Fuzzy logic method-based cardiac arrest risk level prediction method and system

Abstract

The invention relates to a cardiac arrest risk grade prediction method and system based on a fuzzy logic method, and the method comprises the steps: obtaining real-time vital sign data of a patient, the real-time vital sign data comprising at least one of heart rate, blood pressure, oxyhemoglobin saturation and respiratory rate; on the basis of the real-time vital sign data, feature selection and importance evaluation processing are carried out through a machine learning algorithm, and target features most related to the sudden cardiac arrest risk are obtained; performing fuzzification processing based on fuzzy C-means clustering on the basis of the target features so as to convert the accurate numerical data into a fuzzy set; according to a fuzzy logic principle, in combination with a case-based reasoning method, performing matching calculation on the fuzzy set and known sudden cardiac arrest case characteristics to obtain a sudden cardiac arrest risk level; based on the cardiac arrest risk level, risk level evaluation is carried out, and a corresponding early warning mechanism is triggered. The implementation of the method can improve the accuracy and comprehensiveness of prediction.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and medical health technology, and in particular to a method and system for predicting the risk level of cardiac arrest based on fuzzy logic. Background Art

[0002] Cardiac arrest refers to the sudden cessation of heartbeat, which leads to interruption of blood circulation and is a serious disease that threatens human life safety. Timely and accurate risk assessment and prediction are crucial for the prevention and emergency treatment of cardiac arrest. At present, the risk assessment of cardiac arrest mainly relies on the patient's clinical characteristics, medical history, physical examination results, laboratory tests and other information. However, this information often has uncertainty, ambiguity and incompleteness, which brings certain challenges to risk assessment. Fuzzy logic is an effective method to deal with uncertainty, ambiguity and incomplete information. It establishes fuzzy sets and fuzzy rules to reason and judge uncertain information, so as to draw more accurate and reliable conclusions. Considering that the risk of cardiac arrest is determined by multiple factors, the interactions between these factors are often complex and nonlinear. The use of fuzzy logic can effectively integrate these multivariate information, capture the complex associations between variables by defining appropriate membership functions and inference rules, and improve the accuracy and comprehensiveness of prediction. Summary of the invention

[0003] In order to improve the accuracy and comprehensiveness of prediction, the purpose of the present invention is to provide a method and system for predicting the risk level of cardiac arrest based on fuzzy logic. The technical scheme adopted is as follows:

[0004] S1. Acquire real-time vital sign data of the patient, wherein the real-time vital sign data includes at least one of heart rate, blood pressure, blood oxygen saturation, and respiratory rate;

[0005] S2. Based on the real-time vital sign data, feature selection and importance assessment processing are performed through a machine learning algorithm to obtain target features that are most relevant to the risk of cardiac arrest;

[0006] S3, performing fuzzification processing based on fuzzy C-means clustering based on the target features to convert precise numerical data into fuzzy sets;

[0007] S4. According to the fuzzy logic principle and the case-based reasoning method, the fuzzy set is matched and calculated with the known cardiac arrest case characteristics to obtain the cardiac arrest risk level;

[0008] S5. Based on the cardiac arrest risk level, a risk level assessment is performed, and a corresponding early warning mechanism is triggered.

[0009] Furthermore, in step S2, the feature selection and importance assessment process is performed by a machine learning algorithm to obtain the target features most relevant to the risk of cardiac arrest, including:

[0010] S21, obtaining the patient's medical history records and living habit information, and integrating the medical history records and living habit information into the real-time vital sign data to obtain a comprehensive data set;

[0011] S22, using feature crossover technology to process the comprehensive data set, and obtaining an enhanced comprehensive data set by mining the potential associations and interactions between different features;

[0012] S23. Perform importance assessment processing based on the enhanced comprehensive data set to obtain the target features that are most relevant to the risk of cardiac arrest.

[0013] Furthermore, in step S22, the comprehensive data set is processed using the feature crossover technology to obtain an enhanced comprehensive data set by mining the potential associations and interactions between different features, including:

[0014] S221. Based on correlation coefficient analysis, screening out a feature subset strongly correlated with the risk of cardiac arrest from the comprehensive data set;

[0015] S222, based on data characteristics and interactions between features, the feature subset is processed using explicit crossover based on statistical methods to obtain multi-dimensional crossover features;

[0016] S223. Based on the multi-dimensional cross-features, an enhanced comprehensive data set is constructed through feature fusion and data integration.

[0017] Furthermore, in step S23, the importance assessment process is performed based on the enhanced comprehensive data set to obtain the target features most relevant to the risk of cardiac arrest, including:

[0018] S231, selecting an interpretative importance assessment processing method based on the distribution characteristics of the data and the correlation between the features;

[0019] S232, applying the importance evaluation processing method to each feature in the enhanced comprehensive data set to obtain an importance evaluation value of each feature;

[0020] S233. Compare the importance assessment value of each feature with a preset importance assessment threshold, and take the feature that is greater than the threshold as the target feature that is most relevant to the risk of cardiac arrest.

[0021] Further, in step S3, the fuzzification processing based on fuzzy C-means clustering is performed based on the target feature to convert the precise numerical data into a fuzzy set, including:

[0022] S31, initialize cluster center m i and membership matrix U;

[0023] S32, calculate the distance m from each data point to each cluster center through the Euclidean distance measurement algorithm i The distance d ij ;

[0024] S33, using the objective function J based on fuzzy C-means clustering m , combined with the distance d ij and fuzzy weighted index m, updating the membership matrix U, wherein the fuzzy weighted index m is dynamically adjusted according to the clustering effectiveness index in the iterative process;

[0025] S34, based on the updated membership matrix U, recalculate the center m of each cluster i i , and determine the change amount by comparing the new and old values ​​of the objective function;

[0026] S35, repeat steps S32 to S34 until the objective function J m When the change in is less than the preset threshold and the clustering effectiveness index reaches the preset standard, or the maximum number of iterations is reached, the iteration is stopped and the final target cluster center and target membership matrix are input;

[0027] S36. Based on the target membership matrix, each data point is mapped to its corresponding fuzzy set to form a final fuzzy set representation.

[0028] Furthermore, in step S33, the fuzzy weighted index m will be dynamically adjusted according to the direction of the gradient descent and the preset learning rate based on the gradient descent method by detecting the changing trend of the intra-cluster error sum during the iteration process and calculating the gradient value of the current iteration.

[0029] Furthermore, in step S33, the objective function J m The definition is as follows:

[0030]

[0031] Among them, N represents the total number of data points, C represents the total number of cluster centers, m represents the fuzzy weighted index, and u ij Represents data point x i For cluster center c j The membership degree of i -cj ‖ 2 Represents data point x i To cluster center c j The square of the Euclidean distance, λ represents the regularization coefficient determined based on the grid search method, u ij_avg It represents the average membership of all samples relative to cluster center j.

[0032] Further, in step S4, according to the fuzzy logic principle, combined with the case-based reasoning method, the fuzzy set is matched and calculated with the known cardiac arrest case characteristics to obtain the cardiac arrest risk level, including:

[0033] S41, extracting known cardiac arrest cases from the historical database, and performing feature extraction on each case to form a feature vector;

[0034] S42, using fuzzy logic principles, calculating the similarity between the fuzzy set and the feature vector of known cardiac arrest case characteristics, and selecting the most similar case from the known case library based on the similarity calculation result;

[0035] S43. Combined with case-based reasoning methods, analyze the risk level, treatment strategy, and treatment outcome of the selected case, and assess the risk level of cardiac arrest based on this information.

[0036] Further, in step S5, the risk level assessment is performed based on the cardiac arrest risk level, and a corresponding early warning mechanism is triggered, including:

[0037] S51, based on the cardiac arrest risk level, calling a corresponding information generation template to generate corresponding warning information;

[0038] S52: Push the warning information to a terminal device designated by medical staff through a variety of instant messaging methods, wherein the multiple instant messaging methods include at least one of SMS notification, email reminder, and application push.

[0039] In the second aspect, the present application also discloses a cardiac arrest risk level prediction system based on fuzzy logic method, the system includes a data acquisition module, a feature selection and evaluation module, a fuzzy processing module, a risk matching and evaluation module, and an early warning mechanism triggering module, wherein:

[0040] The data acquisition module is used to acquire real-time vital sign data of the patient, wherein the real-time vital sign data includes at least one of heart rate, blood pressure, blood oxygen saturation, and respiratory rate;

[0041] The feature selection and evaluation module is used to perform feature selection and importance evaluation processing based on the real-time vital sign data through a machine learning algorithm to obtain the target feature most relevant to the risk of cardiac arrest;

[0042] The fuzzification processing module is used to perform fuzzification processing based on fuzzy C-means clustering based on the target feature to convert precise numerical data into fuzzy sets;

[0043] The risk matching and assessment module is used to match and calculate the fuzzy set with known cardiac arrest case characteristics according to the fuzzy logic principle and in combination with the case-based reasoning method to obtain the cardiac arrest risk level;

[0044] The early warning mechanism triggering module is used to perform risk level assessment based on the cardiac arrest risk level and trigger a corresponding early warning mechanism.

[0045] The present invention has the following beneficial effects:

[0046] 1) Through feature selection and importance assessment using machine learning algorithms, vital sign data most closely related to the risk of cardiac arrest can be screened out, thereby reducing the interference of redundant information on risk assessment and improving prediction accuracy;

[0047] 2) The use of fuzzy C-means clustering technology can accurately convert numerical data into fuzzy sets. This conversion mechanism cleverly handles the uncertainty and ambiguity in vital sign data. This processing method not only enhances the richness of data expression, but also significantly improves the accuracy and comprehensiveness of risk assessment, enabling the model to better adapt to complex and changing clinical scenarios;

[0048] 3) Combining case-based reasoning methods enables matching calculations based on historical cases and fuzzy features of current patients, thereby more flexibly assessing the risk of cardiac arrest. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] 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.

[0050] Figure 1 A flow chart of a method for predicting cardiac arrest risk level based on fuzzy logic method provided by one embodiment of the present invention;

[0051] Figure 2A system structure diagram of a cardiac arrest risk level prediction system based on fuzzy logic method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] 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 method and system for predicting the risk level of cardiac arrest based on fuzzy logic method proposed by the present invention, its specific implementation method, structure, characteristics and effects, 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.

[0053] 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.

[0054] The specific scheme of a method and system for predicting the risk level of cardiac arrest based on fuzzy logic provided by the present invention is described in detail below in conjunction with the accompanying drawings.

[0055] See also Figure 1 , which shows a method flow chart of a method for predicting cardiac arrest risk level based on fuzzy logic method provided by an embodiment of the present invention, the method comprising:

[0056] Step S1, acquiring real-time vital sign data of the patient, wherein the real-time vital sign data includes at least one of heart rate, blood pressure, blood oxygen saturation, and respiratory rate.

[0057] Step S2: Based on the real-time vital sign data, feature selection and importance assessment processing are performed through a machine learning algorithm to obtain target features that are most relevant to the risk of cardiac arrest.

[0058] Step S3, performing fuzzification processing based on fuzzy C-means clustering based on the target features to convert precise numerical data into fuzzy sets.

[0059] Step S4, according to the fuzzy logic principle and in combination with the case-based reasoning method, the fuzzy set is matched and calculated with the known cardiac arrest case characteristics to obtain the cardiac arrest risk level.

[0060] Step S5: Based on the cardiac arrest risk level, a risk level assessment is performed and a corresponding early warning mechanism is triggered.

[0061] From the above, it can be seen that the present application discloses a method for predicting the risk level of cardiac arrest based on fuzzy logic. By performing feature selection and importance assessment using a machine learning algorithm, the method can screen out the vital sign data that is most closely related to the risk of cardiac arrest, thereby reducing the interference of redundant information on risk assessment and improving prediction accuracy. The fuzzy C-means clustering technology can accurately convert numerical data into fuzzy sets. This conversion mechanism cleverly handles the uncertainty and ambiguity in vital sign data. This processing method not only enhances the richness of data expression, but also significantly improves the accuracy and comprehensiveness of risk assessment, enabling the model to better adapt to complex and changeable clinical scenarios. Combined with the case-based reasoning method, it enables matching calculations based on the fuzzy features of historical cases and current patients, thereby more flexibly assessing the risk of cardiac arrest.

[0062] In one embodiment, in step S2, the feature selection and importance assessment process is performed by a machine learning algorithm to obtain the target features most relevant to the risk of cardiac arrest, including:

[0063] Step S21, obtaining the patient's medical history records and living habit information, integrating the medical history records and living habit information into the real-time vital sign data to obtain a comprehensive data set.

[0064] Specifically, considering the comprehensiveness and in-depth analysis of the data, this application integrates medical history records and lifestyle information into real-time vital signs data to construct a comprehensive data set containing comprehensive health information of patients. Among them, for medical history records and lifestyle information, such as text descriptions of disease type, medical history duration, dietary preferences, exercise frequency, etc., it is necessary to convert them into numerical features (such as unique hot encoding, label encoding, etc.) to facilitate subsequent feature cross-processing. Afterwards, the real-time vital signs data can be fused and matched with the pre-processed medical history records and lifestyle information according to the patient ID to ensure the correspondence of the data.

[0065] Exemplary forms of constructing a comprehensive dataset include:

[0066]

[0067] Step S22, using feature crossover technology to process the comprehensive data set, and obtaining an enhanced comprehensive data set by mining the potential associations and interactions between different features.

[0068] Specifically, this application will first screen out feature subsets that are strongly correlated with the risk of cardiac arrest through correlation coefficient analysis. Then, using the data characteristics and interactions of these features, statistical methods are used to display and cross each feature in the feature subset to generate multi-dimensional cross features. Finally, these cross features are fused with the original features to construct an enhanced comprehensive data set with richer information and stronger predictive capabilities.

[0069] Step S23, performing importance assessment processing based on the enhanced comprehensive data set to obtain the target feature most relevant to the risk of cardiac arrest.

[0070] Specifically, this application selects an importance assessment processing method based on the distribution characteristics of the data (such as the normal distribution characteristics of the data) and the correlation between the features, and applies the importance assessment processing method to each feature in the enhanced comprehensive data set to obtain the importance assessment value of each feature. Finally, through the threshold comparison method, the features with importance assessment values ​​higher than the preset threshold are screened out from the enhanced comprehensive data set and used as the target features most relevant to the risk of cardiac arrest.

[0071] In the above embodiments, on the one hand, by integrating medical history records and living habits information into real-time vital signs data, a closer data association can be established. This association helps to reveal potential health problems and risk factors. On the other hand, through feature cross-talk, more cross-features can be obtained, which may contain information that is not explicitly expressed in the original data. This helps to enhance the information content and predictive ability of the data set. Finally, by performing importance assessment processing based on the enhanced comprehensive data set, the target features that are most relevant to the risk of cardiac arrest can be identified. By identifying key features, the patient's risk of cardiac arrest can be more accurately assessed and the accuracy of the prediction model can be improved.

[0072] In one embodiment, in step S22, the integrated data set is processed using feature crossover technology to obtain an enhanced integrated data set by mining potential associations and interactions between different features, including:

[0073] Step S221, based on correlation coefficient analysis, a feature subset strongly correlated with the risk of cardiac arrest is screened out from the comprehensive data set.

[0074] Specifically, the present application calculates the correlation coefficient (such as the Pearson correlation coefficient) between each feature in the comprehensive data set and the risk of cardiac arrest, which measures the strength of the linear relationship between the feature and the risk of cardiac arrest (which is a binary variable indicating whether cardiac arrest occurs).

[0075] In one embodiment, the present application combines feature weighting and threshold adjustment based on the Pearson correlation coefficient to evaluate the correlation coefficient between each feature in the comprehensive data set and the risk of cardiac arrest. The specific calculation formula is as follows:

[0076]

[0077] Among them, r weighted represents the weighted correlation coefficient, which is used to reflect the importance of different features in the risk assessment of cardiac arrest; ω i represents the weight of the i-th sample feature, which evaluates the correlation between the i-th sample feature and the risk of cardiac arrest based on mutual information, and obtains a weight distribution accordingly; x i represents the eigenvalue of the i-th sample, represents the weighted feature average; y i represents the cardiac arrest risk value corresponding to the i-th sample, which is a binary variable, usually 0 or 1, indicating whether cardiac arrest occurs; Represents the average of all samples' cardiac arrest risk values.

[0078] Furthermore, the present application will also set a correlation coefficient threshold (which can be determined based on the specific circumstances of the data set, and the present application does not limit this), and select features whose correlation coefficients exceed the threshold as strongly correlated features, and construct a feature subset through these strongly correlated features.

[0079] Step S222: Based on the data characteristics and the interactions between the features, the feature subset is processed by explicit crossover based on statistical methods to obtain multi-dimensional crossover features.

[0080] Specifically, this application defines the cross-features that need to be calculated according to the research objectives and the characteristics of the data to obtain new features.

[0081] Where, for each sample:

[0082] 1) When it is determined that there is a multiplication effect between cross-features (i.e., the change of one feature will cause another feature to change in a multiplicative manner), the value of the cross-feature is calculated by direct multiplication. For example, the cross-feature corresponding to "heart rate * blood pressure" reflects the combined effect between heart rate and blood pressure, and the cross-feature corresponding to "blood oxygen saturation * respiratory rate" can also be calculated by direct multiplication.

[0083] 2) When it is determined that there is an additive effect between cross-features (i.e., the change of one feature will cause another feature to change in an additive manner), the value of the cross-feature is calculated by direct summation. For example, "medical history duration + exercise frequency" can be defined as a cross-feature to further reflect the patient's overall health status;

[0084] 3) When it is determined that there is a differential effect between cross-features, the value of the cross-feature is calculated by difference calculation. For example, "diet preference code-neutral diet preference code (assuming a neutral diet as a benchmark)" can be defined as a cross-feature to further reflect the impact of dietary preferences on health.

[0085] Finally, the calculated cross features are added to the original feature set to expand the feature space and obtain multi-dimensional cross features. These multi-dimensional cross features will contain more information and help with subsequent analysis and modeling.

[0086] Step S223, based on the multi-dimensional cross-features, an enhanced comprehensive data set is constructed through feature fusion and data integration.

[0087] Specifically, the present application will fuse the dimensional cross-features obtained based on step S222 with the original features to obtain a feature set containing more information dimensions. Among them, the present application specifically optimizes the feature combination through a feature selection algorithm to fuse information from different sources and different dimensions. Afterwards, based on feature fusion, data integration is performed based on data cleaning and preprocessing rules, including outlier detection, data normalization and other operations to obtain an enhanced comprehensive data set. This data set contains original features, dimensional cross-features and possible other derived features, which can more comprehensively reflect the relevant factors of cardiac arrest risk and provide richer information input for subsequent prediction models.

[0088] In the above embodiment, through correlation coefficient analysis, a feature subset that is strongly correlated with the risk of cardiac arrest can be screened out from the comprehensive data set. This screening method is based on the principles of statistics and can objectively evaluate the correlation between each feature and the risk of cardiac arrest, thereby ensuring the effectiveness and accuracy of the selected features. Through feature fusion and data integration, multi-dimensional cross-features are combined with other relevant information to construct an enhanced comprehensive data set. This data set not only contains useful information in the original data, but also incorporates the interactions and complex relationships between features, providing more comprehensive data support for subsequent data processing.

[0089] In one embodiment, in step S23, the importance assessment process is performed based on the enhanced comprehensive data set to obtain the target features most relevant to the risk of cardiac arrest, including:

[0090] Step S231, selecting an explanatory importance assessment processing method based on the distribution characteristics of the data and the correlation between features.

[0091] Specifically, when it is determined that the data analysis is close to a normal distribution and there is a linear relationship between the feature and the target variable (i.e., the risk of cardiac arrest), the importance assessment processing method based on Lasso regression is selected, and the importance of the feature is judged by analyzing the regression coefficient of each feature in the Lasso regression model. When it is determined that the data distribution characteristics are non-normal or there is a nonlinear relationship, and the relationship between the feature and the target variable is complex, the importance assessment processing method based on the tree model is selected. The tree model is used to capture the nonlinear relationship and complex interactions in the data, and the importance of the feature is judged by the frequency of the feature being used as a split node in the tree model or the information gain of the feature on the split node.

[0092] Step S232: Apply the importance evaluation processing method to each feature in the enhanced comprehensive data set to obtain an importance evaluation value for each feature.

[0093] Specifically, for the importance assessment processing method based on Lasso regression, the Lasso regression model is needed to fit the data, and the best model is selected by adjusting the regularization parameter λ. Finally, the regression coefficient of each feature in the model is analyzed to obtain the importance assessment value of each feature. For the importance assessment processing method based on the tree model, it is necessary to select a suitable tree model (such as random forest, gradient boosting tree, etc.) to train the data. During the training process, the frequency of each feature being used as a split node or the information gain of the feature on the split node is recorded. Finally, the importance assessment value of each feature is calculated based on this information.

[0094] In one of the embodiments, for the selection of the regularization parameter λ, the present application selects the regularization parameter λ that minimizes the model prediction error based on the results of cross-validation, thereby obtaining more reliable prediction results and more meaningful feature importance evaluation.

[0095] Step S233, comparing the importance evaluation value of each feature with a preset importance evaluation threshold, and taking the feature greater than the threshold as the target feature most relevant to the risk of cardiac arrest.

[0096] In the above embodiment, the importance evaluation processing method is selected according to the distribution characteristics of the data and the correlation between the features, which can ensure that the selected method can accurately capture the key information in the data, thereby avoiding the selection of inappropriate or invalid features. The importance evaluation processing method is applied to each feature in the enhanced comprehensive data set to obtain the importance evaluation value of each feature. These evaluation values ​​are not only helpful for feature selection, but also can be used as a basis for model optimization.

[0097] In one embodiment, in step S3, the fuzzification processing based on fuzzy C-means clustering is performed based on the target feature to convert the precise numerical data into a fuzzy set, including:

[0098] Step S31, initializing the cluster center m i And the membership matrix U.

[0099] The present application determines the number of clusters C based on the distribution density of the data set. The element u in the matrix U ij Indicates the degree of membership of data point j to cluster i and satisfies the condition: for all j, there is ∑u ij =1 (i.e., the sum of the membership of each data point to all clusters is 1), for all i and j, 0≤u ij ≤1 (that is, the membership value is between 0 and 1).

[0100] Step S32, using the Euclidean distance metric algorithm, calculate the distance m between each data point and each cluster center. i The distance d ij .

[0101] Specifically, for the data set X = {x 1 ,x 2 ,...,x n} for each data point x j (where j = 1, 2, ..., n), and the cluster center set M = {m 1 ,m 2 ,...,m c Each cluster center m in i (where i = 1, 2, ..., c, c is the number of clusters), this application will calculate x using the following formula j to m i The Euclidean distance of:

[0102]

[0103] Among them, d represents the dimension of the data point, x jk Represents data point x j The value in the kth dimension, m ik Represents the cluster center m i The value in the kth dimension.

[0104] Step S33, using the objective function J based on fuzzy C-means clustering m , combined with the distance d ij and fuzzy weighted index m, updating the membership matrix U, wherein the fuzzy weighted index m is dynamically adjusted according to the clustering effectiveness index in the iterative process;

[0105] Specifically, for the data set X = {x 1 ,x 2 ,...,x n} and cluster center set M = {m 1 ,m 2 ,...,m c}, the membership matrix U is an n*c matrix, where the element u ij Indicates the degree of membership of data point j to cluster i. For the objective function, it is based on the distance d ij The fuzzy weighted index m is determined by calculating the sum of the squares of the weighted distances from all data points to their cluster centers.

[0106] Furthermore, during the iteration process, the present application will dynamically adjust the fuzzy weighted index m based on the clustering effectiveness index. For example, if the current clustering result is determined to be not clear enough according to the clustering effectiveness index, m can be appropriately increased through a preset adjustment step to enhance the fuzziness of the clustering, thereby promoting a more detailed division; conversely, if the clustering result is determined to be too dispersed or the cluster center is unstable, m can be appropriately reduced through a preset adjustment step to promote a clearer clustering division.

[0107] It should be noted that the aforementioned adjustment step size is determined based on a small range of m value variation experiments and by observing the changes in clustering effectiveness indicators. Within this range, each adjustment of the m value can effectively improve the clustering effect without causing drastic fluctuations in the clustering results.

[0108] Step S34: recalculate the center m of each cluster i based on the updated membership matrix U i , and determine the change by comparing the new and old values ​​of the objective function.

[0109] Specifically, for each cluster i, the present application will calculate the weighted average position according to the corresponding column in the membership matrix U (i.e., the membership of all data points to the cluster i) after the update, and use it as the new cluster center m i . Then, based on the new cluster center m i and the updated membership matrix U, by calculating the distance from each data point to its membership cluster center m i The new value of the objective function is obtained by the weighted sum of the squares of the distances, and the change of the objective function is determined by the difference between the new value and the old value of the objective function.

[0110] Step S35, repeat steps S32 to S34 until the objective function J m When the change in is less than the preset threshold and the clustering validity index reaches the preset standard, or the maximum number of iterations is reached, the iteration is stopped and the final target cluster center and target membership matrix are input.

[0111] Step S36, based on the target membership matrix, mapping each data point to its corresponding fuzzy set to form a final fuzzy set representation.

[0112] In the above embodiment, on the one hand, the objective function of fuzzy C-means clustering is used, and combined with the distance and the dynamically adjusted fuzzy weighted index m, the membership matrix can be continuously updated, so that the membership of each data point to different clusters is more accurate. In addition, by setting the preset threshold of the objective function change, the preset standard of the clustering effectiveness index, and the maximum number of iterations and other stop conditions, it is possible to improve convergence and efficiency while ensuring the clustering quality.

[0113] In one embodiment, in step S33, the fuzzy weighted index m is based on the intra-cluster error sum in the iterative process, by detecting the changing trend of the intra-cluster error sum, and by calculating the gradient value of the current iteration based on the gradient descent method, and then dynamically adjusted according to the direction of the gradient descent and the preset learning rate.

[0114] Specifically, in each iteration, the application will calculate the current intra-cluster error sum by calculating the sum of the squares of the distances from each data point to its corresponding cluster center, and record the trend of the intra-cluster error sum changing with the number of iterations. Afterwards, according to the trend of the intra-cluster error sum, the gradient descent method is used to adjust the value of the fuzzy weighted index m by calculating the gradient and following the negative gradient direction. Specifically, it involves calculating the intra-cluster error and the gradient value of the parameter m, and then, according to the preset learning rate (i.e., a hyperparameter that controls the update step size) and the calculated gradient value, the value of the parameter m is updated along the negative gradient direction. The specific calculation formula includes: Among them, m new is the updated value of parameter m, m old is the value of parameter m before update, α is the learning rate, is the gradient value of the intra-cluster error and parameter m. This updating process is repeated in each iteration until the change of the intra-cluster error and parameter m is stopped when the change is less than the preset threshold.

[0115] In the above embodiment, the fuzzy weighted index m is dynamically adjusted based on the gradient descent method, so that the intrinsic structure and characteristics of the data can be better captured, thereby obtaining a more accurate clustering result.

[0116] In one embodiment, in step S33, the objective function J m The definition is as follows:

[0117]

[0118] Among them, N represents the total number of data points, C represents the total number of cluster centers, m represents the fuzzy weighted index, and uij Represents data point x i For cluster center c j The membership degree of i -c j ‖ 2 Represents data point x i To cluster center c j The square of the Euclidean distance, λ represents the regularization coefficient determined based on the grid search method, u ij_avg It represents the average membership of all samples relative to cluster center j.

[0119] Specifically, the objective function J m By introducing a regularization term to control the membership u ij With its average value u ij_avg The difference between them is calculated, so as to minimize the weighted sum of the distances from all data points to their corresponding cluster centers while keeping the membership distribution relatively smooth, so as to achieve a more robust and high-quality clustering result. As a fuzzy weighting factor of membership, it ensures that the contribution of each data point to the cluster center increases with its membership, and is weighted in a nonlinear manner (controlled by the fuzzy weighting index m). This weighting method makes data points with higher membership have a greater impact on the location of the cluster center and the clustering results, while data points with lower membership have less impact, thereby enhancing the stability and accuracy of the clustering results.

[0120] In one embodiment, in step S4, the fuzzy set is matched and calculated with known cardiac arrest case characteristics according to the fuzzy logic principle and combined with the case-based reasoning method to obtain the cardiac arrest risk level, including:

[0121] Step S41, extracting known cardiac arrest cases from the historical database, and performing feature extraction on each case to form a feature vector.

[0122] It should be noted that feature extraction refers to converting key information in a case into a numerical value or vector form that can be used for calculation.

[0123] Step S42, using the fuzzy logic principle, similarity calculation is performed between the fuzzy set and the feature vector of the known cardiac arrest case characteristics, and based on the similarity calculation result, the most similar case is selected from the known case library.

[0124] Specifically, this application utilizes the uncertainty of fuzzy logic processing and finds the historical case that is most similar to the patient's condition through similarity calculation. It should be noted that this application uses the similarity measurement method in fuzzy logic (such as the intersection of fuzzy sets) to calculate the similarity between the fuzzy set of the current patient and the feature vector of the known cardiac arrest case characteristics. Finally, based on the similarity calculation results, the case with the highest similarity is selected from the known case library.

[0125] Step S43, combining case-based reasoning methods, analyzing the risk level, treatment strategy, and treatment results of the selected case, and evaluating the risk level of cardiac arrest based on this information.

[0126] Specifically, the analysis of treatment strategies includes drug therapy, non-drug therapy, and lifestyle adjustments. The treatment outcomes of the selected cases after cardiac arrest were analyzed, including survival rate, neurological prognosis, complications, etc. Based on these case information, the risk level of cardiac arrest was evaluated. The effectiveness of the treatment strategy of the selected cases, the statistical data of the treatment results, and the individual differences of the patients were combined, and the risk level of cardiac arrest was evaluated through risk assessment algorithms (such as deep learning algorithms, integrated learning algorithms, etc.). Among them, the risk level is divided into low risk, medium risk, high risk, and extremely high risk according to the patient's age, underlying disease, inducing factors of cardiac arrest, historical number of attacks, sensitivity of treatment response, and current cardiac function status. Each level represents a different risk level. Low risk means that the patient's risk of cardiac arrest is relatively low, while extremely high risk means that the patient's risk of cardiac arrest is extremely high and emergency intervention measures need to be taken immediately.

[0127] In the above embodiment, the similarity calculation is performed using the fuzzy logic principle, which can handle uncertainty information more flexibly and improve matching accuracy. Combined with the case-based reasoning method, the risk level, treatment strategy and treatment results of the selected case can be used to provide a more personalized risk assessment for the patient. This assessment method not only takes into account the patient's current condition, but also the historical experience and treatment results of similar patients, providing a more reliable data basis for risk assessment.

[0128] In one embodiment, in step S5, the risk level assessment is performed based on the cardiac arrest risk level, and a corresponding early warning mechanism is triggered, including:

[0129] Step S51, based on the cardiac arrest risk level, calling the corresponding information generation template to generate corresponding warning information.

[0130] Specifically, the information generation template can be customized according to different cardiac arrest risk levels. For example, for patients with low risk levels, the generated warning information only contains basic patient information and risk levels; for patients with medium risk levels, the generated warning information not only contains basic patient information and risk levels, but also briefly summarizes possible risk factors and reminds medical staff to monitor the patients more closely; and for patients with high or extremely high risk levels, the generated warning information contains more detailed risk factors, recommended intervention measures, etc. This customized warning information can better meet the needs of patients with different risk levels, improve the effectiveness and pertinence of warnings, and ensure that medical staff can respond quickly and accurately, thereby reducing the risk of adverse events such as cardiac arrest.

[0131] Step S52: Push the warning information to the terminal device designated by the medical staff through a variety of instant messaging methods, wherein the multiple instant messaging methods include at least one of SMS notification, email reminder, and application push.

[0132] Specifically, the transmission of early warning information combines a variety of methods to improve the coverage and reception rate of information. For example, in emergency situations, SMS notifications and application push notifications can be used simultaneously to ensure that medical staff can receive information in a timely manner under any circumstances. In one embodiment, after the early warning information is pushed to the terminal device, the terminal device will feedback a corresponding reception confirmation signal to indicate that the medical staff has successfully received and viewed the early warning information. This confirmation signal can be a message reading receipt, as well as a message reading status, etc. For unconfirmed early warning information, a second push will be automatically performed to ensure the effective transmission of the information.

[0133] In the above embodiment, on the one hand, by calling the corresponding information generation template based on the cardiac arrest risk level, it can ensure that the content of the warning information matches the actual risk status of the patient, thereby improving the accuracy and pertinence of the information. On the other hand, by pushing the warning information to the terminal device designated by the medical staff through a variety of instant messaging methods, it can ensure that the information is conveyed to the relevant personnel in the first time, increasing the accessibility of the information.

[0134] Please refer to Figure 2 The present application discloses a cardiac arrest risk level prediction system based on fuzzy logic method, the system includes a data acquisition module, a feature selection and evaluation module, a fuzzy processing module, a risk matching and evaluation module, and an early warning mechanism triggering module, wherein:

[0135] The data acquisition module is used to acquire real-time vital sign data of the patient, and the real-time vital sign data includes at least one of heart rate, blood pressure, blood oxygen saturation, and respiratory rate.

[0136] The feature selection and evaluation module is used to perform feature selection and importance evaluation processing based on the real-time vital signs data through a machine learning algorithm to obtain the target features most relevant to the risk of cardiac arrest.

[0137] The fuzzification processing module is used to perform fuzzification processing based on fuzzy C-means clustering based on the target feature to convert precise numerical data into fuzzy sets.

[0138] The risk matching and assessment module is used to match and calculate the fuzzy set with known cardiac arrest case characteristics based on fuzzy logic principles and combined with case-based reasoning methods to obtain the cardiac arrest risk level.

[0139] The early warning mechanism triggering module is used to perform risk level assessment based on the cardiac arrest risk level and trigger a corresponding early warning mechanism.

[0140] From the above, it can be seen that the cardiac arrest risk level prediction system based on fuzzy logic method disclosed in the present application can screen out the vital signs data most closely related to the risk of cardiac arrest through feature selection and importance assessment by machine learning algorithm, thereby reducing the interference of redundant information on risk assessment and improving the prediction accuracy; the use of fuzzy C-means clustering technology can accurately convert numerical data into fuzzy sets. This conversion mechanism cleverly handles the uncertainty and ambiguity in vital signs data. This processing method not only enhances the richness of data expression, but also significantly improves the accuracy and comprehensiveness of risk assessment, so that the model can better adapt to complex and changeable clinical scenarios; combined with case-based reasoning methods, it enables matching calculations based on historical cases and fuzzy features of current patients, thereby more flexibly assessing the risk of cardiac arrest.

[0141] In one of the embodiments, the feature selection and evaluation module is also used to obtain the patient's medical history records and living habit information, integrate the medical history records and living habit information into the real-time vital signs data, and obtain a comprehensive data set; use feature cross-talk technology to process the comprehensive data set, and obtain an enhanced comprehensive data set by mining the potential correlation and interaction between different features; perform importance assessment processing based on the enhanced comprehensive data set to obtain the target features that are most relevant to the risk of cardiac arrest.

[0142] In one of the embodiments, the feature selection and evaluation module is also used to screen out a feature subset that is strongly correlated with the risk of cardiac arrest from the comprehensive data set based on correlation coefficient analysis; based on data characteristics and interactions between features, the feature subset is processed using explicit crossover based on statistical methods to obtain multidimensional crossover features; based on the multidimensional crossover features, an enhanced comprehensive data set is constructed through feature fusion and data integration.

[0143] In one of the embodiments, the feature selection and evaluation module is also used to select an explanatory importance assessment processing method based on the distribution characteristics of the data and the correlation between features; apply the importance assessment processing method to each feature in the enhanced comprehensive data set to obtain an importance assessment value for each feature; compare the importance assessment value of each feature with a preset importance assessment threshold, and use features greater than the threshold as target features most relevant to the risk of cardiac arrest.

[0144] In one embodiment, the fuzzy processing module is also used to initialize the cluster center m i And the membership matrix U; calculate the distance from each data point to each cluster center m through the Euclidean distance metric algorithm i The distance d ij ; Using the objective function J based on fuzzy C-means clustering m , combined with the distance d ij and fuzzy weighted index m, update the membership matrix U, wherein the fuzzy weighted index m is dynamically adjusted according to the clustering effectiveness index in the iterative process; based on the updated membership matrix U, recalculate the center m of each cluster i i , and determine the change by comparing the new and old values ​​of the objective function; repeat the above steps until the objective function J m When the change in is less than the preset threshold and the clustering validity index reaches the preset standard, or the maximum number of iterations is reached, the iteration is stopped and the final target cluster center and target membership matrix are input; based on the target membership matrix, each data point is mapped to its corresponding fuzzy set to form the final fuzzy set representation.

[0145] In one of the embodiments, the fuzzy weighted index m is dynamically adjusted according to the direction of the gradient descent and the preset learning rate based on the intra-cluster error sum during the iteration process by detecting the changing trend of the intra-cluster error sum and calculating the gradient value of the current iteration based on the gradient descent method.

[0146] In one embodiment, the objective function J m The definition is as follows:

[0147]

[0148] Among them, N represents the total number of data points, C represents the total number of cluster centers, m represents the fuzzy weighted index, and u ij Represents data point x i For cluster center c j The membership degree of i -c j ‖ 2 Represents data point x i To cluster center c j The square of the Euclidean distance, λ represents the regularization coefficient determined based on the grid search method, u ij_avg It represents the average membership of all samples relative to cluster center j.

[0149] In one of the embodiments, the risk matching and assessment module is also used to extract known cardiac arrest cases from a historical database, and perform feature extraction on each case to form a feature vector; using fuzzy logic principles, the fuzzy set is used to calculate the similarity with the feature vector of the known cardiac arrest case characteristics, and based on the similarity calculation results, the most similar case is selected from the known case library; combined with case-based reasoning methods, the risk level, treatment strategy, and treatment results of the selected case are analyzed, and the risk level of cardiac arrest is assessed based on this information.

[0150] In one of the embodiments, the warning mechanism trigger module is also used to call the corresponding information generation template based on the cardiac arrest risk level to generate corresponding warning information; and push the warning information to the terminal device designated by the medical staff through multiple instant messaging methods, and the multiple instant messaging methods include at least one of SMS notification, email reminder, and application push.

[0151] 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.

[0152] 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 method for predicting cardiac arrest risk level based on fuzzy logic, characterized in that: The method comprises: S1. Acquire real-time vital sign data of the patient, wherein the real-time vital sign data includes at least one of heart rate, blood pressure, blood oxygen saturation, and respiratory rate; S2. Based on the real-time vital sign data, feature selection and importance assessment processing are performed through a machine learning algorithm to obtain target features that are most relevant to the risk of cardiac arrest; S3, performing fuzzification processing based on fuzzy C-means clustering based on the target features to convert precise numerical data into fuzzy sets; S4. According to the fuzzy logic principle and the case-based reasoning method, the fuzzy set is matched and calculated with the known cardiac arrest case characteristics to obtain the cardiac arrest risk level; S5. Based on the cardiac arrest risk level, a risk level assessment is performed, and a corresponding early warning mechanism is triggered.

2. The method according to claim 1, characterized in that In step S2, the feature selection and importance assessment process is performed by a machine learning algorithm to obtain the target features most relevant to the risk of cardiac arrest, including: S21, obtaining the patient's medical history records and living habit information, and integrating the medical history records and living habit information into the real-time vital sign data to obtain a comprehensive data set; S22, using feature crossover technology to process the comprehensive data set, and obtaining an enhanced comprehensive data set by mining the potential associations and interactions between different features; S23. Perform importance assessment processing based on the enhanced comprehensive data set to obtain the target features that are most relevant to the risk of cardiac arrest.

3. The method according to claim 2, characterized in that In step S22, the comprehensive data set is processed by using the feature crossover technology to obtain an enhanced comprehensive data set by mining the potential associations and interactions between different features, including: S221. Based on correlation coefficient analysis, screening out a feature subset strongly correlated with the risk of cardiac arrest from the comprehensive data set; S222, based on data characteristics and interactions between features, the feature subset is processed using explicit crossover based on statistical methods to obtain multi-dimensional crossover features; S223. Based on the multi-dimensional cross-features, an enhanced comprehensive data set is constructed through feature fusion and data integration.

4. The method according to claim 2, characterized in that: In step S23, the importance assessment process is performed based on the enhanced comprehensive data set to obtain the target features most relevant to the risk of cardiac arrest, including: S231, selecting an interpretative importance assessment processing method based on the distribution characteristics of the data and the correlation between the features; S232, applying the importance evaluation processing method to each feature in the enhanced comprehensive data set to obtain an importance evaluation value of each feature; S233. Compare the importance assessment value of each feature with a preset importance assessment threshold, and take the feature that is greater than the threshold as the target feature that is most relevant to the risk of cardiac arrest.

5. The method according to claim 1, characterized in that In step S3, the fuzzification processing based on fuzzy C-means clustering is performed based on the target feature to convert the precise numerical data into a fuzzy set, including: S31, initialize cluster center m i and membership matrix U; S32, calculate the distance m from each data point to each cluster center through the Euclidean distance measurement algorithm i The distance d ij ; S33, using the objective function J based on fuzzy C-means clustering m , combined with the distance d ij and fuzzy weighted index m, updating the membership matrix U, wherein the fuzzy weighted index m is dynamically adjusted according to the clustering effectiveness index in the iterative process; S34, based on the updated membership matrix U, recalculate the center m of each cluster i i , and determine the change amount by comparing the new and old values ​​of the objective function; S35, repeat steps S32 to S34 until the objective function J m When the change in is less than the preset threshold and the clustering effectiveness index reaches the preset standard, or the maximum number of iterations is reached, the iteration is stopped and the final target cluster center and target membership matrix are input; S36. Based on the target membership matrix, each data point is mapped to its corresponding fuzzy set to form a final fuzzy set representation.

6. The method according to claim 5, characterized in that In step S33, the fuzzy weighted index m will be dynamically adjusted according to the direction of gradient descent and the preset learning rate based on the gradient descent method by detecting the changing trend of the intra-cluster error sum during the iteration process and calculating the gradient value of the current iteration.

7. The method according to claim 5, characterized in that In step S33, the objective function J m The definition is as follows: Among them, N represents the total number of data points, C represents the total number of cluster centers, m represents the fuzzy weighted index, and u ij Represents data point x i For cluster center c j The membership degree of i -c j ‖ 2 Represents data point x i To cluster center c j The square of the Euclidean distance, λ represents the regularization coefficient determined based on the grid search method, u ij_avg It represents the average membership of all samples relative to cluster center j.

8. The method according to claim 1, characterized in that In step S4, according to the fuzzy logic principle and the case-based reasoning method, the fuzzy set is matched and calculated with the known cardiac arrest case characteristics to obtain the cardiac arrest risk level, including: S41, extracting known cardiac arrest cases from the historical database, and performing feature extraction on each case to form a feature vector; S42, using fuzzy logic principles, calculating the similarity between the fuzzy set and the feature vector of known cardiac arrest case characteristics, and selecting the most similar case from the known case library based on the similarity calculation result; S43. Combined with case-based reasoning methods, analyze the risk level, treatment strategy, and treatment outcome of the selected case, and assess the risk level of cardiac arrest based on this information.

9. The method according to claim 1, characterized in that: In step S5, based on the cardiac arrest risk level, risk level assessment is performed and a corresponding early warning mechanism is triggered, including: S51, based on the cardiac arrest risk level, calling a corresponding information generation template to generate corresponding warning information; S52: Push the warning information to a terminal device designated by medical staff through a variety of instant messaging methods, wherein the multiple instant messaging methods include at least one of SMS notification, email reminder, and application push.

10. A cardiac arrest risk level prediction system based on fuzzy logic method, characterized in that: The system includes a data acquisition module, a feature selection and evaluation module, a fuzzy processing module, a risk matching and evaluation module, and an early warning mechanism triggering module, wherein: The data acquisition module is used to acquire real-time vital sign data of the patient, wherein the real-time vital sign data includes at least one of heart rate, blood pressure, blood oxygen saturation, and respiratory rate; The feature selection and evaluation module is used to perform feature selection and importance evaluation processing based on the real-time vital sign data through a machine learning algorithm to obtain the target feature most relevant to the risk of cardiac arrest; The fuzzification processing module is used to perform fuzzification processing based on fuzzy C-means clustering based on the target feature to convert precise numerical data into fuzzy sets; The risk matching and assessment module is used to match and calculate the fuzzy set with known cardiac arrest case characteristics according to the fuzzy logic principle and in combination with the case-based reasoning method to obtain the cardiac arrest risk level; The early warning mechanism triggering module is used to perform risk level assessment based on the cardiac arrest risk level and trigger a corresponding early warning mechanism.

Images