A health assessment method and device for chronic disease population

Abstract

The present invention discloses a health assessment method and device for chronic disease population, which belongs to the field of chronic disease treatment monitoring, including: obtaining a knowledge graph for storing and recording historical patient information of chronic disease population; obtaining abnormal physiological state information of target chronic disease patients to obtain a query vector; taking each node of target chronic disease patients in the knowledge graph as a target node, taking nodes with edge weights greater than a first preset value with the target node as neighbor nodes of the target node, and calculating the correlation between the query vector and each target node in combination with the edge weights between the neighbor nodes and the target node; sorting all target nodes in descending order according to the size of the correlation, and extracting the first Z target nodes in the sorting to obtain key medical information; inputting the key medical information into a MedGPT model, and the MedGPT model generates a chronic disease health assessment plan for target chronic disease patients according to the key medical information. The present invention can perform accurate health assessment on chronic disease patients.

Description

Technical Field

[0001] The present invention relates to the technical field of chronic disease treatment monitoring, and in particular to a health assessment method and device for chronic disease population. Background Art

[0002] Chronic diseases are short for chronic non-communicable diseases. They do not refer to a specific disease, but are a general term for a class of diseases that have an insidious onset, a long course of illness, are difficult to cure, lack definite evidence of infectious biological etiology, have complex etiologies, and some have not yet been fully confirmed. Chronic diseases are mostly difficult to cure. Generally speaking, they can be well managed through long-term medication, regular checkups, and daily proper diet and exercise, so that one's condition can be stably controlled, the condition can be prevented from getting worse, and the progression of the disease can be slowed down.

[0003] With the continuous development of science and technology, researchers have designed a system for chronic disease patients to take medicine and review their daily lives, targeting the characteristics of chronic disease patients and the treatment of chronic diseases. For example, a smart medicine box that monitors and reminds chronic disease patients to follow doctor's orders to take medicine, a reminder program that reminds chronic disease patients to have regular medical checkups, and wearable devices that monitor patients' heart rate and blood pressure. However, existing technologies only monitor single indicators to remind chronic disease patients, and cannot provide more accurate health assessments of chronic disease patients based on their health conditions. Summary of the invention

[0004] The present invention provides a health assessment method and an assessment device for chronic disease population, which solves the problem that the prior art cannot accurately assess the health risks of chronic disease patients in combination with their health conditions.

[0005] In order to achieve the above object, the present invention adopts the following technical scheme:

[0006] In a first aspect, the present invention provides a method for health assessment of a chronic disease population, the method comprising:

[0007] Obtain a predetermined knowledge graph for storing and recording historical patient information of a chronic disease population; in the knowledge graph, take the basic information of each chronic disease patient in the historical patient information, each physiological indicator in the historical health data, each historical diagnosis information, and each historical treatment information as a node, and take the degree of association between any two nodes as the edge weight between the two nodes;

[0008] Acquire abnormal physiological state information of a target chronic disease patient, and vectorize the abnormal physiological state information to obtain a query vector; the abnormal physiological state information includes physiological indicators with abnormal risks determined based on the health data of the target chronic disease patient, and risk trend levels determined after risk trend analysis of the physiological indicators with abnormal risks;

[0009] Each node of the target chronic disease patient in the knowledge graph is taken as a target node, and a node whose edge weight with the target node is greater than a first preset value is taken as a neighbor node of the target node, and the correlation between the query vector and each of the target nodes is calculated in combination with the edge weight between the neighbor node and the target node;

[0010] All target nodes are sorted in descending order according to the magnitude of the relevance, and the first Z target nodes in the sorting are extracted to obtain the key medical information of the target chronic disease patient;

[0011] The key medical information is input into the MedGPT model, and the MedGPT model generates a chronic disease health assessment plan for the target chronic disease patient based on the key medical information; the chronic disease health assessment plan includes the risk trend level of the target chronic disease patient and treatment plan adjustment suggestions.

[0012] In a possible implementation, combining the edge weights between the neighbor nodes and the target node and respectively calculating the relevance between the query vector and each of the target nodes specifically includes:

[0013] Determining a first weight of the target node according to a predetermined similarity between the query vector and each of the target nodes, and a similarity between the query vector and the neighboring nodes;

[0014] Calculate the average value of the sum of edge weights between the target node and each of the neighboring nodes to obtain a second weight of the target node;

[0015] The first weight and the second weight are weighted and summed to obtain a comprehensive weight of the target node, and the comprehensive weight is used as the relevance of the target node.

[0016] In a possible implementation manner, determining the first weight of the target node according to a predetermined similarity between the query vector and each of the target nodes, and a similarity between the query vector and the neighboring nodes, specifically includes:

[0017] The cosine similarity algorithm is used to calculate the similarity between the query vector and each of the target nodes, and the similarity between the query vector and the neighboring nodes. The cosine similarity algorithm is specifically as follows:

[0018] ;

[0019] in, represents the query vector, represents the vector of the target node, Represents the similarity between the query vector and the target node;

[0020] The first weight of the target node is calculated according to the softmax function, and the softmax function is specifically:

[0021] ;

[0022] in, represents the first weight of the target node, represents the vector of neighbor nodes, represents the similarity between the query vector and the neighboring nodes, Represents neighbor nodes The node set belonging to the target chronic disease patient in the knowledge graph .

[0023] In a possible implementation, before obtaining the abnormal physiological state information of the target chronic disease patient and performing vectorization processing on the abnormal physiological state information to obtain a query vector, the method further includes:

[0024] Based on the health data of the target chronic disease patient acquired in real time, determining the physiological indicators with abnormal risks in the health data; the health data includes physiological data corresponding to a plurality of physiological indicators for feedback on the health status of the target chronic disease patient;

[0025] Perform risk trend analysis on the physiological indicators with abnormal risks to determine the risk trend level of the chronic disease of the target chronic disease patient.

[0026] In a possible implementation, based on the health data of the target chronic disease patient acquired in real time, determining the physiological indicators with abnormal risks in the health data specifically includes:

[0027] Obtain health data of target chronic disease patients in real time;

[0028] Taking the physiological data corresponding to each physiological indicator of the target chronic disease patient as the original input data, inputting the original input data into the autoencoder trained with the historical health data of the healthy population, and compressing and reconstructing the original input data through the autoencoder to obtain the corresponding reconstructed data;

[0029] A reconstruction error between the original input data and the reconstructed data is calculated, and according to the size of the reconstruction error and a first preset error threshold, the detection result of the physiological indicator corresponding to the original input data is marked to obtain a marked result, and the physiological indicators with abnormal risks in the marked result are screened out; the marked result is used to indicate whether the physiological indicator corresponding to the original input data has abnormal risks.

[0030] In a possible implementation, risk trend analysis is performed on the physiological indicators with abnormal risks to determine the risk trend level of the chronic disease of the target chronic disease patient, specifically including:

[0031] When there are physiological indicators of abnormal risk in the marking results, the health data of the target chronic disease patient and the marking results corresponding to multiple physiological indicators of the health data of the target chronic disease patient are used as target evaluation data, and the target evaluation data are input into a support vector machine model trained with historical chronic disease data of the chronic disease population. The target evaluation data is subjected to risk trend analysis by the support vector machine model to obtain a risk trend level.

[0032] In a possible implementation, before performing risk trend analysis on the target assessment data by using the support vector machine model, the method further includes:

[0033] The original support vector machine model is constructed using historical chronic disease data as the first training sample and the target evaluation data as the sample to be evaluated; the sample to be evaluated includes a first feature matrix composed of physiological data corresponding to multiple physiological indicators of the target chronic disease patient and the labeling results of the physiological indicators, and each of the first training samples includes a second feature matrix composed of physiological data corresponding to multiple physiological indicators of a chronic disease patient and the labeling results of the physiological indicators, and a risk label corresponding to the second feature matrix;

[0034] The decision function of the original support vector machine model is:

[0035] ;

[0036] in, For the return function, when When +1 is returned, it means that the chronic disease risk trend level of the sample to be evaluated is low risk. When -1 is returned, it means that the chronic disease risk trend level of the sample to be evaluated is high risk;

[0037] represents the number of the first training samples, represents the position of the current first training sample, represents the Lagrange multiplier parameter, is the bias parameter;

[0038] Indicates The second feature matrix of the first training sample, Indicates a risk label of the first training sample, when the risk label is +1, it indicates that the chronic disease risk is high risk, and when the risk label is -1, it indicates that the chronic disease risk is low risk;

[0039] represents the sample to be evaluated;

[0040] is a kernel function used to calculate the similarity between the first training sample and the sample to be evaluated.

[0041] In a possible implementation, each of the physiological indicators includes a plurality of indicator features; the kernel function of the decision function of the original support vector machine model is determined according to the radial basis function and the polynomial kernel function;

[0042] The kernel function is specifically:

[0043] ;

[0044] in, represents the adjustment parameter of the radial basis function, which is a preset fixed value;

[0045] Indicates the degree of the polynomial in the polynomial kernel function, which is a preset fixed value;

[0046] Indicates the total number of physiological indicators included in health data, Indicates the current position of physiological indicators;

[0047] Indicates the total number of indicator characteristics included in a physiological indicator. Indicates the location of the current indicator feature;

[0048] Indicates The weight coefficient of each physiological index;

[0049] Indicates the sample to be evaluated. The first physiological index The value of an indicator feature;

[0050] Indicates The first training sample The first physiological index The numerical value of an indicator feature.

[0051] In a possible implementation, the autoencoder includes an encoding unit, an attention unit, and a decoding unit; the autoencoder compresses and reconstructs the original input data to obtain corresponding reconstructed data, specifically including:

[0052] The encoding unit compresses the input high-dimensional original input data into low-dimensional first data through multiple fully connected layers;

[0053] The attention unit extracts important features related to the chronic disease of the target chronic disease patient through an attention mechanism to obtain second data;

[0054] The decoding unit reconstructs the second data into reconstructed data with the same dimension as the original input data through multiple fully connected layers.

[0055] In a second aspect, the present invention provides a device for evaluating the health of a chronic disease population, the device comprising:

[0056] The first processing module is used to obtain a predetermined knowledge graph for storing and recording historical patient information of a chronic disease population; in the knowledge graph, the basic information of each chronic disease patient in the historical patient information, each physiological indicator in the historical health data, each historical diagnosis information, and each historical treatment information are respectively used as a node, and the degree of association between any two nodes is used as the edge weight between the two nodes;

[0057] The second processing module is used to obtain abnormal physiological state information of the target chronic disease patient, and vectorize the abnormal physiological state information to obtain a query vector; the abnormal physiological state information includes physiological indicators with abnormal risks determined based on the health data of the target chronic disease patient, and risk trend levels determined after risk trend analysis of the physiological indicators with abnormal risks;

[0058] A third processing module is used to respectively use each node of the target chronic disease patient in the knowledge graph as a target node, use a node whose edge weight with the target node is greater than a first preset value as a neighbor node of the target node, and calculate the relevance of the query vector to each of the target nodes in combination with the edge weight between the neighbor node and the target node;

[0059] The fourth processing module is used to sort all target nodes in descending order according to the magnitude of the correlation, and extract the first Z target nodes in the sorting to obtain the key medical information of the target chronic disease patient;

[0060] The fifth processing module is used to input the key medical information into the MedGPT model, and the MedGPT model generates a chronic disease health assessment plan for the target chronic disease patient based on the key medical information; the chronic disease health assessment plan includes the risk trend level of the target chronic disease patient and treatment plan adjustment suggestions.

[0061] In a possible implementation, when respectively calculating the relevance between the query vector and each of the target nodes in combination with the edge weights between the neighbor nodes and the target nodes, the third processing module is configured to execute:

[0062] Determining a first weight of the target node according to a predetermined similarity between the query vector and each of the target nodes, and a similarity between the query vector and the neighboring nodes;

[0063] Calculate the average value of the sum of edge weights between the target node and each of the neighboring nodes to obtain a second weight of the target node;

[0064] The first weight and the second weight are weighted and summed to obtain a comprehensive weight of the target node, and the comprehensive weight is used as the relevance of the target node.

[0065] In a possible implementation, when determining the first weight of the target node according to the predetermined similarity between the query vector and each of the target nodes, and the similarity between the query vector and the neighboring nodes, the third processing module is specifically configured to execute:

[0066] The cosine similarity algorithm is used to calculate the similarity between the query vector and each of the target nodes, and the similarity between the query vector and the neighboring nodes. The cosine similarity algorithm is specifically as follows:

[0067] ;

[0068] in, represents the query vector, represents the vector of the target node, Represents the similarity between the query vector and the target node;

[0069] The first weight of the target node is calculated according to the softmax function, and the softmax function is specifically:

[0070] ;

[0071] in, represents the first weight of the target node, represents the vector of neighbor nodes, represents the similarity between the query vector and the neighboring nodes, Represents neighbor nodes The node set belonging to the target chronic disease patient in the knowledge graph .

[0072] In a possible implementation, the chronic disease population health assessment device further includes a sixth processing module; before obtaining abnormal physiological state information of the target chronic disease patient and performing vector processing on the abnormal physiological state information to obtain a query vector, the sixth processing module is configured to execute:

[0073] Based on the health data of the target chronic disease patient acquired in real time, determining the physiological indicators with abnormal risks in the health data; the health data includes physiological data corresponding to a plurality of physiological indicators for feedback on the health status of the target chronic disease patient;

[0074] Perform risk trend analysis on the physiological indicators with abnormal risks to determine the risk trend level of the chronic disease of the target chronic disease patient.

[0075] In a possible implementation, when determining that there are physiological indicators with abnormal risks in the health data based on the health data of the target chronic disease patient acquired in real time, the sixth processing module is specifically configured to execute:

[0076] Obtain health data of target chronic disease patients in real time;

[0077] Taking the physiological data corresponding to each physiological indicator of the target chronic disease patient as the original input data, inputting the original input data into the autoencoder trained with the historical health data of the healthy population, and compressing and reconstructing the original input data through the autoencoder to obtain the corresponding reconstructed data;

[0078] A reconstruction error between the original input data and the reconstructed data is calculated, and according to the size of the reconstruction error and a first preset error threshold, the detection result of the physiological indicator corresponding to the original input data is marked to obtain a marked result, and the physiological indicators with abnormal risks in the marked result are screened out; the marked result is used to indicate whether the physiological indicator corresponding to the original input data has abnormal risks.

[0079] In a possible implementation, when performing risk trend analysis on the physiological indicator with abnormal risk to determine the risk trend level of the chronic disease of the target chronic disease patient, the sixth processing module is specifically configured to execute:

[0080] When there are physiological indicators of abnormal risk in the marking results, the health data of the target chronic disease patient and the marking results corresponding to multiple physiological indicators of the health data of the target chronic disease patient are used as target evaluation data, and the target evaluation data are input into a support vector machine model trained with historical chronic disease data of the chronic disease population. The target evaluation data is subjected to risk trend analysis by the support vector machine model to obtain a risk trend level.

[0081] In a possible implementation, before performing risk trend analysis on the target assessment data through the support vector machine model, the sixth processing module is further configured to execute:

[0082] The original support vector machine model is constructed using historical chronic disease data as the first training sample and the target evaluation data as the sample to be evaluated; the sample to be evaluated includes a first feature matrix composed of physiological data corresponding to multiple physiological indicators of the target chronic disease patient and the labeling results of the physiological indicators, and each of the first training samples includes a second feature matrix composed of physiological data corresponding to multiple physiological indicators of a chronic disease patient and the labeling results of the physiological indicators, and a risk label corresponding to the second feature matrix;

[0083] The decision function of the original support vector machine model is:

[0084] ;

[0085] in, For the return function, when When +1 is returned, it means that the chronic disease risk trend level of the sample to be evaluated is low risk. When -1 is returned, it means that the chronic disease risk trend level of the sample to be evaluated is high risk;

[0086] represents the number of the first training samples, represents the position of the current first training sample, represents the Lagrange multiplier parameter, is the bias parameter;

[0087] Indicates The second feature matrix of the first training sample, Indicates a risk label of the first training sample, when the risk label is +1, it indicates that the chronic disease risk is high risk, and when the risk label is -1, it indicates that the chronic disease risk is low risk;

[0088] represents the sample to be evaluated;

[0089] is a kernel function used to calculate the similarity between the first training sample and the sample to be evaluated.

[0090] In a possible implementation, each of the physiological indicators includes a plurality of indicator features; the kernel function of the decision function of the original support vector machine model is determined according to the radial basis function and the polynomial kernel function;

[0091] The kernel function is specifically configured as:

[0092] ;

[0093] in, represents the adjustment parameter of the radial basis function, which is a preset fixed value;

[0094] Indicates the degree of the polynomial in the polynomial kernel function, which is a preset fixed value;

[0095] Indicates the total number of physiological indicators included in health data, Indicates the current position of physiological indicators;

[0096] Indicates the total number of indicator characteristics included in a physiological indicator. Indicates the location of the current indicator feature;

[0097] Indicates The weight coefficient of each physiological index;

[0098] Indicates the sample to be evaluated. The first physiological index The value of an indicator feature;

[0099] Indicates The first training sample The first physiological index The numerical value of an indicator feature.

[0100] In a possible implementation, the autoencoder includes an encoding unit, an attention unit, and a decoding unit; when the original input data is compressed and reconstructed by the autoencoder to obtain corresponding reconstructed data, the sixth processing module is specifically configured to execute:

[0101] The encoding unit compresses the input high-dimensional original input data into low-dimensional first data through multiple fully connected layers;

[0102] The attention unit extracts important features related to the chronic disease of the target chronic disease patient through an attention mechanism to obtain second data;

[0103] The decoding unit reconstructs the second data into reconstructed data with the same dimension as the original input data through multiple fully connected layers.

[0104] In a third aspect, the present invention provides an electronic device comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement a health assessment method for a chronic disease population as described in any one of the above items.

[0105] In a fourth aspect, the present invention provides a computer-readable storage medium, which stores at least one instruction, at least one program, a code set or an instruction set. The at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by a processor to implement the health assessment method for a chronic disease population as described in any one of the above items.

[0106] In practical applications, the health assessment method for chronic disease population provided by the embodiment of the present invention firstly obtains a pre-constructed knowledge graph for storing and recording historical patient information of chronic disease population. In the knowledge graph, the basic information in the historical patient information of each chronic disease patient, each physiological indicator in the historical health data, each historical diagnosis information, and each historical treatment information are respectively taken as a node, and the degree of association between any two nodes is taken as the edge weight between the two nodes; secondly, the physiological indicators with abnormal risks determined based on the health data of the target chronic disease patient and the risk trend level determined after risk trend analysis of the physiological indicators with abnormal risks are integrated as the abnormal physiological state information of the target chronic disease patient, and the abnormal physiological state information is vectorized to obtain a query vector; thirdly, each node of the target chronic disease patient in the knowledge graph is taken as a target node, and the nodes with edge weights with the target node greater than a first preset value are taken as neighbor nodes of the target node. When calculating the query vector and each node, the node is vectorized to obtain a query vector. When calculating the correlation of a target node, the influence of the neighboring nodes of the target node on the correlation is comprehensively considered to obtain a more accurate correlation between each target node and the query vector; then, all target nodes are sorted in descending order according to the correlation size, and the first Z target nodes are selected as the key medical information of the target patient; finally, the key medical information is processed by the MedGPT model to obtain a chronic disease health assessment plan including the risk trend level of the target chronic disease patient and treatment plan adjustment suggestions; the present invention dynamically integrates the real-time health data and historical patient information of the target patient through the knowledge graph, and screens out the key medical information that best reflects the disease development status of the target chronic disease patient according to the correlation between the query vector and each target node in the knowledge graph, so as to timely discover potential risks and conduct accurate health assessments on chronic disease patients; the MedGPT model can directly generate the text of the chronic disease health assessment plan for the key medical information, which not only improves the generation speed of the treatment plan adjustment suggestions, but also ensures the professionalism and personalization of the suggestions. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] Figure 1 A flowchart of the steps of a method for health assessment of chronic disease population provided by an embodiment of the present invention;

[0108] Figure 2 A structural block diagram of a chronic disease population health assessment device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0109] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0110] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, unless otherwise specified, "multiple" means two or more. In addition, the use of "based on" or "according to" means openness and inclusiveness, because the process, steps, calculations or other actions "based on" or "according to" one or more of the conditions or values ​​may be based on additional conditions or values ​​beyond the described values ​​in practice.

[0111] In order to solve the problem that the prior art cannot accurately assess the health risks of chronic disease patients in combination with their health conditions, an embodiment of the present invention provides a health assessment method and device for chronic disease population.

[0112] like Figure 1 As shown, in the first aspect, an embodiment of the present invention provides a method for health assessment of a chronic disease population, the method comprising:

[0113] Step 101: Obtain a predetermined knowledge graph for storing and recording historical patient information of a chronic disease population.

[0114] In the knowledge graph, the basic information of each chronic disease patient in the historical patient information, each physiological indicator in the historical health data, each historical diagnosis information, and each historical treatment information are taken as a node respectively, and the degree of association between any two nodes is taken as the edge weight between the two nodes.

[0115] Specifically, when the knowledge graph is established, nodes need to be defined first. When defining nodes, the basic information of each chronic disease patient can be used as a node, including the name, gender, age, height, weight and other personal basic information of the chronic disease patient; the physiological indicators of the chronic disease patient such as blood pressure, blood sugar, heart rate and so on can be used as a node; each piece of historical diagnosis information of the chronic disease patient can be used as a node, where the historical diagnosis information can be the disease diagnosis information of the chronic disease patient obtained in the hospital, or the diagnosis information in the historical chronic disease health assessment plan; each piece of historical treatment information of the chronic disease patient can be used as a node, where the historical treatment information includes drug treatment information and other auxiliary treatment information, which can be obtained from the treatment record of the chronic disease patient in the hospital, or from the treatment plan adjustment suggestions in the historical chronic disease health assessment plan.

[0116] After completing the node definition, in order to facilitate subsequent analysis and calculation, each node is converted into a numerical vector through the Embedding function. In the embodiment of the present invention, the Embedding function adopts the Node2Vec function.

[0117] The degree of association between any two nodes is the influence relationship and causal relationship between any two nodes. For example, a patient's high blood pressure may increase the risk of heart disease, and a specific treatment plan can alleviate the patient's symptoms.

[0118] The edge weight between any two nodes can be determined by calculating clinical medical data or by expert scoring.

[0119] Step 102: Acquire abnormal physiological status information of the target chronic disease patient, and vectorize the abnormal physiological status information to obtain a query vector.

[0120] Among them, the abnormal physiological state information includes physiological indicators with abnormal risks determined based on the health data of the target chronic disease patients, and the risk trend level determined after risk trend analysis of the physiological indicators with abnormal risks.

[0121] In the physiological indicators of abnormal risk, for example: the blood sugar fluctuation of the target chronic disease patient obviously exceeds the blood sugar set value, or the heart rate fluctuation of the target chronic disease patient obviously exceeds the heart rate set value. In the risk trend level, for example: according to the preset conditions, the blood sugar value of the target chronic disease patient with abnormal risk is analyzed, and the risk trend level of type 2 diabetes of the target chronic disease patient is determined to be high risk. The above information is integrated to obtain the abnormal physiological state information of the target chronic disease patient. In order to facilitate the analysis of the abnormal physiological state information, the abnormal physiological state information is vectorized through the Embedding function to obtain the query vector corresponding to the abnormal physiological state information.

[0122] Step 103: Take each node of the target chronic disease patient in the knowledge graph as the target node, take the node whose edge weight with the target node is greater than the first preset value as the neighbor node of the target node, and calculate the correlation between the query vector and each target node in combination with the edge weight between the neighbor node and the target node.

[0123] Specifically, in the present invention, the importance of each target node is reflected by calculating the correlation between the query vector and each target node, so as to facilitate the subsequent selection of points with greater importance to determine a more accurate chronic disease health assessment plan.

[0124] In order to improve the accuracy of correlation calculation, when calculating the correlation between each target node and the query vector, not only the association degree between the target node and the query vector should be considered, but also the influence of the neighbor nodes of the target node on the target node should be considered.

[0125] The value of edge weight is generally between 0 and 1. In the present invention, the first preset value can be set to 0.5, that is, when the edge weight between the target node and other nodes is greater than 0.5, the node is the neighbor node of the target node. A target node can have no neighbor nodes, or one or more neighbor nodes.

[0126] Step 104: sort all target nodes in descending order according to the magnitude of the relevance, and extract the first Z target nodes in the sorting to obtain the key medical information of the target chronic disease patient.

[0127] Wherein, Z is a positive integer, and the specific value of Z can be adjusted as needed. For example, when there are 1000 target nodes, the first 10%, that is, the first 100 target nodes can be aggregated to form the key medical information of the target chronic disease patient.

[0128] In this embodiment, a weighted sum is performed on the selected multiple target nodes to obtain an aggregate vector, which is then decoded by a decoding function Decode corresponding to the Embedding function to obtain key medical information of the target chronic disease patient.

[0129] Key medical information is a text that describes the health status and related medical information of the target chronic disease patient.

[0130] For example: "Zhang San; 58 years old; male; blood sugar tested at 12 o'clock this week; abnormal 3 times; blood sugar continued to rise and fluctuated abnormally; heart rate indicators were normal;...; high risk of type 2 diabetes; history of cardiovascular disease; current medications include..., and the drug dosage is..."

[0131] Step 105: Input the key medical information into the MedGPT model, and the MedGPT model generates a chronic disease health assessment plan for the target chronic disease patient based on the key medical information.

[0132] Among them, the chronic disease health assessment plan includes the risk trend level of target chronic disease patients and treatment plan adjustment recommendations.

[0133] Specifically, the MedGPT model is an existing language model designed specifically for the medical field. After training, the model can process medical data and health data, including case records, medical literature, clinical guidelines, etc., enabling it to generate high-quality medical texts and recommendations.

[0134] The core capabilities of the MedGPT model are: understanding complex medical contexts and generating outputs with practical application value based on this, such as disease diagnosis support, treatment plan recommendations, and other medical-related documents.

[0135] The key medical information in step 104 is integrated into a pre-set template with prompt words.

[0136] In this embodiment, the template with prompt words is:

[0137] "Please provide targeted medical advice based on the above information, including medication dosage adjustments, lifestyle changes, and necessary tests or treatments."

[0138] After key medical information is input into a template with prompt words, the filled template is input into the MedGPT model, and the MedGPT model outputs the corresponding chronic disease health assessment plan based on the filled template.

[0139] In this embodiment, the treatment plan adjustment suggestions in the chronic disease health assessment plan are as follows:

[0140] Based on patient Zhang San's current blood sugar monitoring results and health status, the following adjustments to the existing diabetes treatment plan are recommended:

[0141] 1. Medication adjustment:

[0142] The dose of metformin was increased to 850 mg twice daily to more effectively control blood sugar levels.

[0143] Consider introducing a glucocorticoid inhibitor, such as sitagliptin, once daily to improve insulin sensitivity.

[0144] 2. Regular inspection:

[0145] Comprehensive blood glucose monitoring, including fasting and postprandial blood glucose, is recommended once a month.

[0146] It is recommended that the cardiovascular system be examined every six months to monitor for possible development of cardiovascular disease.

[0147] In practical applications, the health assessment method for chronic disease population provided by the embodiment of the present invention firstly obtains a pre-constructed knowledge graph for storing and recording historical patient information of chronic disease population. In the knowledge graph, the basic information in the historical patient information of each chronic disease patient, each physiological indicator in the historical health data, each historical diagnosis information, and each historical treatment information are respectively taken as a node, and the degree of association between any two nodes is taken as the edge weight between the two nodes; secondly, the physiological indicators with abnormal risks determined based on the health data of the target chronic disease patient and the risk trend levels determined after risk trend analysis of the physiological indicators with abnormal risks are integrated as the abnormal physiological state information of the target chronic disease patient, and the abnormal physiological state is analyzed. The information is vectorized to obtain a query vector; again, each node of the target chronic disease patient in the knowledge graph is taken as a target node, and the nodes with an edge weight with the target node greater than a first preset value are taken as neighbor nodes of the target node. When calculating the correlation between the query vector and each target node, the influence of the neighbor nodes of the target node on the correlation is comprehensively considered to obtain a more accurate correlation between each target node and the query vector; then, all target nodes are sorted in descending order by the size of the correlation, and the first Z target nodes are selected as the key medical information of the target patient; finally, the key medical information is processed through the MedGPT model to obtain a chronic disease health assessment plan that includes the risk trend level of the target chronic disease patient and treatment plan adjustment suggestions.

[0148] The present invention dynamically integrates the real-time health data and historical patient information of the target patient through the knowledge graph, and screens out the key medical information that best reflects the disease progression of the target chronic disease patient according to the correlation between the query vector and each target node in the knowledge graph, so as to timely discover potential risks and conduct accurate health assessments on chronic disease patients; the MedGPT model can directly generate the text of the chronic disease health assessment plan based on the key medical information, which not only improves the speed of generating treatment plan adjustment suggestions, but also ensures the professionalism and personalization of the suggestions.

[0149] Furthermore, in step 103, the correlation between the query vector and each target node is calculated respectively in combination with the edge weights between the neighbor nodes and the target node, specifically including:

[0150] A first weight of the target node is determined according to a predetermined similarity between the query vector and each target node, and a similarity between the query vector and a neighboring node.

[0151] Among them, the similarity between the query vector and the neighboring nodes refers to the semantic similarity.

[0152] The average of the sum of the edge weights between the target node and each neighbor node is calculated to obtain the second weight of the target node.

[0153] The first weight and the second weight are weighted and summed to obtain a comprehensive weight of the target node, and the comprehensive weight is used as the relevance of the target node.

[0154] Further, determining a first weight of the target node according to the predetermined similarity between the query vector and each target node, and the similarity between the query vector and the neighboring nodes, specifically includes:

[0155] The cosine similarity algorithm is used to calculate the similarity between the query vector and each target node, as well as the similarity between the query vector and the neighboring nodes. The specific cosine similarity algorithm is:

[0156] ;

[0157] in, represents the query vector, represents the vector of the target node, Represents the similarity between the query vector and the target node;

[0158] The first weight of the target node is calculated according to the softmax function, which is specifically:

[0159] ;

[0160] in, represents the first weight of the target node, represents the vector of neighbor nodes, represents the similarity between the query vector and the neighboring nodes, Represents neighbor nodes The node set belonging to the target chronic disease patient in the knowledge graph .

[0161] Furthermore, before obtaining the abnormal physiological state information of the target chronic disease patient and performing vector processing on the abnormal physiological state information to obtain the query vector, the method further includes:

[0162] Based on the health data of target chronic disease patients obtained in real time, determine the physiological indicators with abnormal risks in the health data.

[0163] Among them, the health data includes physiological data corresponding to multiple physiological indicators used to provide feedback on the health status of target chronic disease patients.

[0164] Physiological indicators used to provide feedback on the health status of target chronic disease patients include not only heart rate and blood pressure, but also calories consumed, exercise steps, exercise distance, resting heart rate, sleeping heart rate, exercise heart rate, light exercise heart rate, sedentary time, aerobic exercise time, aerobic exercise calories consumed, and fat burned by target chronic disease patients in the target time period; physiological indicators such as sleep efficiency, sleep time, sleep onset time, sleep start time, sleep end time, deep sleep time, light sleep time, REM sleep time, and wake-up times of target chronic disease patients.

[0165] A physiological indicator with an abnormal risk means that the physiological data corresponding to the physiological indicator exceeds the normal range of the physiological indicator.

[0166] Conduct risk trend analysis on physiological indicators with abnormal risks to determine the risk trend level of chronic diseases in target chronic disease patients.

[0167] Furthermore, based on the health data of the target chronic disease patients obtained in real time, the physiological indicators with abnormal risks in the health data are determined, including:

[0168] Obtain health data of target chronic disease patients in real time.

[0169] In order to ensure the accuracy and real-time nature of the data, the equipment used to obtain the health data of the target chronic disease patients needs to be suitable for long-term monitoring, such as wearable devices and home medical devices. Wearable devices can be smart bracelets, heart rate monitors, etc.; home medical devices can be blood glucose meters and blood pressure meters.

[0170] The equipment used to obtain health data of target chronic disease patients is not only required to be able to continuously and accurately collect the patient's physiological data, but also to ensure the patient's comfort and convenience.

[0171] In this embodiment, each physiological indicator is time-serialized, that is, each physiological indicator is measured continuously or periodically over time.

[0172] The physiological data corresponding to each physiological indicator of the target chronic disease patient is used as the original input data, and the original input data is input into the autoencoder trained with the historical health data of the healthy population. The original input data is compressed and reconstructed by the autoencoder to obtain the corresponding reconstructed data.

[0173] Specifically, the autoencoder is trained through the historical health data of healthy people, that is, the autoencoder corresponding to each physiological indicator of the healthy people is trained through the historical health data corresponding to the physiological indicator. That is, the compression and reconstruction rules of each autoencoder are determined based on the historical health indicator data of the healthy people corresponding to the physiological indicator of the encoder. Therefore, after the physiological data corresponding to each physiological indicator of the target chronic disease patient is input as the original input data into the trained autoencoder of the corresponding physiological indicator, each autoencoder will compress and reconstruct the original input data according to the compression and reconstruction rules of the historical health data of the corresponding physiological indicator. At this time, the reconstructed data after reconstruction is the theoretical health data of the original input data. By studying the error between the original input data and the reconstructed data, it can be determined whether the physiological indicator corresponding to the original input data is abnormal.

[0174] The reconstruction error between the original input data and the reconstructed data is calculated, and according to the size of the reconstruction error and the first preset error threshold, the detection result of the physiological indicator corresponding to the original input data is marked to obtain the marked result, and the physiological indicators with abnormal risks in the marked result are screened out.

[0175] Among them, the marking result is used to indicate whether the physiological indicators corresponding to the original input data have abnormal risks.

[0176] Specifically, for each set of original input data and reconstructed data, the reconstruction error between the original input data and the reconstructed data is first calculated. If the reconstruction error is greater than or equal to a first preset error threshold, it is considered that the physiological indicator corresponding to the original input data has an abnormal risk, and the physiological indicator is marked as abnormal; if the reconstruction error is less than the first preset error threshold, it is considered that the physiological indicator corresponding to the original input data does not have an abnormal risk, and the physiological indicator is marked as normal; through this method, the detection and labeling of all physiological indicators included in the health data of the target chronic disease patient are completed, and the labeling results are obtained, and the abnormal indicator is represented by 1 and the normal indicator is represented by 0.

[0177] If all physiological indicators in the marking results are normal, the target chronic disease patient does not have any health risk during this time period. If there are physiological indicators marked as abnormal, further health risk assessment of the target chronic disease patient is required during this time period.

[0178] In this embodiment, taking the blood sugar index of the target chronic disease patient as an example, the blood sugar data autoencoder is obtained by training with the historical blood sugar data of the healthy population.

[0179] The blood glucose data detected before and after breakfast, lunch and dinner of the target chronic disease patients for N consecutive days are used as the original input data. The original input data is compressed and reconstructed by the trained blood glucose data autoencoder to obtain the reconstructed data before and after breakfast, lunch and dinner of the target chronic disease patients for N consecutive days.

[0180] The average error between the input blood glucose and the reconstructed blood glucose in N days is calculated. If the calculated average error exceeds a first preset error threshold, the blood glucose index of these N days is marked as abnormal.

[0181] Through the above method, the patient's blood sugar level can be monitored in real time and potential health risks can be identified in time. This monitoring method is particularly suitable for the field of chronic disease management that requires continuous monitoring and immediate response.

[0182] Furthermore, risk trend analysis is performed on physiological indicators with abnormal risks to determine the risk trend level of chronic diseases of target chronic disease patients, including:

[0183] When there are physiological indicators with abnormal risks in the labeling results, the health data of the target chronic disease patient and the labeling results corresponding to multiple physiological indicators of the target chronic disease patient's health data are used as target evaluation data. The target evaluation data are input into a support vector machine model trained with historical chronic disease data of the chronic disease population. The support vector machine model is used to perform risk trend analysis on the target evaluation data to obtain the risk trend level.

[0184] When there are physiological indicators marked as abnormal in the marking results, target evaluation data are constructed through the health data of the target patient and the marking results corresponding to each physiological indicator in the health data.

[0185] For each physiological indicator in the health data of the target patient, statistical features of the physiological data corresponding to the physiological indicator are extracted, and the statistical features include but are not limited to the mean value, standard deviation, etc. of the physiological data.

[0186] The target assessment data is input into the support vector machine model, and the risk trend analysis of the target assessment data is performed through the support vector machine model to obtain the risk trend level corresponding to the target assessment data.

[0187] Specifically, the support vector machine model is trained by the historical chronic disease data of multiple chronic disease patients. The historical chronic disease data of each patient includes the historical health data of the target patient, the labeling result corresponding to each physiological indicator in the historical health data, and the risk trend of the historical chronic disease data.

[0188] The historical health data of the target patient includes physiological data corresponding to a plurality of physiological indicators, and the physiological data corresponding to each physiological indicator is represented by a plurality of extracted statistical features.

[0189] In this embodiment, taking Zhang San's health data for one month as an example, the health data includes multiple physiological indicators such as blood pressure and heart rate; the statistical characteristics of the blood pressure indicator include: the maximum blood pressure this month, the minimum blood pressure this month, the average blood pressure this month, and the standard deviation of blood pressure this month. The chronic disease risk trend corresponding to the health data is high risk or low risk.

[0190] The risk trend of the target patient's historical chronic disease data can be determined through doctor scoring.

[0191] The multiple physiological indicators corresponding to each patient's health data, the labeling results corresponding to each physiological indicator, and the corresponding multiple statistical features are represented by a feature matrix S, and the chronic disease risk trend corresponding to the patient's health data is represented by a label vector Y, where the label indicates that the chronic disease risk trend is classified as high risk or low risk.

[0192] The feature matrix S and label vector Y corresponding to each patient's health data are regarded as a point on the plane. The historical chronic disease data of the chronic disease population include (S1, Y1), (S2, Y2)… (S u , Y u ), where u represents the total number of historical chronic disease data.

[0193] The health data of the target chronic disease patients are input into the trained support vector machine model as new points to be classified. The trained support vector machine model can classify the new points to be classified, determine whether the new points to be classified should be classified into the high-risk category or the low-risk category, and obtain the risk trend level.

[0194] The present invention utilizes autoencoding and support vector machine technologies to conduct accurate health risk assessments on patients with chronic diseases by real-time monitoring of patient health data, so as to timely discover potential risks.

[0195] Furthermore, before performing risk trend analysis on the target assessment data through the support vector machine model, the method further includes:

[0196] The original support vector machine model was constructed using historical chronic disease data as the first training sample and the target evaluation data as the sample to be evaluated.

[0197] The samples to be evaluated include a first feature matrix consisting of physiological data corresponding to multiple physiological indicators of target chronic disease patients and labeling results of the physiological indicators.

[0198] Each first training sample includes a second feature matrix composed of physiological data corresponding to multiple physiological indicators of a chronic disease patient and labeling results of the physiological indicators, and a risk label corresponding to the second feature matrix.

[0199] The decision function of the original support vector machine model is:

[0200] ;

[0201] in, For the return function, when When +1 is returned, it means that the chronic disease risk trend level of the sample to be evaluated is low risk. When -1 is returned, it means that the chronic disease risk trend level of the sample to be evaluated is high risk.

[0202] represents the number of the first training samples, represents the position of the current first training sample, represents the Lagrange multiplier parameter, is the bias parameter.

[0203] Indicates The second feature matrix of the first training sample, Indicates The risk label of the first training sample is +1, which means that the chronic disease risk is high, and when the risk label is -1, which means that the chronic disease risk is low.

[0204] Indicates the sample to be evaluated.

[0205] is a kernel function used to calculate the similarity between the first training sample and the sample to be evaluated.

[0206] Furthermore, each physiological indicator includes multiple indicator features.

[0207] The kernel function of the decision function of the original support vector machine model is determined according to the radial basis function and the polynomial kernel function.

[0208] The kernel function is specifically:

[0209] ;

[0210] in, Represents the adjustment parameter of the radial basis function. It is a preset fixed value used to control the bandwidth of the Gaussian function. The setting of this parameter can affect the smoothness of the model.

[0211] Represents the degree of the polynomial in the polynomial kernel function. It is a preset fixed value used to determine the complexity and fitting ability of the polynomial kernel.

[0212] Indicates the total number of physiological indicators included in health data, Indicates the current position of physiological indicators.

[0213] Indicates the total number of indicator characteristics included in a physiological indicator. Indicates the location of the current indicator feature.

[0214] Indicates The weight coefficient of each physiological indicator can be adjusted according to the importance of the physiological indicator. For example, for diabetic patients, the weight coefficient of blood sugar indicator will be set higher than the weight coefficient of blood pressure indicator.

[0215] Indicates the sample to be evaluated. The first physiological index The numerical value of an indicator feature.

[0216] Indicates The first training sample The first physiological index The numerical value of an indicator feature.

[0217] In this embodiment, when the first training sample is Zhang San's health data for one month as an example, the health data includes three physiological indicators: blood sugar, blood pressure, and heart rate. The indicator characteristics of blood sugar include four indicators: the maximum blood sugar value this month, the minimum blood sugar value this month, the average blood sugar value this month, and the standard deviation of blood sugar this month.

[0218] In the first training sample data, when =1, =1, =1, It represents the maximum blood sugar value of Zhang San this month.

[0219] In the sample data to be evaluated, when =1, =1, It indicates the maximum blood sugar value of the target chronic disease patient this month.

[0220] The kernel function constructed by the present invention combines the advantages of radial basis function and polynomial kernel function, that is, it takes into account the nonlinear characteristics of data and adds high-order interaction terms of polynomials, so that the support vector machine model can better capture the complex relationship of patient health data. At the same time, dynamic weights are introduced The kernel function is adjusted according to the importance of each health indicator, which enhances the responsiveness of the support vector machine model to key health indicators. Finally, the design of the kernel function allows for flexible adjustment in different chronic disease types and patient groups, and by changing the weight coefficients and parameters, it can be optimized for specific chronic diseases and patient characteristics.

[0221] The present invention creates a kernel function specifically for processing health data of chronic disease patients based on radial basis functions and polynomial kernel functions, and adds a weighting strategy for key physiological indicators. The kernel function can help capture the complex patterns and nonlinear relationships in the health data of target chronic disease patients, especially when processing chronic disease indicators such as blood sugar and blood pressure that may have nonlinear characteristics. The kernel function can also enhance the sensitivity to key physiological indicators.

[0222] In this embodiment, before performing risk trend analysis on the target assessment data through the support vector machine model, the method further includes:

[0223] Using historical chronic disease data as the first training sample, the original support vector machine model was trained based on the Lagrange multiplier method to obtain the maximum value of the Lagrange multiplier parameter and the minimum value of the bias parameter;

[0224] Substitute the maximum value of the Lagrange multiplier parameter and the minimum value of the bias parameter into the original support vector machine model to obtain a trained support vector machine model.

[0225] Among them, the training process of the support vector machine model is the process of determining the separation plane of data points labeled as high risk or low risk, that is, the process of determining the classification boundary. This classification boundary can distinguish data points of different categories to the greatest extent to increase the robustness of the support vector machine model.

[0226] Furthermore, the autoencoder includes an encoding unit, an attention unit, and a decoding unit.

[0227] The original input data is compressed and reconstructed through the autoencoder to obtain the corresponding reconstructed data, including:

[0228] The encoding unit compresses the input high-dimensional original input data into low-dimensional first data through multiple fully connected layers.

[0229] Specifically, the encoding unit is used to compress the original input data into a low-dimensional hidden layer representation through multiple fully connected layers.

[0230] Each fully connected layer of the encoding unit is defined as:

[0231] ;

[0232] in, Indicates the first fully connected layers, express The previous fully connected layer, represents the ReLU activation function, and Respectively represent the first The weights and biases of the fully connected layers.

[0233] The attention unit extracts important features related to the chronic diseases of the target chronic disease patients through the attention mechanism to obtain the second data.

[0234] Among them, the attention unit is used to identify and amplify the physiological indicators of chronic diseases. For example, when processing the data of diabetic patients, the attention unit can focus on the pattern of blood sugar fluctuations, insulin responsiveness, etc.

[0235] The second data output by the attention unit can be expressed as:

[0236] ;

[0237] in, The second data representing the output of the attention unit, represents the output of the last fully connected layer of the encoding unit, represents the attention function, and denote the weight and bias of the attention unit respectively.

[0238] The decoding unit reconstructs the second data into reconstructed data having the same dimension as the original input data through multiple fully connected layers.

[0239] The decoding unit is used to reconstruct the output hidden layer into reconstructed data of the same dimension as the original input data. Specifically, the dimension of the reconstructed data is gradually increased through multiple fully connected layers until the dimension of the reconstructed data is the same as the dimension of the original input data.

[0240] Each fully connected layer of the decoding unit is defined as:

[0241] ;

[0242] in, Indicates the decoding unit fully connected layers, express The previous fully connected layer, and Respectively represent the decoding unit The weights and biases of the fully connected layers.

[0243] The reconstructed data output by the decoding unit is expressed as:

[0244] ;

[0245] in, represents the reconstructed data output by the decoding unit, represents the last layer of the decoding unit, It represents the previous fully connected layer of the last fully connected layer of the decoding unit. and They represent the weights and biases of the fully connected layer at the last layer of the decoding unit, respectively.

[0246] Furthermore, before compressing and reconstructing the original input data by the autoencoder, the method further includes:

[0247] The historical health data of healthy people is used as the second training sample, and the constructed initial autoencoder is trained by the second training sample until the error between the reconstructed data output by the autoencoder calculated by the loss function and the input data meets the second preset error threshold, and the trained autoencoder is output.

[0248] Specifically, the process of training the autoencoder is the process of determining the weights and biases of each fully connected layer of the autoencoder encoding unit, the attention unit, and each fully connected layer of the decoding unit.

[0249] Furthermore, the loss function of the autoencoder is the weighted sum of the health parameter sensitive loss function and the anomaly detection penalty loss function.

[0250] The health parameter sensitive loss function is specifically:

[0251] ;

[0252] in, is the total number of physiological indicators in health data, is the current position of the physiological indicator, is the weight coefficient of the current physiological index, For the The original input data of physiological indicators, For the Reconstructed data of physiological indicators.

[0253] The specific anomaly detection penalty loss function is:

[0254] ;

[0255] in, is the adjustment parameter of the anomaly detection penalty loss function, which is a preset fixed value;

[0256] is the preset reconstruction error threshold.

[0257] Specifically, in order to train the autoencoder, it is necessary to minimize the loss between the original input data and the reconstructed data. Therefore, we introduce the health parameter sensitivity loss function and the anomaly detection penalty loss function as the loss function of the autoencoder to improve the sensitivity of the autoencoder to the physiological indicators related to the chronic disease to be studied; for data with large reconstruction errors, an additional penalty term is introduced to reduce the final reconstruction error.

[0258] In the autoencoder of the present invention, in order to improve the application effect of the autoencoder in the anomaly detection process, we introduce an attention unit specifically for processing chronic disease features in the traditional autoencoder, and the attention unit focuses on extracting key features related to the progression of chronic diseases from physiological data. In other words, the attention unit acts as an enhancement module in the autoencoder to enhance the sensitivity and responsiveness of the autoencoder to key indicators of chronic diseases.

[0259] In the embodiment of the present invention, the health risk assessment of type 2 diabetes is taken as an example. The target chronic disease patient is a patient with type 2 diabetes, and the physiological indicators of the acquired health data include: blood sugar data, daily diet data, patient's weight record data, blood pressure and heart rate, etc.

[0260] Use the trained autoencoder and support vector machine models to perform health risk analysis on the health data of target chronic disease patients.

[0261] Specifically, historical health data of normal people or historical health data collected from patients with type 2 diabetes during a stable period of their condition are used as training samples. The autoencoder model is trained through the training samples, so that the autoencoder model learns to reconstruct normal health data patterns.

[0262] By inputting the health data of the target chronic disease patients into the trained autoencoder, the autoencoder model compresses and reconstructs the input data to obtain reconstructed data. According to the reconstruction error between the reconstructed data and the input data and the size of the first preset error threshold, it is determined whether each physiological indicator has potential risks, such as abnormal increase or decrease in blood sugar.

[0263] When there are abnormal risks in the health data of the target chronic disease patients, the data is input into the trained support vector machine model, and the support vector machine model is used to classify the data into risk levels, including low risk and high risk.

[0264] When the risk trend level is low risk, it means that the type 2 diabetes of the target chronic disease patient is well controlled. When the risk trend level is high risk, it is necessary to screen out the key medical information that best reflects the disease progression of the target chronic disease patient based on the correlation between the query vector and each target node in the knowledge graph, so as to timely discover potential risks and conduct accurate health risk assessment for chronic disease patients; then use the MedGPT model to directly generate the text of the chronic disease health assessment plan based on the key medical information.

[0265] The health assessment method for chronic disease population of the present invention uses an autoencoder to learn normal health data patterns, and identifies potential abnormal physiological indicators by comparing the reconstruction errors of the reconstructed data processed by the autoencoder with the original input data, thereby greatly improving the accuracy of early diagnosis.

[0266] After determining that the target patient has physiological indicators with abnormal risks, the present invention performs trend analysis on the patient's long-term data through a support vector machine model, identifies the patient's risk of developing chronic diseases and physiological indicators that require the patient's special attention, and supports personalized chronic disease management plans.

[0267] The support vector machine model of the present invention adopts a kernel function combining radial basis function and polynomial kernel function, which can better handle nonlinear data in health data and the complex relationship between various physiological indicators in health data. By setting different weights for different physiological indicators, the support vector machine model's responsiveness to key features is enhanced, so that the support vector machine model can more accurately identify risks and trends associated with specific chronic diseases.

[0268] The autoencoder of the present invention introduces an attention unit specially designed for chronic diseases, which is used for in-depth learning and extraction of features closely related to the progression of chronic diseases. The attention mechanism is used to focus on processing features related to chronic diseases, thereby improving the autoencoder's ability to detect abnormalities in physiological indicators.

[0269] The loss function of the autoencoder of the present invention dynamically adjusts the loss weight according to the importance of each physiological indicator through a customized health parameter sensitive loss function, ensuring that the autoencoder pays more attention to the physiological indicators that have a greater impact on the development of chronic diseases during the training process. While learning the general characteristics of the sample data through the customized health parameter sensitive loss function, it also strengthens the special learning that has a greater impact on chronic diseases, thereby improving the accuracy of the autoencoder model prediction and monitoring.

[0270] The loss function of the autoencoder of the present invention also adds an abnormal detection penalty term, so that the autoencoder has a higher sensitivity to potential monitoring analysis during the training process, so that the autoencoder can warn of potential health problems in advance in practical applications.

[0271] The loss function design of the autoencoder of the present invention enables the autoencoder to more accurately identify and mark abnormal data, thereby supporting doctors to make more timely and accurate medical decisions.

[0272] The present invention dynamically integrates the real-time health data of target patients with historical patient information through knowledge graphs to form a fully interconnected data structure.

[0273] The present invention utilizes the MedGPT model to automatically generate the text of the chronic disease health assessment plan based on the extracted key medical information, which not only improves the generation speed of the assessment plan, but also ensures the professionalism and personalization of the treatment plan adjustment suggestions in the chronic disease health assessment plan.

[0274] In order to ensure treatment efficiency, all treatment plan adjustment suggestions in the generated chronic disease health assessment plans need to be manually reviewed by doctors and pushed to the target chronic disease patients for application.

[0275] like Figure 2 As shown, in a second aspect, an embodiment of the present invention further provides a health assessment device for chronic disease population, the device comprising:

[0276] The first processing module 201 is used to obtain a predetermined knowledge graph for storing and recording historical patient information of a chronic disease population; in the knowledge graph, the basic information of each chronic disease patient in the historical patient information, each physiological indicator in the historical health data, each historical diagnosis information, and each historical treatment information are respectively used as a node, and the degree of association between any two nodes is used as the edge weight between the two nodes;

[0277] The second processing module 202 is used to obtain abnormal physiological state information of the target chronic disease patient and perform vectorization processing on the abnormal physiological state information to obtain a query vector; the abnormal physiological state information includes physiological indicators with abnormal risks determined based on the health data of the target chronic disease patient, and risk trend levels determined after risk trend analysis of the physiological indicators with abnormal risks;

[0278] The third processing module 203 is used to take each node of the target chronic disease patient in the knowledge graph as the target node, take the node whose edge weight with the target node is greater than the first preset value as the neighbor node of the target node, and calculate the relevance of the query vector and each target node respectively in combination with the edge weight between the neighbor node and the target node;

[0279] The fourth processing module 204 is used to sort all target nodes in descending order according to the magnitude of the relevance, and extract the first Z target nodes in the sorting to obtain the key medical information of the target chronic disease patient;

[0280] The fifth processing module 205 is used to input key medical information into the MedGPT model, and the MedGPT model generates a chronic disease health assessment plan for the target chronic disease patient based on the key medical information; the chronic disease health assessment plan includes the risk trend level of the target chronic disease patient and treatment plan adjustment suggestions.

[0281] Furthermore, when calculating the relevance between the query vector and each target node respectively in combination with the edge weights between the neighbor nodes and the target node, the third processing module 203 is configured to execute:

[0282] Determining a first weight of the target node based on a predetermined similarity between the query vector and each target node, and a similarity between the query vector and a neighboring node;

[0283] Calculate the average of the sum of the edge weights between the target node and each neighbor node to obtain the second weight of the target node;

[0284] The first weight and the second weight are weighted and summed to obtain a comprehensive weight of the target node, and the comprehensive weight is used as the relevance of the target node.

[0285] Further, when determining the first weight of the target node according to the predetermined similarity between the query vector and each target node, and the similarity between the query vector and the neighboring nodes, the third processing module 203 is specifically configured to execute:

[0286] The cosine similarity algorithm is used to calculate the similarity between the query vector and each target node, as well as the similarity between the query vector and the neighboring nodes. The specific cosine similarity algorithm is:

[0287] ;

[0288] in, represents the query vector, represents the vector of the target node, Represents the similarity between the query vector and the target node;

[0289] The first weight of the target node is calculated according to the softmax function, which is specifically:

[0290] ;

[0291] in, represents the first weight of the target node, represents the vector of neighbor nodes, represents the similarity between the query vector and the neighboring nodes, Represents neighbor nodes The node set belonging to the target chronic disease patient in the knowledge graph .

[0292] Furthermore, the chronic disease population health assessment device further includes a sixth processing module; before obtaining the abnormal physiological state information of the target chronic disease patient and performing vector processing on the abnormal physiological state information to obtain the query vector, the sixth processing module is configured to execute:

[0293] Based on the health data of the target chronic disease patient acquired in real time, determine the physiological indicators with abnormal risks in the health data; the health data includes physiological data corresponding to multiple physiological indicators used to feedback the health status of the target chronic disease patient;

[0294] Conduct risk trend analysis on physiological indicators with abnormal risks to determine the risk trend level of chronic diseases in target chronic disease patients.

[0295] Further, when determining physiological indicators with abnormal risks in the health data based on the health data of the target chronic disease patient obtained in real time, the sixth processing module is specifically configured to execute:

[0296] Obtain health data of target chronic disease patients in real time;

[0297] The physiological data corresponding to each physiological indicator of the target chronic disease patient is used as the original input data, and the original input data is input into the autoencoder trained with the historical health data of the healthy population, and the original input data is compressed and reconstructed by the autoencoder to obtain the corresponding reconstructed data;

[0298] The reconstruction error between the original input data and the reconstructed data is calculated, and according to the size of the reconstruction error and the first preset error threshold, the detection result of the physiological indicator corresponding to the original input data is marked to obtain the marked result, and the physiological indicators with abnormal risks in the marked result are screened out; the marked result is used to indicate whether the physiological indicator corresponding to the original input data has abnormal risks.

[0299] Further, when performing risk trend analysis on physiological indicators with abnormal risks to determine the risk trend level of chronic diseases of target chronic disease patients, the sixth processing module is specifically configured to execute:

[0300] When there are physiological indicators with abnormal risks in the labeling results, the health data of the target chronic disease patient and the labeling results corresponding to multiple physiological indicators of the target chronic disease patient's health data are used as target evaluation data. The target evaluation data are input into a support vector machine model trained with historical chronic disease data of the chronic disease population. The support vector machine model is used to perform risk trend analysis on the target evaluation data to obtain the risk trend level.

[0301] Furthermore, before performing risk trend analysis on the target assessment data through the support vector machine model, the sixth processing module is further configured to execute:

[0302] The original support vector machine model is constructed using historical chronic disease data as the first training sample and target evaluation data as the sample to be evaluated; the sample to be evaluated includes a first feature matrix consisting of physiological data corresponding to multiple physiological indicators of the target chronic disease patient and the labeling results of the physiological indicators, and each first training sample includes a second feature matrix consisting of physiological data corresponding to multiple physiological indicators of a chronic disease patient and the labeling results of the physiological indicators, as well as a risk label corresponding to the second feature matrix;

[0303] The decision function of the original support vector machine model is:

[0304] ;

[0305] in, For the return function, when When +1 is returned, it means that the chronic disease risk trend level of the sample to be evaluated is low risk. When -1 is returned, it means that the chronic disease risk trend level of the sample to be evaluated is high risk;

[0306] represents the number of the first training samples, represents the position of the current first training sample, represents the Lagrange multiplier parameter, is the bias parameter;

[0307] Indicates The second feature matrix of the first training sample, Indicates The risk label of the first training sample. When the risk label is +1, it means that the chronic disease risk is high risk, and when the risk label is -1, it means that the chronic disease risk is low risk.

[0308] represents the sample to be evaluated;

[0309] is a kernel function used to calculate the similarity between the first training sample and the sample to be evaluated.

[0310] Furthermore, each physiological index includes multiple index features; the kernel function of the decision function of the original support vector machine model is determined according to the radial basis function and the polynomial kernel function;

[0311] The kernel function is specifically configured as:

[0312] ;

[0313] in, represents the adjustment parameter of the radial basis function, which is a preset fixed value;

[0314] Indicates the degree of the polynomial in the polynomial kernel function, which is a preset fixed value;

[0315] Indicates the total number of physiological indicators included in health data, Indicates the current position of physiological indicators;

[0316] Indicates the total number of indicator characteristics included in a physiological indicator. Indicates the location of the current indicator feature;

[0317] Indicates The weight coefficient of each physiological index;

[0318] Indicates the sample to be evaluated. The first physiological index The value of an indicator feature;

[0319] Indicates The first training sample The first physiological index The numerical value of an indicator feature.

[0320] Furthermore, the autoencoder includes an encoding unit, an attention unit, and a decoding unit; when the original input data is compressed and reconstructed by the autoencoder to obtain corresponding reconstructed data, the sixth processing module is specifically configured to execute:

[0321] The encoding unit compresses the input high-dimensional original input data into low-dimensional first data through multiple fully connected layers;

[0322] The attention unit extracts important features related to the chronic diseases of the target chronic disease patients through the attention mechanism to obtain the second data;

[0323] The decoding unit reconstructs the second data into reconstructed data having the same dimension as the original input data through multiple fully connected layers.

[0324] The chronic disease population health assessment device of the present invention can not only utilize autoencoding and support vector machine technology, but also conduct accurate health risk assessment on chronic disease patients by real-time monitoring of patient health data, so as to timely discover potential risks and facilitate doctors to adjust treatment plans in time according to the patient's actual health status.

[0325] The chronic disease population health assessment device of the present invention can also be used as a patient terminal of the chronic disease management system; the health data, abnormal risk indicators and risk trend levels of each chronic disease patient can be uploaded to the database of the chronic disease management system through the network. Doctors can log in to the chronic disease management system through the doctor terminal, and use the patient number or patient ID number as an index to query the relevant data of the target patient, so that the doctor can accurately understand the patient's physical condition and provide accurate guidance and suggestions to the patient. The chronic disease management system can also centrally manage the data of multiple chronic disease patients, improve the efficiency of chronic disease management, reduce medical costs, and improve the quality of life of patients.

[0326] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0327] In a third aspect, an embodiment of the present invention further provides an electronic device, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, at least one program, a code set or an instruction set is loaded and executed by the processor to implement the health assessment method for a chronic disease population in an embodiment of the present invention.

[0328] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, which stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, at least one program, a code set or an instruction set is loaded and executed by a processor to implement the health assessment method for chronic disease populations in an embodiment of the present invention.

[0329] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the process or function according to the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be accessed by the computer or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0330] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for health assessment of chronic disease population, characterized in that: include: Obtain a predetermined knowledge graph for storing and recording historical patient information of a chronic disease population; In the knowledge graph, the basic information of each chronic disease patient in the historical patient information, each physiological indicator in the historical health data, each historical diagnosis information, and each historical treatment information are respectively regarded as a node, and the degree of association between any two nodes is regarded as the edge weight between the two nodes; Acquire abnormal physiological state information of a target chronic disease patient, and vectorize the abnormal physiological state information to obtain a query vector; the abnormal physiological state information includes physiological indicators with abnormal risks determined based on the health data of the target chronic disease patient, and risk trend levels determined after risk trend analysis of the physiological indicators with abnormal risks; Each node of the target chronic disease patient in the knowledge graph is taken as a target node, and a node whose edge weight with the target node is greater than a first preset value is taken as a neighbor node of the target node, and the correlation between the query vector and each of the target nodes is calculated in combination with the edge weight between the neighbor node and the target node; All target nodes are sorted in descending order according to the magnitude of the relevance, and the first Z target nodes in the sorting are extracted to obtain the key medical information of the target chronic disease patient; the key medical information is a text describing the health status and related medical information of the target chronic disease patient; The key medical information is input into the MedGPT model, and the MedGPT model generates a chronic disease health assessment plan for the target chronic disease patient according to the key medical information; the chronic disease health assessment plan includes the risk trend level of the target chronic disease patient and treatment plan adjustment suggestions; Before obtaining the abnormal physiological state information of the target chronic disease patient and performing vectorization processing on the abnormal physiological state information to obtain a query vector, the method further includes: Based on the health data of the target chronic disease patient acquired in real time, determining the physiological indicators with abnormal risks in the health data; the health data includes physiological data corresponding to a plurality of physiological indicators for feedback on the health status of the target chronic disease patient; Performing risk trend analysis on the physiological indicators with abnormal risks to determine the risk trend level of the chronic disease of the target chronic disease patient; Based on the health data of the target chronic disease patient acquired in real time, determining physiological indicators with abnormal risks in the health data, specifically including: Obtain health data of target chronic disease patients in real time; Taking the physiological data corresponding to each physiological indicator of the target chronic disease patient as the original input data, inputting the original input data into the autoencoder trained with the historical health data of the healthy population, and compressing and reconstructing the original input data through the autoencoder to obtain the corresponding reconstructed data; Calculating a reconstruction error between the original input data and the reconstructed data, and marking the detection result of the physiological indicator corresponding to the original input data according to the size of the reconstruction error and a first preset error threshold, obtaining a marking result, and screening out the physiological indicator with abnormal risk in the marking result; the marking result is used to indicate whether the physiological indicator corresponding to the original input data has abnormal risk; Perform risk trend analysis on the physiological indicators with abnormal risks to determine the risk trend level of the chronic disease of the target chronic disease patient, specifically including: When there are physiological indicators of abnormal risk in the marked results, the health data of the target chronic disease patient and the marked results corresponding to multiple physiological indicators of the health data of the target chronic disease patient are used as target evaluation data, and the target evaluation data is input into a support vector machine model trained with historical chronic disease data of a chronic disease population, and the support vector machine model is used to perform risk trend analysis on the target evaluation data to obtain a risk trend level; Before performing risk trend analysis on the target assessment data through the support vector machine model, the method further includes: The original support vector machine model is constructed using historical chronic disease data as the first training sample and the target evaluation data as the sample to be evaluated; the sample to be evaluated includes a first feature matrix composed of physiological data corresponding to multiple physiological indicators of the target chronic disease patient and the labeling results of the physiological indicators, and each of the first training samples includes a second feature matrix composed of physiological data corresponding to multiple physiological indicators of a chronic disease patient and the labeling results of the physiological indicators, and a risk label corresponding to the second feature matrix; The decision function of the original support vector machine model is: ； in, For the return function, when When +1 is returned, it means that the chronic disease risk trend level of the sample to be evaluated is low risk. When -1 is returned, it means that the chronic disease risk trend level of the sample to be evaluated is high risk; represents the number of the first training samples, represents the position of the current first training sample, represents the Lagrange multiplier parameter, is the bias parameter; Indicates The second feature matrix of the first training sample, Indicates a risk label of the first training sample, when the risk label is +1, it indicates that the chronic disease risk is high risk, and when the risk label is -1, it indicates that the chronic disease risk is low risk; represents the sample to be evaluated; is a kernel function used to calculate the similarity between the first training sample and the sample to be evaluated; The kernel function of the decision function of the original support vector machine model is determined according to the radial basis function and the polynomial kernel function.

2. The method for health assessment of chronic disease population according to claim 1, characterized in that: In combination with the edge weights between the neighbor nodes and the target node, respectively calculating the relevance between the query vector and each of the target nodes, specifically including: Determining a first weight of the target node according to a predetermined similarity between the query vector and each of the target nodes, and a similarity between the query vector and the neighboring nodes; Calculate the average value of the sum of edge weights between the target node and each of the neighboring nodes to obtain a second weight of the target node; The first weight and the second weight are weighted and summed to obtain a comprehensive weight of the target node, and the comprehensive weight is used as the relevance of the target node.

3. The method for health assessment of chronic disease population according to claim 2, characterized in that: Determining a first weight of the target node according to a predetermined similarity between the query vector and each of the target nodes, and a similarity between the query vector and the neighboring nodes, specifically includes: The cosine similarity algorithm is used to calculate the similarity between the query vector and each of the target nodes, and the similarity between the query vector and the neighboring nodes. The cosine similarity algorithm is specifically as follows: ； in, represents the query vector, represents the vector of the target node, Represents the similarity between the query vector and the target node; The first weight of the target node is calculated according to the softmax function, and the softmax function is specifically: ； in, represents the first weight of the target node, represents the vector of neighbor nodes, represents the similarity between the query vector and the neighboring nodes, Represents neighbor nodes The node set belonging to the target chronic disease patient in the knowledge graph .

4. The method for health assessment of chronic disease population according to claim 1, characterized in that: Each of the physiological indicators includes a plurality of indicator features; The kernel function is specifically: ； in, represents the adjustment parameter of the radial basis function, which is a preset fixed value; Indicates the degree of the polynomial in the polynomial kernel function, which is a preset fixed value; Indicates the total number of physiological indicators included in health data, Indicates the current position of physiological indicators; Indicates the total number of indicator characteristics included in a physiological indicator. Indicates the location of the current indicator feature; Indicates The weight coefficient of each physiological index; Indicates the sample to be evaluated. The first physiological index The value of an indicator feature; Indicates The first training sample The first physiological index The numerical value of an indicator feature.

5. The method for health assessment of chronic disease population according to claim 1, characterized in that: The autoencoder includes an encoding unit, an attention unit, and a decoding unit; the original input data is compressed and reconstructed by the autoencoder to obtain corresponding reconstructed data, specifically including: The encoding unit compresses the input high-dimensional original input data into low-dimensional first data through multiple fully connected layers; The attention unit extracts important features related to the chronic disease of the target chronic disease patient through an attention mechanism to obtain second data; The decoding unit reconstructs the second data into reconstructed data with the same dimension as the original input data through multiple fully connected layers.

6. A health assessment device for chronic disease population, characterized in that: include: The first processing module is used to obtain a predetermined knowledge graph for storing and recording historical patient information of a chronic disease population; In the knowledge graph, the basic information of each chronic disease patient in the historical patient information, each physiological indicator in the historical health data, each historical diagnosis information, and each historical treatment information are respectively regarded as a node, and the degree of association between any two nodes is regarded as the edge weight between the two nodes; The second processing module is used to obtain abnormal physiological state information of the target chronic disease patient, and vectorize the abnormal physiological state information to obtain a query vector; the abnormal physiological state information includes physiological indicators with abnormal risks determined based on the health data of the target chronic disease patient, and risk trend levels determined after risk trend analysis of the physiological indicators with abnormal risks; A third processing module is used to respectively use each node of the target chronic disease patient in the knowledge graph as a target node, use a node whose edge weight with the target node is greater than a first preset value as a neighbor node of the target node, and calculate the relevance of the query vector to each of the target nodes in combination with the edge weight between the neighbor node and the target node; The fourth processing module is used to sort all target nodes in descending order according to the magnitude of the correlation, and extract the first Z target nodes in the sorting to obtain the key medical information of the target chronic disease patient; The key medical information is a text describing the health status and related medical information of the target chronic disease patient; A fifth processing module, for inputting the key medical information into the MedGPT model, wherein the MedGPT model generates a chronic disease health assessment plan for the target chronic disease patient according to the key medical information; the chronic disease health assessment plan includes the risk trend level of the target chronic disease patient and a treatment plan adjustment suggestion; The chronic disease population health assessment device further includes a sixth processing module; Before obtaining the abnormal physiological state information of the target chronic disease patient and performing vectorization processing on the abnormal physiological state information to obtain a query vector, the sixth processing module is configured to execute: Based on the health data of the target chronic disease patient acquired in real time, determining the physiological indicators with abnormal risks in the health data; the health data includes physiological data corresponding to a plurality of physiological indicators for feedback on the health status of the target chronic disease patient; Performing risk trend analysis on the physiological indicators with abnormal risks to determine the risk trend level of the chronic disease of the target chronic disease patient; When determining that there are physiological indicators with abnormal risks in the health data based on the health data of the target chronic disease patient acquired in real time, the sixth processing module is specifically configured to execute: Obtain health data of target chronic disease patients in real time; Taking the physiological data corresponding to each physiological indicator of the target chronic disease patient as the original input data, inputting the original input data into the autoencoder trained with the historical health data of the healthy population, and compressing and reconstructing the original input data through the autoencoder to obtain the corresponding reconstructed data; Calculating a reconstruction error between the original input data and the reconstructed data, and marking the detection result of the physiological index corresponding to the original input data according to the magnitude of the reconstruction error and a first preset error threshold, to obtain a marking result, and screening out the physiological index with abnormal risk in the marking result; The marking result is used to indicate whether the physiological indicator corresponding to the original input data has an abnormal risk; When performing risk trend analysis on the physiological indicators with abnormal risk to determine the risk trend level of the chronic disease of the target chronic disease patient, the sixth processing module is specifically configured to execute: When there are physiological indicators of abnormal risk in the marked results, the health data of the target chronic disease patient and the marked results corresponding to multiple physiological indicators of the health data of the target chronic disease patient are used as target evaluation data, and the target evaluation data is input into a support vector machine model trained with historical chronic disease data of a chronic disease population, and the support vector machine model is used to perform risk trend analysis on the target evaluation data to obtain a risk trend level; Before performing risk trend analysis on the target assessment data through the support vector machine model, the sixth processing module is further configured to execute: The original support vector machine model is constructed using historical chronic disease data as the first training sample and the target evaluation data as the sample to be evaluated; the sample to be evaluated includes a first feature matrix composed of physiological data corresponding to multiple physiological indicators of the target chronic disease patient and the labeling results of the physiological indicators, and each of the first training samples includes a second feature matrix composed of physiological data corresponding to multiple physiological indicators of a chronic disease patient and the labeling results of the physiological indicators, and a risk label corresponding to the second feature matrix; The decision function of the original support vector machine model is: ； in, For the return function, when When +1 is returned, it means that the chronic disease risk trend level of the sample to be evaluated is low risk. When -1 is returned, it means that the chronic disease risk trend level of the sample to be evaluated is high risk; represents the number of the first training samples, represents the position of the current first training sample, represents the Lagrange multiplier parameter, is the bias parameter; Indicates The second feature matrix of the first training sample, Indicates a risk label of the first training sample, when the risk label is +1, it indicates that the chronic disease risk is high risk, and when the risk label is -1, it indicates that the chronic disease risk is low risk; represents the sample to be evaluated; is a kernel function used to calculate the similarity between the first training sample and the sample to be evaluated; The kernel function of the decision function of the original support vector machine model is determined according to the radial basis function and the polynomial kernel function.

Images