Osteoporosis secondary screening method and device, computer equipment and storage medium

Abstract

The embodiment of the invention discloses an osteoporosis secondary screening method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring medical data to be screened; preprocessing the medical data to be screened; performing feature extraction and combination on the processing result to obtain data features; inputting the data features into an osteoporosis secondary screening model for prediction to obtain a prediction result; outputting a prediction result; the osteoporosis secondary screening model comprises a primary osteoporosis screening model which is obtained by training a gradient lifting machine learning model through a sample set formed by processing medical data with labels indicating whether sample bone mass is abnormal or not; the second-level osteoporosis screening model is obtained by training a neural network through a sample set formed by performing feature fusion on an output result of the first-level osteoporosis screening model and a medical examination data subset. By implementing the method provided by the embodiment of the invention, the accuracy of screening high-risk groups with osteoporosis in middle-aged and elderly women can be improved.

Description

Technical Field

[0001] This invention relates to artificial intelligence, and more specifically to methods, devices, computer equipment, and storage media for secondary osteoporosis screening. Background Technology

[0002] Osteoporosis is the most common skeletal disease, a systemic skeletal disorder characterized by increased bone fragility and susceptibility to fractures. The incidence of osteoporosis increases with age, and it is most prevalent in middle-aged and elderly people, especially middle-aged and elderly women. Osteoporosis is an insidious disease; because there are no prior symptoms before the first fracture, its diagnosis is often only confirmed during hospitalization for non-violent fractures, hence its nickname "silent killer." However, osteoporosis is preventable and treatable; therefore, early screening and prevention are crucial for improving the quality of life for osteoporosis patients and reducing medical costs.

[0003] DXA (dual-energy X-ray absorptiometry) is the "gold standard" recommended by the World Health Organization for diagnosing osteoporosis. However, its adoption rate is low, and it is mainly concentrated in large hospitals. Osteoporosis patients are mostly concentrated in community and primary healthcare areas. Due to the lack of DXA equipment, many patients cannot receive timely diagnosis, leading to delayed treatment and increased risk of fractures. With the development of artificial intelligence technology, machine learning has been widely applied in various fields, especially the medical industry. In the era of big data, how to use computer technology to analyze existing medical big data and obtain knowledge models to assist clinical diagnosis is a hot topic in medical informatization. Currently, most osteoporosis screening studies use relatively simple input feature types, which cannot cope with different levels of medical scenarios, and there is a lack of relevant feature combinations to fully explore and utilize features to improve screening accuracy, affecting the effectiveness of osteoporosis screening.

[0004] Therefore, it is necessary to design a new method to improve the accuracy of screening for high-risk osteoporosis groups among middle-aged and elderly women. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, device, computer equipment and storage medium for secondary osteoporosis screening.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a secondary osteoporosis screening method, comprising:

[0007] Obtain medical data to be screened;

[0008] The medical data to be screened is preprocessed to obtain the processing results;

[0009] The processing results are then subjected to feature extraction and combination to obtain data features;

[0010] The data features are input into the osteoporosis secondary screening model to predict the probability of osteoporosis and obtain the prediction results.

[0011] Output the prediction result;

[0012] The osteoporosis secondary screening model includes a primary osteoporosis screening model and a secondary osteoporosis screening model. The primary osteoporosis screening model is obtained by training a gradient boosting machine learning model with a sample set formed by preprocessing, feature extraction, and combination of medical data labeled with abnormal bone mass. The secondary osteoporosis screening model is obtained by training a neural network with a sample set formed by feature fusion of the output results of the primary osteoporosis screening model and a subset of medical test data labeled with abnormal bone mass after preprocessing, feature extraction, and combination.

[0013] The further technical solution is as follows: the preprocessing of the medical data to be screened to obtain the processing result includes:

[0014] The medical data to be screened is divided into a non-medical testing data subset and a medical testing data subset based on whether medical testing is required.

[0015] The non-medical test data subset and the medical test data subset are preprocessed by data filtering, data cleaning and transformation, missing value imputation and normalization to obtain the processing results.

[0016] The further technical solution is as follows: the process of extracting and combining features from the processing results to obtain data features includes:

[0017] Feature selection is performed on the processing results based on the information gain ratio of real-time variables calculated by the decision tree to obtain the selection result;

[0018] Genetic transformation is used to generate and combine features from the selection results to obtain data features.

[0019] The further technical solution is as follows: The feature selection based on the information gain ratio of real-time variables calculated using a decision tree to obtain the selection result includes:

[0020] Calculate the feature importance of the non-medical test data subset in the processing result to obtain a first calculation result;

[0021] The feature importance of a subset of medical test data in the processing result is calculated to obtain a second calculation result;

[0022] The features corresponding to the first calculation result being greater than the information gain ratio threshold and the second calculation result being greater than the information gain ratio threshold are determined to obtain the selected features.

[0023] A further technical solution is as follows: the use of genetic transformation to generate and combine features from the selection results to obtain data features includes:

[0024] Genetic programming is used to generate new features from the original feature set using four rules and feature transformations for each feature in the selection results;

[0025] The process involves evolving and repeating the process of adding the new features to the original feature set and adding features with a defined metric to determine the data characteristics.

[0026] The further technical solution is as follows: the evolution and repeated addition of the new features to the original feature set and the addition of features with a set metric to determine data features includes:

[0027] To improve AUC, accept and add new features to the original feature set and add features with set metrics to determine data characteristics.

[0028] Its further technical solution is as follows: The primary osteoporosis screening model is obtained by training a gradient boosting machine learning model with a sample set formed by preprocessing, feature extraction, and combination of medical data labeled with abnormal bone mass, including:

[0029] Medical data labeled with whether or not the sample has abnormal bone mass are preprocessed, feature extracted, and combined to obtain a sample set;

[0030] The sample set is subjected to category attribute processing to obtain the category attribute processing result;

[0031] The CatBoost model is trained by combining the ranking boosting method with the processing results, and the base learners generated in each round are linearly added to obtain a first-level osteoporosis screening model.

[0032] The present invention also provides a secondary osteoporosis screening device, comprising:

[0033] The data acquisition unit is used to acquire the medical data to be screened.

[0034] A preprocessing unit is used to preprocess the medical data to be screened in order to obtain the processing result;

[0035] The feature extraction and combination unit is used to extract and combine features from the processing results to obtain data features;

[0036] The prediction unit is used to input the data features into the osteoporosis secondary screening model to predict the probability of osteoporosis and obtain the prediction result.

[0037] The output unit is used to output the prediction result.

[0038] The present invention also provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the above-described method.

[0039] The present invention also provides a storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0040] The beneficial effects of this invention compared with the prior art are as follows: This invention acquires medical data, performs preprocessing, feature generation and combination, and inputs the combination results into a secondary osteoporosis screening model for prediction. The secondary osteoporosis screening model is a model combining a CatBoost model and a TabNet model. The primary osteoporosis screening model and the secondary osteoporosis screening model formed by these two models are combined to predict the probability of osteoporosis, thereby improving the accuracy of screening for high-risk groups of osteoporosis among middle-aged and elderly women.

[0041] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 A schematic diagram illustrating an application scenario of the osteoporosis secondary screening method provided in this embodiment of the invention;

[0044] Figure 2 A schematic flowchart of a secondary osteoporosis screening method provided in an embodiment of the present invention;

[0045] Figure 3 A schematic diagram of a sub-process of the osteoporosis secondary screening method provided in an embodiment of the present invention;

[0046] Figure 4 A schematic diagram of a sub-process of the osteoporosis secondary screening method provided in an embodiment of the present invention;

[0047] Figure 5A schematic diagram of a sub-process of the osteoporosis secondary screening method provided in an embodiment of the present invention;

[0048] Figure 6 A schematic diagram of a sub-process of the osteoporosis secondary screening method provided in an embodiment of the present invention;

[0049] Figure 7 A schematic diagram of a sub-process of the osteoporosis secondary screening method provided in an embodiment of the present invention;

[0050] Figure 8 A schematic diagram of the genetic programming algorithm provided in an embodiment of the present invention;

[0051] Figure 9 A schematic diagram of the Ordered boosting algorithm provided in an embodiment of the present invention;

[0052] Figure 10 A schematic block diagram of a secondary osteoporosis screening device provided in an embodiment of the present invention;

[0053] Figure 11 A schematic block diagram of a computer device provided for an embodiment of the present invention. Detailed Implementation

[0054] 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, not all, of the embodiments of the present invention. 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.

[0055] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0056] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0057] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0058] Please see Figure 1 and Figure 2 , Figure 1 This is a schematic diagram illustrating an application scenario of the osteoporosis secondary screening method provided in an embodiment of the present invention. Figure 2 This is a schematic flowchart illustrating the secondary osteoporosis screening method provided in this embodiment of the invention. The method is implemented in a server. The server interacts with a terminal, which inputs the medical data to be screened. The server preprocesses, extracts, and combines the data, and then inputs the resulting data features into a secondary osteoporosis screening model to predict the probability of osteoporosis. A hybrid model of GP-CatBoost-TabNet is established to screen middle-aged and elderly women at high risk of osteoporosis. This model learns from experience and mines hidden knowledge within the data to make more effective and rational decisions.

[0059] Figure 2 This is a schematic flowchart of the secondary osteoporosis screening method provided in an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S150.

[0060] S110. Obtain medical data to be screened.

[0061] In this embodiment, the medical data to be screened includes multi-source data based on questionnaires and electronic medical record systems.

[0062] S120. The medical data to be screened is preprocessed to obtain the processing result.

[0063] In this embodiment, the processing result refers to the result formed after dividing the medical data to be screened and performing data filtering, data cleaning and transformation, missing value imputation, normalization and other processes.

[0064] In one embodiment, please refer to Figure 3 The above-mentioned step S120 may include steps S121 to S122.

[0065] S121. The medical data to be screened is divided into a non-medical test data subset and a medical test data subset according to whether medical testing is required.

[0066] In this embodiment, the multi-source medical data mainly includes demographic data, physical examination data, and laboratory test records. Considering that the ease of data acquisition affects the applicability of subsequent methods, the data is divided into two subsets based on whether medical testing is required: the Non-Medical Test Data Subset (NMLS) and the Medical Test Data Subset (MLS). The Non-Medical Test Data Subset includes age, height, weight, waist circumference, hip circumference, waist-to-hip ratio, body mass index, heart rate, menstrual period (days) - lower limit, menstrual period (days) - upper limit, menopause status, age at first marriage, number of full-term births, number of miscarriages, age at menarche, smoking status, alcohol consumption status, and whether MHT treatment has been received, etc. The subset of medical laboratory data includes luteinizing hormone (LH), follicle-stimulating hormone (FSH), estradiol, progesterone, total testosterone, prolactin, fasting blood glucose, insulin, total cholesterol, triglycerides, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), apolipoprotein A1, apolipoprotein B, lipoprotein(a), high-sensitivity C-reactive protein (HCRP), homocysteine, total triiodothyronine (TTI), free triiodothyronine (FTTH), total thyroxine, human thyroid-stimulating hormone (TSH), alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyl transferase (GGT), alkaline phosphatase (ALP), total protein, albumin, total bilirubin, direct bilirubin, indirect bilirubin, creatinine, urea, uric acid, whole blood viscosity (low shear 1), whole blood viscosity (low shear 5), whole blood viscosity (medium shear 30), whole blood viscosity (high shear 200), plasma viscosity, alpha-fetoprotein (AFP), carcinoembryonic antigen (CEA), carbohydrate antigen 19-9, carbohydrate antigen 125, cancer antigen 15-3, and DXA test results, etc.

[0067] Training a level-one osteoporosis screening model requires obtaining dual-energy X-ray absorptiometry (DXA) results from samples, which are then labeled by a professional physician to indicate whether the sample has normal (0) bone mass or abnormal bone mass.

[0068] S122. Perform data filtering, data cleaning and transformation, missing value imputation, and normalization preprocessing on the non-medical test data subset and the medical test data subset to obtain the processing results.

[0069] In this embodiment, data filtering refers to excluding data containing drugs that affect bone density and bone metabolism; data cleaning and transformation refers to converting the remaining data in terms of format, etc.; missing value imputation refers to filling in missing values; and normalization refers to the processing performed to avoid features with large value ranges affecting other features, while improving the convergence speed of the model.

[0070] For example, when training a primary osteoporosis screening model, the following subjects were selected: Middle-aged and elderly women aged 40-74 years who visited a joint outpatient clinic of a hospital and multiple community hospitals in a certain city, province, between January 2014 and May 2022 using the menopause-focused electronic medical record system. All subjects experienced menopausal symptoms, including irregular menstruation or natural menopause. Exclusion criteria included: First, patients with diabetes, hypertension, or other chronic diseases were excluded. Second, certain endocrine disorders such as thyroid and parathyroid diseases were excluded. Additionally, subjects with blood disorders, infectious diseases, malignant tumors, or immune diseases were excluded. Finally, subjects using medications that affect bone mineral density and bone metabolism were excluded.

[0071] When handling missing values, age and menopausal status are important data for the prediction model, so samples with missing age and menopausal status features need to be excluded; features with a missing rate of more than 50% in laboratory test features are deleted, because these features are considered difficult to obtain and will increase the difficulty of applying this method; then records with more than 30% of other missing values ​​in a single record are screened out.

[0072] Previous studies have confirmed a strong correlation between the occurrence of osteoporosis in middle-aged and elderly women and age and menopausal status. Therefore, this invention fully considers this characteristic and adopts a novel approach to imputation of missing data. This method bins the data according to age and chi-square, then groups the data within each bin according to menopausal status, and within each group, applies appropriate imputation methods based on different data types. The specific implementation is as follows:

[0073] Chi-square binning was performed on the data based on age: age was used as the characteristic variable for binning, bone mass status was used as the target variable, the significance level was set at 0.05, and the number of bins was 5.

[0074] Chi-merge binning is a bottom-up binning method. The idea is to initialize the original data into multiple age data intervals, merge samples from adjacent intervals, and calculate the chi-square value of the merged intervals. The chi-square value measures the difference in class distribution between adjacent intervals. A small chi-square value indicates that the class distributions of adjacent intervals are very similar, and interval merging is feasible; conversely, a large chi-square value indicates that the class distributions of adjacent intervals are different, and interval merging should not be performed. The chi-square value calculation formula is as follows: Among them, O ij This represents the observation frequency (the actual number of occurrences) between each age group and the target variable. E ij This represents the expected frequency (theoretically expected number) between each age segment and the dependent variable bone mass status in the absence of association.

[0075] After binning, the data was filled in according to the menopausal status. Five bins were obtained in the previous step. The age range and sample size of each bin are shown in Table 1.

[0076] Table 1. Information on each container in the Chi-square sub-compartment

[0077]

[0078]

[0079] Based on whether menopause has occurred, the samples in each bin were divided into two groups. For non-continuous data features such as number of full-term deliveries and menstrual period (days) minus the lower limit, the median within the group was used for imputation. For continuous data features such as estradiol and progesterone, Lagrange interpolation was used to imput missing values.

[0080] Lagrange interpolation uses the eigenvalues ​​x of known data points. i and label value y i and the feature values ​​x of other known data points j and label value y j We construct the interpolation polynomial L(x) by adding the product of the known data point label values ​​and the Lagrange basis functions. This yields the interpolation polynomial L(x) at the eigenvalue x. We can then use this polynomial to approximate the label values ​​at the eigenvalue x to fill in missing features.

[0081] During normalization, because the sample feature data have different dimensions and units, the differences between values ​​can affect the model. Therefore, it is necessary to normalize the data to prevent features with large value ranges from affecting other features, and to improve the convergence speed of the model. This embodiment uses min-max normalization, mapping the results to the range [0,1]. The formula is as follows:

[0082] Outlier screening employs the Raida criterion: using a given confidence probability of 99.7% as the standard and three times the standard deviation of the data set as the criterion, any measurement whose residual error in the prediction model exceeds three times σ should be removed. Where vb = |xb - x| > 3σ.

[0083] S130. The processing results are subjected to feature extraction and combination to obtain data features.

[0084] In this embodiment, feature extraction refers to determining the subset of data that needs to be feature-generated and combined, and then using genetic programming to generate the features.

[0085] In one embodiment, please refer to Figure 4The above-mentioned step S130 may include steps S131 to S132.

[0086] S131. Feature selection is performed on the information gain ratio of the real-time variables calculated based on the decision tree based on the processing results to obtain the selection results.

[0087] In this embodiment, the selection result refers to selecting variables whose information gain ratio is greater than the information gain ratio threshold for both data subsets.

[0088] In one embodiment, please refer to Figure 5 The above-mentioned step S131 may include steps S1311 to S1313.

[0089] S1311. Calculate the feature importance of the non-medical test data subset in the processing result to obtain the first calculation result.

[0090] In this embodiment, the first calculation result refers to the feature importance of each feature.

[0091] The feature importance of each feature was calculated using KNN (K-Nearest Neighbors), RF (Random Forest), SVM (Support Vector Machines), GBDT (Gradient Boosting Decision Trees), and GBM (Gradient Boosting Machines), and the average value was taken as the gain-ratio of the feature importance in the non-medical test data subset. NMLS Unimportant features are initially removed to improve computation speed and generalization ability. Gain-ratioNMLS represents the final importance index of a feature. KNN Gain-ratio represents the feature importance index derived from the KNN model. RF This represents the importance index of each feature derived from the RF model; the rest are similar.

[0092] S1312. Calculate the feature importance of the medical test data subset in the processing result to obtain a second calculation result.

[0093] In this embodiment, the second calculation result refers to the feature importance of a subset of medical test data.

[0094] Specifically, the feature importance of a subset of medical laboratory data is as follows:

[0095]

[0096] S1313. Determine the features corresponding to the first calculation result being greater than the information gain rate threshold and the second calculation result being greater than the information gain rate threshold, so as to obtain the selected features.

[0097] In this embodiment, for both data subsets, variables with information gain ratios greater than the threshold are selected, and 10% of the variable with the largest information gain is chosen as the threshold. Variables = {V|V Gain-ratio >0.1×max V Gain-rati}. Variables represents the set of variables that satisfy this requirement, and V represents an individual characteristic within the set, that is, one of the members of the set. V Gain-ratio This represents the Gain-ratio value associated with feature V, where Gain-ratio is a measure of feature importance. maxV Gain-ratio This represents the maximum Gain-ratio value among all features in the set.

[0098] S132. Genetic transformation is used to generate and combine features of the selection results to obtain data features.

[0099] In one embodiment, please refer to Figure 6 The above step S132 may include steps S1321 to S1322.

[0100] S1321. Generate new features from the original feature set using genetic programming with four rules and feature transformations for each feature in the selection results;

[0101] S1322. Evolve and repeat the process of adding the new features to the original feature set and adding features with a set metric to determine the data features.

[0102] The original feature set mentioned above refers to the set of selected features. Specifically, to improve AUC, new features are accepted and added to the original feature set, along with features with a defined metric, to determine the data characteristics.

[0103] In this embodiment, GP (Genetic Programming) is a machine learning technique based on evolutionary algorithms, primarily used to generate and optimize computer programs. In feature generation, genetic programming can be used to generate new features or transform existing features to improve the performance of machine learning models. The principle of feature generation through genetic programming is as follows: Figure 8 As shown.

[0104] The mean of the AUC (Average Value of Comparison) used in the five-fold cross-validation of the CatBoost model is taken as the fitness metric. Genetic programming is employed to generate a large number of new features from the original feature set using four rules and feature transformations for each feature. This process of evolution and repetition is repeated, adding these new features to the original dataset and adding features with the best metric. If the AUC does not improve after a specified number of transformations, the generated feature is rejected and not added to the existing feature set, and new features are generated again; only features that can improve the AUC are accepted and added to the variable set. In this way, the AUC will monotonically increase. After feature generation and updates, the non-medical test data subset is V. NMLS-new =V NMLS ∪V GP-NMLS The medical laboratory data subset has been updated to V. MLS-new =V MLS ∪V GP-MLS V NMLS V represents the features retained in NMLS after processing by S1312. GP-NMLS V represents NMLS The new feature obtained after processing by S132, V NMLS-new Indicates fusion V NMLS With V GP-NMLS The resulting new NMLS fusion feature set. V MLS V represents the features retained in MLS after processing by S1312. GP-MLS V represents MLS The new feature obtained after processing by S132, V MLS-new Indicates fusion V MLS With V GP-MLS The resulting new MLS fusion feature set.

[0105] S140. Input the data features into the osteoporosis secondary screening model to predict the probability of osteoporosis and obtain the prediction results.

[0106] In this embodiment, the osteoporosis secondary screening model includes a primary osteoporosis screening model and a secondary osteoporosis screening model. The primary osteoporosis screening model is obtained by training a gradient boosting machine learning model with a sample set formed by preprocessing, feature extraction, and combination of medical data labeled with abnormal bone mass. The secondary osteoporosis screening model is obtained by training a neural network with a sample set formed by feature fusion of the output results of the primary osteoporosis screening model and a subset of medical test data labeled with abnormal bone mass after preprocessing, feature extraction, and combination.

[0107] In one embodiment, please refer to Figure 7The aforementioned primary osteoporosis screening model is obtained by training a gradient boosting machine learning model with a sample set formed by preprocessing, feature extraction, and combination of medical data labeled with abnormal bone mass in the samples. This may include steps S130a to S130c.

[0108] S130a. Medical data labeled with whether or not the sample has abnormal bone mass are preprocessed, feature extracted, and combined to obtain a sample set.

[0109] In this embodiment, the specific process of preprocessing, feature extraction, and combination of medical data labeled with abnormal bone mass can be found in steps S120 to S130, and will not be repeated here.

[0110] S130b: Perform category attribute processing on the sample set to obtain the category attribute processing result.

[0111] In this embodiment, the category attribute processing result refers to the result formed after the sample set has undergone category attribute processing.

[0112] Specifically, NMLS has many categorical features. The general method for handling categorical features is one-hot encoding, but this can lead to overfitting. CatBoost uses a more efficient strategy to reduce overfitting when handling categorical features. A random sequence is generated for the training set, assuming the original order is σ = (σ1, ..., σn). The random sequence is traversed from σ1 to σn times, and the nominal feature values ​​are calculated using the first p records encountered. σp,k is used... The replacement here adds a prior value P and a parameter a > 0. This is a common practice that helps reduce noise from the low-frequency category.

[0113] S130c uses a sorting boosting method to train the CatBoost model in combination with the processing results, and linearly adds the base learners generated in each round to obtain a first-level osteoporosis screening model.

[0114] In this embodiment, predicting offset is often a problematic aspect of modeling. In each iteration of GBDT, the loss function uses the same dataset to calculate the gradient of the current model and then trains the base learner. However, this leads to gradient estimation bias, which in turn causes overfitting. CatBoost replaces the traditional gradient estimation method with ordered boosting, thereby mitigating gradient estimation bias and improving the model's generalization ability. The algorithm flow of ordered boosting is as follows: Figure 9As shown; a separate model Mi, trained using a training set that does not contain samples xi. We use Mi to obtain gradient estimates with respect to the samples, and use these gradients to train base learners to obtain the final model. CatBoost trains a separate model for each sample during training, thus obtaining unbiased gradient estimates. The final CatBoost Level 1 screening model is obtained by linearly summing the base learners generated in each round. F(x) represents the final model's prediction result for the input sample x, f i (x) represents the prediction result of the i-th base learner for sample x.

[0115] In this embodiment, the sample set for the secondary osteoporosis screening model is formed through feature fusion. Specifically, this feature fusion involves taking the new feature y output by the primary osteoporosis screening model. NMLS-pred V is a subset of medical laboratory data that has undergone preprocessing, feature extraction, and combination with medical data labeled with whether or not the samples have abnormal bone mass. MLS-new Feature fusion is performed to obtain the sample set V. final V final =y NMLS-pred ∪V MLS-new .

[0116] For the secondary osteoporosis screening model, the TabNet model is used. TabNet is a network structure specifically designed for tabular data, inheriting the advantages of tree models and DNNs. This allows it to achieve better model performance while maintaining interpretability. Good interpretability is crucial in medical decision support, helping users understand the model's decision-making process, providing decision support and trust, and improving decision transparency. The TabNet model is then used on the feature-fused dataset V. final After training is complete, the same data is used for prediction, and the final output is the prediction result y for osteoporosis. final and the prediction results of feature importance analysis;

[0117] TabNet uses a self-attention mechanism to calculate the importance of each feature. Given an input feature vector x and a set of decision point selections M for the current step (M is a binary vector representing which features are selected for decision-making), the feature importance is calculated as follows: in, Indicates feature x i The importance of step t, g (t) (x i () is a decision point selection network used to predict the outcome of a given feature x. i The probability of choosing this feature as the decision point under certain circumstances. It is an attention weight, representing the feature x i For feature x j The level of attention.

[0118] For the evaluation metrics of the aforementioned primary and secondary osteoporosis screening models, accuracy, precision, recall, and F1 score are specifically used to evaluate the model's screening effectiveness. Accuracy refers to the ratio of correctly classified samples to the total number of samples. Precision represents the proportion of samples predicted as positive that were actually positive. Recall represents the proportion of samples that were actually positive that were predicted as positive. The confusion matrix of the classification results is shown in Table 2.

[0119] Table 2. Confusion Matrix of Classification Results

[0120] Positive Negative True TP TN False FP FN

[0121] In this context, TP represents True Positives, TN represents True Negatives, FP represents False Positives, and FN represents False Negatives.

[0122]

[0123] This secondary osteoporosis screening model can be embedded in an electronic medical record system to achieve intelligent and dynamic screening for osteoporosis in middle-aged and elderly women.

[0124] S150, Output the prediction result.

[0125] In this embodiment, the prediction results are displayed on the terminal so that doctors and others can perform subsequent processing.

[0126] This embodiment employs a novel data bin filling method to achieve targeted data filling. It uses an ensemble model to extract highly relevant features from the dataset and a genetic programming algorithm to generate new features. By fusing features, it improves model accuracy and optimizes screening results. The constructed osteoporosis secondary screening model enables batch screening of osteoporosis in middle-aged and elderly women under different medical conditions, which is of great significance for supporting hierarchical diagnosis and treatment, assisting doctors' decision-making, and improving osteoporosis prognosis. The GP-CatBoost-TabNet model exhibits superior prediction accuracy compared to other prediction methods and has better generalization ability, better serving and supporting medical personnel's decision-making. Furthermore, osteoporosis screening based on electronic medical record datasets has enormous research potential. The method in this embodiment can further explore features closely related to osteoporosis, providing a reference for subsequent research. The method in this embodiment can serve as an outpatient auxiliary system, filling the gap in low-cost and efficient graded screening methods for osteoporosis. It aims to conduct more accurate, reliable, and efficient screening for osteoporosis, achieve early prevention and treatment of osteoporosis, reduce bone loss caused by osteoporosis, lower medical expenses, and improve the quality of life for middle-aged and elderly women.

[0127] Furthermore, the method in this embodiment is based on electronic medical record data, integrating demographic data, physical examination data, and laboratory test data, and utilizing machine learning models to comprehensively mine and analyze the data. It integrates multiple models for feature selection, combines genetic algorithms for feature generation to increase feature expressive power, fully utilizes features, and improves algorithm performance. It combines the advantages of machine learning and deep learning models. A CatBoost-TabNet-based osteoporosis screening and grading model fusion model for middle-aged and elderly women is constructed to optimize tiered diagnosis and treatment and improve screening efficiency. A more accurate and appropriate osteoporosis screening method is proposed, solving the problem of low availability of dual-energy X-ray absorptiometry (DXA) for large-scale screening. This method can achieve different levels of osteoporosis screening under different medical conditions, assisting doctors in decision-making and facilitating personalized assessment of osteoporosis status for high-risk individuals, enabling them to take corresponding prevention and treatment measures.

[0128] The aforementioned secondary osteoporosis screening method acquires medical data, performs preprocessing, feature generation and combination, and inputs the combined results into a secondary osteoporosis screening model for prediction. This secondary osteoporosis screening model is a combination of a CatBoost model and a TabNet model. The primary and secondary osteoporosis screening models formed by these two models are combined to predict the probability of osteoporosis, thereby improving the accuracy of screening for high-risk groups of osteoporosis among middle-aged and elderly women.

[0129] Figure 10This is a schematic block diagram of a secondary osteoporosis screening device 300 provided in an embodiment of the present invention. Figure 10 As shown, corresponding to the above-described osteoporosis secondary screening method, the present invention also provides an osteoporosis secondary screening device 300. This osteoporosis secondary screening device 300 includes a unit for performing the above-described osteoporosis secondary screening method, and the device can be configured in a server. Specifically, please refer to... Figure 10 The osteoporosis secondary screening device 300 includes a data acquisition unit 301, a preprocessing unit 302, a feature extraction and combination unit 303, a prediction unit 304, and an output unit 305.

[0130] The data acquisition unit 301 is used to acquire medical data to be screened; the preprocessing unit 302 is used to preprocess the medical data to be screened to obtain processing results; the feature extraction and combination unit 303 is used to extract and combine features from the processing results to obtain data features; the prediction unit 304 is used to input the data features into the osteoporosis secondary screening model to predict the probability of osteoporosis to obtain prediction results; and the output unit 305 is used to output the prediction results. The osteoporosis secondary screening model includes a primary osteoporosis screening model and a secondary osteoporosis screening model.

[0131] The aforementioned device also includes a primary model generation unit and a secondary model generation unit;

[0132] The first-level model generation unit is used to train the gradient improvement machine learning model by preprocessing, feature extraction, and combining medical data labeled with abnormal bone mass in the samples to form a sample set, thereby forming a first-level osteoporosis screening model.

[0133] The secondary model generation unit is used to train a neural network using the sample set formed by the preprocessing, feature extraction, and combination of a subset of medical test data labeled with abnormal bone mass of the primary osteoporosis screening model and the resulting sample set, to form a secondary osteoporosis screening model.

[0134] In one embodiment, the preprocessing unit 302 includes a partitioning subunit and a processing subunit.

[0135] The data to be screened is divided into a non-medical test data subset and a medical test data subset based on whether medical testing is required. The processing subunit is used to perform preprocessing on the non-medical test data subset and the medical test data subset, including data screening, data cleaning and transformation, missing value imputation, and normalization, to obtain the processing result.

[0136] In one embodiment, the feature extraction and combination unit 303 includes a selection subunit and a generation and combination subunit.

[0137] The selection subunit is used to select features from the information gain ratio of real-time variables calculated based on the decision tree in the processing results to obtain selection results; the generation and combination subunit is used to generate and combine features from the selection results using genetic transformation to obtain data features.

[0138] In one embodiment, the selection subunit includes a first calculation module, a second calculation module, and a determination module.

[0139] A first calculation module is used to calculate the feature importance of the non-medical test data subset in the processing result to obtain a first calculation result; a second calculation module is used to calculate the feature importance of the medical test data subset in the processing result to obtain a second calculation result; a determination module is used to determine the features corresponding to the first calculation result being greater than the information gain ratio threshold and the second calculation result being greater than the information gain ratio threshold to obtain selected features.

[0140] In one embodiment, the generating combination subunit includes a generating module and an evolution module.

[0141] A generation module is used to generate new features from the original feature set through genetic programming using four rules and feature transformations for each feature in the selection results; an evolution module is used to evolve and repeatedly add the new features to the original feature set and add features with a set metric to determine the data features. Specifically, in the case of improving AUC, new features are accepted and added to the original feature set and features with a set metric are added to determine the data features.

[0142] In one embodiment, the first-level model generation unit includes a first sample set acquisition subunit, a sample set processing subunit, and a training subunit.

[0143] The first sample set acquisition subunit is used to preprocess, extract features, and combine medical data labeled with whether or not the sample has abnormal bone mass to obtain a sample set; the sample set processing subunit is used to process the category attributes of the sample set to obtain the category attribute processing results; the training subunit is used to train the CatBoost model by using a ranking boosting method combined with the processing results, and the base learners generated in each round are linearly added to obtain a first-level osteoporosis screening model.

[0144] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned osteoporosis secondary screening device 300 and each unit can be referred to the corresponding description in the foregoing method embodiments. For the sake of convenience and brevity, it will not be repeated here.

[0145] The aforementioned osteoporosis secondary screening device 300 can be implemented as a computer program, which can perform tasks such as... Figure 11 It runs on the computer device shown.

[0146] Please see Figure 11 , Figure 11 This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device 500 can be a server, wherein the server can be a standalone server or a server cluster composed of multiple servers.

[0147] See Figure 11 The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a system bus 501. The memory may include a non-volatile storage medium 503 and internal memory 504.

[0148] The non-volatile storage medium 503 may store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions that, when executed, cause the processor 502 to perform a secondary osteoporosis screening method.

[0149] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0150] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can perform a secondary osteoporosis screening method.

[0151] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 500 to which the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0152] The processor 502 is used to run a computer program 5032 stored in the memory to perform the following steps:

[0153] Acquire medical data to be screened; preprocess the medical data to be screened to obtain processing results; extract and combine features from the processing results to obtain data features; input the data features into the osteoporosis secondary screening model to predict the probability of osteoporosis to obtain prediction results; output the prediction results.

[0154] The osteoporosis secondary screening model includes a primary osteoporosis screening model and a secondary osteoporosis screening model. The primary osteoporosis screening model is obtained by training a gradient boosting machine learning model with a sample set formed by preprocessing, feature extraction, and combination of medical data labeled with abnormal bone mass. The secondary osteoporosis screening model is obtained by training a neural network with a sample set formed by feature fusion of the output results of the primary osteoporosis screening model and a subset of medical test data labeled with abnormal bone mass after preprocessing, feature extraction, and combination.

[0155] In one embodiment, when the processor 502 performs the step of preprocessing the medical data to be screened to obtain the processing result, it specifically implements the following steps:

[0156] The medical data to be screened is divided into a non-medical test data subset and a medical test data subset based on whether medical testing is required. The non-medical test data subset and the medical test data subset are preprocessed by data screening, data cleaning and transformation, missing value imputation and normalization to obtain the processing results.

[0157] In one embodiment, when the processor 502 performs the step of extracting and combining features from the processing results to obtain data features, it specifically implements the following steps:

[0158] The processing results are used to select features based on the information gain ratio of real-time variables calculated by decision trees to obtain selection results; genetic transformation is used to generate and combine features from the selection results to obtain data features.

[0159] In one embodiment, when the processor 502 performs feature selection on the information gain ratio of the real-time variables calculated based on the decision tree to obtain the selection result, the following steps are specifically implemented:

[0160] The importance of features in the non-medical test data subset of the processing result is calculated to obtain a first calculation result; the importance of features in the medical test data subset of the processing result is calculated to obtain a second calculation result; features corresponding to the first calculation result being greater than the information gain ratio threshold and the second calculation result being greater than the information gain ratio threshold are determined to obtain selected features.

[0161] In one embodiment, when the processor 502 performs the step of calculating the generation and combination of features from the selection results using genetic transformation to obtain data features, it specifically implements the following steps:

[0162] Genetic programming is used to generate new features from the original feature set using four rules and feature transformations for each feature in the selection results; the new features are then evolved and repeated to add to the original feature set and to add features with a set metric to determine the data features.

[0163] In one embodiment, when the processor 502 implements the evolution and repeatedly adds the new features to the original feature set and adds features with a set metric to determine the data features, it specifically implements the following steps:

[0164] To improve AUC, accept and add new features to the original feature set and add features with set metrics to determine data characteristics.

[0165] In one embodiment, when the processor 502 implements the step of training a gradient boosting machine learning model using a sample set formed by preprocessing, feature extraction, and combining medical data labeled with abnormal bone mass in the samples, the processor 502 specifically implements the following steps:

[0166] Medical data labeled with abnormal bone mass are preprocessed, feature extracted, and combined to obtain a sample set. The sample set is then subjected to category attribute processing to obtain category attribute processing results. A CatBoost model is trained using a ranking boosting method combined with the processing results, and the base learners generated in each round are linearly added to obtain a first-level osteoporosis screening model.

[0167] It should be understood that in the embodiments of this application, the processor 502 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0168] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0169] Therefore, the present invention also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein when executed by a processor, the computer program causes the processor to perform the following steps:

[0170] Acquire medical data to be screened; preprocess the medical data to be screened to obtain processing results; extract and combine features from the processing results to obtain data features; input the data features into the osteoporosis secondary screening model to predict the probability of osteoporosis to obtain prediction results; output the prediction results.

[0171] The osteoporosis secondary screening model includes a primary osteoporosis screening model and a secondary osteoporosis screening model. The primary osteoporosis screening model is obtained by training a gradient boosting machine learning model with a sample set formed by preprocessing, feature extraction, and combination of medical data labeled with abnormal bone mass. The secondary osteoporosis screening model is obtained by training a neural network with a sample set formed by feature fusion of the output results of the primary osteoporosis screening model and a subset of medical test data labeled with abnormal bone mass after preprocessing, feature extraction, and combination.

[0172] In one embodiment, when the processor executes the computer program to perform the step of preprocessing the medical data to be screened to obtain a processing result, it specifically implements the following steps:

[0173] The medical data to be screened is divided into a non-medical test data subset and a medical test data subset based on whether medical testing is required. The non-medical test data subset and the medical test data subset are preprocessed by data screening, data cleaning and transformation, missing value imputation and normalization to obtain the processing results.

[0174] In one embodiment, when the processor executes the computer program to perform the step of extracting and combining features from the processing results to obtain data features, it specifically implements the following steps:

[0175] The processing results are used to select features based on the information gain ratio of real-time variables calculated by decision trees to obtain selection results; genetic transformation is used to generate and combine features from the selection results to obtain data features.

[0176] In one embodiment, when the processor executes the computer program to perform feature selection on the information gain ratio of the real-time variables calculated based on the decision tree to obtain the selection result, the processor specifically implements the following steps:

[0177] The importance of features in the non-medical test data subset of the processing result is calculated to obtain a first calculation result; the importance of features in the medical test data subset of the processing result is calculated to obtain a second calculation result; features corresponding to the first calculation result being greater than the information gain ratio threshold and the second calculation result being greater than the information gain ratio threshold are determined to obtain selected features.

[0178] In one embodiment, when the processor executes the computer program to implement the step of generating and combining features from the selection results using genetic mutation to obtain data features, it specifically implements the following steps:

[0179] Genetic programming is used to generate new features from the original feature set using four rules and feature transformations for each feature in the selection results; the new features are then evolved and repeated to add to the original feature set and to add features with a set metric to determine the data features.

[0180] In one embodiment, when the processor executes the computer program to implement the evolution and repeatedly adds the new features to the original feature set and adds features with a set metric to determine data features, it specifically implements the following steps:

[0181] To improve AUC, accept and add new features to the original feature set and add features with set metrics to determine data characteristics.

[0182] In one embodiment, when the processor executes the computer program to implement the step of training a gradient boosting machine learning model using a sample set formed by preprocessing, feature extraction, and combination of medical data labeled with abnormal bone mass, the processor specifically implements the following steps:

[0183] Medical data labeled with abnormal bone mass are preprocessed, feature extracted, and combined to obtain a sample set. The sample set is then subjected to category attribute processing to obtain category attribute processing results. A CatBoost model is trained using a ranking boosting method combined with the processing results, and the base learners generated in each round are linearly added to obtain a first-level osteoporosis screening model.

[0184] The storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0185] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0186] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0187] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the device of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0188] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, 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. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0189] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A secondary screening method for osteoporosis, characterized in that, include: Obtain medical data to be screened; The medical data to be screened is preprocessed to obtain the processing results; The processing results are then subjected to feature extraction and combination to obtain data features; The data features are input into the osteoporosis secondary screening model to predict the probability of osteoporosis and obtain the prediction results. Output the prediction result; The osteoporosis secondary screening model includes a primary osteoporosis screening model and a secondary osteoporosis screening model. The primary osteoporosis screening model is obtained by training a gradient boosting machine learning model with a sample set formed by preprocessing, feature extraction, and combination of medical data labeled with abnormal bone mass. The secondary osteoporosis screening model is obtained by training a neural network with a sample set formed by feature fusion of the output results of the primary osteoporosis screening model and a subset of medical test data labeled with abnormal bone mass after preprocessing, feature extraction, and combination.

2. The osteoporosis secondary screening method according to claim 1, characterized in that, The preprocessing of the medical data to be screened to obtain the processing result includes: The medical data to be screened is divided into a non-medical testing data subset and a medical testing data subset based on whether medical testing is required. The non-medical test data subset and the medical test data subset are preprocessed by data filtering, data cleaning and transformation, missing value imputation and normalization to obtain the processing results.

3. The osteoporosis secondary screening method according to claim 1, characterized in that, The step of extracting and combining features from the processing results to obtain data features includes: Feature selection is performed on the processing results based on the information gain ratio of real-time variables calculated by the decision tree to obtain the selection result; Genetic transformation is used to generate and combine features from the selection results to obtain data features.

4. The osteoporosis secondary screening method according to claim 3, characterized in that, The step of performing feature selection on the processing result based on the information gain ratio of real-time variables calculated using a decision tree to obtain the selection result includes: Calculate the feature importance of the non-medical test data subset in the processing result to obtain a first calculation result; The feature importance of a subset of medical test data in the processing result is calculated to obtain a second calculation result; The features corresponding to the first calculation result being greater than the information gain ratio threshold and the second calculation result being greater than the information gain ratio threshold are determined to obtain the selected features.

5. The osteoporosis secondary screening method according to claim 4, characterized in that, The method of using genetic transformation to generate and combine features from the selection results to obtain data features includes: Genetic programming is used to generate new features from the original feature set using four rules and feature transformations for each feature in the selection results; The process involves evolving and repeating the process of adding the new features to the original feature set and adding features with a defined metric to determine the data characteristics.

6. The osteoporosis secondary screening method according to claim 5, characterized in that, The evolution and repetition of adding the new features to the original feature set and adding features with a set metric to determine data features includes: To improve AUC, accept and add new features to the original feature set and add features with set metrics to determine data characteristics.

7. The osteoporosis secondary screening method according to claim 1, characterized in that, The primary osteoporosis screening model is obtained by training a gradient boosting machine learning model with a sample set formed by preprocessing, feature extraction, and combination of medical data labeled with whether or not the samples have abnormal bone mass. This includes: Medical data labeled with whether or not the sample has abnormal bone mass are preprocessed, feature extracted, and combined to obtain a sample set; The sample set is subjected to category attribute processing to obtain the category attribute processing result; The CatBoost model is trained by combining the ranking boosting method with the processing results, and the base learners generated in each round are linearly added to obtain a first-level osteoporosis screening model.

8. A secondary osteoporosis screening device, characterized in that, include: The data acquisition unit is used to acquire the medical data to be screened. A preprocessing unit is used to preprocess the medical data to be screened in order to obtain the processing result; The feature extraction and combination unit is used to extract and combine features from the processing results to obtain data features; The prediction unit is used to input the data features into the osteoporosis secondary screening model to predict the probability of osteoporosis and obtain the prediction result. The output unit is used to output the prediction result.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images