Intelligent state analysis method for anesthetized patients based on physiological data

Abstract

This invention discloses an intelligent state analysis method for anesthetized patients based on physiological data, comprising the following steps: S1, organizing and annotating historical anesthesia data to form executable state recognition rules and establishing a standardized historical case reference library. The beneficial effects of this invention are: by decomposing key interactive substructures such as nerve-circulation, brain-circulation, and respiration-perfusion from dynamic network profiles, coordinating indicators with clear physiological significance, such as baroreflex function index, brain oxygen-blood pressure coupling coefficient, and ventilation-perfusion matching degree, are calculated; these indicators can quantify the coordination of physiological systems under anesthesia from multiple dimensions; the system performs multi-dimensional similarity matching of these real-time indicators with an internalized anesthesia state knowledge base (containing multiple preset state modes), realizing accurate and interpretable independent determination of the anesthesia depth state in three dimensions: sedation, analgesia, and muscle relaxation, overcoming the limitations of traditional single index monitoring.

Description

Technical Field

[0001] This invention relates to the field of anesthesia patient status analysis technology, specifically to an intelligent anesthesia patient status analysis method based on physiological data. Background Technology

[0002] In clinical anesthesia, monitoring the patient's anesthetic status is a core element in ensuring surgical safety, optimizing anesthesia management, and improving patient outcomes. Anesthesia status encompasses three independent yet interconnected dimensions: sedation, analgesia, and muscle relaxation. Accurate assessment and dynamic control of this status directly impact the smooth execution of the surgical procedure, as well as the patient's intraoperative safety and postoperative recovery quality. Over-anesthesia can lead to complications such as respiratory and circulatory depression and delayed postoperative awakening, while insufficient anesthesia may result in increased patient awareness and stress response during surgery, increasing surgical risks. Currently, mainstream clinical methods for monitoring anesthetic status primarily rely on single physiological indicators or subjective assessment, which have several limitations. Firstly, traditional monitoring methods often depend on single physiological parameters (such as heart rate, blood pressure, and bispectral index (BIS)). While BIS is a commonly used indicator for monitoring sedation depth, it only reflects the sedation dimension and cannot simultaneously address analgesia and muscle relaxation. Furthermore, it is susceptible to factors such as anesthetic drugs, individual patient differences, and intraoperative electrical interference. The current methods of monitoring are subject to various factors, including insufficient accuracy and stability. Furthermore, a single indicator cannot reflect the coordinated relationship between multiple physiological systems such as the nervous, circulatory, and respiratory systems, making it difficult to comprehensively characterize the patient's overall anesthetic state. On the other hand, existing anesthetic state assessments rely heavily on the anesthesiologist's clinical experience, which is highly subjective, with varying assessment standards among different doctors. This is especially true in complex surgeries (such as surgery on elderly patients, obese patients, or prolonged large surgeries), where strong intraoperative stimuli (such as skin incision, bone cement implantation, or visceral traction) can cause drastic fluctuations in the patient's physiological indicators. Traditional monitoring methods struggle to quickly capture trends and accurately distinguish the causes of these fluctuations (insufficient anesthesia depth, surgical stimulation, or the influence of underlying medical conditions), leading to delayed warnings and untimely interventions. The visualization methods of existing systems are also relatively simplistic, often presenting numerical values ​​or simple curves, which fail to intuitively reflect the dynamic changes and abnormal locations of the patient's anesthetic state, thus failing to provide anesthesiologists with efficient and intuitive decision support. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent state analysis method for anesthetized patients based on physiological data, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent state analysis of anesthetized patients based on physiological data, comprising the following steps: S1. Organize and annotate historical anesthesia data to form executable status recognition rules and establish a standardized historical case reference library; S2. Core vital signs signals are collected synchronously through a multi-mode sensor network. After preprocessing, frequency domain and nonlinear multidimensional features are extracted and fused to form a high-dimensional physiological feature set. S3. Construct a dynamic relationship network based on high-dimensional physiological characteristics, quantify system coupling and synchronization, and generate a multi-system coordination state profile. S4. Decompose key interaction substructures from dynamic network profiles and calculate interpretable physiological coordination indices. S5. Match the real-time extracted coordination index with the internalized anesthesia status knowledge base, and determine the patient's current sedation, analgesia, and muscle relaxation status in real time through multi-dimensional similarity calculation. S6. Map the identified anesthesia state to a three-dimensional coordinate system of sedation-analgesia-muscle relaxation and present it in the form of a dynamic light sphere; S7. Based on the individual characteristics of the patient, search for similar cases in the historical case database, extract effective anesthesia intervention strategies, and generate personalized anesthesia plans and expected physiological trajectory reference baselines; S8. Combine different stages of surgery and strong stimulating events to dynamically adjust the personalized reference baseline and set dynamic early warning thresholds for the absolute value and trend of the fusion index. S9. During the operation, continuously compare the actual physiological indicators with the personalized reference baseline and warning boundary, and dynamically optimize the status recognition rules and warning parameters according to the deviation pattern to achieve system adaptive evolution. S10. Continuously outputs an individualized monitoring pathway that reflects the patient's current and predicted physiological status; S11. Based on the optimized warning boundary and status determination results, drive the visual status update of augmented reality warning information and three-dimensional dynamic dashboard in real time. S12, integrated individualized monitoring pathways, real-time early warning information, and personalized anesthesia plans.

[0005] Preferably, step S1 specifically includes the following steps: S11. Clean and normalize the physiological data of historical anesthesia surgeries, and standardize and convert the historical medication information. S12. Map the expert evaluation conclusions to preset anesthesia state category labels to form a time-aligned structured historical dataset; S13. Based on the structured historical dataset, extract the physiological feature combination pattern associated with a specific anesthesia state category. The feature combination pattern is defined by the co-occurrence and synergistic change relationship of a set of specific physiological indicators in the time-frequency domain. S14. The extracted feature combination pattern is transformed into an executable anesthesia state recognition rule, which constitutes the internalized anesthesia state recognition logic. S15. Based on the structured historical dataset, extract individualized features of cases, intraoperative physiological trajectories, anesthesia intervention strategies and corresponding outcomes, and construct a standardized historical data reference system; S16. When the surgery is started, load the generated anesthesia state identification logic and the constructed standardized historical data reference system.

[0006] Preferably, step S2 specifically includes the following steps: S21. Sensor nodes are deployed on the patient's body surface through non-invasive flexible electronic patches to automatically and continuously collect core physiological signals, including heart rate, stroke blood pressure, respiratory waveform, prefrontal cortex oxygenation, and skin microcirculation perfusion index, and then generate a synchronous multi-channel raw signal stream. S22. Perform real-time preprocessing on the multi-channel raw signal streams acquired in step S21, including filtering each raw signal to eliminate power frequency interference and motion artifacts, and performing signal quality assessment and bad segment marking. S23. For the signal preprocessed in step S22, extract high-order spectral features in real time by channel. The high-order spectral features include power spectral density, centroid frequency and signal complexity entropy value of a specific frequency band. S24. For the signal preprocessed in step S22, extract nonlinear dynamic parameters in real time by channel. The nonlinear dynamic parameters include approximate entropy, sample entropy and multi-scale entropy. S25. The high-order spectral features extracted in step S23 and the nonlinear dynamic parameters extracted in step S24 are fused by channel and time sequence to generate a high-dimensional original physiological feature set.

[0007] Preferably, step S3 specifically includes the following steps: S31. Using the high-dimensional original physiological feature set generated in step S2 as input, define each physiological indicator as a graph node, calculate the time-varying mutual information and lag correlation between different physiological indicator sequences, use them as dynamic connection weights between nodes, and construct a dynamic coupling relationship graph. S32. Calculate the instantaneous phase difference of the connected node pairs in the dynamic coupling graph, calculate the phase synchronization index based on the instantaneous phase difference, quantify the synchronicity strength between different physiological oscillations, and update the weights of the corresponding connections in the graph. S33. Apply a graph neural network to the dynamic coupling relationship graph constructed in steps S31 and S32 to perform feature propagation and aggregation, learn the aggregated representation of each node and its neighborhood in the graph, and capture the topological structure of the dynamic coupling relationship graph. S34. Based on the node representations learned in step S33, cluster and pool the node representations belonging to the same physiological subsystem of neural, circulatory, and metabolic systems to generate feature vectors that characterize the internal state and inter-system interaction coordination of each physiological subsystem. S35. Aggregate the feature vectors generated in step S34 to construct a dynamic association network profile. Each dimension of the dynamic association network profile corresponds to a learned global or subsystem feature that reflects the coordination state between multiple systems.

[0008] Preferably, step S4 specifically includes the following steps: S41. Based on the physiological system definition, segment the subgraph structure representing the neural-circulatory, brain-circulatory, and respiratory-perfusion interactions from the real-time dynamic association network image constructed in step S3. S42. For each subgraph structure segmented in step S41, calculate the average connection weight and clustering coefficient between internal nodes as the original quantitative value of the interaction strength between the corresponding physiological systems. S43. In the neural-circulatory interaction subgraph structure, identify the node pairs representing heart rate and blood pressure, calculate the power of their interaction intensity in a specific frequency band, and output the standardized power as the baroreflex function index. S44. In the brain-circulation interaction subgraph structure, identify the node pairs representing brain oxygen saturation and blood pressure, and calculate the coherence coefficient of their interaction intensity sequence as the brain oxygen-blood pressure coupling coefficient. S45. In the respiratory-perfusion interaction subgraph structure, identify the node pairs representing the respiratory waveform and the skin perfusion index, calculate the mutual information entropy of their interaction intensity sequence, and output the normalized result as the ventilation-perfusion matching degree. S46. The baroreflex function index calculated in step S43, the cerebral oxygen-blood pressure coupling coefficient calculated in step S44, and the ventilation-perfusion matching degree calculated in step S45 are combined to form an interpretable set of physiological coordination indicators, providing feature input for real-time determination of anesthesia depth.

[0009] Preferably, step S5 specifically includes the following steps: S51. Based on the internalized anesthesia state recognition logic loaded in step S1, extract a variety of preset anesthesia state modes, each of which is defined by a set of preset interpretable physiological coordination index values ​​and allowable fluctuation range. S52. Construct a real-time index vector from the set of interpretable physiological coordination indices generated in step S4, and perform a one-to-one similarity calculation between the real-time index vector and the anesthesia state pattern extracted in step S51. S53. For each anesthesia state mode, calculate the multi-dimensional Euclidean distance or cosine similarity between the real-time index vector and the feature value defined by the mode, as the matching degree between the real-time data and the mode. S54. When the matching degree of any anesthesia state mode exceeds the preset judgment threshold, the patient is determined to be in that anesthesia depth state. If the matching degree of multiple modes exceeds the threshold, the mode with the highest matching degree is selected as the judgment result. If no mode has a matching degree exceeding the threshold, it is determined to be an unknown or transitional state. S55. Output the determination result of step S54 as the patient's current anesthesia depth status, which includes at least the depth grading of sedation, analgesia, and muscle relaxation dimensions.

[0010] Preferably, step S6 specifically includes the following steps: S61. Map the anesthesia depth state identified in step S5 to a Cartesian coordinate system consisting of three dimensions: sedation, analgesia, and muscle relaxation. The value of each dimension represents the depth level of that component, which together determine a unique spatial coordinate point in the coordinate system. S62. In the three-dimensional coordinate system, a dynamic light sphere is generated with the spatial coordinate point determined in step S61 as the center. The color, brightness and radius of the light sphere are dynamically adjusted according to the position and rate of change of the coordinate point and are displayed on the dashboard interface in real time. S63. Continuously analyze the time series of the interpretable physiological coordination index generated in step S4, calculate its short-term trend and acceleration, and identify whether the index shows a continuous deviation pattern toward the abnormal state region. S64. When the deviation pattern identified in step S63 meets the preset conditions, it is predicted that the dynamic light sphere will move into the warning area within a specific time window in the future, and a warning of the corresponding level is triggered according to the prediction result. S65. When an alert is triggered, an augmented reality device is used to overlay and generate a holographic wireframe diagram of the patient's physiological state in the doctor's field of vision. The wireframe diagram outlines the human body with semi-transparent wireframes and marks the location of major organ systems. S66. In the holographic wireframe diagram, the activity of the cardiovascular system is mapped by pulsating red light flow, the ventilation status of the respiratory system is mapped by blue ripples, and the intensity of central nervous system activity is mapped by the flashing frequency and spectral color of the head region; when the indicators of step S4 or the status of step S5 are abnormal, characteristic bright flashing and color alarms are triggered at the corresponding anatomical location in the wireframe diagram.

[0011] Preferably, step S7 specifically includes the following steps: S71. Based on patient physical parameters, cardiopulmonary function classification, and genomic drug metabolism type information, construct a patient-specific feature vector; S72. Perform multi-dimensional similarity calculation between the patient-individualized feature vector constructed in step S71 and the case features in the standardized historical data reference system loaded in step S1, retrieve the preset number of historical cases with the highest similarity, and form the set of the most similar historical cases. S73. From the set of most similar historical cases obtained in step S72, extract the recorded types, dosages, and timing of anesthesia induction and maintenance drugs, as well as the corresponding intraoperative physiological responses and final outcomes, as a verified and effective anesthesia intervention strategy. S74. Based on the validated and effective anesthesia intervention strategies extracted in step S73, with the optimization goal of maintaining physiological stability, a recommended drug infusion plan for the current patient is derived to form a personalized anesthesia plan. S75. Based on the set of most similar historical cases obtained in step S72, the average trajectory and standard deviation range of intraoperative physiological state indicators over time are statistically analyzed to serve as the expected personalized physiological state change path and fluctuation range for the current patient. S76. The personalized physiological state change path and fluctuation range obtained by modeling in step S75 are defined as the personalized monitoring reference baseline for intraoperative monitoring. S77. The personalized monitoring reference baseline defined in step S76 is associated and integrated with the personalized anesthesia plan generated in step S74 to form the personalized intraoperative monitoring framework, providing a benchmark for real-time intraoperative comparison.

[0012] Preferably, step S8 specifically includes the following steps: S81. Based on the surgical plan, divide the surgical process into different stages and mark the types and intensities of strong stimulating events expected to occur in each stage; S82. Based on the current and upcoming surgical stages and strong stimulation events marked in step S81, the personalized monitoring reference baseline defined in step S76 is corrected for temporal offset to generate a corrected dynamic reference baseline. S83. Using the corrected dynamic reference baseline generated in step S82 as the center, for each interpretable physiological coordination index, dynamically calculate the personalized warning upper and lower thresholds within a future time window. The upper and lower thresholds are usually determined based on the dynamic reference baseline value plus a fixed percentage or a preset absolute offset. S84. For each interpretable physiological coordination index, set an allowable range for the short-term rate of change. When the rate of change of the index continues to exceed this range, a trend deterioration warning is triggered. This rate of change threshold is used as the trend deterioration warning threshold. S85. The personalized early warning upper and lower thresholds calculated in step S83 and the trend deterioration early warning thresholds set in step S84 are integrated according to the timeline to form a dynamic early warning boundary that evolves with time and surgical stage. S86. The dynamic early warning boundary synthesized in step S85 is used as the early warning trigger condition, and updated and integrated into the personalized intraoperative monitoring framework constructed in step S77.

[0013] Preferably, step S9 specifically includes the following steps: S91. During anesthesia, continuously receive interpretable physiological coordination indicators generated in step S4 and organize them into actual monitoring pathways according to time sequence. S92. Compare the actual monitoring path organized in step S91 with the personalized monitoring reference baseline defined in step S76 in real time point by point, and calculate the real-time deviation of each indicator at each time point. S93. Compare the actual monitoring path organized in step S91 with the dynamic early warning boundary synthesized in step S85 in real time to see if the monitoring indicators touch or cross the upper or lower threshold of the early warning, or whether they trigger a trend deterioration early warning. S94. Based on the comparison results of steps S92 and S93, extract the systematic deviation pattern, the duration of the deviation, and the trend of the deviation as comparison difference features. S95. Based on the comparison difference features extracted in step S94, adjust the judgment threshold or weight parameters related to the current patient in the internalized anesthesia state recognition logic in step S1 to form individualized state judgment rules. S96. Based on the comparison difference features extracted in step S94, adjust the upper and lower threshold offsets or trend deterioration judgment criteria of the dynamic early warning boundary synthesized in step S85 to make the early warning boundary more in line with the patient's current actual response pattern. S97. Update the individualized state determination rules adjusted in step S95 and the dynamic early warning boundary adjusted in step S96 to the system for subsequent real-time state determination and early warning monitoring, and complete a single adaptive optimization loop.

[0014] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: by decomposing key interactive substructures such as nerve-circulation, brain-circulation, and respiration-perfusion from dynamic network profiles, the system calculates coordination indicators with clear physiological significance, such as the baroreflex function index, brain oxygen-blood pressure coupling coefficient, and ventilation-perfusion matching degree. These indicators can quantify the coordination of physiological systems under anesthesia from multiple dimensions. The system performs multi-dimensional similarity matching between these real-time indicators and the internalized anesthesia state knowledge base (containing multiple preset state modes), achieving accurate and interpretable independent determination of the anesthesia depth state in three dimensions: sedation, analgesia, and muscle relaxation, overcoming the limitations of traditional single-index monitoring. Based on individual patient characteristics and a historical case database, the system generates personalized anesthesia plans and physiological baselines for each patient. Combined with the specific surgical plan, the baseline is dynamically adjusted to a dynamic reference baseline adapted to the surgical stage and intense stimulating events, and dynamic warning thresholds that integrate absolute values ​​and trends are set. During surgery, the system continuously compares the actual physiological trajectory with the personalized baseline, dynamically optimizing state recognition rules and warning parameters based on deviation patterns. This achieves individualized adaptation and adaptive evolution of the monitoring and warning system, significantly improving the accuracy and clinical relevance of warnings. The system maps the determined anesthesia state to a three-dimensional dynamic light sphere, presenting a holographic wireframe of the patient's physiological state through an augmented reality interface, and visually locating abnormalities in the cardiovascular, respiratory, and nervous systems using pulsating optical flow and color flashing. Integrating real-time warning information, individualized monitoring pathways, and personalized anesthesia plans, the system provides anesthesiologists with comprehensive and intuitive decision support, integrating state perception, abnormality localization, trend prediction, and intervention suggestions. Attached Figure Description

[0016] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the overall steps of the method of the present invention; Figure 2 This is a flowchart of the data preprocessing and feature extraction process of the present invention; Figure 3 This is a flowchart of the dynamic network, index calculation, and state determination process of this invention. Figure 4 This is a graph showing the change in early warning trigger frequency before and after adaptive optimization of the system of the present invention; Figure 5 This is a comparison chart of physiological index fluctuations under different strong surgical stimulation events according to the present invention; Figure 6 This is a diagram illustrating the dynamic early warning of surgical progress and the monitoring of physiological indicator changes in this invention. Detailed Implementation

[0018] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses consistent with some aspects of this disclosure as detailed in the appended claims.

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] Example 1: See Figures 1 to 6 As shown in the figure, an intelligent state analysis method for anesthetized patients based on physiological data according to an embodiment of the present invention includes the following steps: S1. Organizing and annotating historical anesthesia data to form executable state recognition rules and establishing a standardized historical case reference library; S2. Synchronously collecting core vital sign signals through a multi-modal sensor network, extracting frequency domain and nonlinear multi-dimensional features after preprocessing, and fusing them to form a high-dimensional physiological feature set; S3. Constructing a dynamic relationship network based on the high-dimensional physiological features, quantifying system coupling and synchronization, and generating a multi-system coordinated state profile; S4. Decomposing key interactive substructures from the dynamic network profile and calculating interpretable physiological coordination indices; S5. Matching the real-time extracted coordination indices with the internalized anesthesia state knowledge base, and determining the patient's current anesthesia state in different dimensions such as sedation, analgesia, and muscle relaxation in real time through multi-dimensional similarity calculation; S6. Mapping the identified anesthesia state to sedation. - Analgesia-muscle relaxation three-dimensional coordinate system, presented in the form of a dynamic light sphere; S7. Based on individual patient characteristics, search similar cases in the historical case database, extract effective anesthetic intervention strategies, and generate personalized anesthesia plans and expected physiological trajectory reference baselines; S8. Combine different surgical stages and strong stimulation events to dynamically adjust the personalized reference baselines and set dynamic early warning thresholds for the absolute value and trend of fusion indicators; S9. Continuously compare actual physiological indicators with personalized reference baselines and early warning boundaries during surgery, and dynamically optimize state recognition rules and early warning parameters according to deviation patterns to achieve system adaptive evolution; S10. Continuously output personalized monitoring paths reflecting the patient's current and predicted physiological states; S11. Based on the optimized early warning boundaries and state judgment results, drive the real-time updating of augmented reality early warning information and the visual state of the three-dimensional dynamic dashboard; S12. Integrate personalized monitoring paths, real-time early warning information, and personalized anesthesia plans.

[0021] Example 2: Step S1 specifically includes the following steps: S11. Clean and normalize the physiological data of historical anesthesia surgeries, and standardize and convert the historical medication information. A systematic preprocessing of physiological data from historical anesthesia surgeries was performed. First, data cleaning was conducted, including removing obvious outliers, imputing reasonable missing values, and correcting timestamp misalignments. Then, the cleaned physiological data (such as time-series signals like heart rate, blood pressure, and EEG) were normalized to eliminate the influence of different dimensions and baseline differences, ensuring comparability. Drug names, dosage forms, and routes of administration were standardized with standardized coding to unify the terminology system. Drug dosages at different units were standardized (e.g., converting infusion rates of different concentrations to g / kg / min) to create standardized drug records. S12. Map the expert evaluation conclusions to preset anesthesia state category labels to form a time-aligned structured historical dataset; The conclusions drawn from retrospective assessments by anesthesiologists in each historical case (such as adequate sedation during induction and insufficient analgesia during incision) are mapped to specific status category labels according to a pre-defined anesthesia status classification system. This covers in-depth grading across multiple dimensions, including sedation, analgesia, and muscle relaxation. These status labels are then strictly aligned chronologically with preprocessed physiological data and standardized medication records, integrating individual patient characteristics to ultimately form a time-aligned structured historical dataset. This structured historical dataset provides a high-quality, standardized data foundation for subsequent feature pattern extraction and recognition rule construction. S13. Based on structured historical datasets, extract physiological feature combination patterns associated with specific anesthesia state categories. The feature combination patterns are defined by the co-occurrence and co-change relationships of a set of specific physiological indicators in the time-frequency domain. S14. The extracted feature combination patterns are transformed into executable anesthesia state recognition rules, forming an internalized anesthesia state recognition logic. S15. Based on structured historical datasets, extract individualized features of cases, intraoperative physiological trajectories, anesthesia intervention strategies and corresponding outcomes, and construct a standardized historical data reference system; S16. When the surgery is initiated, load the generated anesthesia status recognition logic and the constructed standardized historical data reference system.

[0022] Step S2 specifically includes the following steps: S21. Sensor nodes are deployed on the patient's body surface through non-invasive flexible electronic patches to automatically and continuously collect core physiological signals, including heart rate, stroke blood pressure, respiratory waveform, prefrontal cortex oxygenation, and skin microcirculation perfusion index, and then generate a synchronous multi-channel raw signal stream. S22. Perform real-time preprocessing on the multi-channel raw signal streams acquired in step S21, including filtering each raw signal to eliminate power frequency interference and motion artifacts, and performing signal quality assessment and bad segment marking. S23. For the signal preprocessed in step S22, extract high-order spectral features in real time by channel. The high-order spectral features include power spectral density, centroid frequency and signal complexity entropy value of a specific frequency band. S24. Extract nonlinear dynamic parameters from the preprocessed signal in step S22 in real time by channel. The nonlinear dynamic parameters include approximate entropy, sample entropy and multi-scale entropy. S25. The high-order spectral features extracted in step S23 and the nonlinear dynamic parameters extracted in step S24 are fused by channel and time sequence to generate a high-dimensional original physiological feature set. Ensure that the N high-order spectral features extracted in step S23 and the M nonlinear dynamic parameters extracted in step S24 originate from the same multi-channel original signal stream and are strictly aligned based on a unified timestamp to form a synchronous, multi-channel feature time series. For each synchronization time point t, all high-order spectral features (such as power in a specific frequency band, centroid frequency, and complexity entropy) extracted from the same signal channel at that time are concatenated with all nonlinear dynamic parameters (such as approximate entropy and sample entropy) in a predefined order (such as spectral features first and nonlinear parameters second) to form a multidimensional local feature vector characterizing the state of the channel at that time. At the same time t, the local feature vectors constructed from all P physiological signal channels (such as ECG, blood pressure, respiration, cerebral oxygenation, and perfusion channels) are horizontally concatenated in channel order to form a global high-dimensional feature vector F_t at time t. The dimension of this vector F_t is equal to (N+M)*P, which contains feature information of all channels and all types. For all moments within the continuously acquired time window, steps 2 and 3 are repeated to generate a high-dimensional feature vector sequence (F_1, F_2, ..., F_T) arranged in chronological order. This time sequence constitutes a high-dimensional original physiological feature set, including the spatial information of multi-channel physiological signals in terms of spectrum and nonlinear dynamics, as well as the dynamic characteristics of these information evolving over time, providing a comprehensive data foundation for the subsequent construction of dynamic relationship networks.

[0023] Step S3 specifically includes the following steps: S31. Using the high-dimensional original physiological feature set generated in step S2 as input, define each physiological indicator as a graph node, calculate the time-varying mutual information and lag correlation between different physiological indicator sequences, use them as dynamic connection weights between nodes, and construct a dynamic coupling relationship graph. S32. Calculate the instantaneous phase difference of the connected node pairs in the dynamic coupling graph, calculate the phase synchronization index based on the instantaneous phase difference, quantify the synchronicity strength between different physiological oscillations, and update the weights of the corresponding connections in the graph. S33. Apply graph neural networks to the dynamic coupling relationship graph constructed in steps S31 and S32 to perform feature propagation and aggregation, learn the aggregated representation of each node and its neighborhood in the graph, and capture the topological structure of the dynamic coupling relationship graph. Specifically, this includes: taking the dynamic coupling relationship graph optimized by steps S31 and S32 as input; the nodes in the graph represent physiological indicators, and the edge weights contain correlation strength and phase synchronization information; Feature learning is performed using a multi-layer graph neural network; an initial feature vector is created for each node, which can be used directly or transformed through an embedding layer; in each layer of the network, for each node, the attention weights between it and its neighboring nodes are calculated. This weight, calculated using a small neural network, assesses the importance of each neighboring node. The features of neighboring nodes are then weighted and summed to aggregate neighborhood information. After aggregation, a new feature for the node at that layer is generated using a non-linear activation function. To improve training stability, layer normalization or residual connections can be added. By stacking multiple layers, the final feature representation of a node can incorporate multi-hop neighborhood information. After propagation through L layers, the node representation contains a local subgraph topology centered on it with a radius of L, simultaneously encoding the node's own characteristics and its connection patterns and association strengths with other nodes in the network. Since the dynamic graph changes over time, the network can be computed independently at each time step. To further enhance temporal modeling, information transfer in the time dimension can be introduced, such as using the node's features from the previous time step as supplementary input to capture state evolution patterns. After the above processing, each node obtains a vector representation containing contextual information. This representation integrates the node's own features and its association network information with other indicators in the entire physiological system at the current time, thus comprehensively capturing the topology of the dynamically coupled graph. S34. Based on the node representations learned in step S33, cluster and pool the node representations belonging to the same physiological subsystem of neural, circulatory, and metabolic systems to generate feature vectors that characterize the internal state and inter-system interaction coordination of each physiological subsystem. S35. Aggregate the feature vectors generated in step S34 to construct a dynamic association network profile. Each dimension of the dynamic association network profile corresponds to a learned global or subsystem feature that reflects the coordination state between multiple systems.

[0024] Step S4 specifically includes the following steps: S41. Based on the physiological system definition, segment the subgraph structure representing the neural-circulatory, brain-circulatory, and respiratory-perfusion interactions from the real-time dynamic association network image constructed in step S3. S42. For each subgraph structure segmented in step S41, calculate the average connection weight and clustering coefficient between internal nodes, which serve as the original quantitative value of the interaction strength between the corresponding physiological systems. S43. In the neural-circulatory interaction subgraph structure, identify the node pairs representing heart rate and blood pressure, calculate the power of their interaction intensity in a specific frequency band, and output the baroreflex function index after standardization. S44. In the brain-circulation interaction subgraph structure, identify the node pairs representing brain oxygen saturation and blood pressure, calculate the coherence coefficient of their interaction intensity sequence, and use it as the brain oxygen-blood pressure coupling coefficient. S45. In the respiratory-perfusion interaction subgraph structure, identify the node pairs representing respiratory waveforms and skin perfusion index, calculate the mutual information entropy of their interaction intensity sequence, and output the ventilation-perfusion matching degree after normalization. S46. The baroreflex function index calculated in step S43, the cerebral oxygen-blood pressure coupling coefficient calculated in step S44, and the ventilation-perfusion matching degree calculated in step S45 are combined to form an interpretable set of physiological coordination indicators, providing feature input for real-time determination of anesthesia depth. First, based on the definition of physiological systems, three key interaction subgraph structures are segmented from the dynamic relational network profile: neural-circulatory interaction subgraph, brain-circulatory interaction subgraph, and respiratory-perfusion interaction subgraph. Each subgraph contains all nodes and their connecting edges related to the corresponding physiological system. For each subgraph, the average connection weight between all nodes within it is calculated to reflect the overall association strength of the interaction pair. At the same time, the clustering coefficient of the subgraph is calculated to quantify the degree of clustering of node connections. These two values ​​together constitute the basic assessment of the interaction strength between systems. Baroreflex Function Index: In the neuro-circulatory subplot, key node pairs representing heart rate and blood pressure are identified, their interaction intensity sequences are extracted, and the power spectral density in the 0.04-0.15 Hz frequency band (reflecting baroreflex activity) is calculated. After standardization, the index is obtained. The index value reflects the integrity of the baroreflex circuit; a decrease suggests deepening anesthesia. Brain oxygenation-blood pressure coupling coefficient: In the brain-circulation subplot, nodes representing brain oxygen saturation and blood pressure are located. The coherence coefficient of the interaction intensity sequence between the two nodes is calculated to quantify the autoregulatory function of cerebral blood flow; changes in the coefficient reflect the compensatory ability of cerebral blood flow to changes in blood pressure. Ventilation-perfusion matching: In the respiratory-perfusion subplot, respiratory waveforms and skin perfusion index nodes are identified, and the mutual information entropy of their interaction intensity sequences is calculated. After normalization, this index is obtained, which reflects the coordination between respiration and peripheral perfusion and assesses tissue oxygenation efficiency. Indicator integration: The three calculated indicators are combined to form an interpretable set of physiological coordination indicators. This set reflects the coordination status of key physiological systems under anesthesia from different dimensions, providing characteristic inputs with clear physiological significance for subsequent determination of anesthesia depth.

[0025] Step S5 specifically includes the following steps: S51. Based on the internalized anesthesia state recognition logic loaded in step S1, extract the preset multiple anesthesia state modes. Each anesthesia state mode is defined by a set of preset interpretable physiological coordination index values ​​and allowable fluctuation range. S52. Construct a real-time index vector from the set of interpretable physiological coordination indices generated in step S4, and perform a similarity calculation on the real-time index vector and the anesthesia state pattern extracted in step S51. S53. For each anesthesia state mode, calculate the multi-dimensional Euclidean distance or cosine similarity between the real-time index vector and the feature value defined for that mode, as the matching degree between the real-time data and that mode. S54. When the matching degree of any anesthesia state mode exceeds the preset judgment threshold, the patient is determined to be in that anesthesia depth state. If the matching degree of multiple modes exceeds the threshold, the mode with the highest matching degree is selected as the judgment result. If no mode has a matching degree exceeding the threshold, it is determined to be an unknown or transitional state. S55. Output the determination result of step S54 as the patient's current anesthesia depth status. The anesthesia depth status includes at least the depth grading of sedation, analgesia, and muscle relaxation. The system loads the anesthesia state recognition logic internalized in step S1. This logic includes multiple preset anesthesia state modes. Each mode is defined by a set of standard values ​​and allowable fluctuation ranges of interpretable physiological coordination indicators. For example, moderate sedation mode can be defined as a baroreflex function index of 0.6-0.8, a brain oxygen-blood pressure coupling coefficient of 0.7-0.9, and a ventilation-perfusion matching degree of 0.75. After obtaining the real-time interpretable physiological coordination indicators from step S4, they are constructed into a real-time indicator vector in a fixed order. The system compares this vector with each preset mode one by one and calculates the similarity. The similarity calculation adopts a multi-dimensional distance measurement method, including: calculating the Euclidean distance between the real-time indicator vector and the mode feature value (the smaller the distance, the higher the matching degree), and calculating the cosine similarity between the two (the closer the value is to 1, the higher the matching degree). The system can perform weighted fusion of these two measures to obtain a comprehensive matching degree score. A matching degree threshold (such as 0.8) is preset for each mode. When the match between real-time data and a certain pattern exceeds its threshold, the patient is considered to be in that state. If multiple patterns match simultaneously, the pattern with the highest match is selected. If no pattern reaches the threshold, the state is determined to be unknown or transitional. To improve the stability of the determination, the system also requires the match to remain above the threshold for multiple consecutive time windows before finally confirming the state. The system outputs a structured anesthesia depth determination result, including the overall depth grading (superficial / intermediate / deep) and independent grading of the three dimensions of sedation, analgesia, and muscle relaxation. For example, for moderate anesthesia depth: sedation-moderate, analgesia-adequate, muscle relaxation-moderate. The system also records the timestamp of the determination, the match score, and the confidence level for subsequent analysis and evaluation.

[0026] Step S6 specifically includes the following steps: S61. Map the anesthesia depth state identified in step S5 to a Cartesian coordinate system consisting of three dimensions: sedation, analgesia, and muscle relaxation. The value of each dimension represents the depth level of that component, which together determine a unique spatial coordinate point in the coordinate system. S62. In the three-dimensional coordinate system, a dynamic light sphere is generated with the spatial coordinate point determined in step S61 as the center. The color, brightness and radius of the light sphere are dynamically adjusted according to the position and rate of change of the coordinate point and are displayed on the dashboard interface in real time. S63. Continuously analyze the time series of the interpretable physiological coordination index generated in step S4, calculate its short-term trend and acceleration, and identify whether the index shows a continuous deviation pattern toward the abnormal state area. S64. When the deviation pattern identified in step S63 meets the preset conditions, it is predicted that the dynamic light sphere will move into the warning area within a specific time window in the future, and the corresponding level of warning is triggered according to the prediction result. S65. When an alert is triggered, an augmented reality device is used to overlay a holographic wireframe of the patient's physiological state into the doctor's field of vision. The wireframe outlines the human body with semi-transparent wireframes and marks the location of major organ systems. S66. In the holographic wireframe diagram, the activity of the cardiovascular system is mapped by pulsating red light flow, the ventilation status of the respiratory system is mapped by blue ripples, and the intensity of central nervous system activity is mapped by the flashing frequency and spectral color of the head region; when the indicators of step S4 or the status of step S5 are abnormal, characteristic bright flashing and color alarms are triggered at the corresponding anatomical location in the wireframe diagram. The real-time identified sedation, analgesia, and muscle relaxation states are quantified into values ​​from 0 to 10, mapped to coordinate points in a three-dimensional coordinate system, and visualized as a dynamic light sphere on the dashboard. The color of the light sphere changes dynamically according to the coordinate position: green in the safe zone, gradually turning yellow as it approaches the warning zone, and red when entering the danger zone. The radius of the light sphere changes inversely with the stability of the state; when the state fluctuates rapidly, the radius shrinks and the brightness increases to highlight the warning effect. The system continuously analyzes the short-term trends of coordination indicators such as the baroreflex function index, cerebral oxygen-blood pressure coupling coefficient, and ventilation-perfusion matching degree, calculating their slope and acceleration. When multiple indicators are detected to show a continuous deviation from the normal pattern, the system predicts the trajectory of the dynamic light sphere over the next 60 seconds based on the current rate and direction of state change. If the predicted trajectory... If the patient enters a pre-defined danger zone, a corresponding level of warning is triggered. The warning information is displayed as a floating card, including the predicted arrival time, a description of the danger zone, and an intervention time window. After triggering the warning, augmented reality equipment overlays a semi-transparent wireframe of the patient's body onto the doctor's field of vision, clearly marking key anatomical structures such as the heart, lungs, and brain. In the wireframe, the cardiovascular system is represented by a pulsating red light stream emanating from the heart, the respiratory system by blue wavy animation in the lung area, and the central nervous system by dynamic spectral colors in the head area. When specific physiological indicators are abnormal, a location alarm is triggered at the corresponding anatomical location: an abnormal baroreflex index causes the heart area to flash orange, an abnormal brain oxygen-blood pressure coupling coefficient causes a red halo pulse in the head area, and an abnormal ventilation-perfusion mismatch causes both lungs to flash blue alternately. The alarm intensity matches the degree of abnormality, presented in a graded manner from gentle pulsation to intense flashing, providing doctors with intuitive abnormal location and multi-dimensional state perception.

[0027] Step S7 specifically includes the following steps: S71. Based on patient physical parameters, cardiopulmonary function classification, and genomic drug metabolism type information, construct a patient-specific feature vector; The method for constructing individualized feature vectors for patients is as follows: Physical parameter processing: Extract continuous values ​​such as height, weight, and BMI, and perform standardization to eliminate the influence of dimensions; Cardiopulmonary function classification coding: One-hot coding is used to convert ordered categorical variables such as ASA classification into binary vectors; for example, ASA II level is coded as [0,1,0,0]. Genomic information encoding: For drug-metabolizing enzyme phenotypes, such as the rapid metabolizer, intermediate metabolizer, and slow metabolizer of CYP2D6, one-hot encoding is also used; Vector concatenation: The three types of processed features are concatenated in a fixed order to form the final individualized feature vector; For example, the treatment process for a patient is as follows: Standardized physical parameters: [height_norm, weight_norm, BMI_norm]; ASA II level one-hot encoding: [0,1,0,0]; CYP2D6 fast metabolizer encoding: [1,0,0]; Final feature vector: [height_norm, weight_norm, BMI_norm, 0, 1, 0, 0, 1, 0, 0]; S72. Perform multi-dimensional similarity calculation between the patient-individualized feature vector constructed in step S71 and the case features in the standardized historical data reference system loaded in step S1, retrieve the preset number of historical cases with the highest similarity, and form the set of the most similar historical cases. S73. From the set of most similar historical cases obtained in step S72, extract the recorded types, dosages, and timing of anesthesia induction and maintenance drugs, as well as the corresponding intraoperative physiological responses and final outcomes, as a verified and effective anesthesia intervention strategy. S74. Based on the validated and effective anesthesia intervention strategies extracted in step S73, with the optimization goal of maintaining physiological stability, a recommended drug infusion plan for the current patient is derived to form a personalized anesthesia plan. S75. Based on the set of most similar historical cases obtained in step S72, the average trajectory and standard deviation range of intraoperative physiological state indicators over time are statistically analyzed to serve as the expected personalized physiological state change path and fluctuation range for the current patient. S76. The personalized physiological state change path and fluctuation range obtained by modeling in step S75 are defined as the personalized monitoring reference baseline for intraoperative monitoring. S77. The personalized monitoring reference baseline defined in step S76 is linked and integrated with the personalized anesthesia plan generated in step S74 to form a personalized intraoperative monitoring framework, providing a benchmark for real-time intraoperative comparison.

[0028] Step S8 specifically includes the following steps: S81. Based on the surgical plan, divide the surgical process into different stages and mark the types and intensities of strong stimulating events expected to occur in each stage; S82. Based on the current and upcoming surgical stages and strong stimulation events marked in step S81, the personalized monitoring reference baseline defined in step S76 is corrected for temporal offset to generate a corrected dynamic reference baseline. The specific implementation methods of steps S81 and S82 are as follows: Surgical stage division and labeling: The system divides the surgical process into different stages according to the surgical plan, such as anesthesia induction, skin incision, major operation, suturing, and recovery; for each stage, it labels the expected strong stimulating events (such as skin incision, bone manipulation, and visceral traction) and their intensity levels; Dynamic reference baseline correction: Taking a personalized monitoring reference baseline as input, the system corrects for changes based on the current and upcoming surgical phases and strong stimulating events. Stage identification: Determine the current and upcoming surgical stages; Stimulus-Response Mapping: Invokes the built-in stimulus-physiological response model to map the type and intensity of stimulus events to the expected impact patterns (such as the direction and magnitude of change) on various physiological coordination indicators. Baseline calculation: Based on the expected impact, calculate the theoretical offset of each indicator's reference baseline and generate a dynamic correction curve that changes over time; Baseline update: The modified curve is superimposed on the original baseline to form a modified dynamic reference baseline. This baseline will adjust the target value or fluctuation range accordingly during periods of strong stimulation so that it can reflect the expected physiological response under the stimulation. S83. Using the corrected dynamic reference baseline generated in step S82 as the center, for each interpretable physiological coordination index, dynamically calculate the personalized warning upper and lower thresholds within a future time window. The upper and lower thresholds are usually determined based on the dynamic reference baseline value superimposed with a fixed percentage or a preset absolute offset. S84. For each interpretable physiological coordination index, set an allowable range for the short-term rate of change. When the rate of change of the index continues to exceed this range, a trend deterioration warning is triggered. This rate of change threshold is used as the trend deterioration warning threshold. S85. The personalized early warning upper and lower thresholds calculated in step S83 and the trend deterioration early warning thresholds set in step S84 are integrated according to the timeline to form a dynamic early warning boundary that evolves with time and surgical stage. S86. The dynamic early warning boundary synthesized in step S85 is used as the early warning trigger condition, and updated and integrated into the personalized intraoperative monitoring framework constructed in step S77.

[0029] Step S9 specifically includes the following steps: S91. During anesthesia, continuously receive interpretable physiological coordination indicators generated in step S4 and organize them into actual monitoring pathways according to time sequence. S92. Compare the actual monitoring path organized in step S91 with the personalized monitoring reference baseline defined in step S76 in real time point by point, and calculate the real-time deviation of each indicator at each time point. S93. Compare the actual monitoring path organized in step S91 with the dynamic early warning boundary synthesized in step S85 in real time to see if the monitoring indicators touch or cross the upper or lower threshold of the early warning, or whether they trigger a trend deterioration early warning. S94. Based on the comparison results of steps S92 and S93, extract the systematic deviation pattern, the duration of the deviation, and the trend of the deviation as comparison difference features. S95. Based on the comparison difference features extracted in step S94, adjust the judgment threshold or weight parameters related to the current patient in the internalized anesthesia state recognition logic in step S1 to form individualized state judgment rules. S96. Based on the comparison difference features extracted in step S94, adjust the upper and lower threshold offsets or trend deterioration judgment criteria of the dynamic early warning boundary synthesized in step S85 to make the early warning boundary more in line with the patient's current actual response pattern. S97. Update the individualized state determination rules adjusted in step S95 and the dynamic early warning boundary adjusted in step S96 to the system for subsequent real-time state determination and early warning monitoring, and complete a single adaptive optimization loop. During anesthesia, the system continuously receives interpretable physiological coordination indices generated in step S4. These indices are input in real time as streaming data at a fixed frequency (once per second); the system organizes this time-series data into the actual monitoring path as follows: Time alignment: Each indicator (baroreflex function index, cerebral oxygen-blood pressure coupling coefficient, ventilation-perfusion matching degree) is given a precise timestamp to ensure data synchronization among multiple indicators; Sliding window caching: The system maintains a sliding data window (5-10 minutes) with a configurable time length to continuously store the latest time-series data; this window contains both current data and recent historical data; Data structuring: The data within the window is organized into a multi-dimensional time series matrix. Each row corresponds to a time point, and each column corresponds to a physiological coordination index, forming a standardized two-dimensional data structure of [number of time point indices]. Path feature extraction: Several features reflecting changing trends are calculated in real time from this time series data. Short-term trend: Calculate the linear regression slope of each indicator over the past 30 seconds; Volatility: Calculate the standard deviation of each indicator within the window; Change in acceleration: Calculate the second derivative based on the trend slope; Abnormal patterns: Whether the detection indicators show a sudden increase, a sudden decrease, or a plateau; Path quality assessment: Perform quality checks on the data stream at each time point, mark anomalies such as missing signals and noise interference, and ensure the reliability of the actual monitoring path; Through the above processing, the discrete index values ​​are transformed into an actual monitoring path; this path not only records the current state, but also retains the recent change trajectory, providing a complete and standardized time-series data foundation for subsequent real-time comparison with the personalized reference baseline.

[0030] Example 3: Monitoring of patients undergoing routine laparoscopic cholecystectomy: This case involves a 45-year-old female patient, ASA class II, who underwent elective laparoscopic cholecystectomy. Personalized protocol generation: Based on the patient's individual characteristics (BMI 24.3, ASA II, CYP2C19 intermediate metabolizer), the system retrieved 8 highly similar cases from the historical database; a personalized anesthesia protocol was generated: propofol target-controlled infusion (plasma target concentration 3.5-4.5 g / ml), remifentanil 0.1-0.2 g / kg / min, and cisatracurium loading dose followed by maintenance at 0.1 mg / kg / h; the reference baseline was set as follows: baroreflex function index maintained at 0.7-0.15, cerebral oxygen-blood pressure coupling coefficient >0.8, and ventilation-perfusion matching degree >0.75; Intraoperative dynamic monitoring and early warning: Induction period: The system shows that the dynamic photosphere is stable near coordinates (6.2, 7.5, 5.8) (green), and all indicators are in line with the reference baseline; During pneumoperitoneum establishment: The system detected that the baroreflex function index decreased from 0.72 to 0.58 within 30 seconds, and the ventilation-perfusion matching decreased from 0.82 to 0.68; the dynamic photosphere moved towards the area of ​​deepening sedation and potential insufficient analgesia, and its color gradually changed to yellow; the patient's heart and lung areas in the AR interface simultaneously triggered orange pulsating alarms; the system predicted that if the trend continued, the photosphere would enter the red warning zone after 60 seconds, and therefore issued an early warning: the risk of circulatory and respiratory depression is increasing, and adjustments to respiratory parameters and assessment of analgesia depth should be considered; Intervention and optimization: The anesthesiologist reduced the remifentanil to 0.15 g / kg / min as instructed and adjusted the ventilator parameters; after about 2 minutes, the indicators gradually recovered; the system recorded the physiological response pattern that was effective in this intervention, which was used to fine-tune the patient's subsequent warning boundaries; Outcome: The surgery was stable, the patient regained consciousness 5 minutes post-surgery, and extubation was successful; the individualized physiological trajectory and medication adjustment records recorded by the system were desensitized and fed back into the historical case database to enrich the data of similar cases.

[0031] Example 4: Circulatory management of elderly patients who have undergone hip replacement surgery: This case involves a 72-year-old male patient, ASA class III, who underwent hip replacement surgery under general anesthesia. Individualized treatment plan and baseline: The patient feature vector includes information such as age, mild diastolic dysfunction, and CYP2D6 slow metabolizer; the system-generated reference baseline specifically emphasizes the stability of the cerebral oxygen-blood pressure coupling coefficient (>0.75), as the patient's cerebral blood flow autoregulation function may be impaired. The warning threshold is more sensitive to blood pressure fluctuations; Critical Incident Handling: Surgical incision and medullary reaming: According to the surgical plan, the system automatically lowers the reference baseline of analgesia by 10% 1 minute before the incision (anticipating strong stimulation) and temporarily relaxes the warning lower limit of the brain oxygen-blood pressure coupling coefficient from 0.75 to 0.7. During bone cement implantation: The system monitored a 25% drop in blood pressure within 2 minutes, but the brain oxygen-blood pressure coupling coefficient remained at 0.78, indicating that the cerebral blood flow autoregulation function was well compensated. Although the photosphere moved towards the circulatory warning zone due to the drop in blood pressure, no high-level alarms were triggered in the head area of ​​the AR interface. The system's recommendations focused on circulatory support: Blood pressure dropped significantly, but cerebral perfusion was currently compensated. It was recommended to use a single intravenous push of 50-100g of phenylephrine instead of rapid volume expansion. Self-learning optimization: The system found that the patient's pressor response to norepinephrine was more sensitive than predicted by historical similar case models, so it automatically adjusted the parameters of the blood pressure-vasoactive drug response model to make subsequent recommended doses more accurate; Outcome: The patient's circulation remained stable during the operation, and no cerebral hypoxia occurred. Postoperative cognitive function assessment showed no abnormalities.

[0032] Example 5: Respiratory management in obese patients undergoing laparoscopic sleeve gastrectomy: This case involves a 38-year-old female patient with a BMI of 38.5 and an ASA classification of II who underwent laparoscopic sleeve gastrectomy. Individualized settings: The system sets special ventilation-perfusion monitoring weights for high BMI patients. In the reference baseline, the safe range for ventilation-perfusion matching is narrowed (0.78-0.95) because obese patients are more prone to atelectasis and oxygenation impairment; Intraoperative monitoring and early warning: Head-down, feet-up body position and pneumoperitoneum: After the body position was changed, the system monitored that the ventilation-perfusion matching degree gradually decreased from 0.88 to 0.80 within 5 minutes, but the interaction entropy value of the respiratory waveform node and the perfusion index node decreased significantly, indicating a decrease in matching efficiency; the dynamic light ball moved towards the respiratory dimension, and a blue flashing alarm appeared in the bilateral lung base area in the AR interface. Predictive intervention recommendations: The system combines the surgical stage (continuous pneumoperitoneum) with the trend of indicators, predicting that the matching degree will further decrease, and indicating that the ventilation / perfusion matching efficiency is decreasing, which is consistent with the risk of atelectasis related to body position and pneumoperitoneum. It is recommended to consider lung recruitment maneuvers. Adaptive optimization: After the doctor performed lung recruitment, the matching score recovered to 0.90. The system recorded the physiological response trajectory of the effective intervention and automatically relaxed the lower limit of the matching score warning for the patient in similar positions by 5% thereafter, reducing over-warning; Outcome: The patient had good oxygenation during the operation and no postoperative respiratory complications. By studying this case, the system improved the accuracy of its early warning system for managing special postural ventilation in obese patients.

[0033] Table 1. Comparison of Patient Basic Information and Anesthesia Protocols in Three Examples project Example 3: Laparoscopic Cholecystectomy Example 4: Hip replacement surgery in the elderly Example 5: Sleeve gastrectomy in obese patients Patient type 45-year-old woman 72-year-old male 38-year-old woman ASA Classification Level II Level III Level II Merging features BMI 24.3, CYP2C19 intermediate metabolizer Mild diastolic dysfunction, CYP2D6 slow metabolizer BMI 38.5, obese Surgical type Laparoscopic cholecystectomy Hip replacement surgery Laparoscopic sleeve gastrectomy Main anesthetic drugs Propofol, remifentanil, cisatracurium Routine general anesthesia medication Routine general anesthesia ventilation management Key monitoring indicators Pressure reflex function index, ventilation-perfusion matching Brain oxygenation-blood pressure coupling coefficient, circulatory stability Ventilation-perfusion matching Table 2. Key Intraoperative Events and System Early Warning Performance in Three Examples Example Key surgical events Changes in key physiological indicators System Visual Early Warning Performance Example 3 pneumoperitoneum establishment A decreased pressure reflex index and reduced ventilation-perfusion mismatch The dynamic light sphere changes from green to yellow, and the AR heart and lungs display an orange pulsating alarm. Example 4 Skin incision, medullary canal reaming, and bone cement implantation Blood pressure drops by 25% in a short period of time, while brain oxygen-blood pressure coupling coefficient remains stable. The sphere moved toward the cyclical warning zone without any high-level alarm. Example 5 Head-down position with feet-up posture + pneumoperitoneum Ventilation-perfusion mismatch decreases slowly, and respiratory-perfusion interaction entropy decreases. Dynamic light sphere shifts towards the respiratory dimension, double lung base blue flashing alarm

[0034] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0035] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope of this disclosure is indicated by the following claims.

Claims

1. A method for intelligent state analysis of anesthetized patients based on physiological data, characterized in that, Includes the following steps: S1. Organize and annotate historical anesthesia data to form executable status recognition rules and establish a standardized historical case reference library; S2. Core vital signs signals are collected synchronously through a multi-mode sensor network. After preprocessing, frequency domain and nonlinear multidimensional features are extracted and fused to form a high-dimensional physiological feature set. S3. Construct a dynamic relationship network based on high-dimensional physiological characteristics, quantify system coupling and synchronization, and generate a multi-system coordination state profile. S4. Decompose key interaction substructures from dynamic network profiles and calculate interpretable physiological coordination indices. S5. Match the real-time extracted coordination index with the internalized anesthesia status knowledge base, and determine the patient's current sedation, analgesia, and muscle relaxation status in real time through multi-dimensional similarity calculation. S6. Map the identified anesthesia state to a three-dimensional coordinate system of sedation-analgesia-muscle relaxation and present it in the form of a dynamic light sphere; S7. Based on the individual characteristics of the patient, search for similar cases in the historical case database, extract effective anesthesia intervention strategies, and generate personalized anesthesia plans and expected physiological trajectory reference baselines; S8. Combine different stages of surgery and strong stimulating events to dynamically adjust the personalized reference baseline and set dynamic early warning thresholds for the absolute value and trend of the fusion index. S9. During the operation, continuously compare the actual physiological indicators with the personalized reference baseline and warning boundary, and dynamically optimize the status recognition rules and warning parameters according to the deviation pattern to achieve system adaptive evolution. S10. Continuously outputs an individualized monitoring pathway that reflects the patient's current and predicted physiological status; S11. Based on the optimized warning boundary and status determination results, drive the visual status update of augmented reality warning information and three-dimensional dynamic dashboard in real time. S12, integrated individualized monitoring pathways, real-time early warning information, and personalized anesthesia plans.

2. The method for intelligent state analysis of anesthetized patients based on physiological data according to claim 1, characterized in that, Step S1 specifically includes the following steps: S11. Clean and normalize the physiological data of historical anesthesia surgeries, and standardize and convert the historical medication information. S12. Map the expert evaluation conclusions to preset anesthesia state category labels to form a time-aligned structured historical dataset; S13. Based on the structured historical dataset, extract the physiological feature combination pattern associated with a specific anesthesia state category. The feature combination pattern is defined by the co-occurrence and synergistic change relationship of a set of specific physiological indicators in the time-frequency domain. S14. The extracted feature combination pattern is transformed into an executable anesthesia state recognition rule, which constitutes the internalized anesthesia state recognition logic. S15. Based on the structured historical dataset, extract individualized features of cases, intraoperative physiological trajectories, anesthesia intervention strategies and corresponding outcomes, and construct a standardized historical data reference system; S16. When the surgery is started, load the generated anesthesia state identification logic and the constructed standardized historical data reference system.

3. The method for intelligent state analysis of anesthetized patients based on physiological data according to claim 2, characterized in that, Step S2 specifically includes the following steps: S21. Sensor nodes are deployed on the patient's body surface through non-invasive flexible electronic patches to automatically and continuously collect core physiological signals, including heart rate, stroke blood pressure, respiratory waveform, prefrontal cortex oxygenation, and skin microcirculation perfusion index, and then generate a synchronous multi-channel raw signal stream. S22. Perform real-time preprocessing on the multi-channel raw signal streams acquired in step S21, including filtering each raw signal to eliminate power frequency interference and motion artifacts, and performing signal quality assessment and bad segment marking. S23. For the signal preprocessed in step S22, extract high-order spectral features in real time by channel. The high-order spectral features include power spectral density, centroid frequency and signal complexity entropy value of a specific frequency band. S24. For the signal preprocessed in step S22, extract nonlinear dynamic parameters in real time by channel. The nonlinear dynamic parameters include approximate entropy, sample entropy and multi-scale entropy. S25. The high-order spectral features extracted in step S23 and the nonlinear dynamic parameters extracted in step S24 are fused by channel and time sequence to generate a high-dimensional original physiological feature set.

4. The method for intelligent state analysis of anesthetized patients based on physiological data according to claim 3, characterized in that, Step S3 specifically includes the following steps: S31. Using the high-dimensional original physiological feature set generated in step S2 as input, define each physiological indicator as a graph node, calculate the time-varying mutual information and lag correlation between different physiological indicator sequences, and use them as dynamic connection weights between nodes to construct a dynamic coupling relationship graph. S32. Calculate the instantaneous phase difference of the connected node pairs in the dynamic coupling graph, calculate the phase synchronization index based on the instantaneous phase difference, quantify the synchronicity strength between different physiological oscillations, and update the weights of the corresponding connections in the graph. S33. Apply a graph neural network to the dynamic coupling relationship graph constructed in steps S31 and S32 to perform feature propagation and aggregation, learn the aggregated representation of each node and its neighborhood in the graph, and capture the topological structure of the dynamic coupling relationship graph. S34. Based on the node representations learned in step S33, cluster and pool the node representations belonging to the same physiological subsystem of neural, circulatory, and metabolic systems to generate feature vectors that characterize the internal state and inter-system coordination of each physiological subsystem. S35. Aggregate the feature vectors generated in step S34 to construct a dynamic association network profile. Each dimension of the dynamic association network profile corresponds to a learned global or subsystem feature that reflects the coordination state between multiple systems.

5. The method for intelligent state analysis of anesthetized patients based on physiological data according to claim 4, characterized in that, Step S4 specifically includes the following steps: S41. Based on the physiological system definition, segment the subgraph structure representing the neural-circulatory, brain-circulatory, and respiratory-perfusion interactions from the real-time dynamic association network image constructed in step S3. S42. For each subgraph structure segmented in step S41, calculate the average connection weight and clustering coefficient between internal nodes as the original quantitative value of the interaction strength between the corresponding physiological systems. S43. In the neural-circulatory interaction subgraph structure, identify the node pairs representing heart rate and blood pressure, calculate the power of their interaction intensity in a specific frequency band, and output the standardized power as the baroreflex function index. S44. In the brain-circulation interaction subgraph structure, identify the node pairs representing brain oxygen saturation and blood pressure, and calculate the coherence coefficient of their interaction intensity sequence as the brain oxygen-blood pressure coupling coefficient. S45. In the respiratory-perfusion interaction subgraph structure, identify the node pairs representing the respiratory waveform and the skin perfusion index, calculate the mutual information entropy of their interaction intensity sequence, and output the normalized result as the ventilation-perfusion matching degree. S46. The baroreflex function index calculated in step S43, the cerebral oxygen-blood pressure coupling coefficient calculated in step S44, and the ventilation-perfusion matching degree calculated in step S45 are combined to form an interpretable set of physiological coordination indicators, providing feature input for real-time determination of anesthesia depth.

6. The method for intelligent state analysis of anesthetized patients based on physiological data according to claim 5, characterized in that, Step S5 specifically includes the following steps: S51. Based on the internalized anesthesia state recognition logic loaded in step S1, extract a variety of preset anesthesia state modes, each of which is defined by a set of preset interpretable physiological coordination index values ​​and allowable fluctuation range. S52. Construct a real-time index vector from the set of interpretable physiological coordination indices generated in step S4, and perform a one-to-one similarity calculation between the real-time index vector and the anesthesia state pattern extracted in step S51. S53. For each anesthesia state mode, calculate the multi-dimensional Euclidean distance or cosine similarity between the real-time index vector and the feature value defined by the mode, as the matching degree between the real-time data and the mode. S54. When the matching degree of any anesthesia state mode exceeds the preset judgment threshold, the patient is determined to be in that anesthesia depth state. If the matching degree of multiple modes exceeds the threshold, the mode with the highest matching degree is selected as the judgment result. If no mode has a matching degree exceeding the threshold, it is determined to be an unknown or transitional state. S55. Output the determination result of step S54 as the patient's current anesthesia depth status, which includes at least the depth grading of sedation, analgesia, and muscle relaxation dimensions.

7. The method for intelligent state analysis of anesthetized patients based on physiological data according to claim 6, characterized in that, Step S6 specifically includes the following steps: S61. Map the anesthesia depth state identified in step S5 to a Cartesian coordinate system consisting of three dimensions: sedation, analgesia, and muscle relaxation. The value of each dimension represents the depth level of that component, which together determine a unique spatial coordinate point in the coordinate system. S62. In the three-dimensional coordinate system, a dynamic light sphere is generated with the spatial coordinate point determined in step S61 as the center. The color, brightness and radius of the light sphere are dynamically adjusted according to the position and rate of change of the coordinate point and are displayed on the dashboard interface in real time. S63. Continuously analyze the time series of the interpretable physiological coordination index generated in step S4, calculate its short-term trend and acceleration, and identify whether the index shows a continuous deviation pattern toward the abnormal state region. S64. When the deviation pattern identified in step S63 meets the preset conditions, it is predicted that the dynamic light sphere will move into the warning area within a specific time window in the future, and a warning of the corresponding level is triggered according to the prediction result. S65. When an alert is triggered, an augmented reality device is used to overlay and generate a holographic wireframe diagram of the patient's physiological state in the doctor's field of vision. The wireframe diagram outlines the human body with semi-transparent wireframes and marks the location of major organ systems. S66. In the holographic wireframe diagram, the activity of the cardiovascular system is mapped by pulsating red light flow, the ventilation status of the respiratory system is mapped by blue ripples, and the intensity of central nervous system activity is mapped by the flashing frequency and spectral color of the head region; when the indicators of step S4 or the status of step S5 are abnormal, characteristic bright flashing and color alarms are triggered at the corresponding anatomical location in the wireframe diagram.

8. The method for intelligent state analysis of anesthetized patients based on physiological data according to claim 7, characterized in that, Step S7 specifically includes the following steps: S71. Based on patient physical parameters, cardiopulmonary function classification, and genomic drug metabolism type information, construct a patient-specific feature vector; S72. Perform multi-dimensional similarity calculation between the patient-individualized feature vector constructed in step S71 and the case features in the standardized historical data reference system loaded in step S1, retrieve the preset number of historical cases with the highest similarity, and form the set of the most similar historical cases. S73. From the set of most similar historical cases obtained in step S72, extract the recorded types, dosages, and timing of anesthesia induction and maintenance drugs, as well as the corresponding intraoperative physiological responses and final outcomes, as a verified and effective anesthesia intervention strategy. S74. Based on the validated and effective anesthesia intervention strategies extracted in step S73, with the optimization goal of maintaining physiological stability, a recommended drug infusion plan for the current patient is derived to form a personalized anesthesia plan. S75. Based on the set of most similar historical cases obtained in step S72, the average trajectory and standard deviation range of intraoperative physiological state indicators over time are statistically analyzed to serve as the expected personalized physiological state change path and fluctuation range for the current patient. S76. The personalized physiological state change path and fluctuation range obtained by modeling in step S75 are defined as the personalized monitoring reference baseline for intraoperative monitoring. S77. The personalized monitoring reference baseline defined in step S76 is associated and integrated with the personalized anesthesia plan generated in step S74 to form the personalized intraoperative monitoring framework, providing a benchmark for real-time intraoperative comparison.

9. The method for intelligent state analysis of anesthetized patients based on physiological data according to claim 8, characterized in that, Step S8 specifically includes the following steps: S81. Based on the surgical plan, divide the surgical process into different stages and mark the types and intensities of strong stimulating events expected to occur in each stage; S82. Based on the current and upcoming surgical stages and strong stimulation events marked in step S81, the personalized monitoring reference baseline defined in step S76 is corrected for temporal offset to generate a corrected dynamic reference baseline. S83. Using the corrected dynamic reference baseline generated in step S82 as the center, for each interpretable physiological coordination index, dynamically calculate the personalized warning upper and lower thresholds within a future time window. The upper and lower thresholds are usually determined based on the dynamic reference baseline value plus a fixed percentage or a preset absolute offset. S84. For each interpretable physiological coordination index, set an allowable range for the short-term rate of change. When the rate of change of the index continues to exceed this range, a trend deterioration warning is triggered. This rate of change threshold is used as the trend deterioration warning threshold. S85. The personalized early warning upper and lower thresholds calculated in step S83 and the trend deterioration early warning thresholds set in step S84 are integrated according to the timeline to form a dynamic early warning boundary that evolves with time and surgical stage. S86. The dynamic early warning boundary synthesized in step S85 is used as the early warning trigger condition, and updated and integrated into the personalized intraoperative monitoring framework constructed in step S77.

10. The method for intelligent state analysis of anesthetized patients based on physiological data according to claim 9, characterized in that, Step S9 specifically includes the following steps: S91. During anesthesia, continuously receive interpretable physiological coordination indicators generated in step S4 and organize them into actual monitoring pathways according to time sequence. S92. Compare the actual monitoring path organized in step S91 with the personalized monitoring reference baseline defined in step S76 in real time point by point, and calculate the real-time deviation of each indicator at each time point. S93. Compare the actual monitoring path organized in step S91 with the dynamic early warning boundary synthesized in step S85 in real time to see if the monitoring indicators touch or cross the upper or lower threshold of the early warning, or whether they trigger a trend deterioration early warning. S94. Based on the comparison results of steps S92 and S93, extract the systematic deviation pattern, the duration of the deviation, and the trend of the deviation as comparison difference features. S95. Based on the comparison difference features extracted in step S94, adjust the judgment threshold or weight parameters related to the current patient in the internalized anesthesia state recognition logic in step S1 to form an individualized state judgment rule. S96. Based on the comparison difference features extracted in step S94, adjust the upper and lower threshold offsets or trend deterioration judgment criteria of the dynamic early warning boundary synthesized in step S85 to make the early warning boundary more consistent with the patient's current actual response pattern. S97. Update the individualized state determination rules adjusted in step S95 and the dynamic early warning boundary adjusted in step S96 to the system for subsequent real-time state determination and early warning monitoring, and complete a single adaptive optimization loop.

Images