Method and system for carrying out dynamic auxiliary grading on comprehensive chronic disease patients

Abstract

The invention provides a dynamic auxiliary grading method and system for comprehensive chronic disease patients, and the method comprises the steps: (1) obtaining the symptom information of a to-be-examined patient, inputting the symptom information into a grade classifier, and obtaining the comprehensive chronic disease grade of the to-be-examined patient; and (2) the core of the method is that an initial label is obtained by using a semi-nonnegative matrix factorization clustering algorithm, and the initial label is further corrected by using an active learning strategy, so that a grade classifier capable of accurately dividing the chronic disease grade of the patient is trained and obtained. The semi-nonnegative matrix factorization clustering algorithm in the invention can generate an interpretable solution and has a feature selection capability. And obtaining a comprehensive chronic disease grade classification result of the patient to be examined by using a grade classifier, and forming a management system for carrying out nonlinear classification on the comprehensive chronic disease grade.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for dynamically assisting in grading comprehensive chronic disease patients. Background Art

[0002] The prevalence of chronic diseases is on the rise. Chronic diseases are diseases with a long course and usually slow development. They are characterized by a large number of patients, high medical costs, long illness duration and great demand for services, especially comprehensive chronic diseases, which have complex causes and low cure efficiency.

[0003] Moreover, as the number of patients with chronic diseases and the types of chronic diseases are increasing, the specific classification and management of comprehensive chronic diseases still lack reasonable nonlinear classification indicators and a comprehensive chronic disease classification and grading management system. Summary of the invention

[0004] The present invention provides a method and system for dynamic auxiliary grading of comprehensive chronic disease patients, which is used to solve the defects of the prior art in lacking reasonable nonlinear classification indicators and comprehensive chronic disease classification management system. The initial labels are obtained by using a semi-nonnegative matrix decomposition clustering algorithm, and the initial labels are further corrected using an active learning strategy, so as to train and obtain a grade classifier that can accurately classify the chronic disease grades of patients. The grade classifier is used to classify the chronic disease grades of the case data of the patient to be examined, and assist the comprehensive chronic disease treatment and rehabilitation of the patient.

[0005] The present invention provides a method for dynamically assisting in grading comprehensive chronic disease patients, comprising:

[0006] Step 210: Establish a chronic disease management database;

[0007] Step 220: extract data and perform clustering processing;

[0008] Step 230: Based on the clustering results, a grading index system is set;

[0009] Step 240: Based on the initial labels of the samples obtained by clustering, a basic classifier is trained;

[0010] Step 250: Correct the labels of high-uncertainty samples through active learning strategies, and train a hierarchical classifier;

[0011] Step 310: When a new sample is input, the label of the high-uncertainty sample is corrected through the active learning strategy, and the training is updated to obtain a hierarchical classifier;

[0012] Step 320: When the cumulative number of new samples reaches a set number, clustering is performed again and the grading index system is updated based on the clustering results.

[0013] According to a method for dynamic auxiliary grading of comprehensive chronic disease patients provided by the present invention, step 220 is specifically: extracting the patient's disease information from the chronic disease management database to form an initial unlabeled sample; and clustering the initial unlabeled sample using a semi-non-negative matrix decomposition clustering algorithm.

[0014] According to a method for dynamically assisting in grading comprehensive chronic disease patients provided by the present invention, step 220 specifically includes:

[0015] Converting the attribute space of the initial unlabeled sample into a vector space, and constructing a feature matrix based on the vector space;

[0016] Determining an objective function based on the feature matrix and the established basis matrix, the first latent feature matrix, the second latent feature matrix, and the L1 norm of the column vector of the second latent feature matrix, wherein the second latent feature matrix includes importance weight coefficients of basis vectors;

[0017] Obtaining a dimensionality reduction feature based on the objective function, and determining each cluster sample based on the dimensionality reduction feature;

[0018] A clustering operation is performed on the cluster samples.

[0019] According to a method for dynamic auxiliary grading of comprehensive chronic disease patients provided by the present invention, step 230 is specifically: statistics are performed based on the clustering results of the semi-non-negative matrix decomposition clustering algorithm, and the statistical information includes the mean, variance and P value hypothesis test of each category, and a reasonable grading indicator system is set based on the statistical results.

[0020] According to a method for dynamic auxiliary grading of patients with comprehensive chronic diseases provided by the present invention, step 240 is specifically: generating initial labeled samples based on the clustering results of the semi-non-negative matrix decomposition clustering algorithm; using the initial labeled samples to train the model to obtain a basic classifier; the model includes a multilayer perceptron or a boosting model.

[0021] According to a method for dynamic auxiliary grading of patients with comprehensive chronic diseases provided by the present invention, step 250 is specifically: based on an active learning strategy, samples with high uncertainty are screened out from the prediction results of the basic classifier, and their labels are corrected, and the basic classifier is continued to be trained using the corrected labels to obtain the grade classifier.

[0022] According to a method for dynamic auxiliary grading of comprehensive chronic disease patients provided by the present invention, the step 320 is specifically as follows: when the cumulative number of new samples reaches a set number, the new samples and the existing samples are merged into a new data set, the data set is clustered using the semi-non-negative matrix decomposition clustering algorithm, and a reasonable grading index system is reset based on the statistical information of the clustering results, and the statistical information includes the mean, variance and P value hypothesis test of each category.

[0023] According to a method for dynamic auxiliary grading of comprehensive chronic disease patients provided by the present invention, the high uncertainty samples are specifically:

[0024] The clustering results of the semi-non-negative matrix decomposition clustering algorithm are used to form a pseudo-label probability distribution and a normalization factor of the sample;

[0025] Forming a predicted probability distribution of samples based on the prediction results of the model;

[0026] The cross entropy is calculated based on the pseudo-label probability distribution, the normalization factor and the predicted probability distribution, and samples with large cross entropy values ​​are selected as the high uncertainty samples.

[0027] The present invention also provides a system for dynamically assisting in grading comprehensive chronic disease patients, including: a patient information filling and extraction module, a comprehensive chronic disease treatment plan entry module, a critical value definition module, a message reminder module, a health monitoring module, a health record module, a health education module and a patient information query module;

[0028] The patient information filling and extraction module is used to collect sample information into the chronic disease management database;

[0029] The comprehensive chronic disease treatment plan input module is used to obtain a comprehensive chronic disease instruction plan based on the comprehensive chronic disease grade classification result for input;

[0030] The critical value definition module is used for medical experts to grade the rapid interpretability of chronic diseases;

[0031] The message reminder module is used to automatically send a message reminder to the user when the user's comprehensive chronic disease index is abnormal; the user confirms or rechecks the index to obtain accurate disease information;

[0032] The health monitoring module is used to dynamically monitor the health data of the user to obtain real-time disease information;

[0033] The health record module is used to record the user's daily routine data and the user's real-time disease information;

[0034] The health education module pushes personalized health education knowledge to the comprehensive chronic disease patients of different levels based on the health education knowledge graph;

[0035] The patient information query module is used to perform information query based on the user's input conditions to obtain a comprehensive chronic disease patient cohort, personal information, risk level, and treatment plan.

[0036] The method and system for dynamic auxiliary grading of comprehensive chronic disease patients provided by the present invention include: (1) obtaining the symptom information of the patient to be examined, inputting the symptom information into a grade classifier, and obtaining the comprehensive chronic disease grade of the patient to be examined. (2) The core of the present invention is to use a semi-non-negative matrix decomposition clustering algorithm to obtain the initial label, and use an active learning strategy to further correct the initial label, so as to train and obtain a grade classifier that can accurately classify the patient's chronic disease grade. The semi-non-negative matrix decomposition clustering algorithm in the present invention can produce an interpretable solution and has feature selection capabilities. Using the grade classifier, the comprehensive chronic disease grade classification result of the patient to be examined is obtained, forming a management system for nonlinear classification of comprehensive chronic disease grades. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 This is one of the flow charts of the method for dynamic auxiliary grading of comprehensive chronic disease patients provided by the present invention;

[0039] Figure 2 This is the second flow chart of the method for dynamic auxiliary grading of comprehensive chronic disease patients provided by the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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.

[0041] Reference Figure 1 The method and system for dynamically assisting in grading comprehensive chronic disease patients provided by the present invention include:

[0042] Step 110, obtaining the disease information of the patient to be examined;

[0043] Step 120: input the disease information into a grade classifier to obtain the comprehensive chronic disease grade of the patient to be examined.

[0044] This embodiment is a specific chronic disease grade classification method, that is, the application process of the grade classifier.

[0045] First, the medical record data of the patient to be examined is obtained. It should be noted that the case data is the personal specific data of the patient to be examined, which can be directly obtained based on the electronic medical records, test results, etc. of the patient to be examined.

[0046] The acquired medical record data of the patient to be examined is then used to perform chronic disease grade analysis through the trained grade classifier to obtain the chronic disease grade classification result of the patient to be examined.

[0047] The present invention also provides a system for dynamically assisting in grading comprehensive chronic disease patients, including: a patient information filling and extraction module, a comprehensive chronic disease treatment plan entry module, a critical value definition module, a message reminder module, a health monitoring module, a health record module, a health education module and a patient information query module;

[0048] The patient information filling and extraction module is used to collect sample information into the chronic disease management database;

[0049] The comprehensive chronic disease treatment plan input module is used to obtain a chronic disease instruction plan based on the chronic disease grade classification result for input;

[0050] The critical value definition module is used for medical experts to grade the rapid interpretability of chronic diseases;

[0051] The message reminder module is used to automatically send a message reminder to the user when the chronic disease index of the user is abnormal;

[0052] The health monitoring module is used to monitor the health data of the user;

[0053] The health record module is used to record the daily life and routine data of the user;

[0054] The health education module is used to construct a knowledge graph, conduct health education for comprehensive chronic disease patients corresponding to each chronic disease grade classification result based on the knowledge graph, and send suggestions to the comprehensive chronic disease patients;

[0055] The patient information query module is used to perform information query based on the user's input conditions to obtain a comprehensive chronic disease patient cohort, personal information, risk level, and treatment plan.

[0056] It should be noted that the system for dynamically assisting the classification of comprehensive chronic disease patients provided in this embodiment includes a data processing device and the above-mentioned device for dynamically assisting the classification of comprehensive chronic disease patients, and the data processing device includes a user terminal, a doctor PC terminal, an operation management platform, and a chronic disease database;

[0057] The user end includes a patient information filling and extraction module, a health monitoring module, a health record module, and a health education module.

[0058] The doctor's PC terminal includes a patient information query module, a message reminder module, and a chronic disease instruction plan entry and adjustment module;

[0059] The operation and management platform includes a comprehensive chronic disease treatment plan entry module, a critical value definition module, etc.

[0060] Comprehensive chronic disease patients use terminal applications to interact with the chronic disease database. Patients registering for the first time can complete the registration by scanning the code through the terminal application and filling in their personal information, and realize data transmission to the chronic disease database through the patient information filling module.

[0061] Patients can manually enter or automatically obtain numerical values ​​monitored by portable devices, such as blood pressure and blood sugar, through the health monitoring module in the terminal application. At the same time, these indicators of the patient can be recorded to form a trend chart for patients and doctors to understand changes in the patient's condition and make adjustments to the treatment plan.

[0062] Patients can use the health record module of the terminal application to record their daily routines for chronic diseases, such as smoking, drinking, work and rest, exercise, and diet.

[0063] The patient 360 health view module in the terminal application can comprehensively display the patient's inpatient treatment, outpatient treatment, physical examination, single indicators and other information.

[0064] The health education module conducts health education for patients with comprehensive chronic diseases of different levels by constructing a knowledge graph, and pushes appropriate suggestions on living habits and behavioral norms to patients in real time. It also automatically pushes suitable meals to patients based on their diagnosis and treatment conditions.

[0065] On the doctor's PC, use the patient information query module to query and compile statistics on chronic disease patient cohorts according to conditions (patient name, chronic disease type, etc.), query personal information, risk level, treatment plan, etc.

[0066] Doctors can also use the chronic disease treatment plan adjustment module on their PC to establish personalized treatment plans based on the treatment plans maintained by the system and according to the individual differences of each patient.

[0067] The message reminder module on the doctor's PC can obtain patient information in real time through the patient information filling and extraction module, and then call the data in the critical value definition module. When the patient's indicators are abnormal or reach critical values, the system automatically sends a reminder.

[0068] The message reminder module on the doctor's PC is associated with the HIS doctor station interface. When patients visit the doctor, they will be given information such as the type of chronic disease and the risk level of chronic disease. If the patient is not bound, the patient can be informed to bind, and the patient who has visited the doctor can also be added to the comprehensive chronic disease patient queue.

[0069] The chronic disease classification and grading module and information extraction module of the operation management platform apply efficient clustering algorithms and active learning algorithms to the field of comprehensive chronic disease classification, and use the chronic disease questionnaire in the chronic disease database to complete the comprehensive chronic disease classification of patients.

[0070] The comprehensive chronic disease treatment plan entry module of the operation management platform is provided by experts with reasonable grading standards corresponding to the treatment plans for different comprehensive chronic diseases, while the adjustment module is used by doctors to enter chronic disease treatment plans for specific patients; in addition, after each update of the classification and grading standards, the system can prompt the doctor whether to adjust the treatment plan and enter it into the chronic disease management system.

[0071] The critical value definition module of the operation management platform uses statistical indicators and grading systems provided by clustering and active learning algorithms to allow doctor experts to define abnormal value indicators and value ranges such as examination results, test indicators, and vital signs.

[0072] Reference Figure 2 The method and system for dynamically assisting in grading comprehensive chronic disease patients provided by the present invention include:

[0073] Step 210: Establish a chronic disease management database;

[0074] Step 220: extract data and perform clustering processing;

[0075] Step 230: Based on the clustering results, a grading index system is set;

[0076] Step 240: Based on the initial labels of the samples obtained by clustering, a basic classifier is trained;

[0077] Step 250: Correct the labels of high-uncertainty samples through active learning strategies, and train a hierarchical classifier;

[0078] Step 260: Determine whether there is a new sample input;

[0079] If yes, execute step 270, if no, end the process;

[0080] Step 270, determining whether the cumulative number of new samples reaches a set number;

[0081] If yes, execute step 320, if no, execute step 310;

[0082] Step 310: When a new sample is input, the label of the high-uncertainty sample is corrected through the active learning strategy, and the training is updated to obtain a hierarchical classifier;

[0083] Step 320: When the cumulative number of new samples reaches a set number, clustering is performed again and the grading index system is updated based on the clustering results.

[0084] Step 280: Determine whether there is a new sample input;

[0085] If yes, then return to step 270, if no, end the process.

[0086] Among them, step 210 is to establish a chronic disease management database. The chronic disease management database is obtained by regularly extracting data such as HIS, electronic medical records, tests and examinations. Then, all sample comprehensive chronic disease patients' symptom information is extracted from the chronic disease management database, and the symptom information is used as the initial sample set, which can be recorded as

[0087] Among them, step 220 includes: extracting the symptom information of comprehensive chronic disease patients from the chronic disease management database, and clustering the initial sample set based on the semi-non-negative matrix decomposition clustering algorithm. The semi-non-negative matrix decomposition clustering algorithm is faster than the K-Means clustering algorithm in clustering speed, especially when the number of clusters is large, and the semi-non-negative matrix decomposition clustering algorithm does not cause necessary feature loss.

[0088] This embodiment is a specific process of clustering:

[0089] First, according to the extracted disease information, the attribute space of the initial sample set is transformed into a vector space to construct a feature matrix, which can be recorded as X ± , where (X ± ∈R M×M ). Each column represents a sample.

[0090] Then construct the basis matrix, the first latent feature matrix, and the second latent feature matrix, respectively denoted as A ± , W + , Y + Among them, (A ± ∈R M×K ) is the basis matrix, (W + ∈R K×K ) and (Y + ∈R K×N ) are all latent feature matrices, and determine Y + The column vector of the matrix, denoted by y i .

[0091] In order to achieve dimensionality reduction clustering, the objective function is established as follows:

[0092]

[0093] wxya + ≥0,Y + ≥0,0 <y i ≤1

[0094] It should be noted that W + The diagonal elements contain the importance weight coefficients of the basis vectors, and the normalization factors for different clustering categories are

[0095] Furthermore, the objective function output matrix A is used ± , W + , Y + The minimum value of , the number of iterative optimization can be recorded as T. It should be noted that in this embodiment, the Laplace prior is introduced into the semi-nonnegative matrix decomposition clustering algorithm, and the ||y i || 1 It can produce sparse and interpretable solutions and has the ability of feature selection, so some features will not be used.

[0096] According to the output matrix A ± , W + , Y + The minimum value of determines the dimension reduction feature, and determines each cluster sample according to the dimension reduction feature, that is, determines the sample clusters that may be graded.

[0097] Then, step 230 is performed: according to the medical indicators, the classification level of each cluster sample is formed to form a reasonable comprehensive classification indicator system for chronic disease patients.

[0098] Step 230 specifically includes: performing statistics based on the clustering results of the semi-non-negative matrix decomposition clustering algorithm, the statistical information including the mean, variance and P value hypothesis test of each category, and setting a reasonable grading indicator system based on the statistical results.

[0099] Afterwards, step 240 is performed: clustering is performed on the initial sample set based on a semi-non-negative matrix decomposition clustering algorithm to obtain sample clusters of possible levels, thereby determining initial labels of the initial sample set.

[0100] After obtaining the initial label of the sample, it can be recorded as y, combined with the initial sample set Forming a supervised training set Finally, the supervised training set is used to The established initial classifier (eg, initial multi-layer perceptron) is supervisedly trained until the model converges to obtain a basic classifier f. The model includes a multi-layer perceptron or a boosting model.

[0101] Then, step 250 is performed: based on the active learning strategy, samples with high uncertainty are screened out from the prediction results of the basic classifier, and their labels are corrected, and the basic classifier is continued to be trained using the corrected labels to obtain the hierarchical classifier.

[0102] Step 250 uses an active learning strategy to correct the labels of high-uncertainty samples and train a hierarchical classifier.

[0103] Based on the clustering results of the semi-non-negative matrix decomposition clustering algorithm, a pseudo-label probability distribution and a normalization factor of the samples are formed; based on the model prediction results, a predicted probability distribution of the samples is formed; based on the pseudo-label probability distribution, the normalization factor and the predicted probability distribution, the cross entropy is calculated, and samples with large cross entropy values ​​are selected as the high uncertainty samples.

[0104] It should be noted that the active learning strategy in this embodiment is to sample samples with large cross entropy values ​​as high uncertainty samples. The larger the cross entropy, the more information the sample contains.

[0105] The specific formula is as follows:

[0106]

[0107] Among them, S i is the normalization factor of different clustering categories, S i The larger the value, the more information the i-th sample contains. p(x) is the correct probability distribution, q(x) is the newly added sample x * Predicted probability distribution via a multilayer perceptron.

[0108] According to the reasonable interval of hypothesis test p-value≤0.05, based on y i The selected statistical indicators determine new and more reasonable classification indicator estimates.

[0109] When new samples are input and the cumulative number of new samples does not reach the set number, step 310 is executed: the labels of high-uncertainty samples are corrected through active learning strategies, and the level classifier is updated and trained.

[0110] First, obtain new samples. When new medical record data is added to the chronic disease management database, the new medical record data can be used as a new sample set, recorded as And get the new sample set The corresponding classification label can be recorded as y *.

[0111] Furthermore, the initial sample set and the newly added sample set are combined And combine the initial label and the graded label y * + label y, train the basic classifier f to get the final level classifier f * .

[0112] Then, samples with high uncertainty among the newly added samples are selected and their labels are corrected. This process is similar to step 250. The corrected high uncertainty samples are used to classify the classifier f * Continue training to update its parameters and make predictions more accurate.

[0113] Accordingly, the high uncertainty sample is specifically:

[0114] The clustering results of the semi-non-negative matrix decomposition clustering algorithm are used to form a pseudo-label probability distribution and a normalization factor of the sample;

[0115] Forming a predicted probability distribution of samples based on the prediction results of the model;

[0116] The cross entropy is calculated based on the pseudo-label probability distribution, the normalization factor and the predicted probability distribution, and samples with large cross entropy values ​​are selected as the high uncertainty samples.

[0117] When the cumulative number of new samples reaches the set number, step 320 is executed: clustering is performed again and the grading index system is updated according to the clustering results.

[0118] Step 320 is specifically as follows: when the cumulative number of new samples reaches a set number, the new samples and existing samples are merged into a new data set, the data set is clustered using the semi-non-negative matrix decomposition clustering algorithm, and a reasonable grading index system is reset based on the statistical information of the clustering results, wherein the statistical information includes the mean, variance and P value hypothesis test of each category.

[0119] Similarly, combine the initial sample set and the newly added sample set The combined sample set is clustered based on the semi-non-negative matrix decomposition clustering algorithm. Based on the clustering results and medical indicators, the classification level of each cluster sample is formed to form a reasonable comprehensive grading indicator system for chronic disease patients.

[0120] The embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0121] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0122] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic auxiliary grading of patients with comprehensive chronic diseases, characterized in that: include: Step 210: Establish a chronic disease management database; Step 220: extract data and perform clustering processing; Step 230: Based on the clustering results, a grading index system is set; Step 240: Based on the initial labels of the samples obtained by clustering, a basic classifier is trained; Step 250: Correct the labels of high-uncertainty samples through active learning strategies, and train a hierarchical classifier; Step 310: When a new sample is input, the label of the high-uncertainty sample is corrected through the active learning strategy, and the training is updated to obtain a hierarchical classifier; Step 320: When the cumulative number of new samples reaches a set number, clustering is performed again and the grading index system is updated based on the clustering results.

2. The method for dynamic auxiliary grading of comprehensive chronic disease patients according to claim 1, characterized in that: The step 220 specifically includes: extracting the patient's disease information from the chronic disease management database to form an initial unlabeled sample; and clustering the initial unlabeled sample using a semi-non-negative matrix decomposition clustering algorithm.

3. The method for dynamic auxiliary grading of comprehensive chronic disease patients according to claim 2, characterized in that: The step 220 specifically includes: Converting the attribute space of the initial unlabeled sample into a vector space, and constructing a feature matrix based on the vector space; Determining an objective function based on the feature matrix and the established basis matrix, the first latent feature matrix, the second latent feature matrix, and the L1 norm of the column vector of the second latent feature matrix, wherein the second latent feature matrix includes importance weight coefficients of basis vectors; Obtaining a dimensionality reduction feature based on the objective function, and determining each cluster sample based on the dimensionality reduction feature; A clustering operation is performed on the cluster samples.

4. The method for dynamic auxiliary grading of comprehensive chronic disease patients according to claim 1, characterized in that: The step 230 is specifically as follows: performing statistics based on the clustering results of the semi-non-negative matrix decomposition clustering algorithm, the statistical information including the mean, variance and P value hypothesis test of each category, and setting a reasonable grading indicator system based on the statistical results.

5. The method for dynamic auxiliary grading of comprehensive chronic disease patients according to claim 1, characterized in that: The step 240 is specifically as follows: generating initial labeled samples based on the clustering results of the semi-non-negative matrix decomposition clustering algorithm; using the initial labeled samples to train a model to obtain a basic classifier; the model includes a multi-layer perceptron or a boosting model.

6. The method for dynamic auxiliary grading of comprehensive chronic disease patients according to claim 1, characterized in that: The step 250 is specifically as follows: based on the active learning strategy, samples with high uncertainty are screened out from the prediction results of the basic classifier, and their labels are corrected, and the basic classifier is continued to be trained using the corrected labels to obtain the hierarchical classifier.

7. The method for dynamic auxiliary grading of comprehensive chronic disease patients according to claim 1, characterized in that: The step 320 is specifically as follows: when the cumulative number of new samples reaches a set number, the new samples and existing samples are merged into a new data set, the data set is clustered using the semi-non-negative matrix decomposition clustering algorithm, and a reasonable grading indicator system is reset based on the statistical information of the clustering results, wherein the statistical information includes the mean, variance and P value hypothesis test of each category.

8. The method for dynamic auxiliary grading of comprehensive chronic disease patients according to claim 1, characterized in that: The high uncertainty samples are specifically: The clustering results of the semi-non-negative matrix decomposition clustering algorithm are used to form a pseudo-label probability distribution and a normalization factor of the sample; Forming a predicted probability distribution of samples based on the prediction results of the model; The cross entropy is calculated based on the pseudo-label probability distribution, the normalization factor and the predicted probability distribution, and samples with large cross entropy values ​​are selected as the high uncertainty samples.

9. A system for dynamic auxiliary grading of patients with comprehensive chronic diseases, characterized in that: include: Patient information filling and extraction module, comprehensive chronic disease treatment plan entry module, critical value definition module, message reminder module, health monitoring module, health record module, health education module and patient information query module; The patient information filling and extraction module is used to collect sample information into the chronic disease management database; The comprehensive chronic disease treatment plan input module is used to obtain a comprehensive chronic disease instruction plan based on the comprehensive chronic disease grade classification result for input; The critical value definition module is used for medical experts to grade the rapid interpretability of chronic diseases; The message reminder module is used to automatically send a message reminder to the user when the user's comprehensive chronic disease index is abnormal; The user confirms or rechecks the indicators to obtain accurate disease information; The health monitoring module is used to dynamically monitor the health data of the user to obtain real-time disease information; The health record module is used to record the user's daily routine data and the user's real-time disease information; The health education module pushes personalized health education knowledge to the comprehensive chronic disease patients of different levels based on the health education knowledge graph; The patient information query module is used to perform information query based on the user's input conditions to obtain a comprehensive chronic disease patient cohort, personal information, risk level and treatment plan.

Images