that: In step S4, the dose adjustment adopts a closed-loop mechanism of prediction, feedback and optimization. Real-time data is collected every 5 seconds to calculate the risk value change rate. When the risk value rise rate is ≥0.1 / minute, the reinforcement learning agent is triggered to make an emergency adjustment, and the individual metabolic parameters in the PK / PD model are corrected at the same time.
10. The method of using the intelligent prediction and dosage optimization system for post-anesthesia complications based on multimodal data fusion according to claim 8, characterized in that: In step S6, the model adaptive update is achieved through transfer learning. The full-cycle data of new patients is used as incremental samples, and fine-tuning is performed on the pre-trained model in the cloud. This allows the model to adapt to the distribution of patient data in different hospitals without retraining.