Optimized analysis system for simulating anesthesia process

Abstract

The invention discloses an optimization analysis system for simulating an anesthesia process, relates to the technical field of anesthesia simulation optimization analysis, and is used for solving the problem of inaccurate anesthesia individualized simulation decision. By constructing an anesthesia simulation system fusing a pharmacokinetic and pharmacodynamic model and a physiological coupling model, dynamic process simulation under individual physiological states and operative characteristics is realized, state evolution trajectories of an induction period, a maintenance period and a resuscitation period can be accurately deduced, and by introducing deep learning calculation and stage continuity constraints, the accuracy of the state evolution trajectories of the induction period, the maintenance period and the resuscitation period is improved. The method comprises the following steps of: establishing a multi-path drug delivery and ventilation strategy, setting a risk-efficiency scoring function, completing state tracking and index evaluation of the whole process, generating a strategy scoring set with confidence information, further supporting an abnormal probability judgment and risk triggering mechanism by the system, and outputting key monitoring parameters and coping strategies, so as to improve the simulation accuracy and stability; therefore, closed-loop control of strategy optimization and risk early warning is realized, and the credibility of the anaesthesia simulation process is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of anesthesia simulation optimization analysis, and more particularly, to an optimization analysis system for simulating anesthesia process. Background Art

[0002] In anesthesia management scenarios applicable to intraoperative dynamic regulation and individual response prediction, the evolution of the anesthetic state has nonlinear coupling characteristics between the pharmacokinetic and pharmacodynamic processes and physiological feedback such as circulation and respiration. The dynamic switching between the induction period, maintenance period and recovery period is affected by multiple factors such as the intensity of surgical stimulation, physiological differences of patients and strategy adjustment paths. Traditional anesthesia optimization methods are mostly based on simplified pharmacokinetic models, ignoring the linkage mechanism with the physiological system, and often use static rules or staged control logic for parameter setting, which makes it difficult to support the quantitative deduction and risk estimation of the whole process and multi-path schemes. At the same time, existing systems often model the drug efficacy response as a fixed delay structure, and fail to introduce models such as neural networks to learn the complex nonlinear mapping between stimulus and response, which limits their generalization ability in complex surgical scenarios.

[0003] In existing technologies, anesthesia simulation systems fail to integrate pharmacokinetic and pharmacodynamic coupling models with physiological models, and lack a strategy scoring function tailored to procedure characteristics and individual constraints. This makes it impossible to assess the efficacy-risk balance under multi-path medication and ventilation protocols, resulting in simulation outputs lacking the ability to support clinical decision-making. Furthermore, external interference at different stages is generally not modeled, and there is a lack of state continuity constraints and simulation stability assessments between the induction, maintenance, and resuscitation phases. This can easily lead to problems such as non-convergence of simulation results, sudden state changes, and strategy mismatches, thus limiting the effectiveness of the system in real-world clinical pathways. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the following solution is proposed to solve the problem of inaccurate individualized anesthesia simulation decision-making in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions: An optimization and analysis system for simulating anesthesia process, comprising an anesthesia data acquisition module, a simulation scenario construction module, a simulation strategy determination module, and a solution generation module, wherein the modules are connected via data communication; Anesthesia data acquisition module, which is used to construct a multi-source vital sign tensor structure based on historical case data and patient-specific parameters, generate an initial physiological state vector through a unified time reference and quality label, and establish a stimulation intensity curve based on the surgical event stream; The simulation scenario construction module is used to construct the anesthetic state evolution vector based on the pharmacokinetic and pharmacodynamic model and the physiological coupling model. It also performs deep learning simulation calculations based on the combined input of the initial physiological state vector and the stimulation intensity curve to generate dynamic simulation scenarios for the induction period, maintenance period, and recovery period. The simulation strategy determination module is used to generate multi-path medication and ventilation strategies for different surgical procedures and individual constraints, construct a risk-effectiveness indicator scoring function, and perform full-process state tracking and indicator calculation in the simulation scenario to form a strategy score set with confidence information; The plan generation module is used to continuously analyze the strategy score sets of multiple paths and perform sorting and judgment, and generate interactive anesthesia plan deduction results based on high-risk trigger conditions. When the preset risk threshold is met, the plan instruction set containing key monitoring parameters and scenario handling steps is output.

[0006] Furthermore, it is used to construct a multi-source vital sign tensor structure based on historical case data and patient individualized parameters, including: Demographic information, surgical procedure labels, preoperative evaluation records, organ function indicators and perioperative monitoring records were extracted from historical case data. The patient's individualized parameter group includes body composition parameters, liver and kidney function classification, basic cardiopulmonary status, and medication sensitivity labels for feature alignment; Vital sign data are organized according to the time dimension, signal channel dimension, and source dimension. The time dimension uses a sequence with a uniform time step. The signal channel dimension includes EEG-derived depth, arterial pressure waveform, heart rate, blood oxygen saturation, end-tidal carbon dioxide, tidal volume, and anesthetic gas concentration. The source dimension identifies monitoring equipment, infusion pumps, ventilators, and information systems. Perform dimension unification and interval normalization on numerical signs, encode categorical parameters and write them into semantic dictionary, and generate baseline reference vector based on individualized parameters and append it to the first frame of tensor; According to the quality label, the missing or discontinuous fragments are gap-filled or the missing marks are retained and the confidence field is generated. The traceability index and version number are generated for the tensor entry. The processed vital sign data are sorted by timestamp and bound to the traceability index, version identifier and data source identifier, and the multi-source vital sign tensor structure for simulation calculation call is output.

[0007] Furthermore, the initial physiological state vector is generated by unifying the time base and quality label, and the stimulation intensity curve is established in combination with the surgical event flow, including: Obtain the original timestamp information of each signal channel in the multi-source vital sign tensor structure, and extract the time reference parameters of the monitoring equipment, infusion pump, ventilator and event log; Establish a mapping relationship between device time and the system's unified clock, using the surgery start mark, device heartbeat signal, and manual key events as time anchor points to correct time drift across data sources. Resampling or interpolation alignment of the corrected multi-source vital sign sequences according to a uniform time step; Select core vital signs, basic physiological parameters, and individualized features from the time-aligned vital sign data to form an initial physiological state vector, and attach the generation time, source version, and quality label summary to the initial physiological state vector; Parse discrete event records in surgical event streams to identify the types and times of skin incision, traction, body position adjustment, and electrosurgical energy release events; Corresponding stimulation weights and duration parameters are assigned to each type of discrete event, and the discrete events are mapped into continuous stimulation intensity curves along the time axis, and the curve boundaries are smoothed.

[0008] Furthermore, it is used to construct an anesthetic state evolution vector based on the pharmacokinetic and pharmacodynamic model and the physiological coupling model, including: Load the pharmacokinetic and pharmacodynamic model that matches the procedure type and individual parameters, as well as the circulation, respiration, and gas exchange coupling sub-model, and initialize the model parameters and boundary constraints; Map the initial physiological state vector, drug administration sequence, ventilation parameters and stimulation intensity curve as model-driven inputs, and establish a computational relationship from input to state to output; Recursively update plasma concentrations and effect-spatter concentrations, hypnotic and analgesic response quantities, and circulation and ventilation state variables at a uniform time step; The state variables and timestamps of each time step are combined in chronological order to generate the anesthesia state evolution vector.

[0009] Furthermore, deep learning simulation calculations are performed based on the combined input of the initial physiological state vector and the stimulation intensity curve to generate dynamic simulation scenarios for the induction period, maintenance period, and recovery period, including: Perform time alignment, normalization, and feature encoding on the initial physiological state vector and the stimulation intensity curve to form a simulation input sequence; Construct a staged deep model, set up an induction phase encoder, a maintenance phase temporal reasoning network, and a recovery phase decoder, and establish a state transfer mechanism between phases; Recursively calculate at a uniform time step and output the predicted values ​​of anesthesia depth, circulation index and ventilation index at each time step under the conditions of injection drug sequence and ventilation parameters; Continuity and consistency constraints are imposed on stage boundaries, and dynamic simulation scenarios containing timestamps, state trajectories, and event markers are synthesized and attached with beacons.

[0010] Furthermore, it is used to generate multi-path medication and ventilation strategies for different surgical procedures and individual constraints, construct a risk-effectiveness indicator scoring function, and perform full-process state tracking and indicator calculation in simulation scenarios to form a strategy scoring set with confidence information, including: Obtain procedure labels, individualized parameters, and physiological state boundary conditions, and generate multi-pathway drug delivery and ventilation strategies that meet the constraints based on drug dosage limits, ventilation parameter ranges, and surgical staging requirements. A risk-effectiveness scoring function was constructed for each strategy. Risk indicators included circulatory fluctuation amplitude, duration of hypoxia, and probability of delayed awakening; effectiveness indicators included sedation maintenance coverage, analgesic response adequacy, and drug consumption minimization. Each strategy is input into a dynamic simulation scenario to simulate state evolution, and the indicators of each stage are sampled and recorded in different time periods to generate a complete strategy execution trajectory. The confidence interval or abnormal probability is calculated for each indicator to form a scoring structure including the central estimate, upper and lower boundaries and confidence level.

[0011] Furthermore, the process of constructing the risk-effectiveness indicator scoring function includes: For time series indicators, an integral function is used to calculate the cumulative duration or integral area of ​​each indicator value exceeding the set threshold; For volatility indicators, the maximum deviation range of the indicator within the evaluation window is extracted and the ratio is calculated with the standard expected interval to obtain the deviation ratio; For probabilistic indicators, the probability of abnormal occurrence is generated by counting the trigger frequency of abnormal states; Based on the standardized scoring, an indicator weight vector is introduced, and each indicator score and its corresponding weight are linearly or nonlinearly weighted combined to obtain the scoring result of each strategy; The scoring results of each strategy are classified into risk category and effectiveness category respectively, and risk score subsets and effectiveness score subsets are generated; The scoring result records the input parameter range used for calculation, the scoring timestamp, and the version number of the scoring function used, forming a risk-performance scoring structure that includes the central estimate, upper and lower bounds, and confidence levels.

[0012] Furthermore, a confidence interval or anomaly probability is calculated for each indicator to form a scoring structure containing a central estimate, upper and lower bounds, and confidence levels, including: During the path simulation process, subparameter perturbation sampling or belief propagation calculation is performed on each risk and efficiency indicator of each strategy to obtain the fluctuation range of the indicator under different input uncertainties; The central estimate of each indicator was extracted as the average response level, and the numerical interval between the maximum and minimum values ​​was calculated as the upper and lower confidence limits; Based on the trigger frequency of abnormal states during the simulation process, the abnormal occurrence probability of each indicator is calculated; The central estimate, upper and lower bounds, and anomaly probability of each indicator are mapped into confidence level labels and recorded in the corresponding strategy scoring structure.

[0013] Furthermore, it is used to continuously analyze and rank the strategy score sets of multiple paths, generate interactive anesthesia plan deduction results based on high-risk trigger conditions, and output key monitoring parameters and scenario handling plans when the preset risk threshold is met, including: Perform ranking calculation on the confidence score structure of each strategy in the strategy scoring set, and perform weighted evaluation based on the set effectiveness target weight and risk sensitivity parameter; Monitor indicators marked as high abnormal probability or confidence limit exceeded in the scoring structure to determine whether high-risk events have been triggered. High-risk events include sedation interruption, hypoxia threshold, or delayed awakening risk types; When any high-risk event exceeds the preset risk threshold, the input variables and control points corresponding to the indicators are automatically identified; Based on the sorting results and trigger event information, the interactive anesthesia plan deduction results are generated, and the scenario disposal plan including control nodes, key time periods, indicator trajectories and monitoring parameters is output.

[0014] The technical effects and advantages of the optimization analysis system for simulating anesthesia process of the present invention are as follows: The present invention constructs an anesthesia process simulation system driven by a collaborative pharmacokinetic-pharmacodynamic model and a physiological coupling model, thereby achieving full dynamic simulation of surgical procedure characteristics, individual physiological differences, and exogenous stimulus disturbances during the induction, maintenance, and recovery periods. Based on a deep learning framework, it completes nonlinear mapping of physiological state vectors and stimulus intensity curves, generates a continuous state evolution path, and significantly improves the refinement and scenario adaptability of anesthesia process prediction. Compared with traditional scheme adjustment methods based on fixed rules or staged responses, the present invention supports the simultaneous deduction of state trajectories and risk factor response curves under multi-path drug administration and ventilation strategies, has the ability to identify and adapt to strategic risks under complex intraoperative changes in advance, and enhances the level of intelligent response to dynamic changes in the anesthesia process. In addition, an indicator scoring function for risk-effectiveness balance is introduced, which integrates multi-source indicators such as time accumulation, state deviation and abnormal probability to form a scoring structure with upper and lower boundaries and confidence levels. A strategy scoring set sorting and high-risk scenario warning mechanism are constructed to achieve the linkage push of strategy optimization and key monitoring parameters. It supports the generation of strategy scoring sets under the constraints of individualized parameters and surgical requirements, and outputs interactive anesthesia plans containing deduction results, parameter ranges and scoring labels. Finally, an intelligent optimization and analysis system integrating simulation, evaluation and feedback closed loop is constructed to effectively improve the rationality of anesthesia plans under clinical pathways and the ability to ensure intraoperative safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The figure is a schematic structural diagram of an optimization analysis system for simulating anesthesia process according to the present invention. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 creative efforts are within the scope of protection of the present invention.

[0017] In order to achieve the above objectives, Figure 1 The present invention provides a schematic structural diagram of an optimization and analysis system for simulating anesthesia process, which specifically includes an anesthesia data acquisition module, a simulation scenario construction module, a simulation strategy determination module, and a solution generation module. The modules are connected through data communication. Anesthesia data acquisition module, which is used to construct a multi-source vital sign tensor structure based on historical case data and patient-specific parameters, generate an initial physiological state vector through a unified time reference and quality label, and establish a stimulation intensity curve based on the surgical event stream; The simulation scenario construction module is used to construct the anesthetic state evolution vector based on the pharmacokinetic and pharmacodynamic model and the physiological coupling model. It also performs deep learning simulation calculations based on the combined input of the initial physiological state vector and the stimulation intensity curve to generate dynamic simulation scenarios for the induction period, maintenance period, and recovery period. The simulation strategy determination module is used to generate multi-path medication and ventilation strategies for different surgical procedures and individual constraints, construct a risk-effectiveness indicator scoring function, and perform full-process state tracking and indicator calculation in the simulation scenario to form a strategy score set with confidence information; The plan generation module is used to continuously analyze the strategy score sets of multiple paths and perform sorting and judgment, and generate interactive anesthesia plan deduction results based on high-risk trigger conditions. When the preset risk threshold is met, the plan instruction set containing key monitoring parameters and scenario handling steps is output.

[0018] The specific implementation steps of the anesthesia data acquisition module are: The system first retrieves historical case data according to the unique case identifier in the hospital information system and anesthesia information system to which it has access rights, and completes field unification using the semantic dictionary as the only semantic benchmark. The semantic dictionary is a maintainable data structure that records the field name, meaning, unit, allowed value range, coding rules, and reference interval item by item. All subsequent processing is constrained by this dictionary. After the demographic information, surgical procedure labels, preoperative assessment records, organ function indicators, and perioperative monitoring records that need to be extracted from the historical case data are loaded one by one, the system simultaneously collects the target patient's individualized parameter group, which includes four types of information: body composition parameters, liver and kidney function classification, basic cardiopulmonary status, and medication sensitivity labels.

[0019] Based on this, the system performs feature alignment: synonymous fields are mapped to unified field names according to the semantic dictionary; all measurement values ​​are converted to the standard units specified by the semantic dictionary; categorical information (procedure labels, ASA grades, medication sensitivity labels) are converted into computable discrete values ​​according to the coding table in the semantic dictionary; when the same semantic field has multiple sources, the most recent record with higher data quality is used as the primary value, and all sources and selection criteria are retained in the metadata to ensure traceability; After feature alignment is completed, the system organizes the vital sign data according to the time dimension, signal channel dimension, and source dimension to generate a multi-source vital sign tensor structure. The time dimension generates a discrete time series from the unified time step in the system configuration item, covering the entire process from anesthesia preparation, induction, maintenance to resuscitation; the unified time step is configured in seconds and fixedly written into the time step entry of the semantic dictionary as the benchmark for all subsequent resampling and interpolation. The signal channel dimension strictly includes seven channels: EEG-derived depth, arterial pressure waveform, heart rate, blood oxygen saturation, end-tidal carbon dioxide, tidal volume, and anesthetic gas concentration. Each channel has a unit, reference minimum value, reference maximum value, and valid data type definition in the semantic dictionary. The source dimension is used to identify the source of data collection, including at least four types of sources: monitoring equipment, infusion pumps, ventilators, and information systems. Each source entry records the unique identifier of the device or system, interface type, original sampling frequency, and time reference type. For raw data whose sampling frequency is inconsistent with the timestamp, the system first establishes a mapping relationship between the device time and the system unified clock using the time anchor point (surgery start mark, device heartbeat signal, manual key event). After eliminating drift, all channels are resampled or interpolated and aligned according to the unified time step, so that each channel has a one-to-one corresponding data bit at the same time index.

[0020] During the tensor numerical processing phase, the system performs dimensional and interval normalization on all numerical signs. Dimensional normalization is performed according to the target units recorded in the semantic dictionary to ensure consistency across sources. Interval normalization is performed according to the following rules: each raw value is positionally aligned with the reference minimum and maximum values ​​for that channel in the semantic dictionary. Based on the linear positional relationship between the two reference values, the raw value is mapped to a dimensionless value within the target interval. When the raw value exceeds the reference interval, it is bounded within the upper and lower bounds of the target interval according to the boundary truncation rules, and the time point and channel are labeled with an out-of-bounds truncation quality label. All categorical parameters (procedure label, American Society of Anesthesiologists (ASA) grade, and medication sensitivity label) are discretely encoded according to the semantic dictionary's encoding table, and the bidirectional mapping between encoding and semantics is maintained in the tensor's semantic metadata. A baseline reference vector is then generated based on the patient's individualized parameters: the dimensionally unified values ​​of numerical parameters (such as resting heart rate and basal respiratory rate) are taken as baseline components; the coded values ​​of categorical parameters (such as drug sensitivity labels and liver and kidney function grades) are taken, and the semantic interpretation of the code is attached to the metadata. The baseline reference vector is attached to the first frame of the multi-source vital sign tensor structure and used as a fixed reference for subsequent individual state offset calculations and model initialization; the vector also comes with a generation timestamp, an input source list, and a generation algorithm version number to ensure traceability and version control.

[0021] Quality labels and confidence fields are established for each time point and each signal channel, and missing or discontinuous segments are handled. Quality labels are used to identify data availability and anomaly types, including at least the following five types: normal recording (device operation is stable and the value is within the reference range), acquisition anomalies (such as loose leads, sensor detachment, or gas sampling interruption, triggered by the device self-test flag or event log), external interference (such as short-term artifacts caused by electrosurgery energy release, determined by the event timestamp and characteristic waveform), out-of-bounds truncation (values ​​exceed the reference range and are truncated by the boundary), and interpolation generation (values ​​obtained by filling gaps).

[0022] The processing of missing or discontinuous segments follows the following well-publicized rules: When the missing period is short and the preceding and following adjacent segments are all normal records, the system fills the gaps while maintaining trend continuity, and marks the resulting data as interpolated; when the missing period is long or there are acquisition anomalies and external interference labels, the system does not fill in the gaps and directly retains the missing labels to avoid introducing unreliable estimates; the confidence field is written in a comparable graded manner: high confidence indicates that the time-channel data is a normal record and no out-of-bounds truncation or interpolation occurs; medium confidence indicates that there is a short-term interpolation or out-of-bounds truncation but discontinuity; low confidence indicates that there is an acquisition anomaly, external interference, or a long-term missing. The quality label and confidence field jointly participate in the data weighting or avoidance strategy in the subsequent simulation and scoring stages. The relevant usage rules are fixedly written into the system configuration and recorded in the version control; Bind a traceability index, version identifier, and data source identifier to each tensor entry, and output a multi-source vital sign tensor structure after final sorting by timestamp. The traceability index consists of a four-segment path consisting of the original system name - data table name or interface name - original record primary key - original timestamp, which is used to accurately locate the original data. The version identifier also contains the semantic dictionary version number and the current preprocessing process version number, which facilitates difference comparison after semantic or algorithm updates. The data source identifier directly references the unique device or system identifier of the source dimension. The output multi-source vital sign tensor structure is standardized input data that meets the requirements of unified time step, fixed signal channel set, clear source dimension, quality label and confidence field, first frame of baseline reference vector, and traceability index and version identification.

[0023] The specific implementation steps of the simulation scenario construction module are: After obtaining the constructed multi-source vital sign tensor structure, the original timestamp information of each signal channel is first read from the structure. At the same time, the time base parameters recorded in the monitoring device, infusion pump, ventilator, and surgical event log are extracted. The time base parameters include the device's internal clock time, system recording time, sampling period, heartbeat signal timestamp, and event trigger time. The system compares these parameters with the global system time registered in the anesthesia information system for time mapping and drift correction. During the time alignment phase, the system establishes a mapping relationship between the device time and the system unified clock based on a unified time reference. The mapping process uses three types of time anchors: the simulated anesthesia surgery start mark, which is provided by the surgery start event recorded in the anesthesia record sheet or the anesthesia information system, and is the starting point for all device and system time mappings; the device heartbeat signal, which is a system status message timestamp sent by each monitoring device, infusion pump, or ventilator at a fixed period, is used to accurately correct the accumulated drift of the data stream during the acquisition process; artificial key events, such as intubation completion, skin incision start, etc., have clear time records in the event log, which can be used as additional calibration points to correct nonlinear drift. The above anchor points are used to establish a one-to-one corresponding time comparison table, and the original time of each data source is drift calculated and corrected to ensure that all data sources are synchronized under a unified clock; After completing the time base unification, the corrected multi-source vital sign sequences are resampled or interpolated and aligned according to the unified time step defined in the semantic dictionary. For signal channels with a sampling frequency higher than the target time step, equal-interval downsampling or sliding window averaging is used to preserve trend information. For signal channels with a sampling frequency lower than the target time step, linear interpolation or spline interpolation based on adjacent normal values ​​is used to generate missing points, and interpolation-generated quality labels are added to the data metadata. For the time-aligned vital sign data, an initial physiological state vector (IPSV) is generated based on a preset set of core vital signs (including heart rate, systolic blood pressure, diastolic blood pressure, mean arterial pressure, blood oxygen saturation, end-tidal carbon dioxide partial pressure, tidal volume, anesthetic gas concentration, and EEG-derived anesthetic depth proxy), basic physiological parameters (such as basal metabolic rate, body temperature, resting heart rate, and resting respiratory rate), and individual patient characteristics (such as liver and kidney function classification, drug sensitivity labels, and body composition parameters). This IPSV is a multidimensional structure with components arranged in a fixed order. It is supplemented with the generation time, the version number of the data source used, and a quality label summary. The quality label summary is represented by the statistical results of the quality labels of each component within the time period, reflecting the overall data credibility of the IPSV. The system parses discrete event records in the surgical event stream. This stream, provided by the operating room information system or anesthesia record system, includes the event type (skin incision, traction, position adjustment, electrosurgical energy release, etc.), the time of occurrence, and the duration (if any). These event types are categorized and coded, and the standard code, description, reference stimulus intensity, and reference duration for each type of event are recorded in the internal event dictionary. For each discrete event, a corresponding stimulation weight and duration parameter are assigned based on the event dictionary. The stimulation weight is used to characterize the intensity of the event's impact on the patient's physiological state and is usually determined by clinical expert scoring or statistical regression methods based on historical data. For example, the stimulation weight for skin incision events is high, for traction it is medium, and for position adjustment it is low. The duration parameter is the expected duration of the event's impact and can be derived from actual records or the default value in the event dictionary. Each discrete event is mapped along the time axis into a continuous stimulation signal segment with start and end times and intensity. After all event signal segments are superimposed on the time axis, a complete stimulus intensity curve is generated; When generating stimulus intensities, the system calculates the instantaneous stimulus intensity using a maximum intensity composition rule or a weighted summation rule for the overlapping portions of different events. Smoothing is performed at the boundaries between event starts and ends (e.g., using a weighted moving average or cubic spline smoothing) to eliminate numerical oscillations caused by sudden changes. The final output stimulus intensity curve is a time-aligned continuous sequence of values, which, together with the initial physiological state vector, serves as the driving input for the anesthesia simulation calculation.

[0024] The specific implementation steps of the simulation strategy determination module are as follows: According to the surgical procedure type and patient individual parameters, the matching pharmacokinetic-pharmacodynamic model and the circulation, respiration and gas exchange coupling sub-model are loaded from the model library; Pharmacokinetic and pharmacodynamic models are used to simulate the absorption, distribution, metabolism, and excretion of different anesthetic drugs in the body, as well as their pharmacodynamic effects in the effector compartment (usually the central nervous system); The circulation submodel is used to describe the dynamic processes of the cardiovascular system, including cardiac output, blood pressure changes, and blood volume distribution; The respiratory submodel is used to describe respiratory-related variables such as ventilation, airway resistance, and lung compliance; Gas exchange submodels simulate the exchange of oxygen and carbon dioxide between the alveoli and the blood. These submodels are parameterized using data from clinical trials, individualized patient test results, and pharmacokinetic constants from drug instructions. When loading the model, the system maps individual parameters related to the patient's body composition, liver and kidney function, and baseline cardiopulmonary status to the model input, allowing for personalized corrections to drug metabolism rates, respiratory compliance, and other parameters. During initialization, boundary constraints for each model must be set, such as the physiological limits of drug concentration, the permissible blood pressure range for the circulatory system, and upper and lower limits for ventilation, to ensure that simulation results are within physiologically reasonable limits. Before the model-driven calculation begins, the system maps the initial physiological state vector, drug administration sequence, ventilation parameters, and stimulation intensity curve into model input drive signals. The initial physiological state vector provides vital signs and physiological parameters at the start of the simulation; the drug administration sequence includes the administration time point, dosage, and administration method (intravenous push, continuous infusion, or inhalation concentration change curve) of each anesthetic drug; ventilation parameters include tidal volume, respiratory rate, inspiration-expiration ratio, inspired oxygen concentration, etc.; and the stimulation intensity curve reflects the driving effect of external stimuli on the patient's physiological response during surgery. These input signals are transmitted to the driving port of each sub-model through the model interface function. The system establishes a dynamic calculation relationship from input to state to output, ensuring that variables between different sub-models can be exchanged in real time according to the physiological coupling relationship. For example, the blood flow output by the circulatory sub-model will affect the distribution volume calculation in the pharmacokinetic model, and the ventilation volume of the respiratory sub-model will affect the oxygen partial pressure and carbon dioxide partial pressure of the gas exchange sub-model, thereby further affecting the circulatory system and drug efficacy response. Perform recursive update calculations on each sub-model at a uniform time step (e.g., once per second or once every five seconds, depending on the simulation accuracy requirements); For the pharmacokinetic and pharmacodynamic models, the plasma drug concentration and the effect-sac drug concentration are updated at each time step. For the effect size calculation, the hypnotic response and the analgesic response are updated according to the relationship between the effect-sac concentration and the pharmacodynamic curve. The circulatory sub-model updates variables such as cardiac output, systolic blood pressure, diastolic blood pressure, mean arterial pressure, and peripheral resistance at each time step. The respiratory sub-model updates the tidal volume, airway pressure, and alveolar gas volume at the respiratory rate. The gas exchange sub-model updates the arterial oxygen partial pressure and carbon dioxide partial pressure according to the output of the respiratory sub-model. These variables are also affected by the stimulus intensity curve during the calculation process. For example, high-intensity stimulation may cause a short-term increase in heart rate and blood pressure fluctuation in the circulatory system, thereby changing the drug distribution rate and effect intensity. Finally, the system combines the state variables of each time step with the corresponding timestamp in chronological order to form an anesthetic state evolution vector. Each entry of the anesthetic state evolution vector contains the drug concentration, hypnotic and analgesic effect, circulatory state variables, ventilation state variables and gas exchange state variables at that moment, and is attached with the input parameter version number and model version identifier for traceable calculation. The anesthetic state evolution vector will serve as the basic data structure for subsequent dynamic simulation scenario construction and strategy evaluation, ensuring that the simulation of the anesthesia process can reflect the real dynamic changes under the pharmacokinetic and pharmacodynamic coupling mechanism.

[0025] Consider a 50-year-old male patient weighing 75 kg, 175 cm tall, with a Child-Pugh liver function class A, normal renal function, and ASA class II status, scheduled for laparoscopic cholecystectomy. Based on the procedure type (laparoscopic cholecystectomy) and individualized parameters (including body composition, liver and kidney function class, and baseline cardiopulmonary status), the system loads from the model library the pharmacokinetic and pharmacodynamic models of the combination medications that match this procedure and patient type, as well as the cardiovascular circulation sub-model, the pulmonary ventilation sub-model, and the alveolar gas exchange sub-model. During the model initialization phase, the clearance parameters in the pharmacokinetic model are fine-tuned based on the patient's liver function grade (Child-Pugh A); the volume of distribution parameters are corrected based on body weight and BMI; the lung compliance and airway resistance parameters of the respiratory sub-model are set based on the patient's preoperative pulmonary function test results; and the vascular compliance and basal peripheral resistance parameters of the circulatory sub-model are calibrated based on preoperative blood pressure and echocardiogram results. At the same time, safety boundary conditions are set. Subsequently, the system inputs the initial physiological state vector (including heart rate 72 beats / min, mean arterial pressure 92 mmHg, arterial oxygen saturation 99%, end-tidal carbon dioxide 38 mmHg, etc.), drug administration sequence (continuous infusion of anesthetic drugs), ventilation parameters (tidal volume 500 mL, respiratory rate 12 breaths / min, FiO2 0.5, I:E ratio 1:2), and stimulation intensity curve (events such as skin incision, perforation, and traction, as well as their intensity and duration) into the model interface; During the simulation, recursive calculations are performed with a uniform time step of 1 second. The pharmacokinetic model updates the plasma and effect chamber drug concentrations in real time, and calculates the corresponding hypnotic and analgesic responses; the circulatory sub-model changes cardiac output and blood pressure according to drug effects and surgical stimulation; the respiratory and gas exchange sub-model updates alveolar gas volume and blood gas partial pressure according to ventilation settings and patient lung compliance. When a skin incision event occurs, the stimulation intensity curve rises briefly, and the circulatory sub-model experiences a slight increase in heart rate and blood pressure fluctuations. The analgesic effect in the pharmacokinetic model is slightly insufficient. The simulation results show that if the anesthetic drug infusion rate is not adjusted, there may be a risk of a short-term increase in blood pressure beyond the limit. Finally, the system outputs the anesthetic state evolution vector, which includes the drug concentration, hypnotic effect size, analgesic effect size, heart rate, blood pressure, respiratory parameters, blood oxygen partial pressure and carbon dioxide partial pressure every second from the induction period to the maintenance period, and records the input parameter version number and model version identifier for strategy evaluation and risk analysis.

[0026] After the anesthetic state evolution vector is generated, the initial physiological state vector and the stimulation intensity curve are used as joint inputs to enter the deep learning simulation calculation process. The specific steps are as follows: A time alignment operation is performed on the initial physiological state vector and the stimulation intensity curve. That is, based on the timestamp information of the two, the data collected at different times are adjusted to a unified time step by resampling or interpolation, so that each time point can be matched. The aligned data is then normalized to map the values ​​of different physical quantities to a unified dimensionless interval to eliminate the influence of dimensional differences. Feature encoding is then performed to sequentially encode continuous data into time series feature vectors, convert categorical features into one-hot encoding or embedding vectors, and append time step information to the feature set, ultimately forming the input sequence for simulation operations. During the modeling phase, the system constructs a phased deep model, which consists of three main structures: an induction phase encoder, a maintenance phase temporal reasoning network, and a recovery phase decoder. The induction phase encoder is used to receive the initial state and early surgical stimulation data, and extract high-sensitivity features related to rapid changes in anesthesia depth; the maintenance phase temporal reasoning network uses long-short-term memory units or time convolution structures, which are specifically used to process state stability and fluctuation patterns within a longer time span; the recovery phase decoder focuses on predicting the process of drug efficacy disappearance and the recovery trajectory of physiological parameters. The three phases are connected by a state transfer mechanism, that is, when switching between phases, the final state of the previous phase is used as the initial condition of the next phase, and parameters are reset or smoothly transitioned when necessary to ensure the continuity and physical rationality of the prediction results of different phases. During simulation, the system recursively runs the model at a uniform time step (e.g., once per second). At each time step, the drug administration sequence and ventilation parameters are used as conditional inputs, and the predicted values ​​of the current anesthetic depth, circulatory indices (including heart rate, mean arterial pressure, etc.), and ventilation indices (including tidal volume, end-tidal carbon dioxide partial pressure, etc.) are updated in combination with the predicted state of the previous time step. These predicted values ​​not only reflect physiological responses but are also dynamically affected by input control variables and stimulation events. For example, when the stimulation intensity curve reaches a peak, the predicted values ​​of circulatory indices will fluctuate immediately. At the boundaries between the induction and maintenance phases, and between the maintenance and resuscitation phases, the system imposes continuity and consistency constraints. This means smoothing the state variables before and after the boundaries to avoid sudden changes in values. Simultaneously, the system checks whether the trends of key physiological indicators at phase transitions conform to actual clinical patterns (e.g., a stable depth of anesthesia at the end of induction and a gradually increasing level of consciousness at the beginning of resuscitation). Ultimately, the predicted values ​​for all time steps are combined to generate a dynamic simulation scenario. This scenario includes timestamps, trajectories of each state variable, markers for key events, and corresponding confidence beacons (which reflect the confidence level of the prediction result for that time step and are derived from the model's uncertainty estimate). This dynamic simulation scenario can be used for both pre-evaluation of anesthesia plans and preoperative training and intraoperative decision support.

[0027] After obtaining the dynamic simulation scenario, we first generate a multi-path medication and ventilation strategy that meets the constraints based on the procedure label, individualized parameters, and physiological state boundary conditions. The generation process is as follows: Read the surgical staging requirements and common stimulation intensity patterns corresponding to the procedure label, combine the patient's individual parameters with the physiological state boundary conditions, and determine the drug dosage limits (single bolus upper limit, continuous infusion rate upper and lower limits, total upper limit) and ventilation parameter range (tidal volume, respiratory rate, I / O ratio, and the allowable range of inspired oxygen concentration); on this basis, form alternative multi-path drug administration and ventilation strategies through constrained combination and step search. Each strategy clearly defines the drug administration timing, ventilation parameter sequence, and staging linkage rules, and comes with a strategy identifier, applicable staging, and safety boundary description. The above step search is performed at the step granularity defined in the semantic dictionary to ensure that all candidate strategies do not exceed the drug dosage limits and ventilation parameter ranges, and can automatically switch to the corresponding target interval in different stages; When constructing a risk-efficacy indicator scoring function, first determine the indicator set and target interval. The indicator set and its calculation caliber are as follows: Circulatory fluctuation amplitude is measured by the maximum deviation between mean arterial pressure and heart rate within the evaluation interval and the duration of the deviation. The specific steps for constructing the risk-efficacy indicator are as follows: The maximum deviation outside the target interval is calculated and the deviation duration is accumulated; the duration of hypoxia is measured by the cumulative time that the blood oxygen saturation is below the target lower limit; the probability of awakening delay is based on the time distribution of the anesthesia depth proxy reaching the target awakening threshold during the resuscitation period, and the probability of delay is obtained by statistics of multiple simulation trigger frequencies; the sedation maintenance coverage is measured by the proportion of time that the anesthesia depth proxy remains within the target interval during the induction and maintenance periods; the adequacy of the analgesic response is measured by the coverage relationship between the analgesic effect quantity and the stimulation intensity curve, specifically by comparing the analgesic effect quantity at each time step to see if it is not lower than the stimulation intensity requirement at the same time step and the cumulative satisfaction ratio; the degree of drug consumption minimization is inversely measured by the ratio of the cumulative dosage of each drug in the same strategy period to the upper limit allowed for the drug; Each of the aforementioned indicators is first standardized according to its respective target interval or threshold. The standardization rule is: reaching or exceeding the target is scored as excellent, approaching the target is scored as good, and exceeding the target is graded and deducted based on the degree of deviation. The indicators, target intervals, thresholds, and grading boundaries are uniformly stored in a semantic dictionary to ensure consistency across different cases and procedures. The risk-efficacy indicator scoring function uses the aforementioned standardized indicators as input, aggregates them according to preset weights or risk sensitivity, and outputs the sub-item scores and overall scores for each strategy at each stage and throughout the entire process. All weight values ​​and their sources (guideline recommendations, departmental consensus, or historical case regression results) are clearly recorded in the configuration for easy traceability. After inputting each multi-path medication and ventilation strategy into the dynamic simulation scenario, the system executes the full state evolution simulation with a unified time step, and performs time-phase sampling and recording. The time-phase sampling includes both the whole-segment statistics according to the surgical stages (induction period, maintenance period, and resuscitation period), as well as the fine-grained statistics of the fixed-length sliding window. Both are retained to take into account both the stage trend and the instantaneous fluctuation. During the simulation, the system calculates the instantaneous and cumulative values ​​of indicators such as the circulatory fluctuation amplitude, the duration of hypoxia, the sedation maintenance coverage, the adequacy of the analgesic response, and the degree of minimization of drug consumption in real time, and writes the results together with the timestamp, stage mark, input control quantity (medication timing, ventilation parameters), and key event mark (peak and change point from the stimulation intensity curve) into the strategy execution trajectory.

[0028] To form a strategy score set with confidence information, multiple uncertainty simulations were performed under the same strategy and the same dynamic simulation scenario. The sources of uncertainty included the allowable fluctuation range of model parameters, the interval of alternative values ​​of the low-confidence components in the initial physiological state vector, and the segment of the stimulation intensity curve derived from the default estimate. Each uncertainty simulation performs controlled perturbations on the above inputs within the allowable range and performs full-process calculations to obtain a set of values ​​for the same indicator under multiple simulations; The central estimate is defined as the representative statistic of the group of values ​​(using the median or mean, with the standard fixed in the configuration); the upper and lower bounds are defined as the range of quantiles for the group of values ​​at the agreed confidence level (for example, the lower and upper quantiles are used as the boundaries, with the specific percentiles specified in the configuration item); the abnormal probability is defined as the proportion of times in the group of values ​​that trigger an abnormal state. Abnormal states include circulatory fluctuations exceeding the preset safety limit, hypoxia duration exceeding the upper limit allowed, and awakening delay probability exceeding the threshold. Based on the width of the central estimate and the upper and lower bounds, and the probability of anomalies, they are mapped to confidence levels according to pre-defined grading rules (for example, a high confidence level corresponds to a narrow bound and a low probability of anomalies, a medium confidence level corresponds to a moderate bound or a medium probability of anomalies, and a low confidence level corresponds to a wide bound or a high probability of anomalies). The central estimate, upper and lower bounds, and confidence level for each strategy on each indicator are written into the scoring structure and combined with the strategy's sub-scores and overall score to generate a strategy score set. The strategy score set records the input parameter range, weight settings, semantic dictionary version, and simulation engine version used in the calculation, ensuring that scoring results across different time periods and cases are reviewable, comparable, and traceable.

[0029] The construction of the risk-effectiveness indicator scoring function first classifies the indicators according to their types, including time series indicators, fluctuation indicators, and probability indicators. For time series indicators, such as hypoxia duration, sedation maintenance coverage, and analgesia adequacy, a continuous time series of indicator values ​​is extracted from the strategy execution trajectory, and the corresponding target threshold is set. Subsequently, the entire sequence is scanned at a uniform time step. At each time step, it is determined whether the indicator value exceeds (or falls below) the target threshold. If it exceeds, the corresponding time step is accumulated to obtain the cumulative time length. If the integral area needs to be calculated, the difference between the indicator value and the threshold in the super-threshold segment is multiplied by the time step and accumulated to obtain the cumulative excess amount, which is the integral result of the time series indicator. For fluctuation indicators, such as cyclic fluctuation amplitude, the maximum and minimum values ​​of the indicator are first extracted within the specified evaluation window (the window length can be pre-configured by the procedure type), and the difference between them is calculated as the maximum deviation amplitude. The maximum deviation amplitude is then compared with the width of the standard expected interval of the indicator to obtain the deviation ratio. The standard expected interval is set by the normal fluctuation range of clinical guidelines or statistical historical cases to ensure consistency in calculation across different cases. A deviation ratio greater than 1 indicates that the fluctuation exceeds the normal range, while a deviation ratio less than or equal to 1 indicates that the fluctuation is within the acceptable range. For probabilistic indicators, such as the probability of awakening delay, the number of triggering abnormal states (e.g., failure to reach the awakening threshold at the scheduled resuscitation time point) is counted in multiple uncertainty simulation results, and the number of triggering times is divided by the total number of simulations to obtain the abnormal probability of occurrence of the indicator. The abnormality judgment conditions are clearly defined by the threshold configuration file of the corresponding indicator. For example, the judgment of awakening delay can be based on whether the BIS value (bispectral index value) or other anesthesia depth proxy reaches the preset awakening threshold; After completing the original calculation of each indicator, each type of indicator is standardized. The standardization rules are as follows: for indicators that need to be minimized (such as circulatory fluctuation amplitude and duration of hypoxia), when the indicator value is less than or equal to the target value, it is recorded as full score, and when the indicator value is greater than the target value, it is deducted according to the excess ratio; for indicators that need to be maximized (such as sedation maintenance coverage and analgesia response adequacy), when the indicator value is greater than or equal to the target value, it is recorded as full score, and when the indicator value is lower than the target value, it is deducted according to the insufficient ratio; for probability indicators, when the probability of abnormality is lower than the set upper limit, it is recorded as full score, and when it is higher than the upper limit, it is deducted according to the excess ratio; After standardization is completed, the indicator weight vector is introduced, and the standardized score of each indicator is linearly or nonlinearly weighted combined with the corresponding weight to obtain the comprehensive score result of each strategy. The value of the weight vector can be based on three sources: clinical expert experience, guideline recommendations or statistical results of historical big data regression analysis. The source category and specific values ​​are clearly marked in the configuration file. For nonlinear weighting, segmented weighting or multiplication factor method can be used. For example, exponential weights are given to key risk indicators to increase sensitivity.

[0030] The scoring results for each strategy are categorized into risk and efficacy subsets based on their indicator attributes. Detailed sub-scores for risk and efficacy indicators are bundled with the scoring results to facilitate subsequent sorting and strategy selection. Finally, the scoring results record the input parameter ranges used for the calculation (e.g., dosing limits, ventilation parameter ranges), the scoring timestamp, and the version number of the scoring function used. The central estimate, upper and lower bounds, and confidence level are stored together in a risk-efficacy score structure.

[0031] Generate a scoring structure for each risk and performance indicator, including a central estimate, upper and lower bounds, and confidence levels. The specific steps are as follows: First, during the path simulation process, sub-parameter perturbation sampling or belief propagation calculations are performed on each indicator of each strategy. Sub-parameter perturbation sampling refers to the process of performing multiple perturbations on the secondary input parameters that affect the indicator calculation (including individual drug sensitivity coefficients, metabolic half-life time, delayed response time of ventilation equipment, and noise amplitude of vital sign sampling, etc.) within their uncertainty range without changing the main control parameters of the strategy (such as route of administration, basal dose, and ventilation mode). The simulation is then rerun under each set of perturbation values. The perturbation range of each set is determined by the statistical distribution of historical cases, equipment technical specifications, and clinical safety boundaries. For example, the perturbation range of the drug sensitivity coefficient can be ±15% of the patient's individual baseline value. The belief propagation calculation is applicable to input variables for which an error model has been established. The fluctuation range of the result is obtained by transferring the uncertainty distribution of the input to the output indicator distribution. After completing the above-mentioned perturbation sampling or belief propagation calculation, a set of results for each indicator under multiple simulations will be obtained. This result set will be statistically analyzed, and the arithmetic mean of all simulation results will be extracted as the central estimate of the indicator. This central estimate represents the average response level of the indicator under given strategy and input conditions. At the same time, calculate the maximum and minimum values ​​in all simulation results, and subtract the minimum value from the maximum value to obtain the width of the numerical interval. Then use the minimum value as the lower confidence bound and the maximum value as the upper confidence bound to ensure that the interval covers all possible values ​​of the indicator under the uncertainty conditions considered. When calculating the probability of abnormality, the system will count the number of abnormal states in the above multiple simulation results. The rules for determining abnormal states are given by the risk definition file corresponding to the indicator. For example, the abnormal state of hypoxia risk can be defined as the blood oxygen saturation being lower than 90% for more than 10 seconds continuously, the circulatory risk can be defined as the mean arterial pressure being lower than 60 mmHg for more than 15 seconds, and the awakening delay risk can be defined as the awakening threshold not being reached at the preset resuscitation time point (such as BIS value ≥ 80). After counting the number of abnormalities, divide it by the total number of simulations to obtain the abnormal probability of occurrence of the indicator, which ranges from 0 to 1. After obtaining the central estimate, upper and lower confidence bounds, and the probability of anomaly occurrence, the system maps the indicators to confidence level labels based on predefined confidence level classification criteria. Confidence levels can be divided into three levels: high, medium, and low. For example, when the probability of anomaly occurrence is less than 5% and the confidence interval width accounts for less than 10% of the central estimate, it is marked as high confidence. When the probability of anomaly occurrence is between 5% and 15%, or the confidence interval width accounts for between 10% and 25%, it is marked as medium confidence. If it exceeds these ranges, it is marked as low confidence.

[0032] Finally, the central estimate, upper and lower confidence bounds, anomaly probability, and confidence level label of each indicator are recorded in the scoring structure of the corresponding strategy. The scoring structure is stored in the form of structured data, including the strategy identifier, indicator name, central estimate, upper and lower confidence bounds, anomaly probability, confidence level, calculation timestamp, and the perturbation parameter configuration version number used.

[0033] It should be noted that the confidence level standard is formulated by a clinical expert group based on surgical risk management requirements and simulation system performance.

[0034] The specific implementation steps of the solution generation module are: Call the previously generated strategy scoring set, which contains the confidence score structure for each strategy, including fields such as the central estimate, upper and lower confidence bounds, anomaly probability, and confidence level label; When receiving a new round of simulation or historical cumulative scoring data, a ranking calculation is performed to determine the strategy priority; During the ranking calculation process, a weighted comprehensive evaluation is performed on each indicator in the scoring structure based on the user-set efficacy target weight and risk sensitivity parameter. The efficacy target weight is used to reflect the importance of goals such as sedation maintenance, analgesia, and drug consumption control during surgery. Its value is preset by the clinical operator on the system interface and is weighted to amplify efficacy indicators during ranking. The risk sensitivity parameter is used to reflect the tolerance for safety events such as circulatory stability, hypoxia risk, and delayed awakening. The higher its value, the greater the weight the system assigns to risk indicators in the ranking, thereby prioritizing the elimination of potentially high-risk strategies. After completing the weighted evaluation, the system will sort the strategies in descending or ascending order (depending on whether the user prioritizes low risk or high efficiency) and enter the high-risk event monitoring phase. During this process, the system will scan the indicator items marked as "high probability of abnormality" or "confidence boundary exceeded" in each strategy scoring structure. "High probability of abnormality" means that the probability of abnormal occurrence of the indicator exceeds the clinical warning threshold, such as the risk of hypoxia exceeding 20%; "confidence boundary exceeded" means that the confidence interval width of the indicator exceeds the set proportion threshold, such as the confidence interval width exceeds 30% of the central estimate. The system will determine whether these indicators meet the definition of high-risk events. High-risk events include but are not limited to: sedation interruption (sedation level indicators are lower than the set lower limit and continue for more than the clinically allowed time), hypoxia threshold (blood oxygen saturation is lower than 90% and lasts for more than 10 seconds) or delayed awakening (the awakening standard is not met at the preset resuscitation time point); Once the relevant indicator value of any high-risk event is detected to exceed the preset risk threshold, the system will automatically identify the input variables and control points corresponding to the indicator; Input variables include adjustable parameters directly related to the indicator, such as intravenous injection rate, inhaled anesthetic gas concentration, tidal volume setting, respiratory rate, etc. Control points refer to the time window on the surgical timeline that is most relevant to the adjustment of the input variable, such as the start point of drug administration during the induction phase, the circulatory load fluctuation period during the maintenance phase, or the ventilator mode switching point during the resuscitation phase. After identifying the input variables and control points that trigger the event, the system will generate interactive anesthesia plan deduction results based on the sorting results and high-risk event information. The deduction results not only include a strategy sorting list, but also generate a set of scenario-based visualization plans, including control nodes (specific adjustment parameters and their adjustment ranges), key time periods (expected time ranges for intervention), indicator trajectories (predicted curves of indicator changes over time), and key monitoring parameters (such as vital signs that require real-time attention and their thresholds). In addition, the deduction results will also be accompanied by scenario handling plans, clarifying the response measures when risk indicators reach different levels, such as increasing oxygen concentration and adjusting ventilation mode when the risk of hypoxia increases, supplementing sedatives and monitoring circulatory status when the risk of sedation interruption occurs, etc. All simulation results and scenario handling plans will be stored in a structured format and bound to the unique identifier of the corresponding strategy. At the same time, the generation timestamp, risk threshold configuration version number and weight parameters used in the calculation will be recorded to ensure traceability and version consistency of the results during subsequent review and optimization.

[0035] In summary, the present invention realizes the full dynamic simulation of surgical procedure characteristics, individual physiological differences and exogenous stimulation disturbances during the induction period, maintenance period and recovery period by constructing an anesthesia process simulation system driven by the collaborative efforts of the pharmacokinetic-pharmacodynamic model and the physiological coupling model. It completes the nonlinear mapping of the physiological state vector and the stimulation intensity curve based on the deep learning framework, generates a continuous state evolution path, and significantly improves the refinement and scenario adaptability of the anesthesia process prediction. Compared with the traditional scheme adjustment method based on fixed rules or staged response, the present invention supports the simultaneous deduction of state trajectory and risk factor response curve under multi-path drug administration and ventilation strategies, has the ability to identify and adapt to strategic risks under complex intraoperative changes in advance, and enhances the level of intelligent response to dynamic changes in the anesthesia process. In addition, an indicator scoring function for risk-effectiveness balance is introduced, which integrates multi-source indicators such as time accumulation, state deviation and abnormal probability to form a scoring structure with upper and lower boundaries and confidence levels. A strategy scoring set sorting and high-risk scenario warning mechanism are constructed to achieve the linkage push of strategy optimization and key monitoring parameters. It supports the generation of strategy scoring sets under the constraints of individualized parameters and surgical requirements, and outputs interactive anesthesia plans containing deduction results, parameter ranges and scoring labels. Finally, an intelligent optimization and analysis system integrating simulation, evaluation and feedback closed loop is constructed to effectively improve the rationality of anesthesia plans under clinical pathways and the ability to ensure intraoperative safety.

[0036] The above embodiments may be implemented in whole or in part through software, hardware, firmware or any other combination. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0037] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0038] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0039] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0040] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An optimization analysis system for simulating anesthesia process, characterized by: It includes an anesthesia data acquisition module, a simulation scenario construction module, a simulation strategy determination module, and a solution generation module, and each module is connected through data communication; Anesthesia data acquisition module, which is used to construct a multi-source vital sign tensor structure based on historical case data and patient-specific parameters, generate an initial physiological state vector through a unified time reference and quality label, and establish a stimulation intensity curve based on the surgical event stream; The simulation scenario construction module is used to construct the anesthetic state evolution vector based on the pharmacokinetic and pharmacodynamic model and the physiological coupling model. It also performs deep learning simulation calculations based on the combined input of the initial physiological state vector and the stimulation intensity curve to generate dynamic simulation scenarios for the induction period, maintenance period, and recovery period. The simulation strategy determination module is used to generate multi-path medication and ventilation strategies for different surgical procedures and individual constraints, construct a risk-effectiveness indicator scoring function, and perform full-process state tracking and indicator calculation in the simulation scenario to form a strategy score set with confidence information; The plan generation module is used to continuously analyze the strategy score sets of multiple paths and perform sorting and judgment, and generate interactive anesthesia plan deduction results based on high-risk trigger conditions. When the preset risk threshold is met, the plan instruction set containing key monitoring parameters and scenario handling steps is output.

2. The optimization analysis system for simulating anesthesia process according to claim 1, characterized in that: Used to construct a multi-source vital sign tensor structure based on historical case data and patient-individualized parameters, including: Demographic information, surgical procedure labels, preoperative evaluation records, organ function indicators and perioperative monitoring records were extracted from historical case data. The patient's individualized parameter group includes body composition parameters, liver and kidney function classification, basic cardiopulmonary status, and medication sensitivity labels for feature alignment; Vital sign data are organized according to the time dimension, signal channel dimension, and source dimension. The time dimension uses a sequence with a uniform time step. The signal channel dimension includes EEG-derived depth, arterial pressure waveform, heart rate, blood oxygen saturation, end-tidal carbon dioxide, tidal volume, and anesthetic gas concentration. The source dimension identifies monitoring equipment, infusion pumps, ventilators, and information systems. Perform dimension unification and interval normalization on numerical signs, encode categorical parameters and write them into semantic dictionary, and generate baseline reference vector based on individualized parameters and append it to the first frame of tensor; According to the quality label, the missing or discontinuous fragments are gap-filled or the missing marks are retained and the confidence field is generated. The traceability index and version number are generated for the tensor entry. The processed vital sign data are sorted by timestamp and bound to the traceability index, version identifier and data source identifier, and the multi-source vital sign tensor structure for simulation calculation call is output.

3. The optimization analysis system for simulating anesthesia process according to claim 2, characterized in that: The initial physiological state vector is generated by unifying the time base and quality label, and the stimulation intensity curve is established in combination with the surgical event flow, including: Obtain the original timestamp information of each signal channel in the multi-source vital sign tensor structure, and extract the time reference parameters of the monitoring equipment, infusion pump, ventilator and event log; Establish a mapping relationship between device time and the system's unified clock, using the surgery start mark, device heartbeat signal, and manual key events as time anchor points to correct time drift across data sources. Resampling or interpolation alignment of the corrected multi-source vital sign sequences according to a uniform time step; Select core vital signs, basic physiological parameters, and individualized features from the time-aligned vital sign data to form an initial physiological state vector, and attach the generation time, source version, and quality label summary to the initial physiological state vector; Parse discrete event records in surgical event streams to identify the types and times of skin incision, traction, body position adjustment, and electrosurgical energy release events; Corresponding stimulation weights and duration parameters are assigned to each type of discrete event, and the discrete events are mapped into continuous stimulation intensity curves along the time axis, and the curve boundaries are smoothed.

4. The optimization analysis system for simulating anesthesia process according to claim 3, characterized in that: Used to construct an anesthetic state evolution vector based on the pharmacokinetic and pharmacodynamic model and the physiological coupling model, including: Load the pharmacokinetic and pharmacodynamic model that matches the procedure type and individual parameters, as well as the circulation, respiration, and gas exchange coupling sub-model, and initialize the model parameters and boundary constraints; Map the initial physiological state vector, drug administration sequence, ventilation parameters and stimulation intensity curve as model-driven inputs, and establish a computational relationship from input to state to output; Recursively update plasma concentrations and effect-spatter concentrations, hypnotic and analgesic response quantities, and circulation and ventilation state variables at a uniform time step; The state variables and timestamps of each time step are combined in chronological order to generate the anesthesia state evolution vector.

5. The optimization analysis system for simulating anesthesia process according to claim 4, characterized in that: Based on the combined input of the initial physiological state vector and the stimulation intensity curve, deep learning simulation calculations are performed to generate dynamic simulation scenarios for the induction, maintenance, and recovery periods, including: Perform time alignment, normalization, and feature encoding on the initial physiological state vector and the stimulation intensity curve to form a simulation input sequence; Construct a staged deep model, set up an induction phase encoder, a maintenance phase temporal reasoning network, and a recovery phase decoder, and establish a state transfer mechanism between phases; Recursively calculate at a uniform time step and output the predicted values ​​of anesthesia depth, circulation index and ventilation index at each time step under the conditions of injection drug sequence and ventilation parameters; Continuity and consistency constraints are imposed on stage boundaries, and dynamic simulation scenarios containing timestamps, state trajectories, and event markers are synthesized and attached with beacons.

6. The optimization analysis system for simulating anesthesia process according to claim 5, characterized in that: It is used to generate multi-path medication and ventilation strategies for different surgical procedures and individual constraints, construct a risk-effectiveness indicator scoring function, and perform full-process state tracking and indicator calculation in simulation scenarios to form a strategy scoring set with confidence information, including: Obtain procedure labels, individualized parameters, and physiological state boundary conditions, and generate multi-pathway drug delivery and ventilation strategies that meet the constraints based on drug dosage limits, ventilation parameter ranges, and surgical staging requirements. A risk-effectiveness scoring function was constructed for each strategy. Risk indicators included circulatory fluctuation amplitude, duration of hypoxia, and probability of delayed awakening; effectiveness indicators included sedation maintenance coverage, analgesic response adequacy, and drug consumption minimization. Each strategy is input into a dynamic simulation scenario to simulate state evolution, and the indicators of each stage are sampled and recorded in different time periods to generate a complete strategy execution trajectory. The confidence interval or abnormal probability is calculated for each indicator to form a scoring structure including the central estimate, upper and lower boundaries and confidence level.

7. The optimization analysis system for simulating anesthesia process according to claim 6, characterized in that: The process of constructing the risk-effectiveness indicator scoring function includes: For time series indicators, an integral function is used to calculate the cumulative duration or integral area of ​​each indicator value exceeding the set threshold; For volatility indicators, the maximum deviation range of the indicator within the evaluation window is extracted and the ratio is calculated with the standard expected interval to obtain the deviation ratio; For probabilistic indicators, the probability of abnormal occurrence is generated by counting the trigger frequency of abnormal states; Based on the standardized scoring, an indicator weight vector is introduced, and each indicator score and its corresponding weight are linearly or nonlinearly weighted combined to obtain the scoring result of each strategy; The scoring results of each strategy are classified into risk category and effectiveness category respectively, and risk score subsets and effectiveness score subsets are generated; The scoring result records the input parameter range used for calculation, the scoring timestamp, and the version number of the scoring function used, forming a risk-performance scoring structure that includes the central estimate, upper and lower bounds, and confidence levels.

8. The optimization analysis system for simulating anesthesia process according to claim 6, characterized in that: Calculate the confidence interval or abnormal probability for each indicator to form a scoring structure containing the central estimate, upper and lower bounds, and confidence levels, including: During the path simulation process, subparameter perturbation sampling or belief propagation calculation is performed on each risk and efficiency indicator of each strategy to obtain the fluctuation range of the indicator under different input uncertainties; The central estimate of each indicator was extracted as the average response level, and the numerical interval between the maximum and minimum values ​​was calculated as the upper and lower confidence limits; Based on the trigger frequency of abnormal states during the simulation process, the abnormal occurrence probability of each indicator is calculated; The central estimate, upper and lower bounds, and anomaly probability of each indicator are mapped into confidence level labels and recorded in the corresponding strategy scoring structure.

9. The optimization analysis system for simulating anesthesia process according to claim 8, characterized in that: It is used to continuously analyze and rank the strategy score sets of multiple paths, generate interactive anesthesia plan deduction results based on high-risk trigger conditions, and output key monitoring parameters and scenario treatment plans when the preset risk threshold is met, including: Perform ranking calculation on the confidence score structure of each strategy in the strategy scoring set, and perform weighted evaluation based on the set effectiveness target weight and risk sensitivity parameter; Monitor indicators marked as high abnormal probability or confidence limit exceeded in the scoring structure to determine whether high-risk events have been triggered. High-risk events include sedation interruption, hypoxia threshold, or delayed awakening risk types; When any high-risk event exceeds the preset risk threshold, the input variables and control points corresponding to the indicators are automatically identified; Based on the sorting results and trigger event information, the interactive anesthesia plan deduction results are generated, and the scenario disposal plan including control nodes, key time periods, indicator trajectories and monitoring parameters is output.

Images