Kidney disease risk prediction method and system based on machine learning and storage medium

Abstract

The invention relates to the technical field of combination of medical treatment and computers, and discloses a kidney disease risk prediction method and system based on machine learning and a storage medium, and the method comprises the steps: collecting basic data of historical patients, carrying out the preprocessing of the basic data, and adding a label to serve as first learning data; constructing a plurality of first prediction models based on machine learning, training the first prediction models by using the first learning data, obtaining first prediction data, and integrating the first prediction data to generate a first data set; adding the first data set into the first learning data to form second learning data; constructing a plurality of second prediction models, training the second prediction models by using the second learning data, and selecting a target prediction model from the second prediction data; the basic data of the new patient is obtained and input into the target prediction model, the target prediction model outputs the prediction value of the kidney disease risk of the new patient, and the kidney disease risk can be predicted more accurately.

Description

Technical Field

[0001] The present application relates to the technical field of the combination of medicine and computers, and in particular to a method, system and storage medium for predicting kidney disease risk based on machine learning. Background Art

[0002] In today's medical field, early diagnosis and risk prediction of kidney disease are crucial. However, traditional methods for predicting kidney disease risk have many limitations. At present, common prediction methods mainly rely on the clinical experience of doctors and some basic laboratory test indicators, such as blood creatinine and urea nitrogen. Although these methods can provide some references to a certain extent, their accuracy is often unsatisfactory.

[0003] Similar prior art includes a Chinese patent application with publication number CN117542504A, which discloses a method for constructing a chronic kidney disease prediction model, a prediction system and an application thereof, including data acquisition, data screening and model construction. However, the invention only uses decision tree algorithms, generalized linear models and the like as auxiliary means to self-evaluate the future risk of chronic kidney disease through simple calculations, and the accuracy of kidney disease risk prediction is not high.

[0004] Similar prior art also includes a Chinese patent application with publication number CN117711619A, which discloses a system for predicting the risk of chronic kidney disease in diabetic patients, including obtaining the protein level related to the risk of chronic kidney disease in the blood of diabetic patients and the inducing factors of chronic kidney disease; constructing a protein scoring model for diabetic patients, and using the protein scoring model for diabetic patients to obtain the protein scoring results of diabetic patients; obtaining the chronic kidney disease risk index of diabetic patients based on the protein scoring results and the inducing factors of chronic kidney disease; and analyzing the risk of chronic kidney disease in diabetic patients according to the chronic kidney disease risk index. However, this invention can only predict the risk of chronic kidney disease in diabetic patients, and its application scenarios are relatively limited. Summary of the invention

[0005] To solve the above problems, the present application provides a method, system and storage medium for predicting kidney disease risk based on machine learning. By using multiple machine learning algorithms and secondary learning, the risk of kidney disease can be predicted more accurately.

[0006] To achieve the above-mentioned purpose of the invention, in a first aspect, the present application provides a method for predicting kidney disease risk based on machine learning, and the method is implemented by executing the following method, which includes:

[0007] Step S1: Collect basic data of historical patients, including individual characteristic information, examination data and doctor's preliminary diagnosis information;

[0008] Step S2: preprocessing the basic data, and obtaining the basic data after preprocessing, adding a label to each piece of the basic data based on whether the historical patient is finally diagnosed with kidney disease, and using the basic data with the label added as the first learning data;

[0009] Step S3: constructing a plurality of first prediction models based on machine learning, using the first learning data to train each of the first prediction models, and obtaining a plurality of trained first prediction models, wherein each of the first prediction models adopts a different machine learning algorithm;

[0010] Step S4: inputting the first learning data into each of the first prediction models respectively, obtaining the first prediction data output by each of the first prediction models, and integrating the plurality of the first prediction data together to generate a first data set;

[0011] Step S5: adjusting the first learning data based on the first data set, acquiring the adjusted first learning data, and adding the first data set to the first learning data to form second learning data;

[0012] Step S6: construct multiple second prediction models based on machine learning, use the second learning data to train each of the second prediction models, obtain the second prediction data output by each of the second prediction models, and select the second prediction model corresponding to the second prediction data with the highest accuracy from the second prediction data based on the evaluation index as the target prediction model;

[0013] Step S7: Obtain basic data of the new patient and input it into the target prediction model, and the target prediction model outputs a predicted value of the risk of the new patient suffering from kidney disease.

[0014] In combination with the first aspect, in a first implementation of the first aspect of the present application, after step S2 and before step S3 is performed, the method further includes:

[0015] The data volume of the first learning data is calculated, and when the data volume of the first learning data is less than a first threshold, the data volume of the first learning data is increased by using a data enhancement technology.

[0016] In combination with the first aspect, in a second implementation manner of the first aspect of the present application, adjusting the first learning data based on the first data set includes:

[0017] The first learning data in which each of the first prediction data is smaller than the second threshold and the corresponding data label is confirmed is obtained from the first data set as abnormal data, and the abnormal data is checked and processed differently.

[0018] In combination with the first aspect, in a third implementation of the first aspect of the present application, checking the abnormal data and performing different processing includes:

[0019] Verify whether the labeling of the abnormal data is correct, and if the labeling of the abnormal data is incorrect, re-verify the labeling result of the abnormal data;

[0020] Furthermore, when the number of missing items in the basic data corresponding to the abnormal data is greater than or equal to N, the missing items in the basic data corresponding to the abnormal data are processed, where N represents a positive integer greater than or equal to 3.

[0021] In combination with the first aspect, in a fourth implementation of the first aspect of the present application, checking the abnormal data and performing different processing further includes:

[0022] Calculate the frequency of occurrence of the eigenvalue of each of the basic data in the abnormal data; when the frequency of occurrence of a certain eigenvalue is greater than a third threshold and the frequency of occurrence of other eigenvalues ​​is less than a fourth threshold, obtain abnormal data whose frequency of occurrence of a certain eigenvalue is less than or equal to the third threshold and whose frequency of occurrence of other eigenvalues ​​is greater than or equal to the fourth threshold as abnormal learning data, use generative adversarial network or variational autoencoder technology to generate synthetic data with a distribution similar to that of the abnormal learning data, and add the synthetic data to the first learning data.

[0023] In combination with the first aspect, in a fifth implementation of the first aspect of the present application, processing the missing items in the basic data corresponding to the abnormal data includes:

[0024] Obtain missing items in the basic data corresponding to the abnormal data. When the correlation between the missing items and the prediction target is greater than A1, fill the missing items with a data filling method. When the correlation between the missing items and the prediction target is less than A2, delete the abnormal data from the first learning data, wherein the correlation with the prediction target refers to the data items that directly affect the prediction target, A1 is greater than A2, A1 represents a number greater than or equal to 60%, and A2 represents a number less than or equal to 10%.

[0025] In combination with the first aspect, in the sixth implementation method of the first aspect of the present application, the individual characteristic information includes age, gender, time of consultation, living habits, and past medical history; the examination data includes renal function indicators, urine routine indicators, blood routine, liver function, blood lipids, immunology, imaging data, renal puncture data, glomerular filtration rate and tubular function data; and the living habits include smoking history and drinking history.

[0026] In combination with the first aspect, in the seventh implementation method of the first aspect of the present application, the machine learning algorithm includes at least logistic regression, decision tree, random forest, support vector machine, neural network, naive Bayes algorithm, Adaboost algorithm and gradient boosting tree.

[0027] In a second aspect, the present application provides a kidney disease risk prediction system based on machine learning, the system comprising the following modules:

[0028] A data collection unit, used to collect basic data of historical patients, the basic data including individual characteristic information, examination data and preliminary diagnosis information of doctors;

[0029] a first learning data generating unit, configured to preprocess the basic data, obtain the basic data after preprocessing, add a label to each piece of the basic data based on the result of whether the historical patient is finally diagnosed with kidney disease, and use the basic data with the label added as the first learning data;

[0030] a first training unit, configured to construct a plurality of first prediction models based on machine learning, train each of the first prediction models using the first learning data, and obtain a plurality of trained first prediction models, wherein each of the first prediction models adopts a different machine learning algorithm;

[0031] a data set integration unit, configured to input the first learning data into each of the first prediction models, obtain first prediction data output by each of the first prediction models, and integrate a plurality of the first prediction data together to generate a first data set;

[0032] a second learning data generating unit, configured to adjust the first learning data based on the first data set, obtain the adjusted first learning data, and add the first data set to the first learning data to form second learning data;

[0033] A second training unit is used to build multiple second prediction models based on machine learning, use the second learning data to train each of the second prediction models respectively, obtain the second prediction data output by each of the second prediction models, and select the second prediction model corresponding to the second prediction data with the highest accuracy from the second prediction data based on the evaluation index as the target prediction model;

[0034] The result prediction unit is used to obtain the basic data of the new patient and input it into the target prediction model, and the target prediction model outputs the predicted value of the risk of the new patient suffering from kidney disease.

[0035] The third aspect of the present application provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is executed on a computer, the computer executes the above-mentioned method for predicting kidney disease risk based on machine learning.

[0036] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0037] In the technical solution provided by the present application, by collecting the basic data of historical patients, rich data support is provided for the training of subsequent prediction models. And according to whether the patient is finally diagnosed with kidney disease, each piece of data is labeled to form the first learning data, which lays an accurate data foundation for the training of the prediction model. Use machine learning technology to build multiple first prediction models using different algorithms, and use the first learning data for training. Through the application of diversified algorithms, it is possible to explore the potential impact of different modeling methods on the prediction results, increasing the diversity of model selection. The first learning data is input into each trained first prediction model, the first prediction data output by each is obtained, and these data are integrated to form a first data set. It is helpful to comprehensively evaluate the prediction ability of different models and provide a basis for subsequent data adjustment and prediction model optimization. Adjust the first learning data based on the first data set, optimize the data distribution and feature expression, and then merge the first data set with the adjusted first learning data to form a richer and more complete second learning data, so as to further improve the accuracy and generalization ability of the prediction model. The second learning data is used to train multiple second prediction models, and the model with the highest accuracy is selected as the target prediction model based on the evaluation index. The second prediction model is further trained and optimized on the prediction results of multiple first prediction models, and can learn more complex and accurate prediction patterns, thereby further improving the accuracy of kidney disease risk prediction. Finally, the basic data of the new patient is input into the target prediction model, and the model outputs the predicted value of the new patient's kidney disease risk, realizing the practical application of the model; through the coordination between the above steps, using a variety of machine learning algorithms and secondary learning, the risk of kidney disease can be predicted more accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0039] Figure 1 This is a schematic diagram of an embodiment of a method for predicting kidney disease risk based on machine learning in an embodiment of the present application;

[0040] Figure 2The medical record data of patients with a history of kidney disease in the embodiment of this application;

[0041] Figure 3 The medical record data of patients with a history of kidney disease that have been preprocessed and labeled in the embodiment of the present application;

[0042] Figure 4 The first learning data in the embodiment of the present application;

[0043] Figure 5 A first data set obtained by integrating the output results of multiple first prediction models in the embodiment of the present application;

[0044] Figure 6 The second learning data in the embodiment of the present application;

[0045] Figure 7 This is a schematic diagram of an embodiment of a kidney disease risk prediction system based on machine learning in an embodiment of the present application. DETAILED DESCRIPTION

[0046] Embodiments of the present application provide a method, system and storage medium for predicting kidney disease risk based on machine learning. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or precedence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than the content illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, processes, methods, systems, products or equipment comprising a series of steps or units are not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0047] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiments of the present application, an embodiment of the method for predicting kidney disease risk based on machine learning includes:

[0048] Step S1: Collect basic data of historical patients, including individual characteristic information, examination data and doctor's preliminary diagnosis information;

[0049] It is understandable that the execution subject of the present application can be a kidney disease risk prediction system based on machine learning, or a terminal or a server, which is not limited here. The present application embodiment is described by taking the server as the execution subject as an example.

[0050] Specifically, medical record data of patients with a history of kidney disease can be collected from the hospital's electronic medical record system to establish a database to provide a data basis for subsequent data analysis and model training. Figure 2 As shown, it includes the patient's age, gender, lifestyle habits, etc. Lifestyle habits include smoking, drinking, exercise habits (for example, no smoking, moderate drinking, and exercise 3 times a week), past medical history such as diabetes, hypertension or cardiovascular disease, examination data: such as serum creatinine, urine protein, blood sugar level, blood pressure and other laboratory test results, etc. At the same time, the doctor's preliminary diagnosis information of these patients is also collected, such as the preliminary observations recorded by the doctor, such as "no obvious kidney disease" or "consider chronic kidney disease."

[0051] Step S2: preprocessing the basic data, and obtaining the preprocessed basic data, adding a label to each basic data based on the historical result of whether the patient was finally diagnosed with kidney disease, and using the labeled basic data as the first learning data;

[0052] Specifically, the collected basic data is cleaned to remove missing values, outliers or duplicate data to improve the quality of the basic data and reduce the impact of noise on model training. For example, the missing data rows are filled with the mean or deleted. At the same time, the basic data is standardized or normalized to ensure that the dimensions of different features are consistent. After the preprocessing is completed, labels are added, such as Figure 3 As shown in the figure, a label is added to each piece of data according to whether the patient is finally diagnosed with kidney disease (such as 1 for confirmed and 0 for undiagnosed). The addition of labels allows the basic data to be used for supervised learning, providing a clear goal for model training.

[0053] Step S3: constructing a plurality of first prediction models based on machine learning, training each first prediction model using the first learning data, and obtaining a plurality of trained first prediction models, wherein each first prediction model adopts a different machine learning algorithm;

[0054] Specifically, a variety of machine learning algorithms such as logistic regression, support vector machine, decision tree, random forest, gradient boosting tree, etc. are used to build the first prediction model, and these models are trained respectively using the preprocessed first learning data. By comparing multiple algorithms, we can initially understand which algorithm is more suitable for processing the above data, which provides a reference for subsequent model selection.

[0055] Step S4: inputting the first learning data into each first prediction model respectively, obtaining the first prediction data output by each first prediction model, and integrating the plurality of first prediction data together to generate a first data set;

[0056] Specifically, the first learning data is input into each trained first prediction model, the prediction result of each model is obtained, that is, the first prediction data, and the output results of each first prediction model are combined to form a comprehensive prediction data set, that is, the first data set. Specifically, each model will output a prediction result for each sample, usually the probability of predicting that the sample belongs to kidney disease, and then the prediction probability of each model is integrated into a data set to form a new feature set for subsequent learning. For example, we have trained three models, namely logistic regression, random forest and support vector machine (SVM). For a group of patients (for example, 5 patients), the output of each model may be as follows: The probability predicted by the logistic regression model (the probability of suffering from kidney disease): Patient 1: 0.1; Patient 2: 0.8; Patient 3: 0.6; Patient 4: 0.3; Patient 5: 0.9. The probability predicted by the random forest model: Patient 1: 0.2; Patient 2: 0.7; Patient 3: 0.5; Patient 4: 0.4; Patient 5: 0.85. The probabilities predicted by the support vector machine model are: Patient 1: 0.15; Patient 2: 0.75; Patient 3: 0.55; Patient 4: 0.35; Patient 5: 0.88. These outputs are combined into a new dataset, with each column representing the predicted probability of a model and each row representing a different patient, as shown in the following table:

[0057]

[0058]

[0059] After the prediction data is integrated, these new features (prediction probabilities) can be used as input data to train the second layer model (i.e., step S6). For example, these integrated features can be combined with the original basic data (original first learning data) to form the final learning data set. In this way, the new model will be able to be trained based on the prediction results of multiple models, thereby improving the accuracy of the overall prediction.

[0060] Step S5: adjusting the first learning data based on the first data set, acquiring the adjusted first learning data, and adding the first data set to the first learning data to form second learning data;

[0061] Specifically, the first learning data is adjusted based on the first data set to further improve the quality of the data and the generalization ability of the model. The specific adjustment method will be described in detail below. Figure 5 As shown, the new features are added to the adjusted first learning data as Figure 4 As shown, the second learning data is formed, such as Figure 6 As shown, for each sample data in the first learning data, the output probability of each of the above models is added as a feature.

[0062] Step S6: construct multiple second prediction models based on machine learning, use the second learning data to train each second prediction model respectively, obtain the second prediction data output by each second prediction model, and select the second prediction model corresponding to the second prediction data with the highest accuracy from the second prediction data based on the evaluation index as the target prediction model.

[0063] Specifically, the second prediction model is constructed using the same machine learning algorithm as step S3, the models are trained respectively using the second learning data, and the prediction results of each model, i.e., the second prediction data, are obtained, and the model with the best performance is selected as the target prediction model according to evaluation indicators such as accuracy, recall rate, and F1 score. Since the second prediction model is trained based on the prediction results of multiple first prediction models, the model with the best performance selected from the second prediction model, i.e., the target prediction model, can improve the accuracy of the overall prediction.

[0064] Step S7: Obtain basic data of the new patient and input it into the target prediction model, and the target prediction model outputs a predicted value of the new patient's risk of kidney disease.

[0065] Finally, when the basic data of the new patient is obtained, the basic data of the new patient is input into the target prediction model. The output result of the target prediction model can directly obtain the predicted value of the patient's risk of kidney disease, where the predicted value is a probability value. Through the prediction of the machine learning model, early detection and intervention of kidney disease risk can be achieved, improving the treatment effect and quality of life of patients.

[0066] The embodiment of the present application uses a variety of machine learning algorithms and secondary learning through the coordination of the above steps to more accurately predict the risk of kidney disease.

[0067] In a specific embodiment, after step S2 and before step S3 is performed, the method further includes:

[0068] The data volume of the first learning data is calculated, and when the data volume of the first learning data is less than a first threshold, the data volume of the first learning data is increased by using a data enhancement technology.

[0069] Specifically, when the amount of training data is insufficient, the prediction model can easily overfit to a specific pattern of the training set, so it is necessary to reduce the risk of overfitting by increasing the number and diversity of training samples. First, the amount of data of the first learning data is calculated, which can be done by counting the number of samples in the statistical data set. Then set a first threshold, which is determined based on the needs of model training and the adequacy of the amount of data. If the amount of data is less than this threshold, the amount of data is considered insufficient. When the amount of data of the first learning data is less than the first threshold, the data enhancement technology is started. Data enhancement technology can include a variety of methods. In the scenario of kidney disease risk prediction, if the first learning data is a medical image (such as a CT or MRI image of the kidney), an image data enhancement method can be used; if the data is a patient's clinical indicator or text record, a non-image data enhancement method may be required. The enhanced data is merged with the original basic data to form a new, larger data set. Because data enhancement increases the number and diversity of training samples, the prediction model can learn more features about kidney disease risk. It helps to improve the accuracy of the model in predicting kidney disease risk.

[0070] In a specific embodiment, adjusting the first learning data based on the first data set includes:

[0071] The first learning data in which each first prediction data is smaller than the second threshold value and the label of the corresponding data is confirmed is obtained from the first data set as abnormal data, and the abnormal data is checked and processed differently.

[0072] Specifically, after obtaining the first data set, that is, the set of first prediction data output by each first prediction model, we set a second threshold to define which prediction results are abnormal. The second threshold is usually determined based on actual clinical conditions, expert experience or statistical methods, and aims to identify samples whose prediction results significantly deviate from the normal range.

[0073] Next, we traverse the first learning data, and for each piece of data, check all the first prediction data corresponding to different prediction models in the first data set. If all the first prediction data are less than the second threshold and the label of the corresponding data is still confirmed, that is, the result predicted by each prediction model is less than the second threshold, and the label corresponding to the data is still confirmed, we regard it as abnormal data. Abnormal data may represent some special situations, such as some characteristics of the patient are within the normal range, but it is difficult to accurately predict their kidney disease risk based on the existing model.

[0074] For the identified abnormal data, a detailed examination is carried out, including reviewing the original medical records of the patient, confirming the accuracy of the data, and exploring the reasons that may lead to abnormal prediction results. During the inspection process, some data errors or omissions may be found, such as input errors, incomplete examination items, etc., which need to be corrected or supplemented. At the same time, different processing strategies are adopted for different abnormal data. In the case of data errors or omissions, corrections or supplements will be made to ensure the accuracy and completeness of the data. For cases where the model is difficult to predict accurately, attempts may be made to use more complex model structures, increase the number of features, or perform feature engineering to improve the model's predictive ability. In addition, these abnormal data may be studied in depth as special cases to discover new disease characteristics or prediction patterns.

[0075] The above steps firstly improve the accuracy and reliability of data by identifying and processing abnormal data, thus providing a better foundation for subsequent model training and prediction. Secondly, for situations where the model is difficult to predict, targeted measures are taken to improve the model's predictive ability, thereby improving the accuracy of the entire prediction method.

[0076] In a specific embodiment, checking abnormal data and performing different processing include:

[0077] Verify whether the labeling of abnormal data is correct, and re-verify the labeling results of abnormal data if the labeling of abnormal data is incorrect;

[0078] When the number of missing items in the basic data corresponding to the abnormal data is greater than or equal to N, the missing items in the basic data corresponding to the abnormal data are processed, where N represents a positive integer greater than or equal to 3.

[0079] Specifically, first of all, for the identified abnormal data, it is necessary to verify whether its labeling is correct. Labeling refers to the classification or labeling of data, which reflects the characteristics and attributes of the data. In kidney disease risk prediction, labeling is determined based on the patient's diagnosis results. If the labeling of abnormal data is incorrect, the subsequent model training and prediction will be affected. Therefore, it is necessary to carefully verify the labeling results of abnormal data to ensure its accuracy. If labeling errors are found, they need to be re-verified and corrected to ensure the authenticity and reliability of the data.

[0080] Secondly, the missing items in the basic data corresponding to the abnormal data also need to be processed. Basic data refers to all the original data used to build the prediction model, which includes the patient's age, gender, medical history, examination results and other information. During the data collection process, there may be some missing items in the basic data due to various reasons. If the number of missing items in the basic data corresponding to the abnormal data is greater than or equal to N (N represents a positive integer greater than or equal to 3), then these missing items may have a negative impact on the prediction ability of the model. Therefore, these missing items need to be processed. Commonly used methods include filling missing values ​​and deleting samples with missing values. Filling missing values ​​can be done through interpolation, mean method, mode method and other methods, while deleting samples with missing values ​​may result in a reduction in the amount of data and affect the generalization ability of the model. Therefore, in actual operation, it is necessary to choose an appropriate method to handle missing items according to the specific situation.

[0081] By implementing this step, the accuracy and completeness of the data can be further improved, providing a better foundation for subsequent model training and prediction. At the same time, it is also helpful to discover problems and deficiencies in the data, providing improvement directions for subsequent data collection and processing. Ultimately, the accuracy of kidney disease risk prediction methods can be improved, providing patients with better medical services and health protection.

[0082] In a specific embodiment, checking abnormal data and performing different processing further includes:

[0083] Calculate the frequency of occurrence of the eigenvalue of each basic data in the abnormal data. When the occurrence frequency of a certain eigenvalue is greater than the third threshold and the occurrence frequency of other eigenvalues ​​is less than the fourth threshold, obtain abnormal data whose occurrence frequency of a certain eigenvalue is less than or equal to the third threshold and the occurrence frequency of other eigenvalues ​​is greater than or equal to the fourth threshold as abnormal learning data, use generative adversarial network or variational autoencoder technology to generate synthetic data with similar distribution to the abnormal learning data, and add the synthetic data to the first learning data.

[0084] Specifically, this step first calculates the frequency of occurrence of the eigenvalue of each basic data in the abnormal data. Eigenvalue refers to a numerical value or category that can describe the data attribute, such as the patient's age, blood pressure, urine protein content, etc. In the abnormal data, some eigenvalues ​​may appear at a higher frequency, while other eigenvalues ​​may appear less frequently. When the frequency of occurrence of a certain eigenvalue is greater than the set third threshold, and the frequency of occurrence of other eigenvalues ​​is less than the set fourth threshold, this indicates that the eigenvalue occupies a dominant position in the abnormal data, while other eigenvalues ​​are relatively weak.

[0085] Next, based on the comparison results of these eigenvalue frequencies, the abnormal data with a certain eigenvalue frequency less than or equal to the third threshold and other eigenvalue frequencies greater than or equal to the fourth threshold are screened out, and these data are called abnormal learning data. These abnormal learning data have certain uniqueness in the eigenvalue distribution and can reflect certain specific patterns or laws in the abnormal data.

[0086] In order to enhance the learning dataset, the Generative Adversarial Network (GAN) or Variational Autoencoder (VAE) technology is used to generate synthetic data that is similar to the distribution of abnormal learning data. GAN and VAE are two commonly used generative models that can learn the distribution characteristics of data from input data and generate new data similar to the input data. By adding the generated synthetic data to the original first learning data, the diversity and complexity of the learning dataset can be enriched, and the generalization ability and prediction accuracy of the model can be improved.

[0087] Through the above steps, we can first locate specific patterns or regularities in abnormal data more accurately, providing more valuable information for subsequent data processing and model training. Secondly, using GAN or VAE technology to generate synthetic data not only solves the problem of insufficient abnormal data, but also enhances the diversity and complexity of the learning data set, which helps to improve the predictive performance and robustness of the model. Ultimately, the accuracy of the kidney disease risk prediction method is improved.

[0088] In a specific embodiment, processing missing items in basic data corresponding to abnormal data includes:

[0089] Obtain the missing items in the basic data corresponding to the abnormal data. When the correlation between the missing items and the prediction target is greater than A1, use the data filling method to fill the missing items. When the correlation between the missing items and the prediction target is less than A2, delete the abnormal data from the first learning data, where the correlation with the prediction target refers to the data items that directly affect the prediction target, A1 is greater than A2, A1 represents a number greater than or equal to 60%, and A2 represents a number less than or equal to 10%.

[0090] Specifically, we first identify and obtain the missing items in the basic data corresponding to the abnormal data. The missing items may be caused by omissions in the data collection process, equipment failure or other reasons. Then, we evaluate the correlation between these missing items and the prediction target (i.e., kidney disease risk). The so-called prediction target correlation refers to the degree of direct impact of a data item on the prediction target.

[0091] For missing items whose correlation with the prediction target is greater than A1 (A1 represents a threshold greater than or equal to 60%), data filling methods are used for processing. Data filling methods include using statistical values ​​such as mean, median, mode, etc. to fill in, or using more complex algorithms such as regression and interpolation to estimate missing values. Since these missing items are highly correlated with the prediction target, their accurate filling is crucial to improving the accuracy of the prediction model.

[0092] However, for missing items whose correlation with the prediction target is less than A2 (A2 represents a threshold less than or equal to 10%, and A2 is less than A1), the strategy of removing abnormal data containing these missing items from the first learning data is adopted. Because these missing items have little impact on the prediction target, even if they are missing, they will not have a significant impact on the prediction results. At the same time, retaining these data containing low-correlation missing items may introduce noise and reduce the performance of the model.

[0093] By processing missing items in a targeted manner, we can retain key information that is highly relevant to the prediction target while avoiding interference from low-relevance information. This helps to improve the accuracy and stability of the prediction model, making the machine learning-based kidney disease risk prediction method more reliable and effective. In addition, this step also reflects the flexibility and targeting in data preprocessing, and can formulate appropriate data processing strategies based on the specific situation of the data and the needs of the prediction target, thereby improving the prediction accuracy of the prediction model.

[0094] In a specific embodiment, individual characteristic information includes age, gender, time of consultation, lifestyle, and past medical history; examination data includes renal function indicators, urine routine indicators, blood routine, liver function, blood lipids, immunology, imaging data, renal puncture data, glomerular filtration rate and tubular function data; and lifestyle habits include smoking history and drinking history.

[0095] Specifically, individual characteristic information covers many aspects of the patient, including age, gender, height, weight, time of consultation, lifestyle and medical history. It is crucial to assess the patient's risk of kidney disease. For example, age and gender are important influencing factors for the onset of kidney disease, and patients of different ages and genders may have different susceptibility to kidney disease. The time of consultation can reflect the timeliness of the patient's medical treatment and the severity of the disease. Lifestyle habits, especially smoking and drinking history, are important risk factors for the onset of kidney disease, which may directly affect the function and structure of the kidneys. Past medical history provides the patient's past health status and disease experience, which helps to identify potential risk factors and predict the development trend of the disease.

[0096] The examination data reflects the patient's kidney function and health status in more detail. It includes renal function indicators (such as creatinine, urea nitrogen, etc.), urine routine indicators (such as urine protein, urine red blood cells, etc.), blood routine, liver function, blood lipids, immunological indicators (such as antinuclear antibodies, complement, etc.), imaging data (such as B-ultrasound, CT, etc.), renal puncture data (such as pathological type, degree of lesions, etc.), glomerular filtration rate and renal tubular function data, etc. The examination data provides detailed information on the patient's kidney structure and function, and is an important basis for assessing the risk of kidney disease and formulating treatment plans.

[0097] When processing these data, data cleaning and preprocessing are first required, including removing outliers, filling missing values, and standardization to ensure data accuracy and consistency. Then, a prediction model is constructed using a machine learning algorithm, with individual characteristic information and examination data as input features and kidney disease risk as the prediction target. By training and optimizing the model, accurate prediction of kidney disease risk for new patients can be achieved.

[0098] By comprehensively considering the patient's individual characteristic information and examination data, we can more comprehensively assess the patient's kidney disease risk and improve the accuracy of the prediction.

[0099] In a specific embodiment, the machine learning algorithm includes at least logistic regression, decision tree, random forest, support vector machine, neural network, naive Bayes algorithm, Adaboost algorithm and gradient boosting tree.

[0100] In practical applications, appropriate machine learning algorithms can be selected according to the specific situation of the data and the needs of the prediction target. At the same time, multiple algorithms can be combined and integrated to further improve the accuracy and stability of the prediction. The application of these algorithms will provide more comprehensive and reliable support for kidney disease risk prediction. The principles of each algorithm belong to the existing technology and will not be repeated here.

[0101] In the technical solution provided by the present application, by collecting the basic data of historical patients, rich data support is provided for the training of subsequent prediction models. And according to whether the patient is finally diagnosed with kidney disease, each piece of data is labeled to form the first learning data, which lays an accurate data foundation for the training of the prediction model. Use machine learning technology to build multiple first prediction models using different algorithms, and use the first learning data for training. Through the application of diversified algorithms, it is possible to explore the potential impact of different modeling methods on the prediction results, increasing the diversity of model selection. The first learning data is input into each trained first prediction model, the first prediction data output by each is obtained, and these data are integrated to form a first data set. It is helpful to comprehensively evaluate the prediction ability of different models and provide a basis for subsequent data adjustment and prediction model optimization. Adjust the first learning data based on the first data set, optimize the data distribution and feature expression, and then merge the first data set with the adjusted first learning data to form a richer and more complete second learning data, so as to further improve the accuracy and generalization ability of the prediction model. The second learning data is used to train multiple second prediction models, and the model with the highest accuracy is selected as the target prediction model based on the evaluation index. The second prediction model is further trained and optimized on the prediction results of multiple first prediction models, and can learn more complex and accurate prediction patterns, thereby further improving the accuracy of kidney disease risk prediction. Finally, the basic data of the new patient is input into the target prediction model, and the model outputs the predicted value of the new patient's kidney disease risk, realizing the practical application of the model; through the coordination between the above steps, using a variety of machine learning algorithms and secondary learning, the risk of kidney disease can be predicted more accurately.

[0102] The above describes the kidney disease risk prediction method based on machine learning in the embodiment of the present application. The following describes the kidney disease risk prediction system based on machine learning in the embodiment of the present application. Figure 7 In the embodiments of the present application, an embodiment of a kidney disease risk prediction system based on machine learning includes:

[0103] A data collection unit is used to collect basic data of historical patients, including individual characteristic information, examination data, and preliminary diagnosis information of doctors;

[0104] A first learning data generating unit is used to preprocess the basic data and obtain the preprocessed basic data, add a label to each basic data based on the historical result of whether the patient is finally diagnosed with kidney disease, and use the labeled basic data as the first learning data;

[0105] A first training unit is used to construct a plurality of first prediction models based on machine learning, train each first prediction model using the first learning data, and obtain a plurality of trained first prediction models, wherein each first prediction model adopts a different machine learning algorithm;

[0106] A data set integration unit, used to input the first learning data into each first prediction model, obtain the first prediction data output by each first prediction model, and integrate the plurality of first prediction data together to generate a first data set;

[0107] A second learning data generating unit, configured to adjust the first learning data based on the first data set, obtain the adjusted first learning data, and add the first data set to the first learning data to form the second learning data;

[0108] A second training unit is used to construct a plurality of second prediction models based on machine learning, train each second prediction model respectively using the second learning data, obtain the second prediction data output by each second prediction model, and select the second prediction model corresponding to the second prediction data with the highest accuracy from the second prediction data as the target prediction model based on the evaluation index;

[0109] The result prediction unit is used to obtain the basic data of the new patient and input it into the target prediction model, and the target prediction model outputs the predicted value of the risk of kidney disease in the new patient.

[0110] Through the coordinated cooperation of the above components, accurate prediction of kidney disease risk is achieved, and this method also provides strong support for clinical decision-making.

[0111] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the above-mentioned method for predicting kidney disease risk based on machine learning.

[0112] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0113] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0114] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting kidney disease risk based on machine learning, characterized in that: The method comprises: Step S1: Collect basic data of historical patients, including individual characteristic information, examination data and doctor's preliminary diagnosis information; Step S2: preprocessing the basic data, and obtaining the basic data after preprocessing, adding a label to each piece of the basic data based on whether the historical patient is finally diagnosed with kidney disease, and using the basic data with the label added as the first learning data; Step S3: constructing a plurality of first prediction models based on machine learning, using the first learning data to train each of the first prediction models, and obtaining a plurality of trained first prediction models, wherein each of the first prediction models adopts a different machine learning algorithm; Step S4: inputting the first learning data into each of the first prediction models respectively, obtaining the first prediction data output by each of the first prediction models, and integrating the plurality of the first prediction data together to generate a first data set; Step S5: adjusting the first learning data based on the first data set, acquiring the adjusted first learning data, and adding the first data set to the first learning data to form second learning data; Step S6: construct multiple second prediction models based on machine learning, use the second learning data to train each of the second prediction models, obtain the second prediction data output by each of the second prediction models, and select the second prediction model corresponding to the second prediction data with the highest accuracy from the second prediction data based on the evaluation index as the target prediction model; Step S7: Obtain basic data of the new patient and input it into the target prediction model, and the target prediction model outputs a predicted value of the risk of the new patient suffering from kidney disease.

2. The method according to claim 1, characterized in that After step S2, and before step S3 is performed, the method further includes: The data volume of the first learning data is calculated, and when the data volume of the first learning data is less than a first threshold, the data volume of the first learning data is increased by using a data enhancement technology.

3. The method according to claim 1, characterized in that Adjusting the first learning data based on the first data set includes: The first learning data in which each of the first prediction data is smaller than the second threshold and the corresponding data label is confirmed is obtained from the first data set as abnormal data, and the abnormal data is checked and processed differently.

4. The method according to claim 3, characterized in that Check the abnormal data and perform different processing including: Verify whether the labeling of the abnormal data is correct, and if the labeling of the abnormal data is incorrect, re-verify the labeling result of the abnormal data; Furthermore, when the number of missing items in the basic data corresponding to the abnormal data is greater than or equal to N, the missing items in the basic data corresponding to the abnormal data are processed, where N represents a positive integer greater than or equal to 3.

5. The method according to claim 4, characterized in that Checking the abnormal data and performing different processing also includes: Calculate the frequency of occurrence of the eigenvalue of each of the basic data in the abnormal data; when the frequency of occurrence of a certain eigenvalue is greater than a third threshold and the frequency of occurrence of other eigenvalues ​​is less than a fourth threshold, obtain abnormal data whose frequency of occurrence of a certain eigenvalue is less than or equal to the third threshold and whose frequency of occurrence of other eigenvalues ​​is greater than or equal to the fourth threshold as abnormal learning data, use generative adversarial network or variational autoencoder technology to generate synthetic data with a distribution similar to that of the abnormal learning data, and add the synthetic data to the first learning data.

6. The method according to claim 4, characterized in that Processing the missing items in the basic data corresponding to the abnormal data includes: Obtain missing items in the basic data corresponding to the abnormal data. When the correlation between the missing items and the prediction target is greater than A1, fill the missing items with a data filling method. When the correlation between the missing items and the prediction target is less than A2, delete the abnormal data from the first learning data, wherein the correlation with the prediction target refers to the data items that directly affect the prediction target, A1 is greater than A2, A1 represents a number greater than or equal to 60%, and A2 represents a number less than or equal to 10%.

7. The method according to claim 1, characterized in that The individual characteristic information includes age, gender, time of consultation, lifestyle, and past medical history; the examination data includes renal function indicators, urine routine indicators, blood routine, liver function, blood lipids, immunology, imaging data, renal puncture data, glomerular filtration rate and tubular function data; the lifestyle habits include smoking history and drinking history.

8. The method according to claim 1, characterized in that The machine learning algorithms include at least logistic regression, decision tree, random forest, support vector machine, neural network, naive Bayes algorithm, Adaboost algorithm and gradient boosting tree.

9. A kidney disease risk prediction system based on machine learning, used to implement the kidney disease risk prediction method based on machine learning according to any one of claims 1 to 8, characterized in that: The system includes the following modules: A data collection unit, used to collect basic data of historical patients, the basic data including individual characteristic information, examination data and preliminary diagnosis information of doctors; a first learning data generating unit, configured to preprocess the basic data, obtain the basic data after preprocessing, add a label to each piece of the basic data based on the result of whether the historical patient is finally diagnosed with kidney disease, and use the basic data with the label added as the first learning data; a first training unit, configured to construct a plurality of first prediction models based on machine learning, train each of the first prediction models using the first learning data, and obtain a plurality of trained first prediction models, wherein each of the first prediction models adopts a different machine learning algorithm; a data set integration unit, configured to input the first learning data into each of the first prediction models, obtain first prediction data output by each of the first prediction models, and integrate a plurality of the first prediction data together to generate a first data set; a second learning data generating unit, configured to adjust the first learning data based on the first data set, obtain the adjusted first learning data, and add the first data set to the first learning data to form second learning data; A second training unit is used to build multiple second prediction models based on machine learning, use the second learning data to train each of the second prediction models respectively, obtain the second prediction data output by each of the second prediction models, and select the second prediction model corresponding to the second prediction data with the highest accuracy from the second prediction data based on the evaluation index as the target prediction model; The result prediction unit is used to obtain the basic data of the new patient and input it into the target prediction model, and the target prediction model outputs the predicted value of the risk of the new patient suffering from kidney disease.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the method for predicting kidney disease risk based on machine learning as described in any one of claims 1-8 is implemented.

Images