Trachea cannula rescue guiding method and system based on intelligent monitoring and auxiliary decision making

Abstract

The invention relates to a tracheal intubation rescue guiding method and system based on intelligent monitoring and auxiliary decision making, and the method comprises the steps: detecting the real-time vital sign state of a patient based on an intelligent sensor for integrally monitoring vital sign indexes; real-time airway operation and anatomical structure images in the tracheal intubation process are obtained; a machine learning algorithm based on time sequence support vector regression is utilized, and through comparative analysis of historical vital sign data and current real-time data, real-time physiological health state evaluation indexes of the patient are obtained; on the basis of real-time airway operation and anatomical structure images, through edge detection, feature extraction and three-dimensional reconstruction technologies, tracheal intubation operation accuracy evaluation and tracheal anatomical structure anomaly recognition are carried out; real-time operation guidance is performed through intelligent aid decision making based on natural language processing and a knowledge graph by integrating real-time physiological health state evaluation indexes of a patient, accuracy evaluation of tracheal intubation operation and abnormal recognition of a tracheal anatomical structure.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence and medical auxiliary decision-making, and in particular to a tracheal intubation rescue guidance method and system based on intelligent monitoring and auxiliary decision-making. Background Art

[0002] Tracheal intubation is the most effective and reliable method to establish an artificial airway. It refers to the technique of inserting a special tracheal tube into the trachea through the oral or nasal cavity and the glottis. This technology can provide the best conditions for relieving airway obstruction, ensuring airway patency, clearing respiratory secretions, preventing aspiration, and assisting or controlling breathing. At present, in tracheal intubation rescue, relevant researchers have used artificial intelligence technology to assist tracheal intubation operations. Although the application of these technologies can significantly improve the accuracy and efficiency of tracheal intubation and reduce errors in human operation. However, there are still some challenges in practical applications, such as adaptability to complex anatomical structures. Therefore, it is necessary to further develop and optimize artificial intelligence technology and strengthen the ability to identify and process complex anatomical structures to improve the accuracy of auxiliary decision-making. Summary of the invention

[0003] In order to improve the accuracy of auxiliary decision-making, the purpose of the present invention is to provide a tracheal intubation rescue guidance method and system based on intelligent monitoring and auxiliary decision-making. The technical solutions adopted are as follows:

[0004] In a first aspect, the present application discloses a tracheal intubation rescue guidance method based on intelligent monitoring and auxiliary decision-making, the method comprising:

[0005] S1. Based on the intelligent sensor that integrates the vital signs indicators, it detects the real-time vital signs status of the patient;

[0006] S2, obtain real-time airway operation and anatomical structure images during endotracheal intubation;

[0007] S3. Using a machine learning algorithm based on time series support vector regression, by comparing and analyzing historical vital signs data with current real-time data, the patient's real-time physiological health status assessment indicators are obtained;

[0008] S4. Based on real-time airway operation and anatomical structure images, edge detection, feature extraction and 3D reconstruction technology are used to evaluate the accuracy of tracheal intubation and identify abnormal tracheal anatomical structures.

[0009] S5. Comprehensively evaluate the patient's real-time physiological health status, tracheal intubation accuracy, and identification of tracheal anatomical abnormalities, and provide real-time operation guidance through intelligent decision-making based on natural language processing and knowledge graphs.

[0010] Furthermore, in step S3, the machine learning algorithm based on time series support vector regression is used to obtain the patient's real-time physiological health status assessment indicators by comparing and analyzing the historical vital sign data with the current real-time data, including:

[0011] S31, constructing a training set based on the historical vital sign data, and inputting the training set into an SVR model based on time series support vector regression for model training, wherein the SVR model adopts an RBF kernel function, and a width parameter γ of the RBF kernel function is dynamically adjusted based on the distribution of the training data, the characteristics of the time series, and the noise level;

[0012] S32, inputting the real-time vital sign data into the trained SVR model for prediction, and obtaining the predicted value of the vital sign in the future period;

[0013] S33. Based on the predicted value of the real-time vital sign data and in combination with the preset health status assessment standard, the patient's real-time physiological health status assessment index is determined by calculating the comprehensive health index.

[0014] Furthermore, in step S31, the width parameter γ of the RBF kernel function is dynamically adjusted based on the following steps:

[0015] S311. Based on historical experience data, a corresponding initial range is set for the width parameter γ;

[0016] S312, determining the distribution characteristics of the training data based on statistical analysis technology;

[0017] S313, determining the periodicity characteristics of the time series based on autocorrelation analysis and spectrum analysis;

[0018] S314, determining the noise level of the time series by comparing the differences of adjacent data points;

[0019] S315 , dynamically adjusting the width parameter γ of the RBF kernel function within the initial range through a heuristic search strategy based on the distribution characteristics of the training data, the periodic characteristics of the time series, and the noise level of the time series.

[0020] Further, in step S315, the width parameter γ of the RBF kernel function is dynamically adjusted within the initial range by a heuristic search strategy based on the distribution characteristics of the training data, the periodic characteristics of the time series, and the noise level of the time series, including:

[0021] S3151, defining a corresponding search space for a width parameter γ of the RBF kernel function within the initial range;

[0022] S3152: Based on the distribution characteristics of the training data, the periodic characteristics of the time series, and the noise level of the time series, the following heuristic function is set:

[0023] H(γ)=α·DF(γ)-β*OP(γ)-ε*PD(γ)-δ*NR(γ);

[0024] Among them, γ represents the width parameter of the RBF kernel function, DF(γ) represents the degree of fit of the model to the distribution of training data, OP(γ) represents the degree of overfitting of the model, PD(γ) represents the deviation of the model in capturing the periodic characteristics of the time series, NR(γ) represents the robustness of the model to noise, and α, β, ε, and δ are all preset weight coefficients;

[0025] S3153, performing iterative search using a heuristic search algorithm in the search space, and in each iteration, selecting the current optimal width parameter value according to the evaluation result of the heuristic function, and using it as the starting point of the next iteration;

[0026] S354. When it is determined that the iteration end condition is reached, stop searching and output the width parameter γ of the current optimal RBF kernel function.

[0027] Furthermore, in step S33, the predicted value based on the real-time vital sign data is combined with a preset health status assessment standard to determine the patient's real-time physiological health status assessment index through calculation of a comprehensive health index, including:

[0028] S331, determining a threshold range of each vital sign based on the health status assessment standard;

[0029] S332, comparing the predicted value of the real-time vital sign data with the threshold range of each vital sign to assign a corresponding status label to each vital sign, wherein the status label includes normal, slightly abnormal, and severely abnormal;

[0030] S333, integrating the status labels of various vital signs, and obtaining the corresponding comprehensive health index through dynamic weight allocation and weighted calculation;

[0031] S334. According to the comprehensive health index, the patient's real-time physiological health status assessment index is determined through preset health status classification rules.

[0032] Furthermore, in step S4, based on the real-time airway operation and anatomical structure images, the accuracy of tracheal intubation operation is evaluated and abnormal tracheal anatomical structure is identified through edge detection, feature extraction and three-dimensional reconstruction technology, including:

[0033] S41, performing edge detection based on the real-time airway operation and the anatomical structure image to obtain a contour edge image of the airway and endotracheal tube;

[0034] S42, based on the contour edge image, performing feature extraction through a multi-dimensional feature fusion and extraction algorithm to obtain a key feature set for evaluating the accuracy of tracheal intubation operation and identifying abnormal tracheal anatomical structure;

[0035] S43, performing three-dimensional reconstruction based on the key feature set to obtain a corresponding three-dimensional reconstruction model;

[0036] S44. Based on the three-dimensional reconstruction model, the accuracy of tracheal intubation operation is evaluated and tracheal anatomical abnormalities are identified through geometric morphology analysis and rule matching algorithm.

[0037] Furthermore, in step S42, the feature extraction is performed based on the contour edge image by a multi-dimensional feature fusion and extraction algorithm, including:

[0038] S421, extracting morphological features, geometric features, and texture features based on the contour edge image;

[0039] S422, fusing the morphological features, geometric features, and texture features in a feature cross-based fusion manner to obtain a multi-dimensional fused feature vector;

[0040] S423. Based on the correlation evaluation method, features with correlation coefficients higher than a preset threshold are selected from the obtained multi-dimensional fusion feature vectors to obtain a key feature set for evaluating the accuracy of tracheal intubation operation and identifying abnormal tracheal anatomical structure.

[0041] Furthermore, in step S44, based on the three-dimensional reconstruction model, the accuracy of tracheal intubation operation is evaluated and tracheal anatomical structure abnormalities are identified through geometric morphology analysis and rule matching algorithm, including:

[0042] S441, performing geometric analysis on the three-dimensional reconstructed model to obtain geometric parameters of the trachea and the tracheal tube;

[0043] S442, determining the specific position of the tracheal tube in the trachea and its morphological characteristics based on the extracted geometric parameters, and matching and comparing them with preset geometric morphological rules to preliminarily determine the accuracy of the tracheal tube operation and whether there is abnormality in the anatomical structure of the trachea;

[0044] S443. Based on the matching and comparison results, quantitatively evaluate the accuracy of tracheal intubation and classify tracheal anatomical abnormalities.

[0045] Furthermore, in step S5, the comprehensive patient's real-time physiological health status assessment indicators, tracheal intubation operation accuracy assessment, and tracheal anatomical structure abnormality identification are used to provide real-time operation guidance through intelligent auxiliary decision-making based on natural language processing and knowledge graph, including:

[0046] S51, using natural language processing technology to extract and format information on the patient's real-time physiological health status assessment indicators, tracheal intubation operation accuracy assessment, and tracheal anatomical structure abnormality identification to obtain structured data;

[0047] S52, based on the structured data, isomorphic semantic analysis and entity association, construct a knowledge graph including the patient's physiological status, operation evaluation and anatomical structure information;

[0048] S53. Based on the knowledge graph, real-time operation guidance is achieved through template-based reasoning and matching.

[0049] In the second aspect, the present application discloses a tracheal intubation rescue guidance system based on intelligent monitoring and auxiliary decision-making, the system comprising a vital sign status monitoring module, a tracheal intubation image acquisition module, a physiological health status assessment module, an airway assessment and identification module, and an auxiliary decision-making module, wherein:

[0050] The vital signs status monitoring module is used to detect the real-time vital signs status of the patient based on the intelligent sensor that integrates the vital signs indicators;

[0051] The endotracheal intubation image acquisition module is used to acquire real-time airway operation and anatomical structure images during the endotracheal intubation process;

[0052] The physiological health status assessment module is used to obtain the patient's real-time physiological health status assessment index by comparing and analyzing historical vital sign data with current real-time data using a machine learning algorithm based on time series support vector regression;

[0053] The airway assessment and identification module is used to evaluate the accuracy of tracheal intubation and identify abnormal tracheal anatomical structures based on real-time airway operation and anatomical structure images through edge detection, feature extraction and three-dimensional reconstruction technology;

[0054] The auxiliary decision-making module is used to comprehensively evaluate the patient's real-time physiological health status, the accuracy of tracheal intubation operation, and the identification of abnormal tracheal anatomical structure, and provide real-time operation guidance through intelligent auxiliary decision-making based on natural language processing and knowledge graphs.

[0055] The present invention has the following beneficial effects:

[0056] 1) By acquiring real-time airway operation and anatomical structure images during tracheal intubation, and using edge detection, feature extraction, and 3D reconstruction technology, through precise spatial positioning and morphological comparison, the objective evaluation of the quality of tracheal intubation operation and the accurate identification of tracheal anatomical structural abnormalities are achieved, thus improving data reliability;

[0057] 2) Comprehensively combining the patient's real-time physiological health status assessment indicators, tracheal intubation accuracy assessment, and identification results of tracheal anatomical abnormalities, and through intelligent decision-making based on natural language processing and knowledge graphs, medical staff can quickly obtain comprehensive patient information and surgical recommendations, thereby improving the efficiency and accuracy of clinical decision-making;

[0058] 3) Using a machine learning algorithm based on time series support vector regression, by comparing and analyzing historical vital signs data with current real-time data, and by optimizing and adjusting the algorithm model, we can obtain the patient's real-time physiological health status assessment indicators. These indicators can more accurately reflect the patient's health status and provide strong data support for subsequent intelligent decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0060] Figure 1 A method flow chart of a tracheal intubation rescue guidance method based on intelligent monitoring and auxiliary decision-making provided by one embodiment of the present invention;

[0061] Figure 2 A system structure diagram of a tracheal intubation rescue guidance system based on intelligent monitoring and auxiliary decision-making provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a tracheal intubation rescue guidance method and system based on intelligent monitoring and auxiliary decision-making proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0063] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0064] The specific scheme of a tracheal intubation rescue guidance method and system based on intelligent monitoring and auxiliary decision-making provided by the present invention is described in detail below in conjunction with the accompanying drawings.

[0065] See also Figure 1 , which shows a method flow chart of a tracheal intubation rescue guidance method based on intelligent monitoring and auxiliary decision-making provided by an embodiment of the present invention, the method comprising:

[0066] Step S1, detecting the real-time vital sign status of the patient based on the intelligent sensor integrated with the vital sign monitoring indicator.

[0067] Specifically, the vital sign indicator includes at least one of heart rate, blood pressure, blood oxygen saturation, and respiratory rate.

[0068] Step S2, acquiring real-time airway operation and anatomical structure images during tracheal intubation.

[0069] Specifically, the present application is based on the collection of real-time airway operation and anatomical structure images during tracheal intubation by a high-definition camera, wherein the high-definition camera is installed above or to the side of the patient's head to capture the full view and key details of the airway during intubation, ensuring that the image is clear, stable and easy to observe.

[0070] Step S3, using a machine learning algorithm based on time series support vector regression, by comparing and analyzing historical vital sign data with current real-time data, obtain the patient's real-time physiological health status assessment index.

[0071] Specifically, this application will first construct a training set based on historical vital signs data, and use the time series support vector regression SVR model, combined with dynamically adjusted RBF kernel function parameters, to perform model training. Then, the real-time collected vital signs data is input into the trained SVR model to predict the vital signs values ​​in the future. Finally, based on the predicted vital signs values, combined with the preset health status assessment criteria, the comprehensive health index is calculated to determine the patient's real-time physiological health status assessment indicators.

[0072] Step S4, based on the real-time airway operation and anatomical structure images, edge detection, feature extraction and three-dimensional reconstruction technology are used to evaluate the accuracy of tracheal intubation operation and identify abnormal tracheal anatomical structure.

[0073] Specifically, the present application will first obtain the contours of the airway and endotracheal tube through edge detection of real-time airway images, and then use a multi-dimensional feature fusion and extraction algorithm to obtain a key feature set by comprehensively analyzing and fusing the extracted shape features, texture features, etc.; finally, these key features are used for three-dimensional reconstruction, and the accuracy of the endotracheal intubation operation is evaluated and tracheal anatomical abnormalities are identified through geometric morphological analysis and rule matching algorithms.

[0074] Step S5, comprehensively considers the patient's real-time physiological health status assessment indicators, tracheal intubation operation accuracy assessment, and identification of tracheal anatomical structure abnormalities, and provides real-time operation guidance through intelligent auxiliary decision-making based on natural language processing and knowledge graphs.

[0075] Specifically, this application uses natural language processing technology to extract and format information about the patient's real-time physiological health status assessment indicators, tracheal intubation operation accuracy assessment, and tracheal anatomical abnormality identification results. Afterwards, a knowledge graph containing physiological status, operation assessment, and anatomical structure information is constructed, and real-time operation guidance is achieved through template reasoning and matching based on the graph.

[0076] As can be seen from the above, the present application discloses a tracheal intubation rescue guidance method based on intelligent monitoring and auxiliary decision-making. By acquiring real-time airway operation and anatomical structure images during the tracheal intubation process, and using edge detection, feature extraction and three-dimensional reconstruction technology, through precise spatial positioning and morphological comparison, it realizes objective evaluation of the quality of tracheal intubation operation and accurate identification of tracheal anatomical abnormalities, thereby improving data reliability; by integrating the patient's real-time physiological health status evaluation indicators, tracheal intubation operation accuracy evaluation and identification results of tracheal anatomical abnormalities, through intelligent auxiliary decision-making based on natural language processing and knowledge graph, medical staff can quickly obtain comprehensive patient information and surgical recommendations, thereby improving the efficiency and accuracy of clinical decision-making; by using a machine learning algorithm based on time series support vector regression, through comparative analysis of historical vital signs data and current real-time data, and through optimization and adjustment of the algorithm model, the patient's real-time physiological health status evaluation indicators are obtained, and these indicators can more accurately reflect the patient's health status, and provide strong data support for subsequent intelligent auxiliary decision-making.

[0077] In one embodiment, in step S3, the machine learning algorithm based on time series support vector regression is used to obtain the patient's real-time physiological health status assessment index by comparing and analyzing historical vital sign data with current real-time data, including:

[0078] Step S31, constructing a training set based on the historical vital signs data, and inputting the training set into an SVR model based on time series support vector regression for model training, wherein the SVR model adopts an RBF kernel function, and the width parameter γ of the RBF kernel function is dynamically adjusted based on the distribution of the training data, the characteristics of the time series, and the noise level.

[0079] Specifically, the RBF kernel function refers to the radial basis function (RBF) kernel function, which is a commonly used support vector machine (SVM) kernel function that can map the input data to a high-dimensional feature space, thereby solving the nonlinear separability problem. Specifically, when implementing, the application will optimize the parameters according to the heuristic search algorithm based on the distribution of the comprehensive training data, the characteristics of the time series, and the noise level to achieve dynamic adjustment of the width parameter γ.

[0080] Step S32, input the real-time vital sign data into the trained SVR model for prediction, and obtain the predicted value of the vital sign in the future period of time.

[0081] Specifically, the application organizes the real-time vital sign data according to the specified input format, and inputs the organized real-time vital sign data into the trained SVR model, which then forward-propagates the data based on the learned knowledge. After forward-propagation processing, the SVR model outputs the predicted vital sign values ​​for a period of time in the future.

[0082] Step S33, based on the predicted value of the real-time vital sign data and in combination with the preset health status assessment standard, the patient's real-time physiological health status assessment index is determined by calculating the comprehensive health index.

[0083] Specifically, this application will determine the normal, slightly abnormal, and severely abnormal threshold ranges for each vital sign based on the health status assessment standard, and compare the real-time collected vital sign data with the set threshold range to assign a corresponding status label to each vital sign. This status label includes normal, slightly abnormal, and severely abnormal. After that, the status labels of all vital signs are comprehensively considered, and a comprehensive health index reflecting the overall health status of the patient is obtained through dynamic weight allocation and weighted calculation. According to the comprehensive health index, combined with the preset health status classification rules, the patient's real-time physiological health status is evaluated to determine the patient's real-time physiological health status evaluation indicators.

[0084] In the above embodiment, on the one hand, based on the SVR model and using the RBF kernel function, the time series characteristics and nonlinear relationships in the vital signs data are captured to improve the accuracy of prediction of future vital signs. Moreover, the addition of the width parameter of the dynamic adjustment RBF kernel function can also further optimize the model performance, so that it can better adapt to the distribution and noise level of vital signs data of different patients. On the other hand, combined with the preset health status assessment standard, the patient's real-time physiological health status assessment index can be determined by calculating the comprehensive health index. This method not only takes into account the abnormality of a single vital sign, but also integrates the status of multiple vital signs, thereby providing a comprehensive and personalized health assessment result.

[0085] In one embodiment, in step S31, the width parameter γ of the RBF kernel function is dynamically adjusted based on the following steps:

[0086] Step S311, based on historical experience data, a corresponding initial range is set for the width parameter γ.

[0087] Specifically, this application determines the minimum value, maximum value, average value, and common value range of the width parameter γ based on the distribution of the width parameter γ in the historical experience data. According to these statistical information, combined with the actual application scenario, a reasonable and representative initial range is set for the width parameter γ. For example, the initial range can be set to [minimum value + α, maximum value - β], where α and β are adjustment coefficients determined according to statistical characteristics and actual needs, to ensure that most of the width parameter values ​​in the historical data are included, and new situations that may arise in the future can be taken into account.

[0088] Step S312: determining the distribution characteristics of the training data based on statistical analysis techniques.

[0089] Specifically, this application will determine the mean, variance, skewness, kurtosis and other characteristic quantities of the data based on statistical analysis techniques, and based on these statistics, fit the distribution characteristics of the training data through statistical distribution fitting techniques. Specifically, these statistics will be input into a specified probability distribution model (such as normal distribution, lognormal distribution, exponential distribution, etc.), and the parameters of the model will be estimated using parameter estimation methods. By comparing the goodness of fit of different models, the model with the best goodness of fit will be selected to determine the distribution characteristics of the training data.

[0090] Step S313, determining the periodicity characteristics of the time series based on autocorrelation analysis and spectrum analysis.

[0091] Specifically, the present application will determine the peak position of the autocorrelation function of the time series based on the autocorrelation analysis. By analyzing the peak position of the autocorrelation function, the time interval of the main periodic components in the time series can be determined.

[0092] Furthermore, the present application will also convert the time series from the time domain to the frequency domain based on spectrum analysis to reveal the various frequency components contained in the time series.

[0093] After that, the periodicity of the time series is determined by matching and verifying the period length obtained by the autocorrelation analysis with the frequency components obtained by the spectrum analysis. The matching and verification process includes converting the period length corresponding to the peak position of the autocorrelation function into the corresponding frequency (i.e. calculating the inverse of the period length) and comparing it with the peak frequency on the spectrum graph to determine the periodicity of the time series.

[0094] Step S314, determining the noise level of the time series by comparing the differences between adjacent data points.

[0095] Specifically, considering that noise usually manifests itself as random fluctuations or irregular changes in data, this application will calculate the differences between adjacent data points in the time series, which reflect the local fluctuations of the data. Afterwards, by calculating the variance of these differences and based on the relative difference between them and a preset benchmark (this benchmark can be set based on the average noise level of historical data) (i.e. (variance-benchmark) / benchmark), the noise level in the time series is quantitatively evaluated. Among them, the larger the relative difference value, the higher the degree of deviation of the noise level of the current time series from the benchmark, that is, the higher the noise level, and vice versa, the lower the noise level.

[0096] Step S315 , comprehensively considering the distribution characteristics of the training data, the periodic characteristics of the time series, and the noise level of the time series, dynamically adjusting the width parameter γ of the RBF kernel function within the initial range through a heuristic search strategy.

[0097] Specifically, this application will define a reasonable search space within the set initial range. Afterwards, based on the distribution characteristics of the training data, the periodic characteristics of the time series, and the noise level of the time series, by comprehensively considering the impact of these factors on the model performance, a corresponding heuristic function is set (for details, please refer to the subsequent embodiments) to evaluate the markings of different width parameters γ on the training data. Finally, it is necessary to use a heuristic search method to perform an iterative search in the defined search space to find the RBF kernel function width parameter γ that is optimal or approximately optimal for another heuristic function in the current search space.

[0098] In the above embodiment, on the one hand, the distribution characteristics of the training data, the periodic characteristics of the time series and the noise level can be comprehensively understood to understand the essential characteristics of the data, thereby providing a more accurate basis for the selection of the width parameter of the RBF kernel function. On the other hand, the width parameter γ of the RBF kernel function is dynamically adjusted through a heuristic search strategy to achieve the optimal configuration of the parameters. This dynamic adjustment method not only improves the adaptability of the model to the data, but also enhances the generalization performance of the model.

[0099] In one embodiment, in step S315, the width parameter γ of the RBF kernel function is dynamically adjusted within the initial range by a heuristic search strategy based on the distribution characteristics of the training data, the periodic characteristics of the time series, and the noise level of the time series, including:

[0100] Step S3151, defining a corresponding search space for a width parameter γ of the RBF kernel function within the initial range.

[0101] Step S3152, based on the distribution characteristics of the training data, the periodic characteristics of the time series, and the noise level of the time series, the following heuristic function is set:

[0102] H(γ)=α·DF(γ)-β*OP(γ)-ε*PD(γ)-δ*NR(γ);

[0103] Among them, γ represents the width parameter of the RBF kernel function, DF(γ) represents the degree of fit of the model to the distribution of training data, OP(γ) represents the degree of overfitting of the model, PD(γ) represents the deviation of the model in capturing the periodic characteristics of the time series, NR(γ) represents the robustness of the model to noise, and α, β, ε, and δ are all preset weight coefficients.

[0104] Step S3153, in the search space, an iterative search is performed using a heuristic search algorithm. In each iteration, the current optimal width parameter value is selected according to the evaluation result of the heuristic function, and is used as the starting point for the next iteration.

[0105] Specifically, the application of the heuristic search algorithm will generate a set of candidate width parameter γ values ​​in each iteration process; then, these candidate parameters are evaluated based on the heuristic function constructed in step S3152, and the width parameter value with the best current evaluation result is selected as the starting point for the next iteration. Through continuous iterative optimization, the width parameter value gradually approaches the global optimal solution.

[0106] Step S354: when it is determined that the iteration end condition is met, stop searching and output the width parameter γ of the current optimal RBF kernel function.

[0107] Specifically, the iteration end condition can be a preset number of iterations, finding the optimal solution, and reaching the corresponding performance improvement threshold. When the iteration end condition is met, the search will stop and the width parameter value of the RBF kernel function with the best current evaluation result will be output. This optimal parameter value will be used to build the final SVR model to achieve accurate prediction of future vital sign data.

[0108] In one embodiment, in step S33, the predicted value based on the real-time vital sign data is combined with a preset health status assessment standard to determine the patient's real-time physiological health status assessment index through calculation of a comprehensive health index, including:

[0109] Step S331, determining the threshold range of each vital sign based on the health status assessment standard.

[0110] Specifically, this application will set a corresponding threshold range for each vital sign based on practical experience, and these ranges correspond to different health states (such as normal, slightly abnormal, and severely abnormal). These threshold ranges will serve as the basis for subsequent judgment of the vital sign status.

[0111] Step S332, comparing the predicted value of the real-time vital sign data with the threshold range of each vital sign to assign a corresponding status label to each vital sign, wherein the status label includes normal, slightly abnormal, and severely abnormal.

[0112] Specifically, the present application will compare the predicted value of the real-time vital sign data with the threshold range of each vital sign determined in step S331, and assign a corresponding status label to each vital sign based on the comparison result, wherein the status label reflects the current health status of the corresponding vital sign.

[0113] Step S333, integrating the status labels of various vital signs, through dynamic weight allocation and weighted calculation, to obtain the corresponding comprehensive health index.

[0114] Specifically, this application will quantify the status label of each vital sign into a corresponding value. For example, the normal state label is quantified to a value of 1, the mild abnormal state label is quantified to a value of 0.5, and the severe abnormal state label is quantified to a value of 0.1 (this application does not limit the specific quantized values, and different embodiments can be dynamically adjusted according to actual health assessment needs). Then, according to the importance and urgency of each vital sign, the formula W is used. i =α i ×β i , assign a corresponding dynamic weight to each state label. Among them, W i is the weight assigned to the i-th state label, α iis the importance coefficient determined by the expert scoring method according to the importance of vital sign i, β i is the urgency coefficient determined by the expert scoring method according to the urgency of vital sign i. Finally, the corresponding comprehensive health index is obtained through weighted summation calculation.

[0115] Step S334, according to the comprehensive health index, through the preset health status classification rules, determine the patient's real-time physiological health status evaluation index.

[0116] Specifically, the preset health status classification rules will divide the comprehensive health index into different intervals, for example:

[0117] Interval 1: The comprehensive health index is above 0.9 (including 0.9), and the corresponding health status assessment indicator is "good health status", which also means that the patient's overall health status is excellent and all vital signs are within the normal range;

[0118] Interval 2: The comprehensive health index is between 0.6 and 0.9 (excluding 0.9 and including 0.6), and the corresponding health status assessment indicator is "sub-health status", which also means that some of the patient's vital signs may be slightly abnormal, but the overall health status is acceptable;

[0119] Interval 3: The comprehensive health index is between 0.3 and 0.6 (excluding 0.6 and including 0.3), and the corresponding health status assessment indicator is "disease risk exists", which also means that the patient's multiple vital signs are abnormal, the overall health condition is poor, and there is a potential disease risk;

[0120] Range 4: The comprehensive health index is below 0.3 (excluding 0.3), and the corresponding health status assessment indicator is "critical health condition", which also means that the patient's vital signs are seriously abnormal and the overall health condition is extremely poor.

[0121] In this way, by corresponding each interval to a specific health status assessment indicator, a reliable basis can be provided for subsequent auxiliary decision-making.

[0122] In the above embodiment, on the one hand, by combining the status labels of various vital signs, a comprehensive comprehensive health index can be obtained through dynamic weight allocation and weighted calculation. This index can reflect the overall health status of the patient, not just the abnormality of a single vital sign. On the other hand, according to the comprehensive health index, the patient's real-time physiological health status assessment index is determined through the preset health status classification rules, providing a reliable basis for subsequent auxiliary decision-making.

[0123] In one embodiment, in step S4, the tracheal intubation operation accuracy assessment and tracheal anatomical structure abnormality identification are performed based on real-time airway operation and anatomical structure images through edge detection, feature extraction and three-dimensional reconstruction technology, including:

[0124] Step S41 , performing edge detection based on the real-time airway operation and the anatomical structure image to obtain a contour edge image of the airway and the endotracheal tube.

[0125] Specifically, the present application will perform edge detection on the real-time airway operation and anatomical structure images based on an image processing algorithm, and optimize the parameters according to refinement to obtain a high-precision contour edge image, including adjusting the threshold of the edge detection algorithm, filter size, image smoothing, etc., to adapt to the airway and endotracheal tube edge features under different image quality and lighting conditions.

[0126] Step S42, based on the contour edge image, feature extraction is performed through a multi-dimensional feature fusion and extraction algorithm to obtain a key feature set for evaluating the accuracy of tracheal intubation operation and identifying abnormal tracheal anatomical structure.

[0127] Specifically, this application will extract the morphological features, geometric features and texture features of the object from the contour edge image, and combine the above features to form a new multi-dimensional feature vector by cross-fusion. Then, based on the correlation evaluation method, the key feature set that is highly correlated with the accuracy of tracheal intubation operation and the identification of tracheal anatomical abnormalities is screened out, and three-dimensional reconstruction is performed based on the key feature set.

[0128] Step S43, performing three-dimensional reconstruction based on the key feature set to obtain a corresponding three-dimensional reconstruction model.

[0129] Specifically, the present application will extract the surface information of the airway and endotracheal tube from the key features based on the surface rendering technology, and build a corresponding three-dimensional mesh model based on the surface information. Afterwards, the three-dimensional mesh model is rendered using graphics technologies such as texture mapping and illumination models to obtain a realistic three-dimensional reconstruction model.

[0130] Step S44, based on the three-dimensional reconstructed model, the accuracy of the tracheal intubation operation is evaluated and the abnormal anatomical structure of the trachea is identified through geometric morphology analysis and rule matching algorithm.

[0131] Specifically, geometric morphological analysis is performed on the three-dimensional reconstruction model to extract the geometric parameters of the trachea and endotracheal tube, which further determine the position and shape of the endotracheal tube. After that, after the position and shape of the endotracheal tube are determined, it is compared and matched with the preset rules to preliminarily evaluate the accuracy of the operation and identify abnormalities in the anatomical structure of the trachea. Finally, the degree of deviation is calculated to quantitatively evaluate the accuracy of the endotracheal intubation operation, and the abnormalities in the anatomical structure of the trachea are classified based on the preset classification system to accurately identify and distinguish different types of abnormalities.

[0132] In the above embodiment, on the one hand, through the multi-dimensional feature fusion and extraction algorithm, a key feature set can be extracted from the contour edge image, and these features can accurately reflect the accuracy of the tracheal intubation operation and the abnormality of the tracheal anatomical structure, and provide effective data for subsequent geometric morphology analysis and rule matching. On the other hand, based on the three-dimensional reconstruction model and the geometric morphology analysis and rule matching algorithm, the tracheal intubation operation can be further quantitatively evaluated, and the abnormal tracheal anatomical structure can be classified, thereby achieving an objective evaluation of the quality of the tracheal intubation operation and accurate identification of the abnormal tracheal anatomical structure.

[0133] In one embodiment, in step S42, the feature extraction is performed based on the contour edge image by a multi-dimensional feature fusion and extraction algorithm, including:

[0134] Step S421: extracting morphological features, geometric features, and texture features based on the contour edge image.

[0135] Specifically, the morphological features include area, perimeter, shape index, etc., which can describe the overall shape and contour characteristics of the object; the geometric features include the curvature, direction, symmetry, etc. of the contour line, which can reflect the geometric shape and contour details of the object; the texture features include the statistics of the grayscale co-occurrence matrix (such as contrast, energy, entropy, correlation, etc.), local binary patterns and their statistical characteristics, wavelet transform coefficients, etc., which can capture the grayscale distribution and arrangement rules of pixels in the image, thereby describing the texture information of the image.

[0136] Step S422: The morphological features, geometric features, and texture features are fused in a feature cross-based fusion manner to obtain a multi-dimensional fused feature vector.

[0137] Specifically, this application will cross-combine different types of features through feature splicing, feature mapping, and feature transformation and association mining to obtain new features. These new features can capture the interactions and associations between different types of features, thereby providing richer information. Specifically including: 1) combining morphological features (such as area, perimeter) and geometric features (such as the curvature of the contour line) through feature splicing to form a new feature; 2) combining texture features and geometric features through feature mapping and fusion to form a new feature that can reflect the interaction between local details and global shapes of the image; 3) combining morphological features and texture features through feature transformation and association mining to form a new feature that can reflect the comprehensive characteristics of object morphology and texture.

[0138] Step S423, based on the correlation evaluation method, select features with correlation coefficients higher than a preset threshold from the obtained multi-dimensional fusion feature vectors to obtain a key feature set for evaluating the accuracy of tracheal intubation operation and identifying abnormal tracheal anatomical structure.

[0139] Specifically, this application will use a correlation analysis method to evaluate the correlation between each fusion feature and the target variable (i.e., the accuracy of the tracheal intubation operation, the abnormal state of the tracheal anatomical structure). In specific implementation, the correlation coefficient between each fusion feature and the target variable will be calculated first; then, according to the preset threshold (the threshold can be set in the actual application scenario, and this application does not limit this), the features with a correlation coefficient higher than the threshold will be screened out. These screened features will be used for subsequent tracheal intubation accuracy evaluation and identification of tracheal anatomical abnormalities.

[0140] In the above embodiments, different types of features are combined together to form a new multi-dimensional fusion feature vector through feature concatenation, feature mapping, feature transformation and association mining. This fusion method not only enhances the correlation between features, but also mines the potential interactions between features, providing richer and more effective information for subsequent evaluation and recognition tasks.

[0141] In one embodiment, in step S44, the tracheal intubation operation accuracy assessment and tracheal anatomical structure abnormality identification are performed based on the three-dimensional reconstruction model through geometric morphology analysis and rule matching algorithm, including:

[0142] Step S441, performing geometric analysis on the three-dimensional reconstructed model to obtain geometric parameters of the trachea and the endotracheal tube.

[0143] Specifically, this application will first use volume rendering technology to extract the internal structure of the trachea and the positional relationship of the tracheal tube. After that, the surface reconstruction technology is used to generate an accurate three-dimensional surface model of the trachea and the tracheal tube. Finally, the three-dimensional surface model is imported into the three-dimensional measurement tool, and the measurement function provided in the tool is used to measure the geometric parameters of the trachea such as diameter, length, curvature, and key parameters of the tracheal tube such as diameter, insertion depth, and contact area with the tracheal wall.

[0144] Step S442, based on the extracted geometric parameters, the specific position of the tracheal tube in the trachea and its morphological characteristics are determined, and they are matched and compared with the preset geometric morphological rules to preliminarily determine the accuracy of the tracheal intubation operation and whether there are any abnormalities in the anatomical structure of the trachea.

[0145] Specifically, the present application will first analyze the position of the endotracheal tube in the trachea based on the extracted geometric parameters, including the insertion depth, the relative position to the tracheal wall, and the morphological characteristics of the endotracheal tube, such as whether there are bends, twists or deformations. Then, these analysis results are matched and compared with the preset geometric morphological rules. Among them, the geometric morphological rules include the ideal insertion depth range of the endotracheal tube, the matching relationship between the tracheal diameter and the tube diameter, the correct position and direction of the endotracheal tube, etc. Through such a comparison, it is possible to preliminarily judge whether the endotracheal intubation operation is accurate, and whether there are abnormalities in the anatomical structure of the trachea, such as tracheal stenosis, tracheal deviation or tracheal wall damage.

[0146] Step S443, based on the matching and comparison results, quantitatively evaluate the accuracy of the tracheal intubation operation and classify the tracheal anatomical structural abnormalities.

[0147] Specifically, when quantitatively evaluating the accuracy of tracheal intubation, the present application will calculate the deviation between the tracheal intubation and the preset geometric morphology rules, such as the deviation of the insertion depth, the position deviation, etc., and achieve quantitative evaluation by comparing the deviation with the preset deviation threshold. Furthermore, the present application will also classify tracheal anatomical structural abnormalities through a classification system determined based on medical expertise and clinical experience to accurately identify and distinguish different types of abnormalities.

[0148] In one embodiment, in step S5, the comprehensive patient's real-time physiological health status assessment indicators, tracheal intubation operation accuracy assessment, and tracheal anatomical structure abnormality identification are used to provide real-time operation guidance through intelligent auxiliary decision-making based on natural language processing and knowledge graph, including:

[0149] Step S51, using natural language processing technology to extract and format information on the patient's real-time physiological health status assessment indicators, tracheal intubation operation accuracy assessment, and tracheal anatomical structure abnormality identification to obtain structured data.

[0150] Step S52, based on the structured data, isomorphic semantic analysis and entity association, construct a knowledge graph containing the patient's physiological status, operation evaluation and anatomical structure information.

[0151] Step S53, based on the knowledge graph, through template-based reasoning matching, to achieve real-time operation guidance operation.

[0152] Based on step S51 to step S53, it should be noted that the knowledge graph is a graphical data structure used to represent entities (such as patients, physiological states, operations, etc.) and their relationships. Through semantic analysis and entity association, these entities can be further identified and connected to build a complete knowledge graph containing patient physiological states, operation evaluations and anatomical structure information. Before using the knowledge graph for real-time operation guidance, this application will define a series of templates that describe the operations to be taken in different situations. It is specifically based on medical knowledge and operating specifications, which can ensure that the advice given is both accurate and meets clinical requirements. When receiving the patient's real-time information (such as heart rate, blood pressure, blood oxygen saturation, tracheal intubation position, and any abnormalities in the tracheal anatomical structure, etc.), this application will match this information with the entities and relationships in the knowledge graph. Through semantic analysis and entity association, it is possible to identify which template best matches the current situation. Once a matching template is found, reasoning and decision-making can be performed based on the template to achieve real-time operation guidance.

[0153] In the above embodiment, based on the knowledge graph, through template-based reasoning and matching, it is possible to provide medical staff with operational guidance in real time. This guidance is based on the patient's current real-time information and predefined medical knowledge templates, so it can ensure that the advice given is both accurate and meets clinical requirements. This helps to improve the accuracy and safety of medical operations and reduce the occurrence of medical accidents.

[0154] Please refer to Figure 2 The present application discloses a tracheal intubation rescue guidance system based on intelligent monitoring and auxiliary decision-making, the system includes a vital sign status monitoring module, a tracheal intubation image acquisition module, a physiological health status assessment module, an airway assessment and identification module, and an auxiliary decision-making module, wherein:

[0155] The vital sign status monitoring module is used to detect the real-time vital sign status of the patient based on the intelligent sensor that integrates the vital sign indicators.

[0156] The endotracheal intubation image acquisition module is used to acquire real-time airway operation and anatomical structure images during the endotracheal intubation process.

[0157] The physiological health status assessment module is used to obtain the patient's real-time physiological health status assessment index by comparing and analyzing historical vital sign data with current real-time data using a machine learning algorithm based on time series support vector regression.

[0158] The airway assessment and identification module is used to evaluate the accuracy of tracheal intubation operations and identify abnormal tracheal anatomical structures based on real-time airway operations and anatomical structure images through edge detection, feature extraction and three-dimensional reconstruction technology.

[0159] The auxiliary decision-making module is used to comprehensively evaluate the patient's real-time physiological health status, the accuracy of tracheal intubation operation, and the identification of abnormal tracheal anatomical structure, and provide real-time operation guidance through intelligent auxiliary decision-making based on natural language processing and knowledge graphs.

[0160] In one embodiment, the above modules are also used to implement a tracheal intubation rescue guidance method based on intelligent monitoring and auxiliary decision-making as described in any of the aforementioned method embodiments, and this application does not limit this.

[0161] As can be seen from the above, the present application discloses a tracheal intubation rescue guidance system based on intelligent monitoring and auxiliary decision-making. By acquiring real-time airway operation and anatomical structure images during the tracheal intubation process, and using edge detection, feature extraction and three-dimensional reconstruction technology, through precise spatial positioning and morphological comparison, it realizes objective evaluation of the quality of tracheal intubation operation and accurate identification of tracheal anatomical abnormalities, thereby improving data reliability; by integrating the patient's real-time physiological health status evaluation indicators, tracheal intubation operation accuracy evaluation and identification results of tracheal anatomical abnormalities, through intelligent auxiliary decision-making based on natural language processing and knowledge graph, medical staff can quickly obtain comprehensive patient information and surgical recommendations, thereby improving the efficiency and accuracy of clinical decision-making; by using a machine learning algorithm based on time series support vector regression, through comparative analysis of historical vital signs data and current real-time data, and through optimization and adjustment of the algorithm model, the patient's real-time physiological health status evaluation indicators are obtained, and these indicators can more accurately reflect the patient's health status, and provide strong data support for subsequent intelligent auxiliary decision-making.

[0162] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0163] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A tracheal intubation rescue guidance method based on intelligent monitoring and auxiliary decision-making, characterized in that: The method comprises: S1. Based on the intelligent sensor that integrates the vital signs indicators, it detects the real-time vital signs status of the patient; S2, obtain real-time airway operation and anatomical structure images during endotracheal intubation; S3. Using a machine learning algorithm based on time series support vector regression, by comparing and analyzing historical vital signs data with current real-time data, the patient's real-time physiological health status assessment indicators are obtained; S4. Based on real-time airway operation and anatomical structure images, edge detection, feature extraction and 3D reconstruction technology are used to evaluate the accuracy of tracheal intubation and identify abnormal tracheal anatomical structures. S5. Comprehensively evaluate the patient's real-time physiological health status, tracheal intubation accuracy, and identification of tracheal anatomical abnormalities, and provide real-time operation guidance through intelligent decision-making based on natural language processing and knowledge graphs.

2. The method according to claim 1, characterized in that In step S3, the machine learning algorithm based on time series support vector regression is used to obtain the patient's real-time physiological health status assessment indicators by comparing and analyzing historical vital sign data with current real-time data, including: S31, constructing a training set based on the historical vital sign data, and inputting the training set into an SVR model based on time series support vector regression for model training, wherein the SVR model adopts an RBF kernel function, and a width parameter γ of the RBF kernel function is dynamically adjusted based on the distribution of the training data, the characteristics of the time series, and the noise level; S32, inputting the real-time vital sign data into the trained SVR model for prediction, and obtaining the predicted value of the vital sign in the future period; S33. Based on the predicted value of the real-time vital sign data and in combination with the preset health status assessment standard, the patient's real-time physiological health status assessment index is determined by calculating the comprehensive health index.

3. The method according to claim 2, characterized in that In step S31, the width parameter γ of the RBF kernel function is dynamically adjusted based on the following steps: S311. Based on historical experience data, a corresponding initial range is set for the width parameter γ; S312, determining the distribution characteristics of the training data based on statistical analysis technology; S313, determining the periodicity characteristics of the time series based on autocorrelation analysis and spectrum analysis; S314, determining the noise level of the time series by comparing the differences of adjacent data points; S315 , dynamically adjusting the width parameter γ of the RBF kernel function within the initial range through a heuristic search strategy based on the distribution characteristics of the training data, the periodic characteristics of the time series, and the noise level of the time series.

4. The method according to claim 3, characterized in that In step S315, the width parameter γ of the RBF kernel function is dynamically adjusted within the initial range by a heuristic search strategy based on the distribution characteristics of the training data, the periodic characteristics of the time series, and the noise level of the time series, including: S3151, defining a corresponding search space for a width parameter γ of the RBF kernel function within the initial range; S3152: Based on the distribution characteristics of the training data, the periodic characteristics of the time series, and the noise level of the time series, the following heuristic function is set: H(γ)=α·DF(γ)-β*OP(γ)-ε*PD(γ)-δ*NR(γ); Among them, γ represents the width parameter of the RBF kernel function, DF(γ) represents the degree of fit of the model to the distribution of training data, OP(γ) represents the degree of overfitting of the model, PD(γ) represents the deviation of the model in capturing the periodic characteristics of the time series, NR(γ) represents the robustness of the model to noise, and α, β, ε, and δ are all preset weight coefficients; S3153, performing iterative search using a heuristic search algorithm in the search space, and in each iteration, selecting the current optimal width parameter value according to the evaluation result of the heuristic function, and using it as the starting point of the next iteration; S354. When it is determined that the iteration end condition is reached, stop searching and output the width parameter γ of the current optimal RBF kernel function.

5. The method according to claim 2, characterized in that: In step S33, the predicted value based on the real-time vital sign data is combined with the preset health status assessment standard to determine the patient's real-time physiological health status assessment index through calculation of the comprehensive health index, including: S331, determining a threshold range of each vital sign based on the health status assessment standard; S332, comparing the predicted value of the real-time vital sign data with the threshold range of each vital sign to assign a corresponding status label to each vital sign, wherein the status label includes normal, slightly abnormal, and severely abnormal; S333, integrating the status labels of various vital signs, and obtaining the corresponding comprehensive health index through dynamic weight allocation and weighted calculation; S334. According to the comprehensive health index, the patient's real-time physiological health status assessment index is determined through preset health status classification rules.

6. The method according to claim 1, characterized in that In step S4, based on the real-time airway operation and anatomical structure images, edge detection, feature extraction and three-dimensional reconstruction technology are used to evaluate the accuracy of tracheal intubation operation and identify abnormal tracheal anatomical structure, including: S41, performing edge detection based on the real-time airway operation and the anatomical structure image to obtain a contour edge image of the airway and endotracheal tube; S42, based on the contour edge image, performing feature extraction through a multi-dimensional feature fusion and extraction algorithm to obtain a key feature set for evaluating the accuracy of tracheal intubation operation and identifying abnormal tracheal anatomical structure; S43, performing three-dimensional reconstruction based on the key feature set to obtain a corresponding three-dimensional reconstruction model; S44. Based on the three-dimensional reconstruction model, the accuracy of tracheal intubation operation is evaluated and tracheal anatomical abnormalities are identified through geometric morphology analysis and rule matching algorithm.

7. The method according to claim 6, characterized in that In step S42, the feature extraction is performed based on the contour edge image by a multi-dimensional feature fusion and extraction algorithm, including: S421, extracting morphological features, geometric features, and texture features based on the contour edge image; S422, fusing the morphological features, geometric features, and texture features in a feature cross-based fusion manner to obtain a multi-dimensional fused feature vector; S423. Based on the correlation evaluation method, features with correlation coefficients higher than a preset threshold are selected from the obtained multi-dimensional fusion feature vectors to obtain a key feature set for evaluating the accuracy of tracheal intubation operation and identifying abnormal tracheal anatomical structure.

8. The method according to claim 6, characterized in that In step S44, based on the three-dimensional reconstruction model, the accuracy of tracheal intubation operation is evaluated and tracheal anatomical structure abnormalities are identified through geometric morphology analysis and rule matching algorithm, including: S441, performing geometric analysis on the three-dimensional reconstructed model to obtain geometric parameters of the trachea and the tracheal tube; S442, determining the specific position of the tracheal tube in the trachea and its morphological characteristics based on the extracted geometric parameters, and matching and comparing them with preset geometric morphological rules to preliminarily determine the accuracy of the tracheal tube operation and whether there is abnormality in the anatomical structure of the trachea; S443. Based on the matching and comparison results, quantitatively evaluate the accuracy of tracheal intubation and classify tracheal anatomical abnormalities.

9. The method according to claim 1, characterized in that: In step S5, the comprehensive patient's real-time physiological health status assessment indicators, tracheal intubation operation accuracy assessment, and tracheal anatomical structure abnormality identification are used to provide real-time operation guidance through intelligent auxiliary decision-making based on natural language processing and knowledge graph, including: S51, using natural language processing technology to extract and format information on the patient's real-time physiological health status assessment indicators, tracheal intubation operation accuracy assessment, and tracheal anatomical structure abnormality identification to obtain structured data; S52, based on the structured data, isomorphic semantic analysis and entity association, construct a knowledge graph including the patient's physiological status, operation evaluation and anatomical structure information; S53. Based on the knowledge graph, real-time operation guidance is achieved through template-based reasoning and matching.

10. A tracheal intubation rescue guidance system based on intelligent monitoring and auxiliary decision-making, characterized in that: The system includes a vital sign status monitoring module, an endotracheal intubation image acquisition module, a physiological health status assessment module, an airway assessment and identification module, and an auxiliary decision-making module, wherein: The vital signs status monitoring module is used to detect the real-time vital signs status of the patient based on the intelligent sensor that integrates the vital signs indicators; The endotracheal intubation image acquisition module is used to acquire real-time airway operation and anatomical structure images during the endotracheal intubation process; The physiological health status assessment module is used to obtain the patient's real-time physiological health status assessment index by comparing and analyzing historical vital sign data with current real-time data using a machine learning algorithm based on time series support vector regression; The airway assessment and identification module is used to evaluate the accuracy of tracheal intubation and identify abnormal tracheal anatomical structures based on real-time airway operation and anatomical structure images through edge detection, feature extraction and three-dimensional reconstruction technology; The auxiliary decision-making module is used to comprehensively evaluate the patient's real-time physiological health status, the accuracy of tracheal intubation operation, and the identification of abnormal tracheal anatomical structure, and provide real-time operation guidance through intelligent auxiliary decision-making based on natural language processing and knowledge graphs.

Images