Kidney stone prediction and early screening assisting method based on machine learning model

Abstract

The invention discloses a kidney stone prediction and early screening assisting method based on a machine learning model. The kidney stone prediction and early screening assisting method comprises the steps of S1, data integration; s2, preprocessing the data; s3, model training and prediction; and S4, outputting a diagnosis result and a treatment suggestion. Various vital sign data of a patient can be collected, the data are combined to form a personal health information file of the user, the obtained data are adjusted, the model is fitted, the system is used for conducting preliminary diagnosis and screening on kidney stone, the medical history of the patient can be integrated, data classification based on the vital sign data of the user and the model can be conducted, and the diagnosis accuracy of the kidney stone can be improved. And a doctor is helped to adjust the lifestyle of the user so as to avoid the formation of kidney stones.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a method for predicting and assisting early screening of kidney stones based on a machine learning model. Background Art

[0002] Triglycerides (TG) and total cholesterol (TC): High levels of TG and TC may be associated with metabolic syndrome, which increases the risk of kidney stones. Abnormal lipid metabolism may affect the concentrations of calcium, oxalate, and uric acid in urine, thereby contributing to stone formation. Machine learning models are able to automatically learn the complex relationship between these indicators and the risk of kidney stones. Using models such as multiple regression, decision trees, or random forests, the nonlinear effects of TG and TC levels on stone formation can be discovered on larger sample data sets. In addition, through integrated learning algorithms, it is possible to effectively predict how abnormal lipid metabolism affects the concentrations of calcium, oxalate, and uric acid in urine, thereby improving the accuracy of risk assessment.

[0003] Fasting blood glucose (FBG) and TyG index: Increased FBG may indicate insulin resistance, and TyG is an indicator of insulin resistance. Studies have shown that insulin resistance is associated with kidney stone formation, and may affect stone formation by changing the composition of urine (such as calcium, phosphorus, uric acid, etc.). For example, time series analysis based on LSTM (long short-term memory network) can analyze the changing trend of individual blood sugar levels over time, and then predict the long-term impact of insulin resistance on urine composition. This dynamic data analysis can more accurately reflect the potential risk of stone formation than traditional static analysis.

[0004] Gender (SeX): Men and women differ in the incidence and type of kidney stones. Men are generally more likely to form stones than women, which may be related to hormone levels and lifestyle. Classification algorithms (such as support vector machines SVM or XGBoost) can distinguish the differences in stone risk between different genders. Machine learning models can also combine other variables, such as hormone levels and lifestyle habits, to predict the incidence of stones in men and women under different circumstances, thereby providing customized prevention recommendations for specific populations.

[0005] Body mass index (BMI): Being overweight or obese is associated with an increased risk of kidney stones. A higher BMI may be associated with metabolic abnormalities (such as insulin resistance, hypertension), which may affect kidney health and stone formation. Based on clustering algorithms (such as K-means or DBSCAN), people with different body mass indexes can be grouped to discover potential patterns between BMI and stone risk. Combined with multimodal data (such as BMI, blood pressure, urine composition, etc.), machine learning models are able to comprehensively analyze multiple factors and provide personalized intervention strategies based on individual BMI. Summary of the invention

[0006] The purpose of the present invention is to overcome the problems existing in the prior art and provide a method for predicting kidney stones and assisting early screening based on a machine learning model.

[0007] In order to achieve the above technical objectives and the above technical effects, the present invention is implemented through the following technical solutions:

[0008] A method for predicting kidney stones and assisting early screening based on a machine learning model, the method comprising the following steps:

[0009] Step S1: Data integration

[0010] Step S1.1: Collect the patient's medical test data and basic information separately to form a comprehensive patient health text file;

[0011] Step S2: Data preprocessing

[0012] Step S2.1: fill missing values ​​in the collected data using a mean filling strategy;

[0013] Step S2.2: Use the SMOTE algorithm to solve the class imbalance problem;

[0014] Step S2.3: Perform polynomial feature engineering to enhance model performance;

[0015] Step S2.4: Standardization to ensure data consistency and comparability;

[0016] Step S3: Model training and prediction

[0017] Step S3.1: Use one or more machine learning models of Logistic Regression, XGBoost, AdaBoost, ExtraTrees, KNeighborsClassifier, DecisionTreeClassifier, and RandomForestClassifier for training;

[0018] Step S3.2: Validate model performance through cross-validation and independent test sets;

[0019] Step S3.3: Select the best model for final diagnosis prediction;

[0020] Step S4: Output diagnosis results and treatment recommendations

[0021] Step S4.1: Based on the prediction results of the model, output the diagnosis category of kidney stone disease;

[0022] Step S4.2: Generate personalized treatment recommendations based on the patient's detailed information.

[0023] Furthermore, the method further comprises:

[0024] Step S5: Establish a large model evaluation standard system applicable to the medical field to ensure the accuracy and safety of the model.

[0025] Furthermore, in step S5, the evaluation indicators of the large model evaluation standard system include:

[0026] Accuracy: including sensitivity, specificity, positive predictive value, and negative predictive value;

[0027] Security: Ensure the stability of the model in different populations and scenarios;

[0028] The vital sign data were divided into two parts, training set and validation set, in a ratio of 8:2 for cross-validation to evaluate the performance of multiple models;

[0029] Practical application stage: Use independent medical data sets for verification to ensure the applicability and robustness of the model in real scenarios;

[0030] Comparative analysis: Compare the model with existing screening and diagnostic methods to evaluate its strengths and weaknesses.

[0031] Furthermore, in step S1.1, at this stage, the medical test data of the patient is collected from various medical source systems, including:

[0032] Triglyceride TG, total cholesterol TC and fasting blood glucose FBG;

[0033] And basic information, including gender, body mass index (BMI) and age;

[0034] The above data form a comprehensive patient health text file, providing a basis for analysis and modeling.

[0035] Furthermore, in step S2.3, after creating a quadratic or higher order feature combination through polynomial feature engineering, cross validation is used to evaluate the performance.

[0036] Furthermore, in step S2.4, the standardization method includes Z-score standardization and Min-Max normalization.

[0037] Furthermore, in step S3.2, the data set is divided into multiple subsets, and the model is trained on different training and test sets to reduce overfitting.

[0038] Furthermore, in step S3.3, the accuracy, recall rate, and F1-score evaluation indicators of different models are compared, and the model with the best performance is selected as the final diagnostic tool.

[0039] Furthermore, in step S4.1, the confusion matrix and AUC-ROC curve are used to visualize the classification ability of the model.

[0040] Furthermore, in step S4.2, a visual result report is generated to help doctors and patients better understand the decision basis of the model.

[0041] The beneficial effects of the present invention are:

[0042] The present invention can collect various vital signs data of patients, combine the data to form a personal health information file of the user, adjust the obtained data, and fit the model to make a preliminary diagnosis and screening of kidney stones. It can also integrate the patient's medical history and classify the data based on the user's vital signs data and the model to help doctors adjust the user's lifestyle to avoid the formation of kidney stones. DETAILED DESCRIPTION

[0043] The present invention will be described in detail below with reference to examples.

[0044] A method for predicting kidney stones and assisting early screening based on a machine learning model, the method comprising the following steps:

[0045] Step S1: Data integration

[0046] Step S1.1: Collect the patient's medical test data and basic information separately to form a comprehensive patient health text file;

[0047] Step S2: Data preprocessing

[0048] Step S2.1: Fill missing values ​​in the collected data using the mean filling strategy. For numerical data, this method can effectively maintain the overall distribution of the data; however, for some features, it may be more appropriate to use the median or mode filling to reduce the impact of extreme values; in addition, interpolation or more complex filling techniques (such as KNN filling) can be considered to improve the accuracy of filling;

[0049] Step S2.2: Use the SMOTE algorithm to solve the class imbalance problem. The SMOTE (Synthetic Minority Over-sampling Technique) algorithm balances the data set by synthesizing minority class samples. SMOTE generates new samples by finding the neighbors of minority class samples in the feature space and interpolating between these samples. This process can improve the model's prediction ability for minority classes.

[0050] Step S2.3: Perform polynomial feature engineering to enhance model performance;

[0051] Step S2.4: Standardization to ensure data consistency and comparability;

[0052] Step S3: Model training and prediction

[0053] Step S3.1: Use one or more machine learning models selected from Logistic Regression, XGBoost, AdaBoost, ExtraTrees, KNeighborsClassifier, DecisionTreeClassifier, and RandomForestClassifier for training, where:

[0054] Logistic regression is a linear model for binary classification problems. It outputs classification probabilities through a logistic function and is easy to interpret.

[0055] XGBoost is a powerful model based on the gradient boosting algorithm. It performs well in processing high-dimensional data and has regularization function to effectively prevent overfitting.

[0056] AdaBoost improves classification accuracy by weighted combination of multiple weak classifiers (such as decision trees), and is suitable for dealing with noise and complexity;

[0057] ExtraTrees builds classifiers using extreme random trees, which are fast and usually have higher prediction accuracy;

[0058] KNeighborsClassifier (K ​​nearest neighbor classifier) ​​is based on instance learning. It votes according to the categories of the nearest K neighbors during classification. It is sensitive to the distance metric in the feature space, suitable for small-scale data sets, and has a large amount of calculation.

[0059] DecisionTreeClassifier uses a tree structure for classification. Data is divided according to features starting from the root node until it reaches the leaf node. It is easy to visualize and explain, but it is prone to overfitting.

[0060] RandomForestClassifier is composed of multiple decision trees. It uses voting to determine the final category and randomly selects features and samples to reduce overfitting and improve generalization ability.

[0061] Step S3.2: Validate model performance through cross-validation and independent test sets;

[0062] Step S3.3: Select the best model for final diagnosis prediction;

[0063] Step S4: Output diagnosis results and treatment recommendations

[0064] Step S4.1: Based on the prediction results of the model, output the diagnosis category of kidney stone disease;

[0065] Step S4.2: Generate personalized treatment recommendations based on the patient's detailed information.

[0066] The method further includes:

[0067] Step S5: Establish a large model evaluation standard system applicable to the medical field to ensure the accuracy and safety of the model.

[0068] In step S5, the evaluation indicators of the large model evaluation standard system include:

[0069] Accuracy: including sensitivity, specificity, positive predictive value, and negative predictive value;

[0070] Security: Ensure the stability of the model in different populations and scenarios;

[0071] The vital sign data were divided into two parts, training set and validation set, in a ratio of 8:2 for cross-validation to evaluate the performance of multiple models;

[0072] Practical application stage: Use independent medical data sets for verification to ensure the applicability and robustness of the model in real scenarios;

[0073] Comparative analysis: Compare the model with existing screening and diagnostic methods to evaluate its strengths and weaknesses.

[0074] In step S1.1, at this stage, the medical test data of the patient is collected from various medical source systems, including:

[0075] Triglyceride TG, total cholesterol TC and fasting blood glucose FBG;

[0076] And basic information, including gender, body mass index (BMI) and age;

[0077] The above data form a comprehensive patient health text file, providing a basis for analysis and modeling.

[0078] In step S2.3, polynomial feature engineering aims to enhance the expressiveness of the model by generating polynomial combinations of features. This method can capture the nonlinear relationship between features. By creating quadratic or higher-order feature combinations, the flexibility and accuracy of the model can be improved. It should be noted that the increase in the number of features may lead to overfitting, so it is necessary to use cross-validation to evaluate performance.

[0079] In step S2.4, the standardization method includes Z-score standardization (mean is 0, standard deviation is 1) and Min-Max normalization (scaling the data to the interval [0,1]). This step is particularly important in the specific implementation because many machine learning algorithms (such as KNN, support vector machine) are sensitive to the scale of the data, and standardization can improve the convergence speed and performance of the model.

[0080] In step S3.2, after the model training is completed, the model performance will be evaluated through a cross-validation method (such as k-fold cross-validation). By dividing the data set into multiple subsets and training the model on different training and test sets, overfitting can be reduced. The use of an independent test set can ensure the generalization ability of the model on unseen data.

[0081] In step S3.3, the accuracy, recall rate, and F1-score evaluation indicators of different models are compared, and the model with the best performance is selected as the final diagnostic tool to ensure that the selected model can provide the most reliable prediction results.

[0082] In step S4.1, based on the prediction results of the model, the patient's kidney stone disease diagnosis category will be output. In order to improve the credibility of the diagnosis results, the confusion matrix and AUC-ROC curve can be used to visualize the classification ability of the model.

[0083] In step S4.2, based on the generated diagnostic categories, personalized treatment recommendations are provided for the patient in combination with the patient's detailed health information. These recommendations will refer to the latest clinical guidelines to ensure that the patient can obtain scientific and effective treatment plans. In addition, a visual result report is generated to help doctors and patients better understand the decision-making basis of the model and promote communication between doctors and patients.

[0084] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting and assisting early screening of kidney stones based on a machine learning model, characterized in that: The method comprises the following steps: Step S1: Data integration Step S1.1: Collect the patient's medical test data and basic information separately to form a comprehensive patient health text file; Step S2: Data preprocessing Step S2.1: fill missing values ​​in the collected data using a mean filling strategy; Step S2.2: Use the SMOTE algorithm to solve the class imbalance problem; Step S2.3: Perform polynomial feature engineering to enhance model performance; Step S2.4: Standardization to ensure data consistency and comparability; Step S3: Model training and prediction Step S3.1: using one or more machine learning models selected from Logistic regression, XGBoos t, AdaBoost, ExtraTree s, KNeighbor sClassifier, DecisionTreeClassifier, and RandomForestClassifier for training; Step S3.2: Validate model performance through cross-validation and independent test sets; Step S3.3: Select the best model for final diagnosis prediction; Step S4: Output diagnosis results and treatment recommendations Step S4.1: Based on the prediction results of the model, output the diagnosis category of kidney stone disease; Step S4.2: Generate personalized treatment recommendations based on the patient's detailed information.

2. The method for predicting and assisting early screening of kidney stones based on a machine learning model according to claim 1, characterized in that: The method further includes: Step S5: Establish a large model evaluation standard system applicable to the medical field to ensure the accuracy and safety of the model.

3. The method for predicting and assisting early screening of kidney stones based on a machine learning model according to claim 1, characterized in that: In step S5, the evaluation indicators of the large model evaluation standard system include: Accuracy: including sensitivity, specificity, positive predictive value, and negative predictive value; Security: Ensure the stability of the model in different populations and scenarios; The vital sign data were divided into two parts, training set and validation set, in a ratio of 8:2 for cross-validation to evaluate the performance of multiple models; Practical application stage: Use independent medical data sets for verification to ensure the applicability and robustness of the model in real scenarios; Comparative analysis: Compare the model with existing screening and diagnostic methods to evaluate its strengths and weaknesses.

4. The method for predicting and assisting early screening of kidney stones based on a machine learning model according to claim 2, characterized in that: In step S1.1, at this stage, the medical test data of the patient is collected from various medical source systems, including: Triglyceride TG, total cholesterol TC and fasting blood glucose FBG; And basic information, including gender, body mass index (BMI) and age; The above data form a comprehensive patient health text file, providing a basis for analysis and modeling.

5. The method for predicting and assisting early screening of kidney stones based on a machine learning model according to claim 2, characterized in that: In step S2.3, after creating quadratic or higher-order feature combinations through polynomial feature engineering, cross-validation is used to evaluate performance.

6. The method for predicting and assisting early screening of kidney stones based on a machine learning model according to claim 2, characterized in that: In step S2.4, the normalization method includes Z-score normalization and Min-Max normalization.

7. The method for predicting and assisting early screening of kidney stones based on a machine learning model according to claim 2, characterized in that: In step S3.2, the data set is divided into multiple subsets, and the model is trained on different training and test sets to reduce overfitting.

8. The method for predicting and assisting early screening of kidney stones based on a machine learning model according to claim 2, characterized in that: In step S3.3, the accuracy, recall rate, and F1-score evaluation indicators of different models are compared, and the model with the best performance is selected as the final diagnostic tool.

9. The method for predicting and assisting early screening of kidney stones based on a machine learning model according to claim 2, characterized in that: In step S4.1, the confusion matrix and AUC-ROC curve are used to visualize the classification ability of the model.

10. The method for predicting and assisting early screening of kidney stones based on a machine learning model according to claim 2, characterized in that: In step S4.2, a visual result report is generated to help doctors and patients better understand the decision basis of the model.