Multi-model fusion bone mineral density value prediction method and system

Abstract

The invention relates to the technical field of bone mineral density value prediction, and provides a multi-model fusion bone mineral density value prediction method and system, and the method comprises the following steps: obtaining clinical feature data of bone mineral density prediction; constructing a multi-model fusion prediction model comprising base learners and meta learners which are connected in sequence, and inputting the obtained features into the plurality of base learners for prediction identification to obtain a primary prediction result; according to the prediction accuracy of the base learners, adaptively calculating a fusion weight, and fusing the primary prediction results of the base learners based on the fusion weight obtained by calculation; and transmitting the fused primary prediction result to a meta-learner for prediction to obtain a prediction result of the bone mineral density. According to the invention, a plurality of machine learning models are designed as a multi-stage prediction architecture, the model of each stage learns and optimizes the prediction result of the previous stage and the original input features, and the prediction accuracy is improved stage by stage.

Description

Technical Field

[0001] The present disclosure relates to the technical field related to bone density value prediction, and in particular, to a bone density value prediction method and system based on multi-model fusion. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] Bone density refers to the amount of bone per unit area or unit volume. The currently recognized bone density examination method in the academic community is dual energy X-ray absorptiometry (DXA), and its measured value is used as the gold standard for the diagnosis of osteoporosis.

[0004] In recent years, machine learning-based methods have been applied in the field of bone density prediction. These methods collect multimodal data from patients and use machine learning models for prediction in order to provide a convenient and low-cost solution. However, most of the current machine learning-based bone density prediction methods rely on a single model, and common models include linear regression, decision trees, support vector machines (SVMs), or simple neural networks. These models lack sufficient ability to characterize complex factors affecting bone density (such as age, gender, and lifestyle habits), resulting in unsatisfactory prediction accuracy. A single model is difficult to adapt to the diverse feature distribution in the training data, and may perform better for some groups, but have poor prediction effects on other groups. In the face of noisy data or incomplete features, the performance of a single model is prone to significant decline. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a bone density value prediction method and system with multi-model fusion, designs multiple machine learning models as a multi-level prediction architecture, and each level of the model learns and optimizes the prediction results and original input features of the previous stage, thereby improving the prediction accuracy step by step.

[0006] In order to achieve the above objectives, the present disclosure adopts the following technical solutions:

[0007] One or more embodiments provide a multi-model fusion bone density value prediction method, comprising the following steps:

[0008] Obtain clinical characteristic data for bone density prediction;

[0009] Construct a multi-model fusion prediction model including base learners and meta-learners connected in sequence, input the acquired features into multiple base learners for prediction and recognition, and obtain primary prediction results;

[0010] The fusion weight is adaptively calculated according to the prediction accuracy of the base learners, and the primary prediction results of each base learner are fused based on the calculated fusion weight;

[0011] The fused primary prediction results are transmitted to the meta-learner for prediction to obtain the prediction results of bone density.

[0012] One or more embodiments provide a multi-model fusion bone density value prediction system, including:

[0013] An acquisition module is configured to acquire clinical characteristic data for bone density prediction;

[0014] A primary prediction module is configured to construct a multi-model fusion prediction model including base learners and meta learners connected in sequence, input the acquired features into multiple base learners for prediction and recognition, and obtain a primary prediction result;

[0015] A fusion module is configured to adaptively calculate a fusion weight according to a prediction accuracy of a base learner, and fuse the primary prediction results of each base learner based on the calculated fusion weight;

[0016] The secondary prediction module is configured to transmit the fused primary prediction results to the meta-learner for prediction to obtain the prediction results of bone density.

[0017] An electronic device includes a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps in the above-mentioned multi-model fusion bone density value prediction method are completed.

[0018] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the steps in the above-mentioned multi-model fusion bone density value prediction method are completed.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] The present invention makes full use of the strengths of different models and enhances the robustness of prediction through a multi-model fusion method. Based on adaptive weight calculation, the influence of low-precision models on the final prediction results is reduced. The meta-learner further optimizes the fusion results, eliminates the deviation in the prediction of the base learner, ensures the stability of the model's performance on unknown data, and ultimately provides accurate bone density prediction values. This method can be extended to other medical prediction tasks by simply changing the input features and base learner types.

[0021] The advantages of the present disclosure and the advantages of additional aspects will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and the description thereof are used to explain the present disclosure but do not constitute a limitation of the present disclosure.

[0023] Figure 1 It is a schematic diagram of the structure of the multi-model fusion prediction model of Example 1 of the present disclosure;

[0024] Figure 2 is a schematic diagram of the prediction process of Embodiment 1 of the present disclosure;

[0025] Figure 3 is a simulation experiment result diagram of the multi-model fusion prediction model used in Example 1 of the present disclosure; DETAILED DESCRIPTION

[0026] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0027] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0028] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof. It should be noted that, in the absence of conflict, the various embodiments in the present disclosure and the features in the embodiments can be combined with each other. The embodiments will be described in detail below in conjunction with the accompanying drawings.

[0029] Example 1

[0030] In the technical solutions disclosed in one or more embodiments, Figures 1 to 3 As shown, a multi-model fusion bone density value prediction method includes the following steps:

[0031] Step 1, obtaining clinical characteristic data for bone density prediction;

[0032] Step 2: construct a multi-model fusion prediction model including base learners and meta learners connected in sequence, input the acquired features into multiple base learners for prediction and recognition, and obtain primary prediction results;

[0033] Step 3: adaptively calculate the fusion weight according to the prediction accuracy of the base learners, and fuse the primary prediction results of each base learner based on the calculated fusion weight;

[0034] Step 4: The fused primary prediction results are transmitted to the meta-learner for prediction to obtain the prediction results of bone density;

[0035] In this embodiment, by taking advantage of the multi-model fusion technology, multiple base learners can analyze clinical feature data from different angles to generate primary prediction results. Through adaptive weight calculation, the weight of the high-accuracy model is appropriately amplified to improve the reliability of the overall prediction. The meta-learner further optimizes the fusion results, eliminates the deviation in the prediction of the base learner, and finally provides an accurate bone density prediction value.

[0036] In step 1, data collection and preprocessing;

[0037] The bone density prediction method of this embodiment is based on multimodal clinical feature data and forms a data input set by collecting individual features and indicators related to bone density. The specific steps are as follows:

[0038] Specifically, the clinical characteristics required for bone density prediction are obtained through scale data and physical examination data, including but not limited to demographic data, behavioral habit data and other clinical indicators;

[0039] Optionally, demographic data may include age, gender, height, weight, occupation, etc. Behavioral habits may include whether one smokes, drinks, etc. Other clinical indicators may include characteristics related to bone density in medical examinations, such as calcium intake and exercise frequency.

[0040] Furthermore, the acquired data set is preprocessed, including data formatting, missing value processing and encoding; the preprocessing includes the following:

[0041] Step 1.1: Format the data. The above features can be organized into a unified CSV format data file to facilitate subsequent data processing and model training. Each record includes 28 features of an individual and the target variable (T-Value of bone density). You can use Python's pandas library to import and process CSV files.

[0042] Step 1.2: Perform missing value processing, detect missing values, and determine the missing value ratio of each feature; use the mean or median to fill in features with fewer missing values; remove features with more missing values ​​and lower information value to ensure data quality;

[0043] Step 1.3: Perform feature encoding. For the collected multimodal features, use appropriate encoding strategies to process non-numerical features. Binary feature encoding and multi-category feature encoding can be used.

[0044] Binary feature coding uses values ​​of 0 and 1 to represent binary features such as gender, smoking, and drinking, such as:

[0045] Gender: 1 for male and 0 for female.

[0046] Whether to smoke: 1 for yes, 0 for no.

[0047] Drinking or not: 1 for yes, 0 for no.

[0048] Multi-category feature coding, for features with multiple categories such as occupations, an integer coding method of 0-N is used, where N is the number of categories.

[0049] In some embodiments, in step 2, the multi-model fusion prediction model constructed can be as follows: Figure 1 As shown, it includes two layers of models, the first layer model includes multiple base learners, and the second layer model includes a meta learner;

[0050] The first layer model settings include a support vector machine model (SVM model), an XGBoost model, and a GBDT model set in parallel; each base learner is trained and predicted independently, and the prediction results of the first layer;

[0051] The meta-learner uses a random forest model. The meta-learner combines the prediction results of the first layer into a new feature vector for prediction to obtain the final prediction result.

[0052] A feasible technical solution is to build a support vector machine (SVR) model, optimize the objective function through kernel function and Lagrange multiplier method, and use GridSearchCV to tune the kernel function type and regularization parameters.

[0053] This embodiment uses the SVR (Support Vector Regression) model of SVM (Support Vector Machine) to perform the task of predicting bone density values.

[0054] Specifically, for a data set consisting of n samples with m features

[0055]

[0056] SVM uses a linear or nonlinear function to predict the output. The goal of the SVR model is to minimize the following objective function:

[0057]

[0058] Among them, ω is the weight vector, C is the regularization parameter used to control the complexity of the model and the degree of penalty for errors, and ξi is a slack variable, which is used to allow for a certain amount of prediction error. The first term of the objective function is the L2 norm of the weight vector, which is used to control the complexity of the model; the second term is the error term, which is used to measure the difference between the predicted value and the target value.

[0059] In the SVR model of this embodiment, an insensitive loss function is introduced, that is, when the difference between the predicted value and the target value is less than or equal to ε, the prediction is considered correct and no loss is incurred. Therefore, the optimization problem can be further transformed into:

[0060]

[0061] The constraints are:

[0062]

[0063] Among them, ξ i and They respectively represent the extent to which the deviation of the sample point above and below the hyperplane exceeds ε.

[0064] In order to solve this optimization problem, this embodiment further adopts the Lagrange multiplier method and the kernel function. Specifically, the original problem is transformed into a dual problem through the Lagrange function, and the kernel function is used to map the input data to a high-dimensional space, so as to find a better hyperplane in the new space to fit the data.

[0065] During the training process, the SVR model will traverse all sample points, find support vectors (i.e. sample points located near the hyperplane or violating the constraints), and use these support vectors to determine the position and direction of the hyperplane. At the same time, the regularization parameter C and the threshold ε of the insensitive loss function will also have an important impact on the performance of the model.

[0066] In order to prevent the model from overfitting, cross validation, feature scaling and kernel function parameter tuning can also be used for the SVR model. Finally, the trained SVR model can be used to predict new input data to obtain the predicted bone density value. In order to evaluate the performance of the model, various indicators are used to compare the difference between the predicted value and the true value.

[0067] Furthermore, in this embodiment, the GridSearchCV class in the scikit-learn library is used to automatically search for the optimal parameter combination of the support vector regression (SVR) model. Specifically, the following steps are included:

[0068] Step 21.1, define the parameter grid parameters, which is used to store the two key parameters of the SVR model: kernel function type and regularization parameter;

[0069] The kernel function type may include a linear kernel, a radial basis function kernel, and a polynomial kernel; the regularization parameter C has a value ranging from 0.01 to 100 and may be transformed with different step sizes.

[0070] Step 21.2 then creates a GridSearchCV object grid_search_svr, which will train the SVR model using the specified parameter grid and evaluate the performance of each parameter combination through 10-fold cross validation.

[0071] During the training process, the verbose=1 parameter is used to output information about the search process. Finally, by calling the grid_search_svr.fit method, the code uses the training dataset (reg_train as features and target_reg_train as the target variable) to train the model and find the best parameter combination.

[0072] A feasible technical solution is to build an XGBoost model, which uses a greedy algorithm to optimize the objective function of the XGBoost model. During the training process, overfitting is prevented by using strategies such as shrinkage and feature bucketing. The construction process includes the following:

[0073] First, the greedy algorithm is used to optimize the objective function of the XGBoost model;

[0074] Specifically, the XGBRegressor regression model of the XGBoost model is used for model training. For a data set consisting of n samples with m features,

[0075]

[0076] The XGBoost model uses an additive function to predict the output:

[0077]

[0078] Among them, l represents the loss function, which is used to measure the prediction and target y i The second term Ω penalizes the complexity of the model and helps smooth the final learned weights to avoid overfitting. Formula (5) includes the function as a parameter and is trained in an additive manner. Formally, if is the prediction of the i-th instance at the tth iteration, and f needs to be added t To minimize the following objective:

[0079]

[0080] This means greedily adding f according to equation (5) t to optimize the model to the greatest extent possible. The second-order approximation can be used to quickly optimize the objective in a general setting:

[0081]

[0082] Among them, g i and h i are the first and second order gradients with respect to the loss function.

[0083] At the tth iteration, the constant term can be removed to obtain the following simplified objective function:

[0084]

[0085] XGBoost uses a greedy algorithm to traverse all feature partitioning points of all features, and divides the objective function value after the split by the gain of the objective function of the single leaf node. At the same time, in order to limit the tree from growing too deep, a threshold is added, and the split is performed only when the gain is greater than the threshold. The maximum depth of the tree can be set, and the growth will stop when the sample weight is less than the set threshold to prevent overfitting.

[0086]

[0087] Furthermore, during the XGBoost model training process, set up strategies to prevent overfitting, including:

[0088] 1) In each iteration of XGBoost model training, the leaf node score of each tree is multiplied by the reduction weight so that the subsequently generated trees can further optimize the model; before splitting each layer of nodes, some features are randomly selected to determine the optimal split point;

[0089] Specifically, in order to further prevent the model from overfitting, the Shrinkage method is used to multiply the score of each leaf node of the tree by a reduction weight in each iteration, so that the influence of each tree will not be too large, leaving more room for the trees generated later to optimize the model. Column Subsampling is also used for random sampling by layer. Before splitting each node in the same layer, a part of the features is randomly selected, and then only the features of this part need to be traversed to determine the optimal split point.

[0090] 2) When training the XGBoost model, the continuous input features are divided into multiple buckets according to the preset quantiles. The samples between adjacent quantiles are divided into the same bucket. Each quantile is traversed to calculate the optimal division, reduce the range of feature values ​​and optimize the calculation efficiency.

[0091] Specifically, the feature is bucketed, that is, L partition points are found, and the samples between adjacent quantiles are divided into one bucket. When traversing the feature, only each quantile needs to be traversed to calculate the optimal partition, further preventing the overfitting of the model.

[0092] 3) Tree growth limit: Set the maximum tree depth max_depth and sample weight threshold to prevent the tree from growing without limit, and stop splitting when the gain is less than the set threshold;

[0093] Furthermore, during the XGBoost model training process, the optimal parameters of the XGBoost model are selected through grid search and cross validation. The process of parameter tuning and grid search is as follows:

[0094] By defining the parameters dictionary, determine the candidate value ranges of the three key parameters of the XGBoost model: the maximum depth of the tree (max_depth), the number of trees (n_estimators), and the learning rate (eta). Use GridSearchCV to perform a cross-validated grid search to search for the optimal parameter combination. GridSearchCV will try all parameter combinations and use cross-validation to evaluate the performance of each combination. In this embodiment, it is set to 10-fold cross-validation). The verbose=1 parameter is used to output information during the search process. The best parameter combination found in the grid search can be obtained through grid_search_xgb_reg.best_params_, and grid_search_xgb_reg.best_score_ shows the corresponding best cross-validation score. Using the best parameters obtained, initialize a new XGBoost regressor regressor and use the fit method to train on the training set reg_train and target_reg_train. Finally, the feature importance image is plotted through the xgb.plot_importance function. Here, the trained regressor model is used, the ax parameter is specified as ax in fig,ax, and max_num_features=25 is set to display the top 25 most important features.

[0095] A feasible technical solution is to construct the GBDT model by using a gradient boosting decision tree and optimizing the model by reducing the residual to obtain a trained GBDT model.

[0096] GBDT (Gradient Boosting Decision Tree) uses the Boosting concept in ensemble learning to perform bone density regression prediction tasks. It obtains the final prediction result by superimposing the prediction results of multiple decision trees. Each decision tree is optimized based on the prediction results of the previous decision tree. The goal of the GBDT regression model is to achieve regression prediction by continuously reducing the residual error (i.e., the difference between the predicted value and the actual value) during the training process.

[0097] Specifically, the GradientBoostingRegressor class in the scikit-learn library is used as the implementation of the GBDT model. According to the characteristics of the data set and the model requirements, the parameters of the GBDT model are set, and key parameters such as n_estimators (the number of weak learners), learning_rate (learning rate), and max_depth (the maximum depth of the decision tree) are set. The GBDT model is trained using the training set data. During the training process, multiple decision trees are gradually constructed, and the model performance is optimized by gradient boosting. After the training is completed, the GBDT model is evaluated using the test set data. The prediction performance of the model is evaluated by calculating the error between the predicted value and the actual value, such as the mean absolute error MAE, the root mean square error RMSE, etc. For the trained GBDT model, the feature importance can be analyzed to understand which features have the greatest impact on bone density prediction.

[0098] A feasible technical solution is to build a random forest model RandomForest, based on ensemble learning of multiple decision trees, optimize the depth and number of trees through grid search, and use the average prediction results of multiple decision trees as the output of the model.

[0099] In the embodiment, an integrated learning method is used for model training, and the stability and accuracy of the model are improved by constructing multiple decision trees and taking their average prediction results.

[0100] For a dataset of n samples with m features:

[0101]

[0102] The output is predicted by building multiple decision trees, each of which is trained based on a different subset of features and training samples and outputs a prediction result. The final prediction result is the average of the prediction results of all decision trees.

[0103] During the training process, you can optimize the performance of the model by adjusting the parameters of the ensemble learning. You can adjust the n_estimators parameter to specify the number of decision trees, adjust the max_depth parameter to limit the maximum depth of the decision tree, and adjust other related parameters to further optimize the model.

[0104] After the model training is completed, the test set is used for prediction. First, define the parameter dictionary and specify the parameter value range of the RandomForestRegressor model (n_estimators, max_depth, etc.). Then use GridSearchCV to perform a cross-validated grid search to search for the optimal parameter combination. GridSearchCV will try all parameter combinations and use cross-validation to evaluate the performance of each combination (here set to 10-fold cross-validation).

[0105] The best parameter combination found in the grid search can be obtained through grid_search_rf_reg.best_params_, while grid_search_rf_reg.best_score_ shows the corresponding best cross-validation score. Using the best parameters obtained, initialize a new RandomForestRegressor regressor and use the fit method to train it on the training set.

[0106] Finally, use the trained regressor to predict the test set and evaluate the performance of the model. The accuracy of the model can be evaluated by calculating the error between the predicted results and the true values ​​(such as mean absolute error MAE, root mean square error RMSE, etc.). In addition, you can also use the feature_importances_ attribute in the scikit-learn library to view the importance of the features to better understand the working mechanism of the model.

[0107] In the above technical solution, the construction of each model of the base learner and the meta learner and the improvement method of the training process are given. By optimizing the training process, the obtained model can predict bone density more accurately.

[0108] Furthermore, the multi-model fusion prediction model, the training process is as follows:

[0109] Step S1, training base learners: First, multiple different models are trained, which are called base learners or individual learners. Each base learner is trained independently and can use different algorithms or parameters.

[0110] Step S2: Generate first-level prediction results: Use these base learners to predict the training set or another data set to obtain the prediction results of each model.

[0111] Step S3, constructing a meta-learner: inputting the prediction results of the base learner as new features into a meta-learner for training to obtain a meta-learner;

[0112] The goal of the meta-learner is to obtain the final prediction result by combining the predictions of the base learners. During training, the data is first input into the base learner for training. After the training is completed, each algorithm in the base learner outputs the corresponding prediction result, which is pieced together into a new feature matrix and then input into the secondary learner for training.

[0113] Step S4: Use the meta-learner to perform secondary prediction on the prediction results of the base learner to obtain the final prediction results; iterative training is performed until the set accuracy is met to obtain the trained multi-model fusion prediction model.

[0114] Optionally, the accuracy of each model prediction is evaluated based on MAE (mean absolute error) and RMSE (root mean square error) indicators.

[0115] The following is a specific training example to illustrate the above training process;

[0116] (1) Data preparation: Obtain training and test sets, and divide the training set into 10 parts: train1, train2, train3, train4, train5, train6, train7, train8, train9, train10 for subsequent 10-fold cross-validation.

[0117] (2) Select the base model: Select SVM, XGBoost, and GBDT as the base models of the Stacking regression model, and then perform training and prediction on the base models (taking XGBoost as an example).

[0118] (3) Training and prediction of the base model: train1, train2, train3, train4, train5, train6, train7, train8, train9, train10 are used as validation sets, and the remaining 9 sets are used as training sets. In each fold, the XGBoost regression model is trained using the training set, and predictions are made on the validation set to obtain the regression prediction value of the fold. After the 10-fold cross-validation is completed, the prediction values ​​of the 10 validation sets are merged vertically to obtain the prediction value matrix A1 generated by the XGBoost model on the training set (each row corresponds to an original training sample, and each column corresponds to a prediction value of a cross-validation fold. However, in regression tasks, the average or median of the predictions of each fold is usually taken as the final prediction value of the sample to reduce the number of features and avoid overfitting, and to obtain a prediction value vector A1' with the same number of samples in the original training set). A final XGBoost regression model is trained using the entire training set (or the average / ensemble of the 10-fold model), and predictions are made on the test set to obtain the prediction value B1 on the test set.

[0119] (4) Repeat the training and prediction of the base model: Repeat the above steps for the support vector machine (SVM) and GBDT models to obtain the prediction value vectors A2' and A3' on the training set, and the prediction values ​​B2 and B3 on the test set, respectively.

[0120] (5) Constructing a meta-model: Use the random forest regression model as the meta-model, and use the prediction value vectors A1', A2', and A3' of the three base models on the training set as new features, together with the target variable (continuous numerical value) of the original training set, to train the random forest regression model as the meta-model. The feature vector of each training sample consists of the prediction values ​​of the three base models, and the target variable is the original continuous numerical value.

[0121] (6) Metamodel prediction: Use the trained random forest regression metamodel to make predictions on the new feature vector constructed from the prediction values ​​B1, B2, and B3 of the three base models on the test set to obtain the final prediction value (continuous value).

[0122] In step 3, the fusion method of the primary prediction results of the base learners includes the following steps:

[0123] Step 31, obtaining the prediction accuracy of the base learner;

[0124] Specifically, the prediction accuracy can be counted based on the test set. During the use of the model, the data of the test set is updated at the set time. Based on the test results of the test set and the actual values, the prediction accuracy of each base learner is calculated.

[0125] Step 32: The accuracy is used as the fusion weight, and multiplied by the primary prediction result of the corresponding base learner to obtain the fused primary prediction result;

[0126] In this step, when the primary prediction results of the base learner are adjusted according to whether they are accurate, the features input to the meta-learner are adjusted based on the accuracy, which can significantly improve the overall prediction accuracy and robustness of the model.

[0127] The simulation experiments verified that the evaluation index of the fusion model of this embodiment is improved compared with the single prediction model. Figure 3 As shown, Figure 3 This is a comparative statistical chart between the true value and the predicted value of the bone density value prediction. The figure shows the comparison between the true value and the predicted value. It can be seen that the overall predicted value line of the model follows the true value line, and the prediction effect is good. There are some differences in some sample points.

[0128] Table 1 Prediction comparison results

[0129]

[0130]

[0131] In the table, FTTransformer: Transformer-based deep learning model, suitable for tabular data;

[0132] KNN (K-Nearest Neighbors): K nearest neighbor algorithm, used for classification and regression;

[0133] ResNet (Residual Network): Residual neural network, an architecture in deep learning;

[0134] Compared with the prediction results of the existing single model, as shown in Table 1, the mean absolute error MAE of the fusion model is 0.80, and the root mean square error RMSE is 1.02. The Stacking fusion model, which uses the SVM regression prediction model, XGBoost prediction model, and GBDT prediction model as primary learners and the random forest RandomForest prediction model as the secondary learner, has better prediction effect than a single model.

[0135] Example 2

[0136] Based on Example 1, this embodiment provides a multi-model fusion bone density value prediction system, including:

[0137] An acquisition module is configured to acquire clinical characteristic data for bone density prediction;

[0138] A primary prediction module is configured to construct a multi-model fusion prediction model including base learners and meta learners connected in sequence, input the acquired features into multiple base learners for prediction and recognition, and obtain a primary prediction result;

[0139] A fusion module is configured to adaptively calculate a fusion weight according to a prediction accuracy of a base learner, and fuse the primary prediction results of each base learner based on the calculated fusion weight;

[0140] The secondary prediction module is configured to transmit the fused primary prediction results to the meta-learner for prediction to obtain the prediction results of bone density.

[0141] It should be noted here that the various modules in this embodiment correspond one-to-one to the various steps in Example 1, and the specific implementation process is the same, which will not be repeated here.

[0142] Example 3

[0143] This embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps in the bone density value prediction method of multi-model fusion in embodiment 1 are completed.

[0144] Example 4

[0145] This embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps in the multi-model fusion bone density value prediction method of embodiment 1 are completed.

[0146] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

[0147] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A multi-model fusion bone density value prediction method, characterized in that: The steps include: Obtain clinical characteristic data for bone density prediction; Construct a multi-model fusion prediction model including base learners and meta-learners connected in sequence, input the acquired features into multiple base learners for prediction and recognition, and obtain primary prediction results; The fusion weight is adaptively calculated according to the prediction accuracy of the base learners, and the primary prediction results of each base learner are fused based on the calculated fusion weight; The fused primary prediction results are transmitted to the meta-learner for prediction to obtain the prediction results of bone density.

2. The bone density value prediction method of multi-model fusion according to claim 1, characterized in that: The constructed multi-model fusion prediction model includes two layers of models, the first layer of the model includes multiple base learners, and the second layer of the model includes a meta learner; The first-layer model setting includes a support vector machine model, an XGBoost model, and a GBDT model set in parallel; each base learner is trained and predicted independently to obtain primary prediction results; The meta-learner uses a random forest model. The meta-learner combines the prediction results of the first layer into a new feature vector for prediction to obtain the final prediction result.

3. The bone density value prediction method of multi-model fusion as claimed in claim 2, characterized in that: The support vector machine model is constructed, the objective function is optimized through the kernel function and Lagrange multiplier method, and GridSearchCV is used to tune the kernel function type and regularization parameters.

4. The bone density value prediction method of multi-model fusion as claimed in claim 2, characterized in that: The support vector machine model is constructed by optimizing the objective function through the kernel function and the Lagrange multiplier method. Specifically, the original problem is transformed into a dual problem through the Lagrange function, and the kernel function is used to map the input data to a high-dimensional space in order to find a better hyperplane in the new space to fit the data.

5. The bone density value prediction method of multi-model fusion as claimed in claim 2, characterized in that: The XGBoost model is constructed and the greedy algorithm is used to optimize the objective function of the XGBoost model.

6. The bone density value prediction method of multi-model fusion as claimed in claim 2, characterized in that: During the XGBoost model training process, set up strategies to prevent overfitting, including: In each iteration of XGBoost model training, the leaf node score of each tree is multiplied by the reduction weight so that the subsequently generated trees can further optimize the model; before splitting each layer of nodes, some features are randomly selected to determine the optimal split point; When training the XGBoost model, the continuous input features are divided into multiple buckets according to the preset quantiles. The samples between adjacent quantiles are divided into the same bucket, and each quantile is traversed to calculate the optimal division. Set the maximum tree depth and sample weight threshold to prevent the tree from growing without limit, and stop splitting when the gain is less than the set threshold.

7. The bone density value prediction method of multi-model fusion as claimed in claim 1, characterized in that: The fusion method of the primary prediction results of the base learners includes the following steps: Get the prediction accuracy of the base learner; The prediction accuracy is calculated based on the test set, the test set data is updated at a set time, and the accuracy of the prediction of each base learner is calculated based on the test results of the test set and the actual value; The accuracy is used as the fusion weight and multiplied by the primary prediction result of the corresponding base learner to obtain the primary prediction result after fusion.

8. A multi-model fusion bone density value prediction system, characterized in that: include: An acquisition module is configured to acquire clinical characteristic data for bone density prediction; A primary prediction module is configured to construct a multi-model fusion prediction model including base learners and meta learners connected in sequence, input the acquired features into multiple base learners for prediction and recognition, and obtain a primary prediction result; A fusion module is configured to adaptively calculate a fusion weight according to a prediction accuracy of a base learner, and fuse the primary prediction results of each base learner based on the calculated fusion weight; The secondary prediction module is configured to transmit the fused primary prediction results to the meta-learner for prediction to obtain the prediction results of bone density.

9. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the method for predicting bone density values ​​by fusion of multiple models as described in any one of claims 1 to 7 are completed.

10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps in the bone density value prediction method of multi-model fusion described in any one of claims 1-7.

Images