Method and system for predicting hemoglobin concentration of postoperative patient

By using extreme gradient lifting tree model and fusion filling method in patients after renal transplant surgery, combined with multiple machine learning algorithms, the prediction accuracy of hemoglobin concentration is improved, and the inaccurate prediction of hemoglobin concentration after renal transplantation in the prior art is solved. The problem of this achieves an accurate assessment of postoperative risks.

CN119028473APending Publication Date: 2024-11-26TONGJI HOSPITAL ATTACHED TO TONGJI MEDICAL COLLEGE HUAZHONG SCI TECH
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411102560.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-12
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The prediction accuracy of the hemoglobin concentration after renal transplantation is not high, and the lack of targeted improvements for organ transplantation surgery has led to the prediction of the hemoglobin concentration after surgery.

Method used

A method for predicting hemoglobin concentration in postoperative patients is adopted. By obtaining the patient's basic information and physical examination data, the pre-trained extreme gradient lifting tree model is used to predict. The missing data values ​​are filled by fusion filling method, and sample equalization is performed, combined with Multiple machine learning algorithm models are trained to improve prediction accuracy.

Benefits of technology

It improves the accuracy of prediction of postoperative hemoglobin concentration, helps doctors quickly identify patients' situation, predict and evaluate surgical risks in advance, and improves the prognostic effect of kidney transplant surgery.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119028473A_ABST
    Figure CN119028473A_ABST
Patent Text Reader

Abstract

The invention relates to a postoperative patient hemoglobin concentration prediction method and system. The method comprises the following steps: acquiring basic information and physical examination data of a patient; inputting the basic information and the physical examination data of the patient into the target classification prediction model to obtain a hemoglobin concentration prediction result of the patient; the target classification prediction model is a classification prediction model of which model parameters are determined after multiple machine learning algorithm models are trained and evaluated based on sample data sets of multiple cases in advance and meet preset conditions, and the machine learning algorithm model corresponding to the target classification prediction model comprises an extreme gradient boosting tree model; the construction of the sample data set comprises the steps that multiple pieces of original case data of multiple sample cases are obtained, preprocessing, namely data missing value filling and / or sample equalization processing is carried out on the multiple pieces of original case data, and data missing value filling adopts a fusion filling method; according to the method, the relationship between the same feature data of different cases and the relationship between different feature data of the same case are considered at the same time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The disclosed embodiments relate to the field of auxiliary medical technology, and in particular, to a method and system for predicting hemoglobin concentration in postoperative patients. Background Art

[0002] Kidney transplantation is one of the most common organ transplantation surgeries. It has become the preferred treatment for patients with end-stage renal disease. The correct prediction of various physical indicators of patients after kidney transplantation can estimate and predict the risk of complications after surgery, and has become one of the important contents of kidney transplantation prediction research. Anemia is one of the common complications after kidney transplantation, with an incidence of about 30%. There are many causes of anemia after kidney transplantation, the most common of which is delayed recovery of transplanted kidney function after kidney transplantation. The hemoglobin concentration of patients after kidney transplantation is a key medical test indicator for diagnosing anemia after kidney transplantation. Therefore, the correct prediction of hemoglobin concentration after kidney transplantation is of great significance for assisting clinical diagnosis and risk assessment after kidney transplantation.

[0003] In recent years, with the rapid development of computer technology and artificial intelligence technology, related technologies in the field of artificial intelligence have been widely used in the medical field. In particular, machine learning and deep learning methods have achieved relatively excellent results in the practice of medical prediction and have played an important role in assisting medical clinical diagnosis. Among the related technologies, there are schemes that use machine learning-related algorithms to predict patients' postoperative physical examination indicators, complications, and other conditions. However, for some organ transplant surgeries, such as kidney transplantation and liver transplantation, the model establishment considerations are still relatively simple, and there is a lack of targeted improvements for organ transplantation applications, which makes the prediction accuracy of postoperative hemoglobin concentration still not high, and needs further improvement. Summary of the invention

[0004] In order to solve the above technical problems or at least partially solve the above technical problems, the embodiments of the present disclosure provide a method and system for predicting the hemoglobin concentration of a postoperative patient.

[0005] In a first aspect, the present disclosure provides a method for predicting hemoglobin concentration in a postoperative patient, comprising: Obtain the patient's basic information and physical examination data; Inputting the basic information and physical examination data of the patient into a target classification prediction model to obtain a prediction result of the hemoglobin concentration of the patient; wherein the target classification prediction model is a classification prediction model whose model parameters meet preset conditions after training and evaluating multiple machine learning algorithm models based on sample data sets of multiple cases, and the machine learning algorithm model corresponding to the target classification prediction model includes an extreme gradient boosting tree model; And wherein, the construction of the sample data set includes the following methods: obtaining multiple original case data corresponding to multiple sample cases; preprocessing the multiple original case data respectively, and the preprocessing includes filling in missing data values ​​and / or sample balancing processing; wherein, the missing data value filling adopts a fusion filling method, and the fusion filling method simultaneously considers the relationship between the same feature data between different cases and the relationship between different feature data of the same case.

[0006] In one embodiment, the fusion filling method includes: determining the missing attribute of the missing feature data in each of the original case data; based on the missing attribute, using a corresponding filling algorithm to fill the missing data value; Among them, different missing attributes correspond to different filling algorithms, the missing attributes include a first missing attribute and a second missing attribute, the first missing attribute represents that the missing probability of the corresponding missing feature data is less than a first preset probability, and the second missing attribute represents that the missing probability of the corresponding missing feature data is greater than a second preset probability; the filling algorithm corresponding to the first missing attribute includes at least a K-nearest neighbor algorithm, and the filling algorithm corresponding to the second missing attribute includes at least a multi-layer perceptron filling algorithm.

[0007] In one embodiment, the construction of the sample data set further includes: For each original case data after preprocessing, the feature importance score ranking result of different feature data in each original case data after preprocessing is determined based on the extreme gradient boosting tree model, and the feature data with feature importance scores lower than the preset score are screened and removed based on the feature importance score ranking result to obtain preliminary case data; further add a recursive feature elimination algorithm to the extreme gradient boosting tree model, and screen the feature data in the preliminary case data based on this model to obtain the target case data and use it as the sample data in the sample data set.

[0008] In one embodiment, the method further comprises: Determine ideal clustering results and actual clustering results corresponding to sample data in the sample data set under different numbers of classification categories; Determine the ideal clustering results corresponding to different numbers of classification categories and the clustering distribution similarity between actual clustering and ideal clustering, and select the classification category number corresponding to the highest clustering distribution similarity and the ideal clustering result that meets the preset conditions as the optimal classification category number for actual classification of the sample data set.

[0009] In one embodiment, the plurality of machine learning algorithm models include a random forest model, an extreme gradient boosting tree model, a lightweight gradient boosting machine model, a linear support vector machine model, and a binomial support vector machine model; When the multiple machine learning algorithm models are trained, error correction output coding is used to optimize the linear support vector machine model and the binomial support vector machine model so that they can perform multi-classification prediction tasks; in addition, error correction output coding is added to the random forest model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model to improve the classification prediction accuracy of the model.

[0010] In one embodiment, the model parameters satisfy preset conditions including: the classification prediction accuracy of the model is greater than a first preset value, and the recall rate corresponding to the sample data with a specified label of the model is greater than a second preset value; wherein the specified label indicates an abnormal hemoglobin concentration.

[0011] In one embodiment, the sample equalization process adopts an adaptive comprehensive oversampling method.

[0012] In a second aspect, the present disclosure provides a system for predicting hemoglobin concentration in a postoperative patient, comprising: The acquisition module is used to obtain the patient's basic information and physical examination data; A prediction module, used for inputting the basic information and physical examination data of the patient into a target classification prediction model to obtain a prediction result of the hemoglobin concentration of the patient; wherein the target classification prediction model is a classification prediction model whose model parameters meet preset conditions after training and evaluating multiple machine learning algorithm models based on sample data sets of multiple cases, and the machine learning algorithm model corresponding to the target classification prediction model includes an extreme gradient boosting tree model; And wherein, the construction of the sample data set includes the following methods: obtaining multiple original case data corresponding to multiple sample cases; preprocessing the multiple original case data respectively, and the preprocessing includes filling in missing data values ​​and / or sample balancing processing; wherein, the missing data value filling adopts a fusion filling method, and the fusion filling method simultaneously considers the relationship between the same feature data between different cases and the relationship between different feature data of the same case.

[0013] In a third aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for predicting the hemoglobin concentration of a postoperative patient as described in any of the above embodiments.

[0014] In a fourth aspect, an embodiment of the present disclosure provides an electronic device, including: Processor; and Memory for storing computer programs; Wherein, the processor is configured to execute the method for predicting the hemoglobin concentration of a postoperative patient described in any of the above-mentioned embodiments by executing the computer program.

[0015] Compared with the prior art, the technical solution provided by the embodiments of the present disclosure has the following advantages: The method and system for predicting the hemoglobin concentration of a postoperative patient provided in the embodiments of the present disclosure obtain the basic information and physical examination data of the patient, and input the basic information and physical examination data of the patient into a target classification prediction model to obtain a prediction result of the hemoglobin concentration of the patient; wherein the target classification prediction model is a classification prediction model whose model parameters meet preset conditions after training and evaluating multiple machine learning algorithm models based on sample data sets of multiple cases, and the machine learning algorithm model corresponding to the target classification prediction model includes an extreme gradient boosting tree model; and wherein the construction of the sample data set includes the following methods: obtaining multiple original case data corresponding to multiple sample cases; preprocessing the multiple original case data respectively, and the preprocessing includes filling in missing data values ​​and / or sample balancing processing; wherein the missing data value filling adopts a fusion filling method, and the fusion filling method simultaneously considers the relationship between the same feature data between different cases and the relationship between different feature data of the same case, that is, considering the vertical correlation information and horizontal correlation information of the feature data. In this embodiment, targeted improvements are made for the application of organ transplant surgery, and the characteristics of the characteristic data of the case and the characteristics of the algorithm during model training are fully considered, so that the characteristics and the algorithm are adapted to each other, and the longitudinal correlation information and the horizontal correlation information of the characteristic data are considered simultaneously, so that the model can mine more abundant and accurate characteristic information to improve the prediction accuracy of the postoperative hemoglobin concentration of the trained model; in addition, during classification prediction, the imbalance of the sample data set will lead to the problem of classification result deviation, so the uneven data distribution will affect the classification prediction accuracy of the classification prediction model. In this embodiment, the data set can be simultaneously processed for data balancing to improve the problem of classification result deviation, and the best model that meets the conditions can be selected by combining the training of multiple models. On the whole, the prediction accuracy of postoperative hemoglobin concentration can be further improved, and then it can assist doctors to quickly identify and focus on the patient's condition according to the prediction results, accurately predict and evaluate the surgical risks in advance, and improve the prognosis of kidney transplant surgery. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0018] Figure 1 This is a flow chart of a method for predicting hemoglobin concentration in a postoperative patient according to an embodiment of the present disclosure; Figure 2 is a comparison chart of feature importance scores of data before and after preprocessing in the embodiment of the present disclosure; Figure 3 The feature retention results and the scores of each feature data in the feature screening stage of the embodiment of the present disclosure; Figure 4 is a graph showing the correlation of each feature of Lasso regression with the penalty term coefficient in the embodiment of the present disclosure; Figure 5 is the classification AUC value of the Lasso regression at different penalty coefficients in the embodiment of the present disclosure; Figure 6 The K-means clustering elbow method of the embodiment of the present disclosure determines the optimal number of clusters; Figure 7 is the silhouette coefficient of each number of clusters in K-means clustering of the embodiment of the present disclosure; Figure 8 is the test set confusion matrix of the classification prediction model selected in the embodiment of the present disclosure; Fig. 9 Schematic diagram of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0019] In order to more clearly understand the above-mentioned objectives, features and advantages of the present disclosure, the scheme of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0021] It should be understood that, in the following, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0022] Figure 1 The flowchart of a method for predicting the hemoglobin concentration of a postoperative patient according to an embodiment of the present disclosure is as follows. The method for predicting the hemoglobin concentration of a postoperative patient may include the following steps: Step S101: Obtain the patient's basic information and physical examination data.

[0023] Exemplarily, the patient in this embodiment may be a patient after an organ transplant, such as a patient after a kidney transplant or a liver transplant, but is not limited thereto. The following description takes a patient after a kidney transplant as an example, and is used for predicting abnormal hemoglobin concentration 30 days after a kidney transplant.

[0024] The patient's basic information may include, but is not limited to, gender, age, height, weight, body mass index (BMI), weight change, medical history, etc. Physical examination data may include examination data related to the prediction of hemoglobin concentration, such as preoperative blood routine, blood routine mean, extreme value, medication status within 30 days after surgery, time when each item of blood routine reaches the extreme value, urine test results within the last week, etc., without limitation.

[0025] Step S102: Input the basic information and physical examination data of the patient into the target classification prediction model to obtain the hemoglobin concentration prediction result of the patient; wherein the target classification prediction model is a classification prediction model whose model parameters meet the preset conditions after training and evaluating multiple machine learning algorithm models based on sample data sets of multiple cases, and the machine learning algorithm model corresponding to the target classification prediction model includes an extreme gradient boosting tree (XGBoost) model. And wherein, the construction of the sample data set includes the following methods: obtaining multiple original case data corresponding to multiple sample cases; preprocessing the multiple original case data respectively, and the preprocessing includes data missing value filling and / or sample balancing processing; wherein the data missing value filling adopts a fusion filling method, and the fusion filling method simultaneously considers the relationship between the same feature data between different cases and the relationship between different feature data of the same case.

[0026] Exemplarily, case screening is performed on real-world patient data of kidney transplantation, and case data that meet the use conditions are retained to determine the sample data set of the model. For example, based on data integrity and availability, the electronic medical record data of 958 kidney transplant patients from the Affiliated Hospital of Tongji Medical College of Huazhong University of Science and Technology from January 2017 to June 2019 can be included. In order to ensure data quality and the pertinence of the research, within the above selected scope, the following data screening conditions are defined, including but not limited to: (1) Both kidney transplant donors and recipients are adults; (2) The kidney transplantation operation is a single kidney transplantation for a single person; (3) The graft survival time after the operation exceeds 365 days; (4) The follow-up time for the patient after the operation is greater than 30 days; (5) The number of missing features for each case does not exceed 30%; Finally, 854 patient data, i.e., data sets, were included in the modeling.

[0027] Afterwards, a new fusion filling method can be used to fill the missing values ​​of each feature item of each case data in the data set to address the data missing problem in the collected real data set, and the effect of the fusion method used can be evaluated by comparing it with the classical method. In addition, an adaptive oversampling method is used to balance the data in the data set to avoid the impact of data imbalance on the classification prediction results.

[0028] In the sample data set, the predicted labels can be pre-classified into three categories: (1) the hemoglobin concentration 30 days after kidney transplantation is within the normal range, that is, the hemoglobin concentration is in the range of 115-150 g / L, and the label is set to 0; (2) the hemoglobin concentration 30 days after kidney transplantation is lower than the normal level, that is, the hemoglobin concentration is less than 115 g / L, and the label is set to 1; (3) the hemoglobin concentration 30 days after kidney transplantation is higher than the normal level, that is, the hemoglobin concentration is greater than 150 g / L, and the label is set to 2.

[0029] After setting the labels, the sample distribution of 854 data is normal: low: high = 390: 447: 17. The unbalanced sample distribution will make it difficult to predict the minority sample category in the subsequent classification prediction model establishment process. In this embodiment, the sample balance processing can use the adaptive comprehensive oversampling algorithm to generate samples to obtain a data set with a more balanced sample distribution. The sample distribution after processing by the adaptive comprehensive oversampling method is normal: low: high = 390: 447: 454, and the total data volume is expanded to 1291 cases, and the sample balance of the data is also relatively good.

[0030] In this embodiment, a prediction model for abnormal hemoglobin concentration in patients 30 days after kidney transplant surgery can be established. Five machine learning algorithm models, namely random forest, extreme gradient boosting tree, lightweight gradient boosting machine, linear support vector machine, and binomial support vector machine, are used to establish a multi-classification prediction model. The sample data set can include a training set and a test set. The ratio of the data volume of the training set to the test set is 8:2, but it is not limited to this. After that, each model can be trained and tested. During the model training process, the model effect can be comprehensively evaluated by using concrete methods such as confusion matrix and model indicator parameters such as recall rate, precision rate, F1 value, and accuracy rate combined with clinical actual significance and needs. The best model that meets the conditions is selected as the target classification prediction model.

[0031] At the end of the training, when the specific application is completed, the patient's basic information and physical examination data are input into the target classification prediction model to obtain the patient's hemoglobin concentration prediction results such as normal, low or high, where low or high means abnormal hemoglobin concentration.

[0032] The above scheme in this embodiment can improve the prediction accuracy of postoperative hemoglobin concentration as a whole, and then assist doctors to quickly identify and focus on the patient's condition based on the prediction results, accurately predict and evaluate the surgical risks in advance, and improve the prognosis of kidney transplant surgery.

[0033] In one embodiment, the fusion filling method includes: determining the missing attributes of the missing feature data in each of the original case data; based on the missing attributes, using a corresponding filling algorithm to fill in the missing data values; wherein different missing attributes correspond to different filling algorithms, the missing attributes include a first missing attribute and a second missing attribute, the first missing attribute represents that the missing probability of the corresponding missing feature data is less than a first preset probability, and the second missing attribute represents that the missing probability of the corresponding missing feature data is greater than a second preset probability; the filling algorithm corresponding to the first missing attribute includes at least a K-nearest neighbor algorithm, and the filling algorithm corresponding to the second missing attribute includes at least a multi-layer perceptron filling algorithm.

[0034] Exemplarily, the first preset probability and the second preset probability can be preset, and there is no limitation on this. Referring to Table 1, the missing attribute indicates the feature type corresponding to the missing feature data, and the specific implementation process of the fusion filling method, i.e., the new fusion method, can be: first, according to the clinical significance of each feature data in each case data and the missing probability of each feature data, all feature data of each case are divided into three categories: "non-missing", "not easy to miss", and "more likely to miss". "Non-missing" features include basic information of the case and preoperative medical examination information, which are rarely missing and are not the object of missing data filling; "not easy to miss" and "more likely to miss" feature data are filled in with missing values ​​using the K nearest neighbor algorithm in machine learning and the multi-layer perceptron algorithm in deep learning, respectively, based on their respective missing probabilities and taking into account the characteristics of the filling algorithm.

[0035] The specific filling process is as follows: first, the “not easy to miss” feature is filled using the K nearest neighbor algorithm. At this time, the longitudinal information of the feature data of different cases is considered. At the same time, due to the small amount of missing data of the “not easy to miss” feature data, it can better avoid the problem of low accuracy of the K nearest neighbor algorithm in filling predictions for rare categories when the reference data is small and the samples are unbalanced; then, the missing values ​​of the remaining “easy to miss” feature data are filled using the multi-layer perceptron algorithm. This method considers the horizontal information of different feature data of the same case. On the basis of the previous filling, the data filled by the K nearest neighbor algorithm is used as the feature variable to predict and fill the data of the remaining “easy to miss” features. When the multi-layer perceptron is used for filling, the feature filled in the last time becomes the feature variable for this filling, that is, the data column of the ath feature filled in the ath time is included as a new feature variable when predicting and filling the a+1th feature to predict and fill the a+1th feature.

[0036] This fusion method considers the relationship between the same feature data of different cases, that is, the longitudinal relationship information, and the relationship between different feature data of a case, that is, the horizontal relationship information. At the same time, the features are classified according to the clinical significance and missing probability of each feature, and filled with different algorithms, fully considering the characteristics of the features and the characteristics of the algorithms, so that the features and algorithms are adapted to each other, and the longitudinal and horizontal association information of the data table are considered at the same time. Table 1 shows the specific classification of the three feature types. The new fusion method is the KNN-MLP (K-Nearest Neighbor and Multilayer Perceptron) method. In some embodiments, cases with no missing data can be extracted from the sample data set, and 10% of the data are randomly missing to simulate the most serious case of missing data. That is, when the data are randomly missing, all feature data of a case except for rarely missing feature data such as basic information and preoperative medical examination information are deleted to simulate the most serious case of missing data. The random missing data set is used to test the effects of the three methods: mean filling, K nearest neighbor filling, and KNN-MLP filling.

[0037] For example, the above-mentioned 10% randomly missing data set can be used to test the effects of the three methods. The mean of the Euclidean distance between the real data bars of the missing data cases and the filled data bars in the n-dimensional space is selected as the evaluation criterion. n is the number of features that need to be filled. The smaller the Euclidean distance, the better the effect of the corresponding method. The test results are shown in Table 2.

[0038] Table 2 Comparison of the effects of different missing value filling methods. It can be seen that after using the KNN-MLP (K-Nearest Neighbor and Multilayer Perceptron) method, the missing data filling effect is better, so that the sample data set contains richer feature information, which can improve the prediction accuracy of postoperative hemoglobin concentration of the final trained model.

[0039] In order to further improve the prediction accuracy of the postoperative hemoglobin concentration of the trained model, in one embodiment, the construction of the sample data set may also include: for each original case data after preprocessing, determining the feature importance score ranking result of different feature data in each original case data after preprocessing based on the extreme gradient boosting tree model, and filtering and removing feature data with feature importance scores lower than a preset score based on the feature importance score ranking result to obtain preliminary case data; further adding a recursive feature elimination algorithm to the extreme gradient boosting tree model, and filtering the feature data in the preliminary case data based on this model to obtain target case data and use it as sample data in the sample data set.

[0040] Exemplarily, the extreme gradient boosting tree model is first used to score the importance of each feature data, and features with importance lower than a certain value are deleted. Based on this implementation scheme, the specific importance score of each feature can be obtained, providing an experimental basis for the doctor's decision-making and the study of the intrinsic mechanism of features and labels. At the same time, the effectiveness of the data preprocessing stage is verified according to the change of importance scores before and after data preprocessing, and it is also used as a comparison of the screening results of the main method (recursive feature elimination + extreme gradient boosting tree); then, a recursive feature elimination link is added to the extreme gradient boosting tree model to eliminate unnecessary features in turn, for example, based on the characteristics of recursive feature elimination, its characteristic is that only one feature with the lowest criticality is selected in one round, it is deleted, and then the next round is entered, iterating again and again. Based on this implementation scheme, the feature data finally included in the modeling is obtained, and the target case data is obtained and used as the sample data in the sample data set.

[0041] like Figure 2 As shown, in this embodiment, feature screening can be performed based on feature importance scores. Specifically, the extreme gradient boosting tree model can be used to rank the feature importance of each feature data of each case, for example, the feature importance of the data before and after the missing value filling and before and after the data balancing process is ranked, and the effectiveness and necessity of the missing value filling and data balancing process can be verified by observing whether there is a significant change in the importance score of each feature.

[0042] like Figure 3 As shown in the figure, the extreme gradient boosting tree model was used to filter and reduce the dimension of each feature data of each case based on the classification accuracy, retaining as little feature data as possible while retaining more information. The best feature retention number for the example obtained was 22, and the initial classification accuracy was 82.24%. The recursive feature elimination link was further added to optimize the extreme gradient boosting tree to further filter and reduce the feature data. The final result was to retain the top 25 feature data in importance, and the initial classification accuracy reached 83.73%.

[0043] like Figure 4 , Figure 5 As shown in Table 3, the feature selection methods commonly used in medical statistics, such as Lasso regression, are used to qualitatively verify the above feature retention results. Since Lasso regression can only be used in binary classification problems, the results of this method are only used as a basis for qualitative verification of the feature selection results. The verification results are shown in Table 3. It can be concluded from Table 3 that 21 of the 25 feature data selected appear at least once in the screening results of the other two feature screening methods used for comparison, and 10 of them appear in the screening results of the other two methods. It can be verified that the 25 selected features have a certain degree of universality. In this embodiment, 25 feature data are extracted from the 95 feature data of each case initially included as sample data to form a sample data set for inclusion in modeling, so that overfitting can be avoided while reducing computing and storage space and improving training efficiency. Avoiding overfitting can improve the generalization of the classification prediction model, thereby further improving the prediction accuracy of the postoperative hemoglobin concentration of the trained model.

[0044] In one embodiment, the method further includes: determining the ideal clustering results and actual clustering results corresponding to the sample data in the sample data set under different numbers of classification categories; determining the ideal clustering results corresponding to different numbers of classification categories and the clustering distribution similarity between the actual clustering and the ideal clustering, and selecting the number of classification categories corresponding to the highest clustering distribution similarity and the ideal clustering result when the ideal clustering result meets the preset conditions as the optimal number of classification categories when the sample data set is actually classified.

[0045] For example, the ideal clustering result satisfies the preset conditions, for example, the ideal clustering effect is the best or the ideal clustering is the best. Before the classification prediction model is established, the number of classification categories needs to be determined first, such as Figure 6 , Figure 7 As shown in Figure 4, the K-means clustering algorithm is used to cluster the sample data set, and the optimal number of classification categories for ideal clustering is obtained by the elbow method and the silhouette coefficient. The optimal number of classification categories finally determined by the elbow method and the silhouette coefficient is 3. Then, according to the clinical significance, the sample data set is set with two-category labels, three-category labels, and four-category labels to obtain the actual clustering results. Then, the clustering distribution similarity of the ideal clustering and the actual clustering under the three classification categories is compared using indicators such as mutual information and V-measure. Table 4 shows the clustering distribution similarity measurement indicators of the actual clustering and the ideal clustering in two-category, three-category, and four-category. As can be seen from Table 4, when the number of classification categories is 3, the clustering distribution similarity between the ideal clustering and the actual clustering is the highest. In this embodiment, the data set is ideally clustered without considering the actual clinical significance, and the clustering effect is best when the three classifications are obtained. Then, actual clustering is performed according to the clinical significance, and the actual clustering of the three classifications has the highest similarity with the corresponding ideal clustering. Therefore, when the data set is actually classified, the optimal number of classification categories should be determined to be 3, that is, three categories corresponding to label 0, label 1, and label 2. This method for determining the number of classification categories is named "optimal approximation method".

[0046] Table 4 Similarity between actual clustering and ideal clustering distribution for each classification number The accurate determination of the number of classification categories of the sample data set is also an influencing factor for the final prediction result of the model. In this embodiment, the appropriate number of classification categories is accurately determined and subsequent model training is performed based on the number of categories. In one embodiment, the multiple machine learning algorithm models include a random forest model, an extreme gradient boosting tree model, a lightweight gradient boosting machine model, a linear support vector machine model, and a binomial support vector machine model; when the multiple machine learning algorithm models are trained, error correction output coding is used to optimize the linear support vector machine model and the binomial support vector machine model to enable them to perform multi-classification prediction tasks; in addition, error correction output coding is added to the random forest model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model to improve the classification prediction accuracy of the model.

[0047] For example, the original linear support vector machine and binomial support vector machine can only perform binary classification predictions, so error correction output coding is needed here to optimize these two models so that they can perform multi-classification prediction tasks. As for the three tree models of random forest, extreme gradient boosting tree, and lightweight gradient boosting machine, they can complete multi-classification tasks by themselves. First, without adding error correction output coding, these three tree models are used to establish classification prediction models, and then error correction output coding is added to these three tree models to observe whether the classification prediction accuracy of the tree models after adding error correction output coding is improved.

[0048] Error-Correcting Output Codes is mainly used to solve multi-classification problems by applying binary classification algorithms. The method is to convert a multi-classification problem into multiple binary classification problems, and then apply the spatial distance method to obtain the final result. In this embodiment, the optimization link is migrated and applied before the multi-classification tree model to verify its effect. Error-Correcting Output Codes preprocesses the multi-class target class labels into binary codes (arrays of 0 and 1), and uses this strategy to represent the target class labels in the Euclidean space of binary codes. Assuming that M-bit binary codes are used to represent the category labels, it means that the multi-classification problem is decomposed into M binary classification problems. First, M binary classification models are trained separately, and then M models are used to predict the samples to be predicted. The prediction result of each binary classification model will obtain a 1-bit binary code (0 or 1). The final prediction result is an M-bit binary code. After that, the spatial distance between the binary code of the prediction result and the binary code of each classification label is calculated in the M-dimensional space, and the classification label with the smallest spatial distance to the binary code of the prediction result is taken as the final classification result of the prediction sample.

[0049] Table 5 shows the classification accuracy of various model algorithms after parameter adjustment in the classification model establishment stage. It can be clearly seen that the classification prediction effect of the tree model is significantly better than that of the linear model, and the error correction output coding also has a good optimization effect on the tree model. The classification prediction accuracy of the three tree models, random forest, extreme gradient boosting tree, and lightweight gradient boosting machine, after error correction output coding optimization, increased from 86.87%, 87.26%, and 87.64% to 87.64%, 89.19%, and 88.42%, respectively. The extreme gradient boosting tree optimized with error correction output coding, which has the best classification prediction effect, is selected as the final prediction model, and its classification accuracy is 89.19%.

[0050] Table 5 Comparison of classification prediction accuracy of each model This embodiment proposes a model construction method and implementation process for the goal of predicting abnormal hemoglobin concentration in renal transplant patients 30 days after surgery, further refines and improves the construction process of the medical clinical prediction model based on machine learning, and makes certain improvements to related technologies. A classification prediction model for abnormal hemoglobin concentration in patients 30 days after renal transplant surgery is constructed, and the accuracy of the final model reaches 89.19%.

[0051] In this embodiment, the best model that meets the conditions is selected from the training results of multiple models, that is, the extreme gradient boosting tree model with error correction output coding added as the final model, so as to further improve the prediction accuracy of postoperative hemoglobin concentration.

[0052] In one embodiment, the model parameters satisfying preset conditions may include: the classification prediction accuracy of the model is greater than a first preset value, and the recall rate corresponding to the sample data with a specified label of the model is greater than a second preset value; wherein the specified label indicates an abnormal hemoglobin concentration.

[0053] like Figure 8 As shown in the figure, after determining and constructing the classification prediction model, the model effect can be comprehensively evaluated by using concrete methods such as confusion matrix and indicators such as recall rate, precision rate, F1 value, and accuracy rate in combination with clinical practical significance and needs. The application demand of this embodiment is to predict the risk of adverse outcomes after kidney transplantation, so as to assist doctors to quickly identify and focus on the patient's condition, and to predict and evaluate the surgical risks in advance. Therefore, for the classification effect indicators of the model, in addition to the overall classification prediction accuracy, the recall rate of the abnormal group also needs to be focused on. It can be seen from Table 6 that the overall classification accuracy of the solution of this embodiment reaches 89%, and the recall rates of the abnormal low group (label 1) and high group (label 2) are as high as 89% and 99% respectively, which is good, and also meets the clinical needs targeted by the model.

[0054] Table 6 Evaluation indicators of classification prediction model In this embodiment, for the prediction of hemoglobin concentration of patients after kidney transplantation, the model parameters that are considered and the preset conditions that are met are specifically set, so that the final trained model can further improve the prediction accuracy of postoperative hemoglobin concentration.

[0055] The prediction method of hemoglobin concentration 30 days after renal transplantation in the above scheme in this embodiment can be roughly divided into 6 stages. First, the real-world data in the electronic medical record is sampled, and a relatively complete case sample that meets the prediction object is selected, and 854 case data are finally included. Subsequently, the data is preprocessed, mainly including data missing value filling and data balancing processing. The data missing values ​​are filled using mean filling, K nearest neighbor algorithm, and fused KNN-MLP method, and the filling effect is analyzed. Finally, the KNN-MLP method with the best effect is selected to process the data missing values, and the adaptive comprehensive oversampling method is used to balance the data. After that, the extreme gradient boosting tree with recursive feature elimination is used to analyze the importance of the features and screen the features at the same time. Finally, 25 features are retained for modeling, and Lasso regression is used to qualitatively verify the included features. Before the model is established, the number of classification categories is first determined, and the best actual clustering classification category number is selected based on the best ideal clustering and the clustering distribution similarity between the actual clustering and the ideal clustering, that is, two bases. A classification model for abnormal hemoglobin concentration 30 days after renal transplantation was established. Five machine learning models were initially selected, and three of the tree models were improved by adding error correction output coding. The classification prediction accuracy of random forest, extreme gradient boosting tree, and lightweight gradient boosting machine increased from 86.87%, 87.26%, and 87.64% to 87.64%, 89.19%, and 88.42%, respectively. The extreme gradient boosting tree with the best prediction effect and improved error correction output coding was selected as the final classification prediction model algorithm, and the classification prediction accuracy reached 89.19%. Finally, the classification prediction effect of the established classification model was evaluated in combination with clinical practical significance and needs.

[0056] It should be noted that, although the steps of the method in the present disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step for execution, and / or one step may be decomposed into multiple steps for execution, etc. In addition, it is also easy to understand that these steps may be, for example, executed synchronously or asynchronously in multiple modules / processes / threads.

[0057] The present disclosure provides a system for predicting hemoglobin concentration in postoperative patients, comprising: The acquisition module is used to obtain the patient's basic information and physical examination data; A prediction module, used for inputting the basic information and physical examination data of the patient into a target classification prediction model to obtain a prediction result of the hemoglobin concentration of the patient; wherein the target classification prediction model is a classification prediction model whose model parameters meet preset conditions after training and evaluating multiple machine learning algorithm models based on sample data sets of multiple cases, and the machine learning algorithm model corresponding to the target classification prediction model includes an extreme gradient boosting tree model; And wherein, the construction of the sample data set includes the following methods: obtaining multiple original case data corresponding to multiple sample cases; preprocessing the multiple original case data respectively, and the preprocessing includes filling in missing data values ​​and / or sample balancing processing; wherein, the missing data value filling adopts a fusion filling method, and the fusion filling method simultaneously considers the relationship between the same feature data between different cases and the relationship between different feature data of the same case.

[0058] In one embodiment, the fusion filling method includes: determining the missing attributes of the missing feature data in each of the original case data; based on the missing attributes, using a corresponding filling algorithm to fill in the missing data values; wherein different missing attributes correspond to different filling algorithms, the missing attributes include a first missing attribute and a second missing attribute, the first missing attribute represents that the missing probability of the corresponding missing feature data is less than a first preset probability, and the second missing attribute represents that the missing probability of the corresponding missing feature data is greater than a second preset probability; the filling algorithm corresponding to the first missing attribute includes at least a K-nearest neighbor algorithm, and the filling algorithm corresponding to the second missing attribute includes at least a multi-layer perceptron filling algorithm.

[0059] In one embodiment, the construction of the sample data set also includes: for each original case data after preprocessing, determining the feature importance score ranking result of different feature data in each original case data after preprocessing based on the extreme gradient boosting tree model, and filtering and removing feature data with feature importance scores lower than a preset score based on the feature importance score ranking result to obtain preliminary case data; further adding a recursive feature elimination algorithm to the extreme gradient boosting tree model, and filtering the feature data in the preliminary case data based on this model to obtain target case data and use it as sample data in the sample data set.

[0060] In one embodiment, the system may also include a determination module, which is used to: determine the ideal clustering results and actual clustering results corresponding to the sample data in the sample data set under different numbers of classification categories; determine the ideal clustering results corresponding to different numbers of classification categories and the clustering distribution similarity between the actual clustering and the ideal clustering, and select the number of classification categories corresponding to the highest clustering distribution similarity and the ideal clustering result that meets the preset conditions as the optimal number of classification categories when the sample data set is actually classified.

[0061] In one embodiment, the multiple machine learning algorithm models include a random forest model, an extreme gradient boosting tree model, a lightweight gradient boosting machine model, a linear support vector machine model, and a binomial support vector machine model; when the multiple machine learning algorithm models are trained, error correction output coding is used to optimize the linear support vector machine model and the binomial support vector machine model to enable them to perform multi-classification prediction tasks; in addition, error correction output coding is added to the random forest model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model to improve the classification prediction accuracy of the model.

[0062] In one embodiment, the model parameters satisfy preset conditions including: the classification prediction accuracy of the model is greater than a first preset value, and the recall rate corresponding to the sample data with a specified label of the model is greater than a second preset value; wherein the specified label indicates an abnormal hemoglobin concentration.

[0063] In one embodiment, the sample equalization process adopts an adaptive comprehensive oversampling method.

[0064] Regarding the system in the above embodiment, the specific manner in which each module performs operations and the corresponding technical effects brought about have been described in detail in the embodiment of the method, and will not be elaborated here.

[0065] It should be noted that although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiment of the present disclosure, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of a module or unit described above can be further divided into multiple modules or units for concretization. The components displayed as modules or units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. Those of ordinary skill in the art can understand and implement it without paying creative work.

[0066] The present disclosure also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for predicting the hemoglobin concentration of a postoperative patient described in any one of the above-mentioned embodiments is implemented.

[0067] Exemplarily, the readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0068] The computer readable storage medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The readable storage medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by an instruction execution system, an apparatus, or a device or used in combination with it. The program code contained on the readable storage medium may be transmitted with any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0069] The present disclosure also provides an electronic device, including a processor and a memory, wherein the memory is used to store a computer program, wherein the processor is configured to execute the method for predicting the hemoglobin concentration of a postoperative patient in any of the above embodiments by executing the computer program.

[0070] Refer to the following Fig. 9 The electronic device 600 according to this embodiment of the present invention is described. Fig. 9 The electronic device 600 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0071] like Fig. 9 As shown, the electronic device 600 is in the form of a general computing device. The components of the electronic device 600 may include but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different system components (including the storage unit 620 and the processing unit 610), a display unit 640, etc.

[0072] The storage unit stores program codes, which can be executed by the processing unit 610, so that the processing unit 610 performs the steps according to various exemplary embodiments of the present invention described in the above method embodiment section of this specification. For example, the processing unit 610 can perform the following steps: Figure 1 The steps of the method shown in .

[0073] The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 6201 and / or a cache memory unit 6202 , and may further include a read-only memory unit (ROM) 6203 .

[0074] The storage unit 620 may also include a program / utility 6204 having a set (at least one) of program modules 6205, such program modules 6205 including but not limited to: an operating system, one or more application programs, other program modules and program data, each of which or some combination may include the implementation of a network environment.

[0075] Bus 630 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0076] The electronic device 600 may also communicate with one or more external devices 700 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 600, and / or communicate with any device that enables the electronic device 600 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed through an input / output (I / O) interface 650. In addition, the electronic device 600 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) through a network adapter 660. The network adapter 660 may communicate with other modules of the electronic device 600 through a bus 630. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0077] Through the description of the above implementation methods, it is easy for those skilled in the art to understand that the example implementation methods described here can be implemented by software, or by combining software with necessary hardware. Therefore, the technical solution according to the implementation methods of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes a number of instructions to enable a computing device (which can be a personal computer, a server, or a network device, etc.) to execute the method steps of the above-mentioned embodiments according to the implementation methods of the present disclosure.

[0078] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0079] The above description is only a specific embodiment of the present disclosure, so that those skilled in the art can understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to the embodiments described herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting hemoglobin concentration in postoperative patients, characterized in that: include: Obtain the patient's basic information and physical examination data; Inputting the basic information and physical examination data of the patient into a target classification prediction model to obtain a prediction result of the hemoglobin concentration of the patient; wherein the target classification prediction model is a classification prediction model whose model parameters meet preset conditions after training and evaluating multiple machine learning algorithm models based on sample data sets of multiple cases, and the machine learning algorithm model corresponding to the target classification prediction model includes an extreme gradient boosting tree model; And wherein, the construction of the sample data set includes the following methods: obtaining multiple original case data corresponding to multiple sample cases; preprocessing the multiple original case data respectively, and the preprocessing includes filling in missing data values ​​and / or sample balancing processing; wherein, the missing data value filling adopts a fusion filling method, and the fusion filling method simultaneously considers the relationship between the same feature data between different cases and the relationship between different feature data of the same case.

2. The method according to claim 1, characterized in that The fusion filling method comprises: determining the missing attribute of the missing feature data in each of the original case data; based on the missing attribute, using a corresponding filling algorithm to fill the missing data value; Among them, different missing attributes correspond to different filling algorithms, the missing attributes include a first missing attribute and a second missing attribute, the first missing attribute represents that the missing probability of the corresponding missing feature data is less than a first preset probability, and the second missing attribute represents that the missing probability of the corresponding missing feature data is greater than a second preset probability; the filling algorithm corresponding to the first missing attribute includes at least a K-nearest neighbor algorithm, and the filling algorithm corresponding to the second missing attribute includes at least a multi-layer perceptron filling algorithm.

3. The method according to claim 1 or 2, characterized in that: The construction of the sample data set also includes: For each original case data after preprocessing, the feature importance score ranking result of different feature data in each original case data after preprocessing is determined based on the extreme gradient boosting tree model, and the feature data with feature importance scores lower than the preset score are screened and removed based on the feature importance score ranking result to obtain preliminary case data; further add a recursive feature elimination algorithm to the extreme gradient boosting tree model, and screen the feature data in the preliminary case data based on this model to obtain the target case data and use it as the sample data in the sample data set.

4. The method according to claim 3, characterized in that The method further includes: Determine ideal clustering results and actual clustering results corresponding to the sample data in the sample data set under different numbers of classification categories; Determine the ideal clustering results corresponding to different numbers of classification categories and the clustering distribution similarity between actual clustering and ideal clustering, and select the classification category number corresponding to the highest clustering distribution similarity and the ideal clustering result that meets the preset conditions as the optimal classification category number for actual classification of the sample data set.

5. The method according to claim 4, characterized in that The multiple machine learning algorithm models include a random forest model, an extreme gradient boosting tree model, a lightweight gradient boosting machine model, a linear support vector machine model, and a binomial support vector machine model; When the multiple machine learning algorithm models are trained, error correction output coding is used to optimize the linear support vector machine model and the binomial support vector machine model so that they can perform multi-classification prediction tasks; in addition, error correction output coding is added to the random forest model, the extreme gradient boosting tree model, and the lightweight gradient boosting machine model to improve the classification prediction accuracy of the model.

6. The method according to claim 5, characterized in that The model parameters satisfying preset conditions include: the classification prediction accuracy of the model is greater than a first preset value, and the recall rate corresponding to the sample data with a specified label of the model is greater than a second preset value; wherein the specified label indicates an abnormal hemoglobin concentration.

7. The method according to claim 5, characterized in that The sample equalization process adopts an adaptive comprehensive oversampling method.

8. A system for predicting hemoglobin concentration in postoperative patients, characterized in that: include: The acquisition module is used to obtain the patient's basic information and physical examination data; A prediction module, used for inputting the basic information and physical examination data of the patient into a target classification prediction model to obtain a prediction result of the hemoglobin concentration of the patient; wherein the target classification prediction model is a classification prediction model whose model parameters meet preset conditions after training and evaluating multiple machine learning algorithm models based on sample data sets of multiple cases, and the machine learning algorithm model corresponding to the target classification prediction model includes an extreme gradient boosting tree model; And wherein, the construction of the sample data set includes the following methods: obtaining multiple original case data corresponding to multiple sample cases; preprocessing the multiple original case data respectively, and the preprocessing includes filling in missing data values ​​and / or sample balancing processing; wherein, the missing data value filling adopts a fusion filling method, and the fusion filling method simultaneously considers the relationship between the same feature data between different cases and the relationship between different feature data of the same case.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting the hemoglobin concentration of a postoperative patient according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: include: processor; as well as Memory for storing computer programs; Wherein, the processor is configured to execute the method for predicting the hemoglobin concentration of a postoperative patient according to any one of claims 1 to 7 by executing the computer program.

Citation Information

Cited By

  • Data missing value filling method and device, equipment, medium and program

    CN119943244A

  • Data Missing Value Filling Method, Device, Equipment, Medium and Program

    CN119943244B