CKD full-process information management system based on active learning

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a CKD full-process information management system based on active learning, which comprises the following steps: collecting diet information, blood pressure information and blood sugar information of a plurality of patients; wherein the diet information comprises a time stamp of food intake, food materials of the food intake, a cooking mode of the food materials and intake of the food materials; the blood pressure information and the blood sugar information respectively comprise blood pressure values and blood sugar values at 0, 1 and 2 after the time stamp of food intake; clustering the diet information of the patients to obtain typical diet of each patient in a preset confidence time period; acquiring blood pressure information and blood sugar information corresponding to the typical diet of the patient, and performing trend analysis to obtain a characteristic trend of the patient; constructing an abnormal verification model based on the typical diet information and the characteristic trend; and performing abnormity verification on the diet information, the blood pressure information and the blood sugar information of the target patient in a second preset time period to obtain confidence judgment of the target patient.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and more specifically, to a CKD full-process information management system based on active learning. Background Art

[0002] CKD (Chronic Kidney Disease) is a common and difficult-to-cure chronic disease in contemporary times, requiring long-term monitoring of potential patients. Patients with diabetes and hypertension are usually the main potential population. Monitoring methods include hospital physical examinations and information collection, but hospital physical examinations are time-consuming and costly, so information collection monitoring has gradually become the mainstream. Information collection mainly covers blood pressure, blood sugar and dietary information. However, for hospitals, only reliable data can serve as an effective basis for monitoring. Therefore, the accuracy of information collection is particularly important. However, some patients may provide inaccurate dietary data due to subjective or objective reasons, resulting in the problem of invalid information collected. Summary of the invention

[0003] The present invention provides a CKD full-process information management system based on active learning, which solves the technical problems raised in the background technology.

[0004] The present invention provides a CKD full-process information management system based on active learning, comprising:

[0005] The data collection module is used to collect the dietary information, blood pressure information and blood sugar information of several patients within a preset time period;

[0006] Among them, the dietary information includes: the timestamp of food intake, the ingredients of the food intake, the cooking method of the ingredients, and the intake of the ingredients; the blood pressure information and blood sugar information include: the blood pressure value and blood sugar value at time 0, time 1, and time 2 after the timestamp of food intake, respectively;

[0007] A diet clustering module is used to cluster the diet information of patients and obtain the typical diet of each patient within a preset time period;

[0008] An information analysis module is used to obtain the blood pressure information and blood sugar information corresponding to the patient's typical diet, and perform trend analysis to obtain the patient's characteristic trend;

[0009] The model building module is used to build an abnormal verification model based on typical dietary information and characteristic trends; the abnormal verification model is used to perform abnormal verification on the target patient's dietary information, blood pressure information and blood sugar information within a second preset time period to obtain a confidence judgment of the target patient.

[0010] Furthermore, the patient's dietary information is clustered, including:

[0011] Step 21, establishing the initial coding of the dietary information, including:

[0012] The timestamp of food intake in the dietary information was used as the first element of the initial coding;

[0013] Query the food ingredients in the dietary information to obtain characteristic parameters; the characteristic parameters include the carbohydrate value, protein value, fat value and sodium value of the unit intake of the food ingredients, as well as the glycemic value and glycemic load value;

[0014] Using each parameter in the characteristic parameter as the second to seventh elements of the initial code;

[0015] The cooking method of the ingredients in the dietary information is encoded by a real number as the eighth element of the initial encoding;

[0016] The intake of ingredients in the dietary information was used as the ninth element of the initial coding;

[0017] Step 22, normalizing the initial code to obtain a feature code;

[0018] Step 23, initialize N encoding cluster centers, calculate the similarity between each feature encoding and each encoding cluster center, and the similarity calculation formula is as follows: ;

[0019] in, Indicates feature encoding and The similarity of the encoded cluster centers, , express The index of Indicates The feature encodes the ath element, Indicates The a-th element of the encoded cluster center, represents the index weight, express The absolute value of represents the linear weight, Indicates feature encoding and The Manhattan distance of the encoded cluster centers;

[0020] Step 24, if The feature encoding and If the similarity of the encoding cluster centers is the smallest, the i-th feature code is assigned to the j-th encoding cluster center;

[0021] Step 25, based on step 24, N clusters are obtained, and the average value of the feature codes in each cluster is encoded as the updated encoded cluster center of the corresponding cluster;

[0022] Step 26, repeating step 25 for a preset number of times to obtain N final clusters;

[0023] Step 27, setting a cluster merging threshold, if the similarity of the updated coded cluster centers of any two final clusters is less than a preset threshold, then the corresponding final clusters are merged to obtain a merged cluster, and the updated coded cluster of the merged cluster is calculated;

[0024] Step 28, repeat step 27 until the similarity of the updated coded clusters between any two remaining final clusters or merged clusters is greater than a preset threshold, then the final cluster or merged cluster with the largest number of feature codes is taken as the typical cluster, and the dietary information corresponding to the feature codes in the typical cluster is taken as the typical diet.

[0025] Further, trend analysis is conducted, including:

[0026] Construct a data set using the blood pressure information and blood sugar information corresponding to each typical diet;

[0027] Sort the data set in chronological order to obtain feature sorting;

[0028] The characteristic trend of each typical diet is calculated based on the characteristic ranking. The calculation formula of the characteristic trend is as follows: ; ; , ; , ;

[0029] in, Indicates the characteristic trends of a typical diet, Indicates the trend of blood pressure, Indicates the trend of blood sugar, The weight representing the trend of blood pressure, The weight representing the trend of blood sugar, The interaction factor representing the trend of blood pressure and the trend of blood sugar, represents the i-th data set, represents the standard blood pressure value of the i-th data set, represents the standard blood glucose value of the ith data set, represents the blood pressure value at time 0 of the i-th data set, represents the blood pressure value at time 1 of the i-th data set, represents the blood pressure value at time 2 of the i-th data set, represents the blood glucose value at time 0 of the i-th data set, represents the blood glucose value at time 1 of the i-th data set, represents the blood glucose value at 2 moments of the i-th data set, express Activation function.

[0030] Furthermore, an abnormal verification model is constructed, including;

[0031] Construct training samples for the anomaly validation model and construct sample labels for the anomaly validation model;

[0032] The training samples include: obtaining the physical examination data of each patient at the start and end of a preset confidence period, obtaining the health score of each patient at the start and end of the preset confidence period based on expert analysis of the physical examination data, and calculating the characteristic difference of the health score of each patient at the start and end of the physical examination data; performing confidence division on the characteristic difference based on experts, and the confidence division includes: excellent, good, poor and very poor;

[0033] Match excellence to characteristic trend values interval, will be well matched to the characteristic trend value interval, matching the poorly matched characteristic trend value interval, will be poorly matched to the characteristic trend value interval;

[0034] If the characteristic trend of the patient in the preset confidence period matches the confidence division, the corresponding patient's diet information, blood pressure information and blood sugar information are judged as confidence data, and the characteristic encoding of the corresponding patient's diet information is used as a training sample, and the corresponding patient's characteristic trend is used as a sample label;

[0035] If the characteristic trend of the patient in the preset confidence period does not match the confidence division, the corresponding patient's diet information, blood pressure information and blood sugar information are judged as untrusted data, and the corresponding patient's diet information and characteristic trend are discarded;

[0036] Based on the training samples and sample labels of the patient in a preset period of time, an abnormality verification model is trained, and the abnormality verification model includes: an input layer, a hidden layer and a classifier.

[0037] Furthermore, the input layer, hidden layer and classifier include:

[0038] An input layer, used for stacking a plurality of feature codes of the target patient in the second preset time period into a feature matrix in chronological order, and inputting the feature matrix into a hidden layer;

[0039] The hidden layer is used to perform nonlinear mapping on the feature matrix to obtain the hidden state of the feature matrix;

[0040] A classifier is used to input a hidden state, and a classification space of the classifier represents a predicted characteristic trend of the target patient within a second preset time period.

[0041] Furthermore, the calculation formula of the hidden layer includes: ; ; ; ;

[0042] in, Indicates The output of the reset gate of the hidden layer is Indicates The output of the update gate of the hidden layer is Indicates candidate hidden states of hidden layers, Indicates The hidden states of the hidden layers are express Activation function, express Activation function, , and Indicates The first weight matrix, the second weight matrix and the bias matrix of the reset gate of the hidden layer, , and Indicates The first weight matrix, the second weight matrix and the bias matrix of the update gate of the hidden layer, , and Indicates The first weight matrix, the second weight matrix and the bias matrix of the candidate hidden states, Indicates The feature matrix of the hidden layer input The feature encoding of the row, Indicates The hidden state of the hidden layer output, Represents element-wise multiplication.

[0043] Furthermore, the anomaly verification model updates the weight matrix and bias matrix of the hidden layer through back propagation based on the mean error variance loss function. The calculation formula of the mean error variance loss function is as follows: ;

[0044] in, Indicates The mean error variance loss value for each patient is Indicates The value of the characteristic trend corresponding to the sample label of each patient, Indicates The values ​​of the predictive characteristics trends for each patient.

[0045] Furthermore, abnormal verification is performed, including:

[0046] Obtaining a characteristic trend of the target patient in a second preset time period, and a predicted characteristic trend of the target patient in the second preset time period by the abnormal verification model;

[0047] If the difference between the characteristic trend and the predicted characteristic trend is less than the preset confidence threshold, it means that the target patient's diet information, blood pressure information and blood sugar information in the second preset time period are credible;

[0048] If the difference between the characteristic trend and the predicted characteristic trend is greater than a preset confidence threshold, it means that the target patient's diet information, blood pressure information, and blood sugar information in the second preset time period are not credible.

[0049] The beneficial effects of the present invention are as follows: through active learning mechanism and multi-dimensional data fusion, the system realizes accurate monitoring and analysis of CKD patients' diet, blood pressure, blood sugar and other information, can effectively extract typical dietary patterns, predict health trends, and build an abnormal verification model to judge the credibility of the target patient's information, thereby greatly improving the reliability of information monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a module diagram of a CKD full-process information management system based on active learning of the present invention. DETAILED DESCRIPTION

[0051] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0052] like Figure 1 As shown, a CKD full-process information management system based on active learning includes:

[0053] The data collection module is used to collect the dietary information, blood pressure information and blood sugar information of several patients within a preset time period;

[0054] Among them, the dietary information includes: the timestamp of food intake, the ingredients of the food intake, the cooking method of the ingredients, and the intake of the ingredients; the blood pressure information and blood sugar information respectively include: the blood pressure value and blood sugar value at time 0, time 1, and time 2 after the timestamp of food intake;

[0055] A diet clustering module is used to cluster the diet information of patients and obtain the typical diet of each patient within a preset time period;

[0056] An information analysis module is used to obtain the blood pressure information and blood sugar information corresponding to the patient's typical diet, and perform trend analysis to obtain the patient's characteristic trend;

[0057] The model building module is used to build an abnormal verification model based on typical dietary information and characteristic trends; the abnormal verification model is used to perform abnormal verification on the target patient's dietary information, blood pressure information and blood sugar information within a second preset time period to obtain a confidence judgment of the target patient.

[0058] In one embodiment of the present invention, clustering the patient's dietary information includes:

[0059] Step 21, establishing the initial coding of the dietary information, including:

[0060] The timestamp of food intake in the dietary information was used as the first element of the initial coding;

[0061] Query the food ingredients in the dietary information to obtain characteristic parameters; the characteristic parameters include the carbohydrate value, protein value, fat value and sodium value of the unit intake of the food ingredients, as well as the glycemic value and glycemic load value;

[0062] Using each parameter in the characteristic parameter as the second to seventh elements of the initial code;

[0063] The cooking method of the ingredients in the dietary information is encoded by a real number as the eighth element of the initial encoding;

[0064] The intake of ingredients in the dietary information was used as the ninth element of the initial coding;

[0065] Step 22, normalizing the initial code to obtain a feature code;

[0066] Step 23, initialize N encoding cluster centers, calculate the similarity between each feature encoding and each encoding cluster center, and the similarity calculation formula is as follows: ;

[0067] in, Indicates feature encoding and The similarity of the encoded cluster centers, , express The index of Indicates The feature encodes the ath element, Indicates The a-th element of the encoded cluster center, represents the index weight, express The absolute value of represents the linear weight, Indicates feature encoding and The Manhattan distance of the encoded cluster centers;

[0068] Step 24, if The feature encoding and If the similarity of the encoding cluster centers is the smallest, the i-th feature code is assigned to the j-th encoding cluster center;

[0069] Step 25, based on step 24, N clusters are obtained, and the average value of the feature codes in each cluster is encoded as the updated encoded cluster center of the corresponding cluster;

[0070] Step 26, repeating step 25 for a preset number of times to obtain N final clusters;

[0071] Step 27, setting a cluster merging threshold, if the similarity of the updated coded cluster centers of any two final clusters is less than a preset threshold, then the corresponding final clusters are merged to obtain a merged cluster, and the updated coded cluster of the merged cluster is calculated;

[0072] Step 28, repeat step 27 until the similarity of the updated coded clusters between any two remaining final clusters or merged clusters is greater than a preset threshold, then the final cluster or merged cluster with the largest number of feature codes is taken as the typical cluster, and the dietary information corresponding to the feature codes in the typical cluster is taken as the typical diet.

[0073] It should be noted that you can query ingredients through the Mint Health APP. For example, you can query the content of various nutrients, glycemic value and glycemic load value of a unit weight of bread (100g per serving) through the Mint Health APP.

[0074] Specifically, this scheme conducts cluster analysis by initializing as many coding cluster centers as possible, each of which represents an artificially fabricated feature code. The construction of feature codes is based on statistical analysis of each element of the feature codes obtained in the historical time period to determine the range of possible values ​​for each element. Subsequently, according to the range of possible values, several different feature codes are generated using randomization or systematic methods, and these feature codes are initialized as cluster centers. The initialization process of each cluster center fully considers the distribution characteristics of the feature space to ensure that all potential dietary information patterns can be reasonably covered and represented. After initialization, the patient's dietary information feature code is calculated for similarity with these cluster centers, and each feature code is classified into the most similar cluster center according to the size of the similarity. The similarity calculation formula used here comprehensively considers the element difference, Manhattan distance and weight factors between feature codes, and can accurately measure the difference between feature codes and cluster centers. The K-means clustering algorithm was used in the clustering process, which specifically includes the following steps: calculating the similarity between the feature code and the cluster center, assigning the feature code to the nearest cluster center, updating the cluster center to the average value code of the cluster, and iterating continuously until the preset convergence condition is met or the specified number of iterations is reached. After the clustering is completed, in order to further optimize the clustering results, the clusters are subsequently merged by setting a merging threshold. Specifically, if the similarity of any two cluster centers is lower than the preset merging threshold, they are merged into a new cluster, and the process is repeated until the similarity between all cluster centers is higher than the merging threshold. Finally, the cluster containing the largest number of feature codes is selected from all clusters as a typical cluster, and the dietary information corresponding to all feature codes in the typical cluster is extracted as the typical diet of the patient. The typical diet obtained in this way can more realistically reflect the patient's dietary habits in a preset time period, because the typical diet is a representative pattern extracted from a large amount of dietary information, which can comprehensively reflect the patient's dietary preferences and habit characteristics. Based on the typical diet, its impact on the patient's blood pressure, blood sugar and other characteristic parameters can be further analyzed, so as to infer and judge the patient's characteristic trend, and provide a basis for subsequent health management and intervention measures. This method not only has high accuracy and representativeness, but also can reduce the interference of noisy data, laying the foundation for scientific and precise patient monitoring.

[0075] In one embodiment of the present invention, trend analysis is performed, including:

[0076] Construct a data set using the blood pressure information and blood sugar information corresponding to each typical diet;

[0077] Sort the data set in chronological order to obtain feature sorting;

[0078] The characteristic trend of each typical diet is calculated based on the characteristic ranking. The calculation formula of the characteristic trend is as follows: ; ; , ; , ;

[0079] in, Indicates the characteristic trends of a typical diet, Indicates the trend of blood pressure, Indicates the trend of blood sugar, The weight representing the trend of blood pressure, The weight representing the trend of blood sugar, The interaction factor representing the trend of blood pressure and the trend of blood sugar, represents the i-th data set, represents the standard blood pressure value of the ith data set, represents the standard blood glucose value of the ith data set, represents the blood pressure value at time 0 of the i-th data set, represents the blood pressure value at time 1 of the i-th data set, represents the blood pressure value at time 2 of the i-th data set, represents the blood glucose value at time 0 of the i-th data set, represents the blood glucose value at time 1 of the i-th data set, represents the blood glucose value at 2 moments of the i-th data set, express Activation function.

[0080] Specifically, blood pressure information and blood sugar information based on a typical diet are analyzed to infer the patient's physical health characteristic trends within a specific time period. This process combines the nutritional components of the ingredients in the patient's diet for systematic fusion analysis and calculation, aiming to fully reveal the long-term impact of eating habits on the patient's physical health and quantify it as characteristic trend values. Characteristic trends are used to reflect the patient's physical health trends within a preset time period, and can effectively describe the dynamic impact of eating habits on key health indicators such as blood pressure and blood sugar. Through scientific calculations, the system combines dietary characteristics with blood pressure and blood sugar characteristics to form a comprehensive trend model. For example, for patients who prefer high-fat, high-carbohydrate foods (such as fried chicken, beer, etc.) in their diet, after the system analyzes their blood pressure and blood sugar information, it may conclude that their characteristic trend values ​​fall within This indicates that this diet structure may have an adverse effect on the patient's health. For patients with a scientific diet structure and balanced nutrition (such as a scientifically proportioned high-fiber, low-fat diet), the health trend of their blood pressure and blood sugar may be more positive, and the system calculates its characteristic trend value as , thus reflecting the positive effect of good eating habits on health. Based on the analysis results of typical eating patterns, the changing trend of the patient's future health characteristics is inferred. Therefore, if the predicted trend does not match the actual trend, it means that the patient's diet information contains untrue content and is judged to be abnormal. The calculation of characteristic trends combines the blood pressure and blood sugar information of a typical diet with scientific nutritional analysis, converting complex health data into intuitive trend values, which can truly reflect the relationship between the patient's health status and diet.

[0081] In one embodiment of the present invention, constructing an abnormality verification model includes:

[0082] Construct training samples for the anomaly validation model and construct sample labels for the anomaly validation model;

[0083] The training samples include: obtaining the physical examination data of each patient at the start and end of a preset confidence period, obtaining the health score of each patient at the start and end of the preset confidence period based on expert analysis of the physical examination data, and calculating the characteristic difference of the health score of each patient at the start and end of the physical examination data; performing confidence division on the characteristic difference based on experts, and the confidence division includes: excellent, good, poor and very poor;

[0084] Match excellence to characteristic trend values interval, will be well matched to the characteristic trend value interval, matching the poorly matched characteristic trend value interval, will be poorly matched to the characteristic trend value interval;

[0085] If the characteristic trend of the patient in the preset confidence period matches the confidence division, the corresponding patient's diet information, blood pressure information and blood sugar information are judged as confidence data, and the characteristic encoding of the corresponding patient's diet information is used as a training sample, and the corresponding patient's characteristic trend is used as a sample label;

[0086] If the characteristic trend of the patient in the preset confidence period does not match the confidence division, the corresponding patient's diet information, blood pressure information and blood sugar information are judged as untrusted data, and the corresponding patient's diet information and characteristic trend are discarded;

[0087] Based on the training samples and sample labels of the patient in a preset period of time, an abnormality verification model is trained, and the abnormality verification model includes: an input layer, a hidden layer and a classifier.

[0088] Specifically, in order to train an effective abnormal verification model, it is necessary to use the real data of the patient in the confidence period. However, since the data provided by the patient may be untrue due to subjective or objective reasons, such as omission of dietary information, incorrect recording or deliberate concealment of certain details, if these untrue data are used to train the abnormal verification model, it may cause the model to be biased and reduce its accuracy. Therefore, the present invention proposes a method for distinguishing real data from untrue data by matching the relationship between the patient's health changes and characteristic trends, so as to ensure that the data used for model training has high credibility. First, during the confidence period, the patient's data collection not only includes dietary information, blood pressure information and blood sugar information, but also a comprehensive physical health examination of the patient at the start and end time of the confidence period. The physical examination data includes the changes in key health indicators such as blood pressure and blood sugar of the patient, which are objective and accurate and can be used as a reference for the patient's health status. Subsequently, the patient's health changes are graded by introducing a grading system annotated by experts. The grading system divides the patient into several levels, such as "excellent", "good", "poor" and "very poor", according to the degree of health changes of the patient during the confidence period. These classifications can quantitatively describe the degree of good or bad of the patient's health changes and provide standards for subsequent feature trend matching. Based on the patient's health change classification, the system matches the feature trend value extracted within the confidence period with the classification result. The feature trend value is a quantitative indicator used to reflect the comprehensive impact of the patient's dietary information on his blood pressure, blood sugar and other health indicators within the confidence period. Assuming that the feature trend value falls within a reasonable range (for example, consistent with the health change classification result), it can be judged that the calculation results of these feature trends are correct, and the corresponding diet information, blood pressure information and blood sugar information are identified as real data. On the contrary, if the feature trend value does not match the health change classification result, it means that there may be an error in the feature trend, and it is further inferred that the patient's diet information, blood pressure information and blood sugar information may not be true. In this case, the present invention marks these untrue data as invalid data and discards them to prevent these data from interfering with the training of the abnormal verification model. Through the above process, a high-quality and reliable data set can be obtained for the training of the abnormal verification model. Based on these real data, the abnormal verification model can learn accurate feature trend patterns to help the system effectively judge whether the new data is abnormal in subsequent operations. For example, when the model is actually applied, the characteristic trend value of the input target patient is compared with the abnormal verification model. If the characteristic trend value meets the model prediction range, the data is considered to be credible; otherwise, it is considered to be unreliable. In summary, the present invention screens out the real data of the patient in the confidence period through the matching relationship between the health change classification and the characteristic trend value, and uses this data to train the abnormal verification model.This method not only ensures the authenticity and high quality of the model training data, but also effectively improves the accuracy and robustness of the anomaly verification model, thereby providing scientific support for the subsequent anomaly data detection and processing of the system.

[0089] In one embodiment of the present invention, the input layer, the hidden layer and the classifier include:

[0090] An input layer, used for stacking a plurality of feature codes of the target patient in the second preset time period into a feature matrix in chronological order, and inputting the feature matrix into a hidden layer;

[0091] The hidden layer is used to perform nonlinear mapping on the feature matrix to obtain the hidden state of the feature matrix;

[0092] A classifier is used to input a hidden state, and a classification space of the classifier represents a predicted characteristic trend of the target patient within a second preset time period.

[0093] Specifically, the internal structure of the abnormal verification model is based on the design of neural networks, which can extract deep features from the input data and make efficient judgments on the credibility of the data. The main function of the input layer is to accept the feature data extracted by the patient during the confidence period. The feature data includes: The feature code is converted from the patient's dietary information (intake, nutritional ingredients, cooking methods, etc.). The feature trend value is a trend indicator obtained based on the time series analysis of blood pressure and blood sugar data. The input data is organized into a feature matrix, where each row represents the feature data at a time point and the column represents the specific dimension of the feature.

[0094] In one embodiment of the present invention, the calculation formula of the hidden layer includes: ; ; ; ;

[0095] in, Indicates The output of the reset gate of the hidden layer is Indicates The output of the update gate of the hidden layer is Indicates candidate hidden states of hidden layers, Indicates The hidden states of the hidden layers are express Activation function, express Activation function, , and Indicates The first weight matrix, the second weight matrix and the bias matrix of the reset gate of the hidden layer, , and Indicates The first weight matrix, the second weight matrix and the bias matrix of the update gate of the hidden layer, , and Indicates The first weight matrix, the second weight matrix and the bias matrix of the candidate hidden states, Indicates The feature matrix of the hidden layer input The feature encoding of the row, Indicates The hidden state of the hidden layer output, Represents element-wise multiplication.

[0096] Specifically, the anomaly verification model can flexibly balance the weight between historical information and current input by introducing a dynamic adjustment mechanism, thereby achieving accurate processing of complex time series data. The model deeply interacts and extracts the historical correlation and current features of the input data by designing multiple layers of computing units, including a reset mechanism, an update mechanism, and a candidate state generation mechanism, and finally generates a hidden state that can reflect the dynamic changes of the features. This design enhances the model's ability to express nonlinear features, and improves the robustness and sensitivity to abnormal data through dynamic adjustment layer by layer, ensuring accurate judgment of abnormal feature data.

[0097] In one embodiment of the present invention, the abnormal verification model updates the weight matrix and bias matrix of the hidden layer through back propagation based on the mean error variance loss function. The calculation formula of the mean error variance loss function is as follows: ;

[0098] in, Indicates The mean error variance loss value for each patient is Indicates The value of the characteristic trend corresponding to the sample label of each patient, Indicates The values ​​of the predictive characteristics trends for each patient.

[0099] It should be noted that the back propagation of the mean error variance loss function is updated to the existing technology, so it will not be repeated here.

[0100] In one embodiment of the present invention, performing abnormal verification includes:

[0101] Obtaining a characteristic trend of the target patient in a second preset time period, and a predicted characteristic trend of the target patient in the second preset time period by the abnormal verification model;

[0102] If the difference between the characteristic trend and the predicted characteristic trend is less than the preset confidence threshold, it means that the target patient's diet information, blood pressure information and blood sugar information in the second preset time period are credible;

[0103] If the difference between the characteristic trend and the predicted characteristic trend is greater than a preset confidence threshold, it means that the target patient's diet information, blood pressure information, and blood sugar information in the second preset time period are not credible.

[0104] Specifically, the core of abnormal verification is to judge the credibility of patient data by comparing the characteristic trend of the target patient with the predicted characteristic trend of the abnormal verification model. Specifically, abnormal verification includes the following steps. Obtaining characteristic trends: The characteristic trend is calculated based on the target patient's diet information, blood pressure information and blood sugar information in the second preset time period. This trend reflects the changing characteristics of the patient's health status over time in this time period, such as the magnitude of blood sugar increase or decrease, the range of blood pressure fluctuations, etc. Model predicts characteristic trends: The abnormal verification model is constructed by training high-quality data in the confidence time period, and can predict the characteristic trend of the patient in the second preset time period based on the historical characteristic data and patterns of the target patient. The predicted characteristic trend represents the reasonable expected value of the model for the patient's health data. Comparing the difference: Compare the actual characteristic trend with the characteristic trend predicted by the model, and calculate the difference between the two. The difference represents the degree of deviation between the actual health changes of the target patient and the health changes predicted by the model.

[0105] Make a credibility judgment. The difference is less than the preset credibility threshold: If the difference is small, it means that the characteristic trend of the target patient is closer to the prediction of the model, indicating that the patient's diet information, blood pressure information and blood sugar information are consistent with his health changes, and these data can be considered to be credible. The difference is greater than the preset credibility threshold: If the difference is large, it means that there is a large deviation between the characteristic trend of the target patient and the model prediction, which may be caused by inaccurate diet information records, abnormal blood pressure or blood sugar information, etc. The system judges these data to be unreliable. In this way, the present invention can automatically screen the data of the target patient without relying on human intervention, ensuring that the data entering subsequent analysis and application has high reliability.

[0106] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A CKD full-process information management system based on active learning, characterized in that: include: The data collection module is used to collect the dietary information, blood pressure information and blood sugar information of several patients within a preset time period; Among them, the dietary information includes: the timestamp of food intake, the ingredients of the food intake, the cooking method of the ingredients, and the intake of the ingredients; the blood pressure information and blood sugar information respectively include: the blood pressure value and blood sugar value at time 0, time 1, and time 2 after the timestamp of food intake; A diet clustering module is used to cluster the diet information of patients and obtain the typical diet of each patient within a preset time period; An information analysis module is used to obtain the blood pressure information and blood sugar information corresponding to the patient's typical diet, and perform trend analysis to obtain the patient's characteristic trend; The model building module is used to build an abnormal verification model based on typical dietary information and characteristic trends; the abnormal verification model is used to perform abnormal verification on the target patient's dietary information, blood pressure information and blood sugar information within a second preset time period to obtain a confidence judgment of the target patient.

2. According to claim 1, a CKD full-process information management system based on active learning is characterized in that: Clustering of patients’ dietary information, including: Step 21, establishing the initial coding of the dietary information, including: The timestamp of food intake in the dietary information was used as the first element of the initial coding; Query the food ingredients in the dietary information to obtain characteristic parameters; the characteristic parameters include the carbohydrate value, protein value, fat value and sodium value of the unit intake of the food ingredients, as well as the glycemic value and glycemic load value; Using each parameter in the characteristic parameter as the second to seventh elements of the initial code; The cooking method of the ingredients in the dietary information is encoded by a real number as the eighth element of the initial encoding; The intake of ingredients in the dietary information was used as the ninth element of the initial coding; Step 22, normalizing the initial code to obtain a feature code; Step 23, initialize N encoding cluster centers, calculate the similarity between each feature encoding and each encoding cluster center, and the similarity calculation formula is as follows: ； in, Indicates feature encoding and The similarity of the encoded cluster centers, , express The index of Indicates The feature encodes the ath element, Indicates The a-th element of the encoded cluster center, represents the index weight, express The absolute value of represents the linear weight, Indicates feature encoding and The Manhattan distance of the encoded cluster centers; Step 24, if The feature encoding and If the similarity of the encoding cluster centers is the smallest, the i-th feature code is assigned to the j-th encoding cluster center; Step 25, based on step 24, N clusters are obtained, and the average value of the feature codes in each cluster is encoded as the updated encoded cluster center of the corresponding cluster; Step 26, repeating step 25 for a preset number of times to obtain N final clusters; Step 27, setting a cluster merging threshold, if the similarity of the updated coded cluster centers of any two final clusters is less than a preset threshold, then the corresponding final clusters are merged to obtain a merged cluster, and the updated coded cluster of the merged cluster is calculated; Step 28, repeat step 27 until the similarity of the updated coded clusters between any two remaining final clusters or merged clusters is greater than a preset threshold, then the final cluster or merged cluster with the largest number of feature codes is taken as the typical cluster, and the dietary information corresponding to the feature codes in the typical cluster is taken as the typical diet.

3. The CKD full-process information management system based on active learning according to claim 2 is characterized in that: Conduct trend analysis, including: Construct a data set using the blood pressure information and blood sugar information corresponding to each typical diet; Sort the data set in chronological order to obtain feature sorting; The characteristic trend of each typical diet is calculated based on the characteristic ranking. The calculation formula of the characteristic trend is as follows: ； ； ， ； ， ； in, Indicates the characteristic trends of a typical diet, Indicates the trend of blood pressure, Indicates the trend of blood sugar, The weight representing the trend of blood pressure, The weight representing the trend of blood sugar, The interaction factor representing the trend of blood pressure and the trend of blood sugar, represents the i-th data set, represents the standard blood pressure value of the i-th data set, represents the standard blood glucose value of the ith data set, represents the blood pressure value at time 0 of the i-th data set, represents the blood pressure value at time 1 of the i-th data set, represents the blood pressure value at time 2 of the i-th data set, represents the blood glucose value at time 0 of the i-th data set, represents the blood glucose value at time 1 of the i-th data set, represents the blood glucose value at 2 moments of the i-th data set, express Activation function.

4. The CKD full-process information management system based on active learning according to claim 3 is characterized in that: Construct anomaly verification model, including; Construct training samples for the anomaly validation model and construct sample labels for the anomaly validation model; The training samples include: obtaining the physical examination data of each patient at the start and end of a preset confidence period, analyzing the physical examination data based on experts to obtain the health score of each patient at the start and end of the preset confidence period, and calculating the characteristic difference of the health score of the physical examination data of each patient at the start and end; performing confidence division on the characteristic difference based on experts, and the confidence division includes: excellent, good, poor and very poor; Match excellence to characteristic trend values interval, will be well matched to the characteristic trend value interval, matching the difference to the characteristic trend value interval, will be poorly matched to the characteristic trend value interval; If the characteristic trend of the patient in the preset confidence period matches the confidence division, the corresponding patient's diet information, blood pressure information and blood sugar information are judged as confidence data, and the characteristic encoding of the corresponding patient's diet information is used as a training sample, and the corresponding patient's characteristic trend is used as a sample label; If the characteristic trend of the patient in the preset confidence period does not match the confidence division, the corresponding patient's diet information, blood pressure information and blood sugar information are judged as untrusted data, and the corresponding patient's diet information and characteristic trend are discarded; Based on the training samples and sample labels of the patient in a preset period of time, an abnormality verification model is trained, and the abnormality verification model includes: an input layer, a hidden layer and a classifier.

5. The CKD full-process information management system based on active learning according to claim 4 is characterized in that: Input layer, hidden layer and classifier, including: An input layer, used for stacking a plurality of feature codes of the target patient in the second preset time period into a feature matrix in chronological order, and inputting the feature matrix into a hidden layer; The hidden layer is used to perform nonlinear mapping on the feature matrix to obtain the hidden state of the feature matrix; A classifier is used to input a hidden state, and a classification space of the classifier represents a predicted characteristic trend of the target patient within a second preset time period.

6. The CKD full-process information management system based on active learning according to claim 5 is characterized in that: The calculation formula for the hidden layer includes: ； ； ； ； in, Indicates The output of the reset gate of the hidden layer is Indicates The output of the update gate of the hidden layer is Indicates candidate hidden states of hidden layers, Indicates The hidden states of the hidden layers are express Activation function, express Activation function, , and Indicates The first weight matrix, the second weight matrix and the bias matrix of the reset gate of the hidden layer, , and Indicates The first weight matrix, the second weight matrix and the bias matrix of the update gate of the hidden layer, , and Indicates The first weight matrix, the second weight matrix and the bias matrix of the candidate hidden states, Indicates The feature matrix of the hidden layer input The feature encoding of the row, Indicates The hidden state of the hidden layer output, Represents element-wise multiplication.

7. The CKD full-process information management system based on active learning according to claim 5 is characterized in that: The anomaly verification model updates the weight matrix and bias matrix of the hidden layer through back propagation based on the mean error variance loss function. The calculation formula of the mean error variance loss function is as follows: ； in, Indicates The mean error variance loss value for each patient is Indicates The value of the characteristic trend corresponding to the sample label of each patient, Indicates The values ​​of the predictive characteristics trends for each patient.

8. The CKD full-process information management system based on active learning according to claim 5 is characterized in that: Perform exception verification, including: Obtaining a characteristic trend of the target patient in a second preset time period, and a predicted characteristic trend of the target patient in the second preset time period by the abnormal verification model; If the difference between the characteristic trend and the predicted characteristic trend is less than the preset confidence threshold, it means that the target patient's diet information, blood pressure information and blood sugar information in the second preset time period are credible; If the difference between the characteristic trend and the predicted characteristic trend is greater than a preset confidence threshold, it means that the target patient's diet information, blood pressure information, and blood sugar information in the second preset time period are not credible.

Images