Search

CN-122000018-A - Anesthesia response characteristic analysis system based on visual recognition

CN122000018ACN 122000018 ACN122000018 ACN 122000018ACN-122000018-A

Abstract

The invention relates to the technical field of medical data analysis, in particular to an anesthesia response characteristic analysis system based on visual identification. The system comprises an anesthesia multisource acquisition module, an anesthesia characterization analysis module, an anesthesia reaction analysis module, an anesthesia characteristic analysis module and an anesthesia state evaluation module, wherein multidimensional visual acquisition equipment and an associated medical data interface in an anesthesia scene can be obtained, human visual characteristic data and synchronous medical monitoring data in an anesthesia process are acquired, noise reduction, time sequence alignment and format standardization preprocessing are carried out, extraction and characterization processing are carried out at the same time, anesthesia state visual characterization data are generated, association analysis is carried out by combining the anesthesia state visual characterization data, an anesthesia reaction characteristic dynamic model is constructed, anesthesia reaction characteristic analysis data are generated through iterative optimization, and anesthesia state evaluation reports and clinical intervention suggestions are output according to the anesthesia reaction characteristic analysis data. The invention can avoid the intervention time lag caused by lack of specific guidance.

Inventors

  • YAN JUAN

Assignees

  • 严娟

Dates

Publication Date
20260508
Application Date
20251210

Claims (10)

  1. 1. An anesthesia response characteristic analysis system based on visual recognition is characterized by comprising the following modules: The anesthesia multisource acquisition module is used for acquiring multidimensional visual acquisition equipment and an associated medical data interface in an anesthesia scene, acquiring human visual characteristic data and synchronous medical monitoring data in the anesthesia process based on the equipment and the interface, and performing noise reduction, time sequence alignment and format standardization preprocessing to generate integrated anesthesia scene multisource fusion data; The anesthesia characterization analysis module is used for extracting and characterizing visual features in the anesthesia scene multisource fusion data to generate anesthesia state visual characterization data; The anesthesia reaction analysis module is used for carrying out association analysis by combining the anesthesia state visual representation data and a preset clinical anesthesia reaction evaluation index to generate an anesthesia reaction preliminary analysis result containing reaction type probability distribution; The anesthesia characteristic analysis module is used for constructing an anesthesia reaction characteristic dynamic model based on an anesthesia reaction preliminary analysis result and generating anesthesia reaction characteristic analysis data through iterative optimization; the anesthesia state evaluation module is used for outputting an anesthesia state evaluation report and clinical intervention advice according to the anesthesia response characteristic analysis data.
  2. 2. The visual recognition-based anesthesia response characteristic analysis system according to claim 1, wherein the anesthesia multisource acquisition module comprises the following functions: Acquiring a medical data interface corresponding to a facial expression acquisition device, an eye dynamic capture device, a limb posture monitoring device, an anesthesia monitor and a vital sign monitor under an anesthesia scene; The visual information corresponding to the facial muscle movement characteristics, eyelid opening and closing states, eyeball movement tracks, limb movement amplitude and posture changes of the human body in the anesthesia process is continuously collected through the visual collection equipment, and medical monitoring data corresponding to heart rate, blood pressure, respiratory frequency and anesthetic drug concentration are synchronously collected through a medical data interface, so that original anesthesia related data are formed together; Performing image denoising, fuzzy correction and characteristic point enhancement processing on visual information in the original anesthesia related data, and performing outlier rejection and missing value completion processing on the medical monitoring data; And (3) aligning the processed visual information with the medical monitoring data in time sequence according to the time stamp, unifying the data format and the sampling frequency, and generating integrated anesthesia scene multisource fusion data.
  3. 3. The visual identification-based anesthesia response characterization system of claim 1 wherein the anesthesia characterization module includes the following functionality: Separating visual characteristic data and medical monitoring associated data from the anesthesia scene multisource fusion data; extracting layered features of the visual feature data, firstly extracting basic visual features corresponding to facial contours and limb contours, and then digging deep dynamic features corresponding to facial micro-expressions, eye tremor frequency and limb micro-twitches to generate a multi-dimensional visual feature set; performing feature screening and dimension reduction processing on the multi-dimensional visual feature set, removing redundant features and interference features, and reserving core visual features highly related to anesthesia reaction; Performing relevance characterization on the core visual features and the corresponding medical monitoring relevant data, constructing a mapping relation between the visual features and the anesthesia physiological state, and generating anesthesia state visual characterization data; and (3) verifying the validity of the visual representation data of the anesthesia state, removing invalid representation information, and ensuring that the visual representation data of the anesthesia state can reflect the relevant probability distribution of the reaction type in the anesthesia process.
  4. 4. The visual recognition-based anesthesia response analysis system of claim 1 wherein the anesthesia response analysis module includes the following functions: extracting clinical anesthesia response evaluation indexes from clinical anesthesia diagnosis and treatment specifications, wherein the clinical anesthesia response evaluation indexes cover key dimensions corresponding to sedation depth grade, analgesia effect evaluation, adverse reaction occurrence signs and neuromuscular blocking degree; Matching the core visual characteristics in the visual representation data of the anesthesia state with each index in the clinical anesthesia response evaluation indexes one by one, and establishing a corresponding relation between the visual characteristics and the evaluation indexes; The association strength between the core visual characteristics and each evaluation index is quantified through the pearson correlation coefficient analysis and mutual information calculation, and the characteristic-index combination with the association strength higher than a preset threshold value is screened out; And carrying out multidimensional association analysis based on the screened characteristic-index combination, mining a dynamic rule between visual characteristic change and anesthesia reaction index fluctuation, calculating occurrence probabilities of different reaction types, and generating an anesthesia reaction preliminary analysis result containing reaction type probability distribution.
  5. 5. The visual recognition-based anesthesia response characteristic analysis system according to claim 1, wherein the anesthesia response characteristic analysis module comprises the following functions: Collecting a large amount of clinical anesthesia case data, covering anesthesia response data, synchronous visual characteristic data and relevant probability statistical data corresponding to different anesthesia modes, different patient physique and different operation types, and constructing an anesthesia response characteristic sample library; determining the dimension of the input features of the model based on the primary analysis result of the anesthesia reaction, taking the core visual features, the associated medical monitoring data and the feature associated probability as input variables, and taking the actual value and the occurrence probability of the clinical anesthesia reaction evaluation index as output variables to construct an initial anesthesia reaction characteristic dynamic model; the initial anesthesia reaction characteristic dynamic model is trained by adopting a gradient descent algorithm and combining a cross verification method, and model parameters are continuously adjusted to minimize prediction errors; Clinical verification is carried out on the anesthesia response characteristic analysis data output by the model, and the anesthesia response characteristic dynamic model is continuously optimized by combining the evaluation feedback of an anesthesiologist.
  6. 6. The visual recognition-based anesthesia response performance analysis system of claim 5 wherein the evaluation feedback continuously optimizes an anesthesia response performance dynamic model comprising: Visual feature sequence data in the anesthetic scene multisource fusion data are obtained, time window division is carried out, and visual feature change trend, mutation point information and feature occurrence probability in each time window are extracted; Analyzing time sequence association between the visual characteristic mutation points and vital sign fluctuation and anesthetic drug infusion nodes by combining synchronously acquired medical monitoring data, calculating occurrence probability of association events, and generating characteristic-physiological-drug association data containing the time sequence association probability; Constructing an anesthesia reaction time sequence change map based on the associated data, wherein the map comprises a visual characteristic track, a physiological index curve, a drug infusion curve, an associated mark among the visual characteristic track, the physiological index curve, the drug infusion curve and related probability information, mining a dynamic rule in the anesthesia reaction time sequence change map, and integrating the mined dynamic rule and corresponding probability information into an anesthesia reaction characteristic dynamic model to continuously optimize anesthesia reaction characteristic analysis data.
  7. 7. The visual recognition-based anesthesia response characteristic analysis system of claim 6 wherein the generating feature-physiology-drug associated data comprising a time-series associated probability comprises: Preprocessing visual characteristic sequence data in the multi-source fusion data of the anesthetic scene, removing abnormal sequence fragments caused by equipment jitter and light change, complementing the missing sequence data by adopting an interpolation method, and ensuring the continuity, the integrity and the probability calculation accuracy of the sequence data; dividing time windows according to preset time intervals, dynamically adjusting the length of each time window according to different stages of the anesthesia process, shortening the window length of the anesthesia induction period and the resuscitation period, and properly prolonging the window length of the maintenance period; Carrying out statistical analysis on the visual feature data in each time window, calculating feature mean value, variance, change rate, peak value and feature occurrence probability, extracting feature change trend, identifying visual feature mutation points in the time windows through an anomaly detection algorithm, and recording the time positions, feature change amplitude and mutation occurrence probability of the visual feature mutation points; the visual characteristic change trend, the mutation point information and the related probability parameters in each time window are in one-to-one correspondence with the heart rate, the blood pressure, the blood oxygen saturation and the anesthetic concentration acquired synchronously, and a correlation index in the time dimension is established; And analyzing the causal relationship among the visual characteristic mutation points, the physiological index fluctuation nodes and the drug infusion nodes by calculating the time difference between the visual characteristic mutation points and the physiological index fluctuation nodes to generate characteristic-physiological-drug association data comprising time sequence association strength, response delay time and association event occurrence probability.
  8. 8. The visual recognition-based anesthesia response characteristic analysis system according to claim 6, wherein the dynamic law comprises a time difference in which the visual characteristic changes ahead of the fluctuation of the physiological index, a response period of the visual characteristic after the adjustment of the drug, and occurrence probabilities of various laws.
  9. 9. The visual recognition-based anesthesia response characteristic analysis system according to claim 8, wherein the dynamic rule mining process comprises: Respectively constructing a visual characteristic track layer, a physiological index graph layer and a drug infusion graph layer by taking a time axis as a horizontal axis, wherein the visual characteristic track layer takes the numerical variation and the occurrence probability of a core visual characteristic as a vertical axis, the physiological index graph layer takes the values of all vital sign parameters and the normal probability range as a vertical axis, and the drug infusion graph layer takes the infusion rate, the accumulated dose and the dose adjustment probability of anesthetic drugs as a vertical axis; Marking key nodes in each layer, marking mutation points, characteristic peak points and corresponding probabilities in a visual characteristic track layer, marking normal range threshold lines, abnormal fluctuation points and abnormal occurrence probabilities in a physiological index curve layer, and marking drug type switching points, dose adjusting points and adjustment probabilities in a drug infusion curve layer; Carrying out connection marking on time-synchronous key nodes in the three layers through the association indexes, and defining the corresponding relation and association probability between the visual characteristic change and the physiological index fluctuation as well as the drug adjustment, so as to form a complete anesthesia reaction time sequence change map; analyzing the anesthesia reaction time sequence change map, identifying repeated characteristic-physiology-medicament change modes, calculating the lead time of visual characteristic change relative to physiological index fluctuation in different modes, the response period of the visual characteristic reaching a stable state after medicament adjustment and the occurrence probability of various modes, and merging the obtained mining including the lead time range, the response period mean value and the mode occurrence probability as supplementary characteristics into the training process of the anesthesia reaction characteristic dynamic model to continuously optimize the anesthesia reaction characteristic analysis data.
  10. 10. The visual recognition-based anesthesia response performance analysis system of claim 9 wherein the training process incorporated into the anesthesia response performance dynamic model to continuously optimize anesthesia response performance analysis data comprises: the parameterized supplementary features, the original core visual features and the associated medical monitoring data together form a model input feature set, and the input dimension of the anesthesia response characteristic dynamic model is expanded; adjusting the network structure of the anesthesia reaction characteristic dynamic model, adding a time sequence characteristic processing layer and a probability calculating unit, and learning and fusing time sequence rule parameters and probability information corresponding to the supplementary characteristics; retraining the optimized anesthesia response characteristic dynamic model by adopting the newly added supplementary features, verifying the influence of time sequence rules and probability information in the supplementary features on the model performance by comparing prediction errors, accuracy, recall rate and probability prediction precision indexes of the anesthesia response characteristic dynamic model before and after training, and generating anesthesia response characteristic analysis data comprising probability evaluation optimization by dynamic iterative computation of the anesthesia response characteristic dynamic model.

Description

Anesthesia response characteristic analysis system based on visual recognition Technical Field The invention relates to the technical field of medical data analysis, in particular to an anesthesia response characteristic analysis system based on visual identification. Background In recent years, along with the continuous improvement of clinical medicine on the requirements of anesthesia precision and individuation, the requirements on real-time monitoring, characteristic analysis and intervention guidance of patient reactions in the anesthesia process are urgent, and higher requirements on timeliness, accuracy and data fusion analysis capability of anesthesia reaction identification are provided. The traditional anesthesia reaction monitoring method is mainly based on fixed clinical threshold judgment, does not form a dynamic and personalized anesthesia reaction characteristic analysis model, and is easy to cause delay of intervention time or judgment deviation. Disclosure of Invention Accordingly, there is a need for an anesthesia response characteristic analysis system based on visual identification, which solves at least one of the above-mentioned problems. In order to achieve the above object, an anesthesia response characteristic analysis system based on visual recognition comprises the following modules: The anesthesia multisource acquisition module is used for acquiring multidimensional visual acquisition equipment and an associated medical data interface in an anesthesia scene, acquiring human visual characteristic data and synchronous medical monitoring data in the anesthesia process based on the equipment and the interface, and performing noise reduction, time sequence alignment and format standardization preprocessing to generate integrated anesthesia scene multisource fusion data; The anesthesia characterization analysis module is used for extracting and characterizing visual features in the anesthesia scene multisource fusion data to generate anesthesia state visual characterization data; The anesthesia reaction analysis module is used for carrying out association analysis by combining the anesthesia state visual representation data and a preset clinical anesthesia reaction evaluation index to generate an anesthesia reaction preliminary analysis result containing reaction type probability distribution; The anesthesia characteristic analysis module is used for constructing an anesthesia reaction characteristic dynamic model based on an anesthesia reaction preliminary analysis result and generating anesthesia reaction characteristic analysis data through iterative optimization; the anesthesia state evaluation module is used for outputting an anesthesia state evaluation report and clinical intervention advice according to the anesthesia response characteristic analysis data. Further, the anesthesia multisource acquisition module comprises the following functions: Acquiring a medical data interface corresponding to a facial expression acquisition device, an eye dynamic capture device, a limb posture monitoring device, an anesthesia monitor and a vital sign monitor under an anesthesia scene; The visual information corresponding to the facial muscle movement characteristics, eyelid opening and closing states, eyeball movement tracks, limb movement amplitude and posture changes of the human body in the anesthesia process is continuously collected through the visual collection equipment, and medical monitoring data corresponding to heart rate, blood pressure, respiratory frequency and anesthetic drug concentration are synchronously collected through a medical data interface, so that original anesthesia related data are formed together; Performing image denoising, fuzzy correction and characteristic point enhancement processing on visual information in the original anesthesia related data, and performing outlier rejection and missing value completion processing on the medical monitoring data; And (3) aligning the processed visual information with the medical monitoring data in time sequence according to the time stamp, unifying the data format and the sampling frequency, and generating integrated anesthesia scene multisource fusion data. Further, the anesthesia characterization analysis module includes the following functions: Separating visual characteristic data and medical monitoring associated data from the anesthesia scene multisource fusion data; extracting layered features of the visual feature data, firstly extracting basic visual features corresponding to facial contours and limb contours, and then digging deep dynamic features corresponding to facial micro-expressions, eye tremor frequency and limb micro-twitches to generate a multi-dimensional visual feature set; performing feature screening and dimension reduction processing on the multi-dimensional visual feature set, removing redundant features and interference features, and reserving core visual features highly related to anesthesia react