Disease risk prediction system
By establishing a sample database in the disease risk prediction system, performing variable screening and data processing, constructing and evaluating multiple machine learning models, and especially combining convolutional neural networks and long short-term memory networks, the problem of inaccurate predictions caused by data errors was solved, achieving higher prediction accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-16
- Publication Date
- 2026-04-03
AI Technical Summary
In existing disease risk prediction systems, the prediction results are not accurate enough due to incorrect data input and measurement errors.
A sample database is established using a data acquisition module. Data is then filtered and processed using a variable selection module and a data processing module. Multiple models are constructed, selected, and evaluated to obtain the optimal prediction model, which includes models such as support vector machine, extreme gradient boosting, and random forest. Convolutional neural network and long short-term memory network are then combined for feature extraction and prediction.
This improved the accuracy of data input and enhanced the accuracy of disease risk prediction.
Smart Images
Figure CN121789964A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of disease risk prediction technology, and in particular to a disease risk prediction system. Background Technology
[0002] A disease risk prediction system is a system that utilizes modern information technology, biostatistics, genetics, and clinical medicine to predict an individual's future risk of developing a specific disease by analyzing various data, including genetic information, lifestyle, environmental factors, and clinical indicators. The development and application of such systems are important components of precision medicine and personalized medicine, aiming to improve the efficiency and effectiveness of disease prevention, diagnosis, and treatment.
[0003] Bioinformatics combines multiple disciplines such as computer science, biology, and mathematics to collect, process, store, analyze, and interpret biological data. Genetics studies the laws governing biological inheritance and variation. In disease risk prediction systems, by analyzing an individual's gene sequence, genetic markers (such as single nucleotide polymorphisms, SNPs) associated with specific diseases are identified, providing a basis for disease risk assessment.
[0004] Biostatistics is the science that applies statistical methods to the fields of biomedicine, public health, and biology. In disease risk prediction, biostatistical methods are used to analyze and interpret data obtained from biological experiments, clinical trials, and epidemiological surveys to determine the association between certain factors and disease risk.
[0005] Machine learning, a branch of artificial intelligence, enables computers to learn from data and make predictions or decisions. Data mining, on the other hand, is the process of extracting useful information from large amounts of data. In disease risk prediction systems, machine learning and data mining techniques are used to build models that can identify patterns and associations of disease risk from an individual's genetic information, lifestyle, environmental factors, and clinical indicators.
[0006] Clinical medical knowledge refers to the theoretical and practical knowledge of medicine related to disease diagnosis, treatment, and prevention. In disease risk prediction systems, clinical medical knowledge is used to interpret the results of prediction models and guide how to conduct effective disease prevention and management based on these results.
[0007] Information technology plays a central role in disease risk prediction systems, encompassing all aspects of data collection, storage, processing, and analysis. High-performance computing, cloud computing, and big data technologies have made it possible to process and analyze massive amounts of genetic and clinical data, enabling faster and more accurate disease risk prediction.
[0008] In summary, disease risk prediction systems are an interdisciplinary field that combines knowledge and technologies from multiple fields such as genetics, bioinformatics, biostatistics, machine learning, data mining, clinical medicine, and information technology. The aim is to provide individuals with more accurate disease risk assessments to guide disease prevention, early diagnosis, and personalized treatment.
[0009] However, while the application of machine learning and data mining in disease risk prediction systems provides powerful tools, it also has some shortcomings and challenges. For example, incorrect data input and measurement errors can affect the performance of the model, resulting in inaccurate prediction results. Summary of the Invention
[0010] In order to overcome the shortcomings of the prior art, the purpose of this invention is to provide a disease risk prediction system.
[0011] To achieve the above objectives, the present invention provides the following solution:
[0012] A disease risk prediction system, comprising:
[0013] The data acquisition module is used to establish a sample database based on the relevant clinical data and automated CT body composition measurement data of the target research subjects;
[0014] The variable filtering module is used to filter variables in the sample database to obtain predictive variables;
[0015] The data processing module is used to process the predicted variables to obtain binary variables;
[0016] The model selection module is used to construct and train multiple models based on the binary classification variables, and to select and evaluate each model to obtain the optimal prediction model; the optimal prediction model is used to predict disease risk.
[0017] Preferably, the plurality of models includes a support vector machine model, a limit gradient boosting model, a random forest model, a linear discriminant analysis model, a logistic regression model, a Naive Bayes model, a k-nearest neighbor model, a multilayer perceptron, a long short-term memory model, and a convolutional neural network model.
[0018] Preferably, the variable filtering module includes:
[0019] The initial screening unit is used to reduce the dimensionality of the data in the training set of the sample database using the Spearman correlation coefficient. It traverses all features and calculates the correlation coefficient of each feature pairwise. When the correlation coefficient is greater than or equal to 0.8, one of the features is randomly removed to select the initial feature values.
[0020] The variable acquisition unit is used to further filter the initial feature values using ridge regression and a five-fold cross-validation method to obtain the predicted variable.
[0021] Preferably, the data processing module includes:
[0022] The first processing unit is used to change the attributes of non-numerical data in the predictor variables to numerical values.
[0023] The second processing unit is used to fill in the missing values of the predicted variables;
[0024] The third processing unit is used to determine the optimal critical value of each continuous predictor variable using ROC curves, calculate the Youden index of each indicator, and the value corresponding to the maximum Youden index is the optimal critical value. It also transforms all the predictor variables into binary variables.
[0025] Preferably, the filling process includes:
[0026] If the first-level variable is selected as "no", the second-level variable will be automatically filled with 0;
[0027] Remove variables with more than 30% missing values.
[0028] For missing values, categorical variables are filled with 99;
[0029] For continuous variables, missing values are filled with the mean.
[0030] Preferably, the indicators for screening and evaluating each model include: prediction accuracy, sensitivity, specificity, negative predictive value, and positive predictive value.
[0031] Preferably, the convolutional neural network model is a CNN neural network sub-model and an LSTM network connected in sequence;
[0032] The CNN neural network sub-model includes two stacked convolutional layers, a ReLU activation function, and a pooling layer arranged sequentially. The CNN neural network sub-model is used to output the data after feature extraction. The LSTM network is used to receive the data after feature extraction and obtain the disease risk prediction result through the LSTM unit update and transmission process and the linear layer transformation process.
[0033] Preferably, the numerical ranges of the Sigmoid function and the tanh activation function in the CNN neural network sub-model and the LSTM network are [0,1] and [-1,1], respectively.
[0034] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0035] This invention provides a disease risk prediction system, comprising: a data acquisition module for establishing a sample database based on relevant clinical data and automated CT body composition measurement data of the target research subjects; a variable screening module for screening variables in the sample database to obtain predictive variables; a data processing module for processing the predictive variables to obtain binary variables; and a model screening module for constructing and training multiple models based on the binary variables, and screening and evaluating each model to obtain the optimal prediction model; the optimal prediction model is used to predict disease risk. This invention, through the variable screening module and the data processing module, improves the accuracy of data input and the accuracy of prediction results by screening and processing data. Attached Figure Description
[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0037] Figure 1 This is a schematic diagram of the system structure provided in an embodiment of the present invention;
[0038] Figure 2 This is a schematic diagram of the technical route provided for an embodiment of the present invention. Detailed Implementation
[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0040] The purpose of this invention is to provide a disease risk prediction system that improves the accuracy of data input and prediction results by filtering and processing data through a variable screening module and a data processing module.
[0041] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0042] Figure 1 This is a schematic diagram of the system structure provided in an embodiment of the present invention, such as... Figure 1 As shown, the present invention provides a disease risk prediction system, comprising:
[0043] The data acquisition module is used to establish a sample database based on the relevant clinical data and automated CT body composition measurement data of the target research subjects;
[0044] The variable filtering module is used to filter variables in the sample database to obtain predictive variables;
[0045] The data processing module is used to process the predicted variables to obtain binary variables;
[0046] The model selection module is used to construct and train multiple models based on the binary classification variables, and to select and evaluate each model to obtain the optimal prediction model; the optimal prediction model is used to predict disease risk.
[0047] like Figure 2 As shown, the technical approach of this embodiment is as follows:
[0048] (1) Selection of machine training samples and models
[0049] First, a sample database was established using the relevant clinical data and automated body composition measurement data from CT scans of the aforementioned research subjects. This database was divided into 70% training samples and 30% test samples. The training samples were used for model building, and the test samples were used for model evaluation. Ten algorithms were used for model building: Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), Random Forest (RF), Linear Discriminant Analysis (LDA), Logistic Regression (LR), Naive Bayes (NB), k-Nearest Neighbors (KNN), Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN) models.
[0050] (2) Variable selection
[0051] First, Spearman Correlation Coefficients (PCC) are used to reduce the dimensionality of the data in the training set. This involves iterating through all features and calculating the correlation coefficient for each pair. When a coefficient is greater than or equal to 0.8 (|r|≥0.8, highly correlated), one feature is randomly removed. This method ensures that the reduced-dimensional features do not have high similarity. While this method can filter out some feature values, the number of resulting feature values is still relatively large, which can easily lead to overfitting in subsequent modeling. To mitigate overfitting, Ridge Regression is used with five-fold cross-validation to further filter features. Ridge Regression makes the coefficient values more even, resulting in a more stable model for the selected features, unlike Lasso Regression where coefficients fluctuate significantly with minor data changes.
[0052] (3) Processing of predictor variables
[0053] Machine learning algorithms require feature attributes to be numerical; therefore, non-numerical feature attributes such as gender and family history need to be converted to numerical values (0, 1). Missing values are handled in three ways: A. Jump variables: If the primary variable is selected as "no", the secondary variable is automatically filled with 0; B. Variables with more than 30% missing values are deleted; C. For remaining missing values, categorical variables are filled with 99, and continuous variables are filled with the mean. ROC curves are used to determine the optimal critical value for each continuous variable, and the Youden index for each indicator is calculated. The value corresponding to the maximum Youden index is the optimal critical value. All data are then converted into binary variables.
[0054] (4) Machine learning modeling
[0055] Based on the model functions provided by the R package, XGBoost and several common models were modeled, with the model parameters configured according to the default settings provided by the functions. A total of 10 models were constructed, including SVM, XGBoost, Random Forest (RF), Linear Discriminant Analysis (LDA), Logistic Regression (LR), Naive Bayes (NB), k-Nearest Neighbors (KNN), Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN).
[0056] (5) Selection of machine learning models
[0057] The accuracy of each model was determined by 10-fold cross-validation, and the best prediction model was selected based on the "horse race mechanism". The prediction model with the highest prediction accuracy was selected as the final prediction model.
[0058] (6) Validation of machine learning models
[0059] In the validation set, ROC curve analysis was used to assess the predictive performance of the prediction model. A higher AUC value indicates higher predictive value. The prediction accuracy, sensitivity, specificity, negative predictive rate, and positive predictive rate of the prediction model were analyzed. Accuracy, Precision, Recall, and F1 score were calculated.
[0060] (7) Statistical methods
[0061] Sample size was calculated using PASS software, and data were analyzed using SPSS 25.0 statistical software. Count data were expressed as "n(%)", and a parallel chi-square test was performed with a two-sided α = 0.05. Non-normally distributed continuous data were expressed as median (interquartile range) "M(QR)", and the Mann-Whitney U test was used for comparisons between groups. Normally distributed continuous data were expressed as "x±s", and an independent samples t-test was used for comparisons between groups. Logistic regression analysis was used to analyze influencing factors, and Spearman or Pearson correlation analysis was used for correlations. A two-sided α = 0.05 was used for the significance level. P < 0.05 was considered statistically significant. Machine learning model processing was performed using R 4.0.4 software, primarily using R packages such as 'mlr', 'nnet', 'rpart', 'e1071', 'catboost', 'stats', 'pcalg', 'ggplot2', and 'caret'.
[0062] Preferably, the convolutional neural network model is a CNN neural network sub-model and an LSTM network connected in sequence;
[0063] The CNN neural network sub-model includes two stacked convolutional layers, a ReLU activation function, and a pooling layer arranged sequentially. The CNN neural network sub-model is used to output the data after feature extraction. The LSTM network is used to receive the data after feature extraction and obtain the disease risk prediction result through the LSTM unit update and transmission process and the linear layer transformation process.
[0064] Preferably, the numerical ranges of the Sigmoid function and the tanh activation function in the CNN neural network sub-model and the LSTM network are [0,1] and [-1,1], respectively.
[0065] Specifically, this invention explores the performance differences of various models (CNN, LSTM, and CNN-LSTM combined neural network models) under different prediction step sizes during multi-step prediction.
[0066] As the prediction step size increases, the CNN neural network performs poorly among the three models, especially when the prediction step size is greater than 1, where the error increases significantly. The CNN-LSTM combined neural network model has lower error than the other two models in most cases, especially when the step size is no higher than 5. The CNN-LSTM combined neural network model exhibits excellent and stable prediction performance, proving that CNN-LSTM can combine the advantages of two basic neural networks.
[0067] The beneficial effects of this invention are as follows:
[0068] This invention uses a variable filtering module and a data processing module to filter and process data, which can improve the accuracy of data input and the accuracy of prediction results.
[0069] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0070] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A disease risk prediction system, characterized in that, include: The data acquisition module is used to establish a sample database based on the relevant clinical data and automated CT body composition measurement data of the target research subjects; The variable filtering module is used to filter variables in the sample database to obtain predictive variables; The data processing module is used to process the predicted variables to obtain binary variables; The model selection module is used to construct and train multiple models based on the binary classification variables, and to select and evaluate each model to obtain the optimal prediction model; the optimal prediction model is used to predict disease risk.
2. The disease risk prediction system according to claim 1, characterized in that, The multiple models include support vector machine model, extreme gradient boosting model, random forest model, linear discriminant analysis model, logistic regression model, Naive Bayes model, k nearest neighbor model, multilayer perceptron, long short-term memory model and convolutional neural network model.
3. The disease risk prediction system according to claim 1, characterized in that, The variable filtering module includes: The initial screening unit is used to reduce the dimensionality of the data in the training set of the sample database using the Spearman correlation coefficient. It traverses all features and calculates the correlation coefficient of each feature pairwise. When the correlation coefficient is greater than or equal to 0.8, one of the features is randomly removed to select the initial feature values. The variable acquisition unit is used to further filter the initial feature values using ridge regression and a five-fold cross-validation method to obtain the predicted variable.
4. The disease risk prediction system according to claim 1, characterized in that, The data processing module includes: The first processing unit is used to change the attributes of non-numerical data in the predictor variables to numerical values. The second processing unit is used to fill in the missing values of the predicted variables; The third processing unit is used to determine the optimal critical value of each continuous predictor variable using ROC curves, calculate the Youden index of each indicator, and the value corresponding to the maximum Youden index is the optimal critical value. It also transforms all the predictor variables into binary variables.
5. The disease risk prediction system according to claim 4, characterized in that, The filling process includes: If the first-level variable is selected as "no", the second-level variable will be automatically filled with 0; Remove variables with more than 30% missing values. For missing values, categorical variables are filled with 99; For continuous variables, missing values are filled with the mean.
6. The disease risk prediction system according to claim 1, characterized in that, The metrics used to screen and evaluate the various models include: prediction accuracy, sensitivity, specificity, negative predictive value, and positive predictive value.
7. The disease risk prediction system according to claim 2, characterized in that, The convolutional neural network model consists of a CNN neural network sub-model and an LSTM network connected in sequence. The CNN neural network sub-model includes two stacked convolutional layers, a ReLU activation function, and a pooling layer arranged sequentially. The CNN neural network sub-model is used to output the data after feature extraction. The LSTM network is used to receive the data after feature extraction and obtain the disease risk prediction result through the LSTM unit update and transmission process and the linear layer transformation process.
8. The disease risk prediction system according to claim 7, characterized in that, The numerical ranges of the Sigmoid function and the tanh activation function in the CNN neural network sub-model and the LSTM network are [0,1] and [-1,1], respectively.