Method, system and storage medium for early warning of lupus nephritis

Abstract

The present invention discloses a method, system and storage medium for early warning of lupus nephritis, which belongs to the field of data analysis technology, including: establishing multiple analysis models, inputting learning data into the analysis model corresponding to any data type, and outputting analysis information; constructing a neural network model, inputting all combined information into the neural network model, and outputting classification results, setting the classification results with a first value greater than a first preset value as first-class information; setting the second value as the importance of the learning data, setting the learning data with an importance greater than the second preset value as necessary data, re-inputting all necessary data into a cascade model for training and learning to generate an early warning model; collecting monitoring data of a monitored object, inputting the monitoring data into an early warning model, and outputting monitoring results, completing early warning of lupus nephritis of the monitored object. The present invention can improve the accuracy and timeliness of early warning of lupus nephritis.

Description

Technical Field

[0001] The present invention belongs to the technical field of data analysis, and in particular relates to an early warning method, system and storage medium for lupus nephritis. Background Art

[0002] Lupus nephritis (LN) is one of the most serious complications of systemic lupus erythematosus (SLE). It is an autoimmune disease that can cause kidney damage. Early diagnosis and treatment of lupus nephritis are crucial to prevent renal failure and improve the long-term prognosis of patients.

[0003] Similar prior art includes a Chinese patent application with publication number CN113707323A, which discloses a disease prediction method based on machine learning, including: performing structured processing on pre-collected user case data to obtain case sample data; inputting the case sample data into a neural network model for training to obtain a disease feature vector, a memory vector, and a distance value; during the training process, using the distance value to continuously adjust the ratio parameter of the disease feature vector and the memory vector, and constructing a disease prediction model based on the feature vector of the user in each disease category determined by the ratio parameter; in response to the triggering of the disease prediction instruction, using the disease prediction model to predict the case data of the target user, and outputting the probability of the target user in each disease category. This invention can improve the accuracy of disease prediction results in the case of unbalanced sample data by transferring the knowledge learned in the majority class samples to the prediction of the minority class samples during the model training process. There is also a Chinese patent application with publication number CN111986814A, which provides a modeling method for a prediction model of lupus nephritis in lupus patients, including the following steps: data collection, using natural language processing methods to extract phenotypic data and demographic data from the electronic medical records of systemic lupus erythematosus patients; converting the phenotypic data and demographic data into eigenvectors through 01 matrices; model training, using part of the eigenvectors to train the lupus nephritis prediction model to obtain the lupus nephritis prediction model. The invention provides a modeling method for a prediction model of lupus nephritis in lupus patients, uses a large amount of existing electronic medical record data to establish a prediction model of lupus nephritis, and applies the prediction model of lupus nephritis to the study of complications in lupus patients, which can take appropriate measures for high-risk groups, assist doctors in diagnosis and treatment, and reduce the incidence of lupus nephritis.

[0004] However, the above-mentioned prior art uses a large variety of case sample data, which may lead to inaccurate prediction of disease categories. When obtaining a lupus nephritis prediction model, the diversity of medical record data is not classified and analyzed, which will reduce the accuracy of model prediction. Summary of the invention

[0005] To solve the above problems, the present invention provides an early warning method, system and storage medium for lupus nephritis to solve the problems in the prior art.

[0006] In order to achieve the above-mentioned purpose of the invention, the present invention proposes an early warning method for lupus nephritis, comprising:

[0007] Collect label data of lupus nephritis patients based on preset labels, combine all label data into a historical data set, obtain the data types contained in the historical data set, establish multiple analysis models based on the data types, set the label data corresponding to any data type as learning data, input the learning data into the analysis model corresponding to any data type, and output analysis information;

[0008] Combining all analysis information corresponding to any lupus nephritis patient to generate combined information, constructing a neural network model, inputting all combined information into the neural network model, and outputting a classification result, calculating a first value between any classification result and the combined information based on a first preset algorithm, and setting the classification result whose first value is greater than a first preset value as first category information;

[0009] Calculate a second value between any first type of information and the learning data based on a second preset algorithm, set the second value as the importance of the learning data, set the learning data whose importance is greater than the second preset value as necessary data, combine all the analysis models and the neural network model and set them as a cascade model, re-input all necessary data into the cascade model for training and learning, so as to generate an early warning model;

[0010] The preset label corresponding to the necessary data is obtained and set as the necessary label, the monitoring data of the monitored object is collected based on the necessary label, the monitoring data is input into the early warning model, and the monitoring result is output, and the early warning of lupus nephritis of the monitored object is completed based on the monitoring result.

[0011] Further, the analysis information is output based on the following steps:

[0012] The analysis model includes a first sub-model and a second sub-model. If the data type of the learning data is an image type, the learning data is input into the first sub-model, the first sub-model arranges pixels of the learning data to generate pixel information, extracts feature points of the learning data, arranges all the feature points based on corresponding frequencies to generate statistical information, combines the pixel information and the statistical information to generate image type information, outputs a feature vector corresponding to the image type information, and sets it as the analysis information;

[0013] Otherwise, the learning data is input into the second sub-model, the second sub-model splits the learning data into multiple sub-data, converts each character in the sub-data into character information, calculates the variance of the character information based on the dimension of the character information, calculates the ratio between the sum of all variances and the number of sub-data, sets the ratio as unit information, outputs the feature vector corresponding to the unit information, and sets it as the analysis information.

[0014] Further, the classification result is output based on the following steps:

[0015] Assigning weight values ​​to the analysis information, and performing weighted summation on all the analysis information based on the weight values ​​to generate the combination information;

[0016] A target category is set based on the severity of lupus nephritis, the neural network model is trained based on the combined information, and a classification estimate corresponding to the combined information is output based on the target category prediction, all classification estimate values ​​are normalized, and a probability value distribution is generated, the target category with the maximum probability is extracted from the probability value distribution and set as the predicted category corresponding to the combined information, and any combined information contained in the same predicted category is set as the classification result.

[0017] Furthermore, the neural network model is trained based on the combined information, comprising the following steps:

[0018] Setting a standard label for the combined information corresponding to the lupus nephritis patient based on the severity of the lupus nephritis, and combining the same standard labels to generate a plurality of standard categories;

[0019] During the training process of the neural network model, the first similarity between the standard category and the predicted category corresponding to the severity of any lupus nephritis is calculated. If the first similarity is greater than or equal to a third preset value, the training of the neural network model is stopped. Otherwise, part of the combination information is eliminated based on a preset ratio, and the remaining combination information is input into the neural network model. If the first similarity shows an increasing trend, the remaining combination information is retained and the neural network model is trained. Otherwise, part of the combination information is re-eliminated from all the combination information based on the preset ratio, and this step is repeated until the first similarity shows an increasing trend.

[0020] Furthermore, the early warning model is generated based on the following steps:

[0021] The cascade model obtains the feature vector of the necessary data and sets it as the theme feature, performs training based on the theme feature, outputs a state estimate, obtains a loss function corresponding to the state estimate, adjusts the parameters of the cascade model until the loss function is reduced to a minimum value, and sets the adjusted cascade model as the early warning model.

[0022] Further, the monitoring result is output based on the following steps:

[0023] The early warning model checks whether the monitoring data is missing based on the necessary labels. If so, the monitoring data is supplemented. If not, the monitoring data is converted into sequence data based on time information. The early warning model integrates all the sequence data, extracts a first association sequence between the sequence data and the necessary data, and a second association sequence between the first association sequence and the state estimation based on an attention mechanism algorithm, obtains association features of the second association sequence, obtains an estimation result of the association feature based on the state estimation, and predicts and outputs a change sequence of the estimation result within a prediction time based on the second association sequence, and sets the change sequence as the monitoring result;

[0024] The estimation result, the predicted time and the monitoring result are transmitted to medical personnel to complete the early warning of lupus nephritis of the monitored object.

[0025] Furthermore, a second similarity between the monitoring data corresponding to any necessary tag and the necessary data is calculated. If all the second similarities are greater than or equal to a fourth preset value, it is determined that the monitoring data is not missing; otherwise, it is determined that the monitoring data is missing.

[0026] Further, the supplementing of the monitoring data comprises the following steps:

[0027] The necessary label whose second similarity is less than the fourth preset value is obtained, and is set as the label to be supplemented. Based on the encoder, the encoder obtains the degree of fit between the label to be supplemented in the necessary data and the data corresponding to the remaining necessary labels. The encoder generates the data corresponding to the label to be supplemented in the monitoring data based on the degree of fit, and supplements it into the monitoring data.

[0028] The present invention also provides an early warning system for lupus nephritis, which is used to implement the above-mentioned early warning method for lupus nephritis. The system mainly includes:

[0029] A data collection module collects label data of lupus nephritis patients based on preset labels, combines all label data into a historical data set, obtains the data types contained in the historical data set, establishes multiple analysis models based on the data types, sets the label data corresponding to any data type as learning data, inputs the learning data into the analysis model corresponding to any data type, and outputs analysis information;

[0030] a data classification module, combining all analysis information corresponding to any lupus nephritis patient to generate combined information, constructing a neural network model, inputting all combined information into the neural network model, and outputting a classification result, calculating a first value between any classification result and the combined information based on a first preset algorithm, and setting the classification result whose first value is greater than a first preset value as first category information;

[0031] A model generation module, which calculates a second value between any first type of information and the learning data based on a second preset algorithm, sets the second value as the importance of the learning data, sets the learning data whose importance is greater than the second preset value as necessary data, combines all the analysis models and the neural network models into a cascade model, and inputs all necessary data into the cascade model for training and learning to generate an early warning model;

[0032] The early warning analysis module obtains the preset label corresponding to the necessary data and sets it as the necessary label, collects the monitoring data of the monitored object based on the necessary label, inputs the monitoring data into the early warning model, and outputs the monitoring results, and completes the early warning of lupus nephritis of the monitored object based on the monitoring results.

[0033] The present invention also provides a computer storage medium storing program instructions, wherein when the program instructions are executed, the device where the computer storage medium is located is controlled to execute the above-mentioned early warning method for lupus nephritis.

[0034] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0035] The present invention first collects label data about patients with lupus nephritis, and extracts analysis information corresponding to the learning data through different analysis models, which can improve the accuracy and completeness of data preprocessing. Then, the combined information is classified through a neural network model, and the classification results are output. The target category can be set according to the severity of lupus nephritis to improve the diversity of the classification results. Finally, the necessary data is screened out through the first value and the second value, and the early warning model is generated through training, which can improve the accuracy of the early warning model.

[0036] The present invention also performs data preprocessing and early warning analysis on the monitoring data of the monitored object through the early warning model, generates monitoring results, and completes the early warning of lupus nephritis of the monitored object based on the monitoring results, which can improve the accuracy and timeliness of the early warning of the monitored object. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A flowchart of the steps of an early warning method for lupus nephritis according to the present invention;

[0038] Figure 2 The present invention is a structural diagram of an early warning system for lupus nephritis. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0040] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0041] like Figure 1 As shown, an early warning method for lupus nephritis comprises:

[0042] S1: Collect label data of lupus nephritis patients based on preset labels, combine all label data into a historical data set, obtain the data types contained in the historical data set, establish multiple analysis models based on the data types, set the label data corresponding to any data type as learning data, input the learning data into the analysis model corresponding to any data type, and output the analysis information.

[0043] Specifically, in this embodiment, lupus nephritis patients refer to patients with systemic lupus erythematosus and kidney inflammation, including but not limited to patients with lupus nephritis of various severity levels. Preset labels refer to disease examination type labels associated with lupus nephritis, including but not limited to patient information, urine, blood, immunology, renal biopsy and imaging, etc. Label data refers to the examination result data corresponding to the preset labels. The historical data set is a data set after all label data corresponding to lupus nephritis patients are summarized and combined based on the preset labels. Data type refers to the storage method of data, including but not limited to numerical type, character type and structural type, etc., wherein the structural type belongs to the image type, and the numerical type and character type belong to the non-image type. The analysis model refers to a model composed of various neural networks, and different neural networks can be set according to different data types to extract the feature vector of the learning data, that is, the analysis information, wherein the analysis information is the vector information generated by each learning data after feature analysis. The essential characteristics and structure of the learning data can be captured to facilitate subsequent training and learning.

[0044] S2: Combine all analysis information corresponding to any lupus nephritis patient to generate combined information, build a neural network model, input all combined information into the neural network model, and output the classification result, calculate the first numerical value between any classification result and the combined information based on the first preset algorithm, and set the classification result whose first numerical value is greater than the first preset value as the first category of information.

[0045] Specifically, in this embodiment, the analysis information corresponding to different lupus nephritis patients has different symptom manifestations. Therefore, the analysis information contained in the same lupus nephritis patient is combined, and the combined information can be used to describe the comprehensive symptom vector of the lupus nephritis patient, rather than determining the symptoms of the lupus nephritis patient through a single analysis information. The neural network model is a computational model composed of a large number of neurons. By using the combined information for training, the function of the feature vector contained in the combined information can be estimated, and the classification result can be obtained according to the function vector, that is, the category to which the combined information belongs is determined, wherein the classification result refers to the combined information contained in the classification label corresponding to the predicted category. The first preset algorithm refers to the algorithm of the neural network model using the chain rule in the first back propagation, which is used to calculate the derivative of the composite function, wherein the first back propagation refers to the first step of reversely transmitting the error signal from the output layer of the network to the input layer in the process of building and training the model, and the first numerical value refers to the gradient value of the output classification result relative to the combined information, and the first numerical value can reflect the degree of influence of the data element corresponding to the combined information on the classification result. Therefore, the first type of information is screened out by the first preset value, wherein the first type of information refers to the combined information with a greater degree of influence in the classification result.

[0046] S3: Calculate the second value between any first category information and the learning data based on the second preset algorithm, set the second value as the importance of the learning data, set the learning data with an importance greater than the second preset value as necessary data, combine all analysis models and neural network models and set them as a cascade model, re-input all necessary data into the cascade model for training and learning to generate an early warning model.

[0047] Specifically, in this embodiment, the second numerical value refers to the gradient value calculated when the first type of information is back-propagated to each input learning data, and the second preset algorithm refers to the algorithm of the neural network model using the chain rule in the second back propagation, wherein the second back propagation refers to the process of further back-propagating from the feature vector corresponding to the integrated first type of information to each input learning data based on the first numerical value obtained by the first back propagation. This process is to calculate the second numerical value, that is, the relationship between the change in the learning data and the change in the output first type of information. The larger the second numerical value, the greater the contribution of the learning data, that is, the higher the importance, therefore, the learning data is screened by the second preset value to generate necessary data. The above-mentioned analysis model and the neural network model both constitute a cascade model, and the cascade model is trained by all necessary data to accurately generate an early warning model associated with lupus nephritis.

[0048] S4: Obtain preset labels corresponding to necessary data and set them as necessary labels, collect monitoring data of the monitored objects based on the necessary labels, input the monitoring data into the early warning model, and output the monitoring results, and complete the early warning of lupus nephritis of the monitored objects based on the monitoring results.

[0049] Specifically, in this embodiment, the monitoring data of the monitored object is collected by preset labels corresponding to the necessary data, which can improve the integrity and necessity of the monitoring data, wherein the monitored object refers to a patient who needs to be monitored for lupus nephritis. The monitoring data is input into the trained early warning model, and the early warning model can perform predictive analysis based on the data features contained in the monitoring data to output the monitoring results, wherein the monitoring results include the result categories and change sequences output after analyzing the monitoring data, and the result categories include but are not limited to the absence of lupus nephritis risk, early lupus nephritis, mid-stage lupus nephritis and late-stage lupus nephritis, and the change sequence refers to the time period corresponding to the potential change trend of lupus nephritis. The early warning function of lupus nephritis can be accurately completed by the monitoring results.

[0050] As a preferred technical solution of the present invention, the analysis information is output based on the following steps:

[0051] The analysis model includes a first sub-model and a second sub-model. If the data type of the learning data is an image type, the learning data is input into the first sub-model, the first sub-model arranges the pixels of the learning data to generate pixel information, extracts the feature points of the learning data, arranges all the feature points based on the corresponding frequencies to generate statistical information, combines the pixel information and the statistical information to generate image type information, outputs the feature vector corresponding to the image type information, and sets it as the analysis information.

[0052] Otherwise, the learning data is input into the second sub-model, the second sub-model splits the learning data into multiple sub-data, converts each character in the sub-data into character information, calculates the variance of the character information based on the dimension of the character information, calculates the ratio between the sum of all variances and the number of sub-data, sets the ratio as unit information, outputs the feature vector corresponding to the unit information, and sets it as analysis information.

[0053] Specifically, in this embodiment, in order to more accurately analyze the learning data corresponding to different data types, different types of analysis models are used, wherein the first sub-model is a neural network model for processing image data, including but not limited to GoogLeNet and ResNet. The second sub-model is a model composed of neural networks for processing non-image data, including but not limited to one-dimensional convolutional neural networks (1D CNN) and Word2Vec. If the data type of the learning data is an image type, the first sub-model is used to statistically generate pixel information after arranging the learning data by pixel values, wherein the pixel information is a histogram distribution of pixel values. Feature points refer to specific locations with unique attributes in the learning data, which can help distinguish different parts of the image corresponding to the learning data or establish matching relationships between different images. The feature points are arranged by the frequency of occurrence to generate statistical information, wherein the statistical information is a histogram distribution of the feature points. Image type information is generated after combining pixel information and statistical information, and the first sub-model then outputs the corresponding feature vector from the image type information, which can be set as the analysis information of the learning data. The analysis information can capture the intrinsic properties of the learning data comprehensively and at multiple levels.

[0054] If the learning data is not of image type, it may be data types such as text characters, table data, and voice text. In this case, the learning data is split into multiple sub-data according to the symbol interval. The character information refers to the vector representation corresponding to each character, where the vector representation refers to the vector of the text in the high-dimensional space, so that the meaning of each character is not represented by a single symbol or value, but is distributed in the entire vector space. First, the dimension of the character information is obtained, and for each dimension, the square of the deviation is calculated, and then the average of these square deviations is calculated to give the variance of each dimension. Then, all the variances are accumulated and divided by the number of sub-data to obtain the ratio, which is the unit variance vector of the character and is set as the unit information. The unit information is input into the second sub-model to output the corresponding feature vector and set as the analysis information of the learning data.

[0055] As a preferred technical solution of the present invention, the classification results are output based on the following steps:

[0056] A weight value is assigned to the analysis information, and all the analysis information is weighted and summed based on the weight value to generate combined information.

[0057] The target category is set based on the severity of lupus nephritis, the neural network model is trained based on the combined information, and the classification estimate corresponding to the combined information is output based on the target category prediction. All classification estimates are normalized and a probability value distribution is generated. The target category with the maximum probability is extracted from the probability value distribution and set as the prediction category corresponding to the combined information, and any combined information contained in the same prediction category is set as the classification result.

[0058] Specifically, in this embodiment, when assigning weight values ​​to analysis information, the chi-square test method can be used to evaluate the correlation between the features corresponding to the analysis information and the category label, and then all the analysis information is weighted and summed to generate combined information, wherein the chi-square test method is a statistical hypothesis test used to determine whether the difference between the observed data and the expected data is significant. It is often used to test whether two pieces of analysis information are independent, that is, to determine whether there is a correlation or relevance between them. The category label refers to the label information corresponding to the target category set in the next paragraph.

[0059] In clinical medicine, the severity of lupus nephritis can be divided into mild lupus nephritis, moderate lupus nephritis and severe lupus nephritis, which correspond to early lupus nephritis, middle lupus nephritis and late lupus nephritis respectively. Therefore, the target category can be set to three categories corresponding to the severity of lupus nephritis. All combined information is input into the neural network model for training, wherein all combined information contains combined information corresponding to different lupus nephritis patients. And based on the set target category, the combined information is classified and labeled, and the classification estimate is output, wherein the classification estimate indicates the degree of effect of the combined information in the corresponding target category.

[0060] Normalization can scale the classification estimate to a specific interval, revealing which features of the combined information have a greater impact on the classification of the neural network model, and thus assist in feature selection. Therefore, normalization can generate a probability value distribution corresponding to all classification estimates, and set the target category corresponding to the maximum probability as the prediction category corresponding to the combined information, where the prediction category is the classification prediction function achieved after continuous training of the neural network model. The combined information contained in the same prediction category has associated features, which are set as classification results.

[0061] As a preferred technical solution of the present invention, the neural network model is trained based on the combined information, comprising the following steps:

[0062] The combined information corresponding to the lupus nephritis patients is set as a standard label based on the severity of lupus nephritis, and the same standard labels are combined to generate a plurality of standard categories.

[0063] During the training process of the neural network model, the first similarity between the standard category and the predicted category corresponding to the severity of any lupus nephritis is calculated. If the first similarity is greater than or equal to a third preset value, the training of the neural network model is stopped. Otherwise, part of the combination information is eliminated based on a preset ratio, and the remaining combination information is input into the neural network model. If the first similarity shows an increasing trend, the remaining combination information is retained and the neural network model is trained. Otherwise, part of the combination information is eliminated again from all the combination information based on the preset ratio, and this step is repeated until the first similarity shows an increasing trend.

[0064] Specifically, in this embodiment, the quality of the training of the neural network model directly affects the accuracy of the classification results, so the reliability of the training of the neural network model needs to be verified. The standard label refers to the correct label information set according to the diagnosis results of the lupus nephritis patient, wherein the standard category corresponding to the lupus patient in the standard label is the same as the target category corresponding to the severity of lupus nephritis.

[0065] The first similarity refers to the degree of overlap between the classification result contained in the predicted category and the standard label of the combination information corresponding to the classification result. The first similarity P1 can be calculated using the first formula, where the first formula is: , where N1 is the number of identical combination information contained in the prediction category and the standard category, and N2 is the total number of corresponding combination information in the standard category. If the first similarity is less than the third preset value, it means that there is a large difference between the combination information contained in the prediction category and the combination information contained in the standard category, and the training effect of the neural network model is poor. Therefore, the accuracy of the neural network model can be improved by eliminating some combination information. For example, if the preset ratio is set to 5%, 5% of the partial combination information is extracted from all the combination information for elimination, and then 95% of the remaining combination information is input into the neural network model, and the first similarity is calculated through the above steps. If the first similarity before elimination is less than that after elimination, it means that the first similarity is increasing, and the neural network model is trained with the remaining combination information. Otherwise, the other 5% of partial combination information is extracted again. If the first similarities corresponding to all the extracted partial combination information do not show an increasing trend, the preset ratio is adjusted, and this step is repeated until the first similarity shows an increasing trend. For example, the preset ratio is adjusted from 5% to 10%. This process belongs to the process of model training verification, which can improve the training accuracy of the neural network model.

[0066] As a preferred technical solution of the present invention, the early warning model is generated based on the following steps:

[0067] The cascade model obtains the feature vector of the necessary data and sets it as the topic feature, performs training based on the topic feature, outputs the state estimate, obtains the loss function corresponding to the state estimate, adjusts the parameters of the cascade model until the loss function is reduced to the minimum value, and sets the adjusted cascade model as the early warning model.

[0068] Specifically, in this embodiment, the cascade model generates an early warning model after training. Similarly, the training process of the cascade model needs to be verified. Inputting necessary data into the cascade model can improve the speed and accuracy of model training and reduce unnecessary data for model training. The cascade model is generated by combining multiple analysis models and a neural network model. Therefore, the feature vector of the necessary data, that is, the topic feature, is extracted by the same method as above, and all the topic features are trained and learned. The state estimation refers to the possibility of the category of lupus nephritis in the topic feature, wherein the category of lupus nephritis refers to the target category. The loss function represents the difference between the category possibility estimated by the cascade model and the category corresponding to the necessary data. The parameters of the cascade model are optimized by minimizing the loss function, and therefore, the adjusted cascade model is set as the early warning model.

[0069] As a preferred technical solution of the present invention, the monitoring results are output based on the following steps:

[0070] The early warning model checks whether the monitoring data is missing based on the necessary labels. If so, the monitoring data is supplemented. If not, the monitoring data is converted into sequence data based on time information. The early warning model integrates all sequence data, extracts the first association sequence between the sequence data and the necessary data, and the second association sequence between the first association sequence and the state estimation based on the attention mechanism algorithm, obtains the association features of the second association sequence, obtains the estimation results of the association features based on the state estimation, and predicts and outputs the change sequence of the estimation results within the prediction time based on the second association sequence, and sets the change sequence as the monitoring result.

[0071] The estimated results, predicted time and monitoring results are transmitted to medical staff to complete the early warning of lupus nephritis in the monitored subjects.

[0072] Specifically, in this embodiment, checking whether the monitoring data is missing refers to data preprocessing before early warning analysis of the monitoring data. If missing, the monitoring data is supplemented, otherwise, subsequent processing is performed through the early warning model. Time information refers to the characteristics of the monitoring data corresponding to the time point, and sequence data refers to the monitoring data with time information being arranged in sequence, for example, the urine and other data of the monitored object are checked at different times, and the data is arranged based on the time information to generate sequence data corresponding to the urine examination. Integration processing refers to combining sequence data corresponding to different necessary labels. The early warning model is generated by training the necessary data, and the first associated sequence refers to a sequence with a higher attention score between the sequence data and the necessary data, wherein the attention score is generated by the attention mechanism algorithm, indicating the potential importance of the information of the sequence data to the elements contained in the target category, and the attention mechanism algorithm can help the early warning model understand the association between the sequence data corresponding to different modalities and the necessary data in multimodal learning. Using the same method, extract the sequence data with a higher attention score between the first associated sequence and the state estimate, that is, generate a second associated sequence. The associated feature refers to the feature vector of the second associated sequence, and the state estimate corresponding to the second associated sequence is estimated by the associated feature, and set as the estimation result. The estimation result refers to the score corresponding to each state estimation of the sequence data, which represents the size of the possibility. The change sequence of the estimation result refers to the change trend of the score corresponding to each state estimation of the sequence data within the prediction time. In this way, a monitoring result can be generated, in which the monitoring result not only includes whether the monitored object belongs to which state estimation, but also predicts the time change trend by the potential second associated sequence, so as to achieve the purpose of early warning of lupus nephritis.

[0073] As a preferred technical solution of the present invention, the second similarity between the monitoring data corresponding to any necessary label and the necessary data is calculated. If all the second similarities are greater than or equal to the fourth preset value, it is determined that the monitoring data is not missing, otherwise, it is determined that the monitoring data is missing.

[0074] Specifically, in this embodiment, the second similarity refers to the cosine similarity result between the monitoring data and the necessary data through the feature vector. If the second similarity corresponding to any necessary label is greater than or equal to the fourth preset value, it means that the monitoring data corresponding to the necessary label has a high consistency with the necessary data, that is, the monitoring data is not missing. Otherwise, it is determined that the monitoring data corresponding to a certain necessary label is missing.

[0075] As a preferred technical solution of the present invention, supplementing monitoring data includes the following steps:

[0076] Obtain a necessary label whose second similarity is less than a fourth preset value, and set it as a label to be supplemented. Based on the encoder, the encoder obtains the degree of fit between the label to be supplemented in the necessary data and the data corresponding to the remaining necessary labels. The encoder generates data corresponding to the label to be supplemented in the monitoring data based on the degree of fit, and supplements it into the monitoring data.

[0077] Specifically, in this embodiment, the monitoring data can be supplemented based on the degree of fit between the labels to be supplemented and the remaining necessary labels in the necessary data. Since the necessary data corresponding to the necessary labels are all training sets for training and generating early warning models, and there is a correlation in the diagnosis of lupus nephritis, the encoder can obtain the degree of fit between the data, and then reversely infer the data corresponding to the labels to be supplemented in the monitoring data. Among them, the encoder specifically refers to a neural network that converts input data (such as images, text or sound clips) into a vector of a fixed length, which captures the important features of the input data. The encoder itself does not directly calculate the degree of fit, but assists in calculating the degree of fit by providing a feature representation of the data. The degree of fit refers to the correlation between the data.

[0078] like Figure 2 As shown, the present invention also provides an early warning system for lupus nephritis, which is used to implement the above-mentioned early warning method for lupus nephritis. The system mainly includes:

[0079] The data collection module collects label data of lupus nephritis patients based on preset labels, combines all label data into a historical data set, obtains the data types contained in the historical data set, establishes multiple analysis models based on the data types, sets the label data corresponding to any data type as learning data, inputs the learning data into the analysis model corresponding to any data type, and outputs the analysis information.

[0080] The data classification module combines all analysis information corresponding to any lupus nephritis patient to generate combined information, constructs a neural network model, inputs all combined information into the neural network model, and outputs the classification result, calculates a first numerical value between any classification result and the combined information based on a first preset algorithm, and sets the classification result whose first numerical value is greater than a first preset value as the first category of information.

[0081] The model generation module calculates the second value between any first-category information and the learning data based on a second preset algorithm, sets the second value as the importance of the learning data, sets the learning data whose importance is greater than the second preset value as necessary data, combines all analysis models and neural network models and sets them as a cascade model, inputs all necessary data into the cascade model for training and learning, so as to generate an early warning model.

[0082] The early warning analysis module obtains the preset labels corresponding to the necessary data and sets them as necessary labels, collects the monitoring data of the monitored objects based on the necessary labels, inputs the monitoring data into the early warning model, and outputs the monitoring results, and completes the early warning of lupus nephritis of the monitored objects based on the monitoring results.

[0083] The present invention also provides a computer storage medium, which stores program instructions, wherein when the program instructions are executed, the device where the computer storage medium is located is controlled to execute the above-mentioned early warning method for lupus nephritis.

[0084] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0085] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the above-mentioned program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0086] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0087] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

[0088] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for early warning of lupus nephritis, characterized in that: The method comprises the following steps: Collect label data of lupus nephritis patients based on preset labels, combine all label data into a historical data set, obtain the data types contained in the historical data set, establish multiple analysis models based on the data types, set the label data corresponding to any data type as learning data, input the learning data into the analysis model corresponding to any data type, and output analysis information; The analysis information is output based on the following steps: The analysis model includes a first sub-model and a second sub-model. If the data type of the learning data is an image type, the learning data is input into the first sub-model, the first sub-model arranges pixels of the learning data to generate pixel information, extracts feature points of the learning data, arranges all the feature points based on corresponding frequencies to generate statistical information, combines the pixel information and the statistical information to generate image type information, outputs a feature vector corresponding to the image type information, and sets it as the analysis information; Otherwise, the learning data is input into the second sub-model, the second sub-model splits the learning data into multiple sub-data, converts each character in the sub-data into character information, calculates the variance of the character information based on the dimension of the character information, calculates the ratio between the sum of all variances and the number of the sub-data, sets the ratio as unit information, outputs a feature vector corresponding to the unit information, and sets it as the analysis information; Combining all analysis information corresponding to any lupus nephritis patient to generate combined information, constructing a neural network model, inputting all combined information into the neural network model, and outputting a classification result, calculating a first value between any classification result and the combined information based on a first preset algorithm, and setting the classification result whose first value is greater than a first preset value as first category information; The classification result is output based on the following steps: Assigning weight values ​​to the analysis information, and performing weighted summation on all the analysis information based on the weight values ​​to generate the combination information; A target category is set based on the severity of lupus nephritis, the neural network model is trained based on the combined information, and a classification estimate corresponding to the combined information is output based on the target category prediction, all the classification estimate values ​​are normalized, and a probability value distribution is generated, the target category with the maximum probability is extracted from the probability value distribution and set as the predicted category corresponding to the combined information, and the combined information contained in any of the same predicted categories is set as the classification result; it also includes: setting a standard label for the combined information corresponding to the lupus nephritis patient based on the severity of the lupus nephritis, and combining the same standard labels to generate multiple standard categories; During the training of the neural network model, a first similarity between the standard category and the predicted category corresponding to the severity of any lupus nephritis is calculated; if the first similarity is greater than or equal to a third preset value, the training of the neural network model is stopped; otherwise, part of the combination information is eliminated based on a preset ratio, and the remaining combination information is input into the neural network model; if the first similarity shows an increasing trend, the remaining combination information is retained and the neural network model is trained; otherwise, part of the combination information is eliminated again from all the combination information based on the preset ratio, and this step is repeated until the first similarity shows an increasing trend; Calculate a second value between any first type of information and the learning data based on a second preset algorithm, set the second value as the importance of the learning data, set the learning data whose importance is greater than the second preset value as necessary data, combine all the analysis models and the neural network model and set them as a cascade model, re-input all necessary data into the cascade model for training and learning, so as to generate an early warning model; The early warning model is generated based on the following steps: The cascade model obtains a feature vector of the necessary data and sets it as a topic feature, performs training based on the topic feature, outputs a state estimate, obtains a loss function corresponding to the state estimate, adjusts the parameters of the cascade model until the loss function is reduced to a minimum value, and sets the adjusted cascade model as the early warning model; The preset label corresponding to the necessary data is obtained and set as the necessary label, the monitoring data of the monitored object is collected based on the necessary label, the monitoring data is input into the early warning model, and the monitoring result is output, and the early warning of lupus nephritis of the monitored object is completed based on the monitoring result.

2. The method according to claim 1, characterized in that The monitoring results are output based on the following steps: The early warning model checks whether the monitoring data is missing based on the necessary labels. If so, the monitoring data is supplemented. If not, the monitoring data is converted into sequence data based on time information. The early warning model integrates all the sequence data, extracts a first association sequence between the sequence data and the necessary data, and a second association sequence between the first association sequence and the state estimation based on an attention mechanism algorithm, obtains association features of the second association sequence, obtains an estimation result of the association feature based on the state estimation, and predicts and outputs a change sequence of the estimation result within a prediction time based on the second association sequence, and sets the change sequence as the monitoring result; The estimation result, the predicted time and the monitoring result are transmitted to medical personnel to complete the early warning of lupus nephritis of the monitored object.

3. The method according to claim 2, characterized in that The second similarity between the monitoring data corresponding to any necessary tag and the necessary data is calculated. If all the second similarities are greater than or equal to a fourth preset value, it is determined that the monitoring data is not missing; otherwise, it is determined that the monitoring data is missing.

4. The method according to claim 3, characterized in that The supplementing of the monitoring data comprises the following steps: The necessary label whose second similarity is less than the fourth preset value is obtained, and is set as the label to be supplemented. Based on the encoder, the encoder obtains the degree of fit between the label to be supplemented in the necessary data and the data corresponding to the remaining necessary labels. The encoder generates the data corresponding to the label to be supplemented in the monitoring data based on the degree of fit, and supplements it into the monitoring data.

5. An early warning system for lupus nephritis, used to implement the method according to any one of claims 1 to 4, characterized in that: The system includes the following modules: The data acquisition module collects label data of lupus nephritis patients based on preset labels, combines all label data into a historical data set, obtains the data type contained in the historical data set, establishes multiple analysis models based on the data type, sets the label data corresponding to any data type as learning data, inputs the learning data into the analysis model corresponding to any data type, and outputs analysis information; the analysis model includes a first sub-model and a second sub-model, if the data type of the learning data is an image type, the learning data is input into the first sub-model, the first sub-model arranges the pixels of the learning data to generate pixel information, extracts the characteristics of the learning data feature points, arranging all the feature points based on corresponding frequencies to generate statistical information, combining the pixel information and the statistical information to generate image type information, outputting a feature vector corresponding to the image type information, and setting it as the analysis information; otherwise, inputting the learning data into the second sub-model, the second sub-model splits the learning data into multiple sub-data, converting each character in the sub-data into character information, calculating the variance of the character information based on the dimension of the character information, calculating the ratio between the sum of all variances and the number of the sub-data, setting the ratio as unit information, outputting a feature vector corresponding to the unit information, and setting it as the analysis information; a data classification module, combining all analysis information corresponding to any lupus nephritis patient to generate combined information, constructing a neural network model, inputting all combined information into the neural network model, and outputting a classification result, calculating a first value between any classification result and the combined information based on a first preset algorithm, and setting the classification result whose first value is greater than a first preset value as first category information; Assigning weight values ​​to the analysis information, and performing weighted summation on all the analysis information based on the weight values ​​to generate the combination information; The target category is set based on the severity of lupus nephritis, the neural network model is trained based on the combined information, and the classification estimation value corresponding to the combined information is output based on the target category prediction, all the classification estimation values ​​are normalized, and a probability value distribution is generated, the target category with the maximum probability is extracted from the probability value distribution and set as the prediction category corresponding to the combined information, and the combined information contained in any same prediction category is set as the classification result; and it also includes: setting a standard label for the combined information corresponding to the lupus nephritis patient based on the severity of the lupus nephritis, combining the same standard labels to generate multiple standard labels, and Quasi-category; during the training process of the neural network model, the first similarity between the standard category corresponding to the severity of any lupus nephritis and the predicted category is calculated; if the first similarity is greater than or equal to a third preset value, the training of the neural network model is stopped; otherwise, part of the combination information is eliminated based on a preset ratio, and the remaining combination information is input into the neural network model; if the first similarity shows an increasing trend, the remaining combination information is retained and the neural network model is trained; otherwise, part of the combination information is re-eliminated from all the combination information based on the preset ratio, and this step is repeated until the first similarity shows an increasing trend; A model generation module, which calculates a second value between any first-category information and the learning data based on a second preset algorithm, sets the second value as the importance of the learning data, sets the learning data whose importance is greater than the second preset value as necessary data, sets all the analysis models and the neural network models as a cascade model after combination, inputs all necessary data into the cascade model for training and learning to generate an early warning model; the cascade model obtains a feature vector of the necessary data and sets it as a topic feature, performs training based on the topic feature, outputs a state estimate, obtains a loss function corresponding to the state estimate, adjusts the parameters of the cascade model until the loss function is reduced to a minimum value, and sets the adjusted cascade model as the early warning model; The early warning analysis module obtains the preset label corresponding to the necessary data and sets it as the necessary label, collects the monitoring data of the monitored object based on the necessary label, inputs the monitoring data into the early warning model, and outputs the monitoring results, and completes the early warning of lupus nephritis of the monitored object based on the monitoring results.

6. A computer storage medium, characterized in that: The computer storage medium stores program instructions, wherein when the program instructions are executed, the device where the computer storage medium is located is controlled to execute the method according to any one of claims 1 to 4.

Images