Ovarian tumor recurrence risk probability prediction method based on multi-model integration

Abstract

The invention discloses an ovarian tumor recurrence risk probability prediction method based on multi-model integration, and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring clinical detection data of a to-be-predicted target; wherein the clinical detection data comprises clinical detection data of ovarian cancer and clinical detection data of border ovarian tumor; inputting the clinical detection data into a pre-trained recurrence risk probability integration model, and outputting and obtaining an ovarian tumor recurrence risk probability prediction result of a to-be-predicted target; wherein the recurrence risk probability prediction integrated model is obtained by training a basic model and a meta-learning model through a training set and a test set, the training set and the test set are obtained by dividing a standard data set, and the standard data set is obtained by preprocessing multi-source clinical data. The accuracy and the stability of the ovarian tumor recurrence risk probability can be improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method for predicting the probability of ovarian tumor recurrence based on multi-model integration. Background Technology

[0002] Ovarian cancer and borderline ovarian tumors (BOT) belong to the same epithelial ovarian tumor spectrum and are common types of gynecological tumors. Ovarian cancer is one of the deadliest malignant tumors of the female reproductive system. Its early symptoms are insidious, and more than 70% of patients are already in an advanced stage at the time of diagnosis. Although the remission rate of first-line treatment with surgery combined with platinum-based chemotherapy can reach more than 80%, 70% of ovarian cancer patients will relapse within 2 to 3 years after initial treatment, and 70% of patients do not survive for more than 5 years, which significantly impacts patients' survival rate and quality of life.

[0003] In related technologies, traditional methods for predicting recurrence risk, such as logistic regression or Cox regression, rely on linear assumptions and are difficult to characterize the complex nonlinear relationships between multidimensional variables. Their prediction accuracy and generalization ability are limited. While single machine learning models can capture nonlinear features, they are easily affected by training data and may overfit. They also have poor adaptability to different central data or feature sets and lack stability.

[0004] Therefore, there is an urgent need for a method based on multi-model ensemble to predict the probability of ovarian tumor recurrence to solve the above-mentioned technical problems. Summary of the Invention

[0005] This invention provides a method for predicting the probability of ovarian tumor recurrence based on multi-model ensemble, which can improve the accuracy and stability of the probability prediction of ovarian tumor recurrence. The technical solution is as follows: On the one hand, a method for predicting the probability of ovarian tumor recurrence based on multi-model ensemble is provided, the method comprising: Acquire clinical testing data for the target to be predicted; wherein, the clinical testing data includes clinical testing data for ovarian cancer and clinical testing data for borderline ovarian tumors; The clinical test data is input into a pre-trained ensemble model for recurrence risk probability, and the output is a prediction result of the recurrence risk probability of the target ovarian tumor. The ensemble model for recurrence risk probability prediction is obtained by training a basic model and a meta-learning model on a training set and a test set. The training set and the test set are obtained by dividing a standard dataset. The standard dataset is obtained by preprocessing multi-source clinical data.

[0006] On the other hand, a multi-model ensemble-based device for predicting the probability of ovarian tumor recurrence risk is provided, the device comprising: The acquisition module is used to acquire clinical test data of the target to be predicted; wherein, the clinical test data includes clinical test data of ovarian cancer and clinical test data of borderline ovarian tumors; The prediction module is used to input the clinical test data into a pre-trained recurrence risk probability ensemble model and output the prediction result of the recurrence risk probability of the target ovarian tumor; wherein, the recurrence risk probability ensemble model is obtained by training the base model and the meta-learning model on the training set and the test set, the training set and the test set are obtained by dividing the standard dataset, and the standard dataset is obtained by preprocessing multi-source clinical data.

[0007] On the other hand, a computer device is provided, the computer device including a memory and a processor, the memory for storing computer programs, and the processor for executing the computer programs stored in the memory to implement the steps of the above-described method for predicting the probability of ovarian tumor recurrence based on multi-model integration.

[0008] On the other hand, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when the computer program is executed by a processor, it implements the steps of the above-described method for predicting the probability of ovarian tumor recurrence based on multi-model integration.

[0009] On the other hand, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the above-described method for predicting the probability of ovarian tumor recurrence based on multi-model integration.

[0010] The technical solution provided by this invention offers at least the following beneficial effects: First, by standardizing and preprocessing multi-center clinical data, a differentiated ensemble model targeting different tumor subtypes is constructed, thereby robustly capturing nonlinear relationships within complex multidimensional features and significantly improving the accuracy and stability of predictions. Second, through an ensemble learning framework, the advantages of multiple single machine learning models are effectively integrated, overcoming the shortcomings of traditional statistical methods, such as reliance on linear assumptions and limited prediction accuracy, as well as the tendency of single machine learning models to overfit and insufficient generalization ability. This method achieves efficient and automated real-time assessment of recurrence risk without increasing the burden on patients, providing clinicians with a high-value decision support tool and possessing significant potential for widespread application. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. 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.

[0012] Figure 1 This is a flowchart of a method for predicting the probability of ovarian tumor recurrence based on multi-model integration, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram comparing the area under the subject operating characteristic curves of ten machine learning algorithms for the BOT internal queue test set provided in an embodiment of the present invention. Figure 3 This is a schematic diagram comparing the area under the subject operating characteristic curves of ten machine learning algorithms for the BOT external queue test set provided in an embodiment of the present invention; Figure 4 This is a schematic diagram comparing the areas under the receiver operating characteristic curves of ten machine learning algorithms for an internal cohort test set of ovarian cancer according to an embodiment of the present invention. Figure 5 This is a schematic diagram comparing the areas under the receiver operating characteristic curves of ten machine learning algorithms for an internal cohort test set of ovarian cancer according to an embodiment of the present invention. Figure 6 This is a schematic diagram of the integrated model structure for recurrence risk probability provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the SHAP value and feature interaction score in a gradient boosting algorithm model for predicting BOT recurrence provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the SHAP value and feature interaction score in a gradient boosting algorithm model for predicting ovarian cancer recurrence provided in an embodiment of the present invention; Figure 9 This is a structural diagram of an ovarian tumor recurrence risk probability prediction device based on multi-model integration provided in an embodiment of the present invention; Figure 10 This is a hardware architecture diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0014] As mentioned earlier, traditional statistical methods struggle to handle complex nonlinear relationships, and single learning models are unable to integrate multidimensional information, resulting in inaccurate predictions.

[0015] Based on this, the concept of the present invention is to improve the prediction accuracy of the model by constructing a comprehensive prediction model that integrates multiple models and training it with multidimensional data.

[0016] The following describes the specific implementation of the above concept.

[0017] Please refer to Figure 1 This invention provides a method for predicting the probability of ovarian tumor recurrence based on multi-model ensemble, the method comprising: Step 100: Obtain clinical testing data for the target to be predicted; wherein, the clinical testing data includes clinical testing data for ovarian cancer and clinical testing data for borderline ovarian tumors; Step 102: Input the clinical test data into the pre-trained recurrence risk probability ensemble model and output the prediction result of the recurrence risk probability of the target ovarian tumor; wherein, the recurrence risk probability prediction ensemble model is obtained by training the basic model and the meta-learning model with the training set and the test set, the training set and the test set are obtained by dividing the standard dataset, and the standard dataset is obtained by preprocessing the multi-source clinical data.

[0018] In this embodiment of the invention, firstly, standardized preprocessing of multi-center clinical data is performed to construct differentiated ensemble models for different tumor subtypes, thereby robustly capturing nonlinear relationships in complex multidimensional features and significantly improving the accuracy and stability of predictions. Then, through an ensemble learning framework, the advantages of multiple individual machine learning models are effectively integrated, overcoming the shortcomings of traditional statistical methods, such as reliance on linear assumptions and limited prediction accuracy, as well as the tendency of individual machine learning models to overfit and insufficient generalization ability. This method achieves efficient and automated real-time assessment of recurrence risk without increasing the burden on patients, providing clinicians with a high-value decision support tool and possessing significant potential for widespread application.

[0019] The following description Figure 1 The execution method for each step is shown.

[0020] First, for step 100, obtain the clinical test data of the target to be predicted.

[0021] In this embodiment of the invention, the data used for prediction comes from real electronic medical record systems of multiple clinical centers, ensuring the diversity and representativeness of the sample. The data covers clinical data from multiple authoritative medical institutions. All data has undergone uniform de-identification and standardization processing to ensure consistency in variable definitions, entry formats, and data quality.

[0022] The collected information included patients' basic demographic characteristics, clinical features, laboratory test results, and pathological indicators, covering multidimensional characteristics of borderline ovarian tumors and ovarian cancer before, during, and after surgery. Variables covered key medical characteristics such as recurrence (overall, within 1 year, within 3 years, and within 5 years), tumor size, FIGO stage, serum CA125 and CA19-9 levels, surgical approach, and histological subtype.

[0023] Then, for step 102, the clinical test data is input into the pre-trained recurrence risk probability integrated model, and the output is the prediction result of the recurrence risk probability of the ovarian tumor of the target to be predicted.

[0024] In this embodiment of the invention, the recurrence risk probability ensemble model is obtained by training the base model and the meta-learning model using a training set and a test set. The training set and the test set are obtained by partitioning a standard dataset, which is obtained by preprocessing multi-source clinical data.

[0025] As described in step 100, the multi-source clinical data used in this step also covers clinical data from multiple authoritative medical institutions and includes key medical features used to train the prediction model.

[0026] In this embodiment of the invention, preprocessing of multi-source clinical data to obtain a standard dataset includes: Missing values ​​were imputed in the multi-source clinical data to obtain complete data.

[0027] Specifically, for continuous variables (such as age, tumor size, CA125 level, etc.), a neighbor interpolation algorithm based on KNNImputer is used for imputation, and the calculation formula is as follows: in, For missing sample estimates, Given j similar samples, The feature space Euclidean distance. For distance weights.

[0028] For categorical variables (such as surgical type, FIGO stage, pathological type, etc.), SimpleImputer's mode imputation method is used to restore missing terms to maintain the stability of the category distribution.

[0029] Furthermore, the complete data is subjected to feature standardization processing to obtain standard data that eliminates dimensional differences.

[0030] Specifically, to eliminate the dimensional differences between different features, all continuous variables are standardized using the Z-score, as shown in the following formula: Where x is the original feature value, The mean, The standard deviation is denoted as . Standardized parameters are saved during the training phase and applied to the validation and test sets to ensure cross-center consistency.

[0031] Furthermore, the standard data is subjected to sample balancing processing to obtain a standard dataset containing relapsed sample data that meets the proportion requirements.

[0032] Specifically, to address the issue of a low proportion of recurrent samples, the SMOTE oversampling algorithm is introduced. This algorithm performs linear interpolation on minority class samples in the feature space to generate new recurrent samples, thereby achieving class balance. The formula is as follows: in, For the original minority class samples, Its nearest neighbor samples This is a random disturbance factor.

[0033] After the above processing, a standard dataset with complete data structure, unified dimensions, and balanced samples is obtained, which can provide high-quality input for subsequent multi-model integrated training.

[0034] In this embodiment of the invention, the recurrence risk probability ensemble model is trained in the following manner: A grid search is performed on all preset base models, and the core hyperparameters of each model are tuned based on the search results.

[0035] Specifically, to address the issues of insufficient generalization ability of single algorithms, complex feature dimensions, and significant differences in data distribution across different centers in the prediction of recurrence of ovarian cancer and borderline ovarian tumors, a Stacking ensemble model is proposed. Through a multi-layered structure design, feature complementarity and information fusion among models are achieved, thereby improving prediction accuracy and cross-center stability.

[0036] This embodiment selects nine single basic machine learning models, including Decision Tree, RandomForest, Extra Trees, Gradient Boosting, AdaBoost, XGBoost, LightGBM, CatBoost, and MLP, and optimizes the core hyperparameters of each model through grid search.

[0037] Furthermore, the standard dataset is divided into a training set and a test set in a 3:1 ratio. The core hyperparameters of several preset basic learning models are optimized using the training set through a combination of grid search and three-fold cross-validation.

[0038] Furthermore, based on preset scoring indicators, all optimized basic models are screened, and the basic models that meet the preset requirements are determined as the underlying models in the stacking ensemble model.

[0039] In this embodiment, the model with the best performance is selected as the underlying model based on metrics such as AUC, recall, and specificity.

[0040] Specifically, since Extra Trees and Random Forest perform most evenly in terms of nonlinear feature expression and recall, based on the nonlinear feature expression and recall in the validation results, the Extremely Random Trees model and Random Forest model were determined as the underlying models of the stacking ensemble model for ovarian borderline tumors in order to extract diverse structural information of features. Since CatBoost, Extra Trees, Random Forest, and XGBoost demonstrate superior performance in handling class imbalance, nonlinear relationships, and high-dimensional features, these four models were selected as the underlying models for subsequent stacking of the ovarian cancer stacking ensemble model based on the class imbalance, nonlinear relationships, and high-dimensional feature indices in the validation results.

[0041] Finally, a Stacking ensemble model is constructed based on the underlying model and the preset meta-learning model, and the Stacking ensemble model is tested using a test set to obtain the final relapse risk probability ensemble model. In this embodiment of the invention, the relapse risk probability ensemble model is obtained through the following process: First, the standard dataset is input into the underlying model, and then the prediction results of each model are output sequentially.

[0042] Furthermore, the prediction results are fused according to a preset meta-learning model, and a loss function for optimizing the parameters of the top-level meta-learning model is constructed based on the fused prediction results.

[0043] Specifically, Stacking is an advanced ensemble machine learning technique. Its core idea is to intelligently combine the prediction results of multiple base learners through a meta-learner in order to achieve superior performance compared to any single model. In this embodiment, the meta-learning model of the ensemble model fuses the prediction outputs of each underlying model using the following formula: in, This indicates that the nth underlying model is related to the i-th input data X. i The predicted output, This represents the mapping function of the top-level meta-learner, used to synthesize multiple prediction results to generate the final prediction value. .

[0044] Subsequently, in order to obtain the optimal fusion parameters of the top-level meta-learning model, this embodiment trains the meta-learner by minimizing the binary cross-entropy loss, constructing the loss function L as shown in the following formula: In the formula, N is the number of data points in the standard dataset; This represents the true label of the i-th data point, where 0 indicates no recurrence and 1 indicates recurrence. This represents the fusion result of the top-level meta-learning model on the i-th data.

[0045] The loss function is minimized to obtain the top-level meta-learning model that optimizes the fused prediction result. In other words, by minimizing the loss function, the meta-learner can automatically learn the optimal combination of the outputs of the lower-level learners, thereby achieving accurate generation of the final prediction result.

[0046] The trained underlying model and the trained meta-learning model are saved together to obtain the Stacking ensemble model, and the test set is used to test it to obtain the recurrence risk probability ensemble model.

[0047] Specifically, to ensure the reproducibility of results, all model training used a fixed random seed and uniform standardized parameters. The final optimal model, after validation, was serialized and saved in PKL format for subsequent multi-center validation and clinical deployment.

[0048] It is worth noting that, since the underlying models used to predict ovarian cancer and BOT are different, the final recurrence risk probability ensemble models include the recurrence risk probability ensemble model for predicting ovarian cancer and the recurrence risk probability ensemble model for predicting BOT. Apart from the difference in the underlying models used, the overall structure and training process are the same, which will not be elaborated here.

[0049] In this embodiment of the invention, after the model is trained on the training set, its performance will be systematically evaluated on an independent test set. This embodiment visualizes the model's discriminative ability by plotting an ROC curve. The ROC curve uses the True Positive Rate (TPR) as the ordinate and the False Positive Rate (FPR) as the abscissa, depicting the model's overall performance at different classification thresholds (e.g., ...). Figures 2-5 (As shown in Table 1). To comprehensively evaluate the model performance, this approach uses multiple metrics, including AUC, sensitivity, specificity, and accuracy, and the statistical results are listed in Tables 1 and 2.

[0050] Table 1. Comparison of the sensitivity and specificity of ten machine learning algorithms in predicting BOT recurrence.

[0051] Table 2. Comparison of the sensitivity and specificity of ten machine learning algorithms in predicting ovarian cancer recurrence.

[0052] By calculating these metrics, this embodiment can comprehensively measure the predictive performance of each model. After multiple model training and validation cycles, the model with the best overall performance, Stacking Ensemble, is selected as the optimal predictive model, i.e., the ensemble model for predicting recurrence risk probability, and deployed in a real-world clinical setting. The Stacking Ensemble model structure is as follows: Figure 6 As shown.

[0053] To enhance the interpretability of the model, this approach employs the Shapley Additive Interpretation (SHAP) framework. This method calculates the marginal contribution of a feature when added to the model, then considers the different marginal contributions of that feature across all feature sequences, and takes the average as the feature's SHAP baseline value. The contribution of a feature to the model output is characterized by calculating its SHAP value. Finally, the features are ranked by their average absolute SHAP values ​​to understand the relative importance of each feature in the model. The visualization results are shown below. Figure 7 and Figure 8 As shown, Figure 7 The 14 features used to predict BOT recurrence are sorted from highest to lowest importance; Figure 8 Fourteen features used to predict ovarian cancer recurrence are ranked from most important to least important.

[0054] Please refer to Figure 9 This invention provides a device for predicting the probability of ovarian tumor recurrence based on multi-model ensemble, the device comprising: The acquisition module 900 is used to acquire clinical test data of the target to be predicted; wherein, the clinical test data includes clinical test data of ovarian cancer and clinical test data of borderline ovarian tumors; The prediction module 902 is used to input the clinical test data into a pre-trained recurrence risk probability ensemble model and output the prediction result of the recurrence risk probability of the target ovarian tumor; wherein, the recurrence risk probability ensemble model is obtained by training the basic model and the top-level meta-learning model on the training set and the test set, the training set and the test set are obtained by dividing the standard dataset, and the standard dataset is obtained by preprocessing multi-source clinical data.

[0055] In this embodiment of the invention, preprocessing of multi-source clinical data to obtain a standard dataset includes: Missing values ​​were imputed in multi-source clinical data to obtain complete data; The complete data is subjected to feature standardization processing to obtain standard data that eliminates dimensional differences; The standard data is subjected to sample balancing to obtain a standard dataset containing relapsed sample data that meets the proportion requirements.

[0056] In this embodiment of the invention, the recurrence risk probability ensemble model is trained in the following manner: The core hyperparameters of several pre-defined basic learning models are optimized using a combination of grid search and three-fold cross-validation on the training set. Based on the preset scoring criteria, all optimized basic models are screened, and the basic models that meet the preset requirements are determined as the underlying models. A Stacking ensemble model is constructed based on the underlying model and the preset meta-learning model, and the Stacking ensemble model is tested using a test set to obtain the final recurrence risk probability ensemble model.

[0057] In this embodiment of the invention, the basic model that satisfies the preset requirements is the underlying model, including: Based on the nonlinear feature expression and recall index in the validation results, the extremely random tree model and the random forest model were determined as the underlying models of the Stacking ensemble model for borderline ovarian tumors. Based on the class imbalance, nonlinear relationships, and high-dimensional feature indicators in the validation results, CatBoost, Extra Trees, Random Forest, and XGBoost were identified as the underlying models of the ovarian cancer Stacking ensemble model.

[0058] In this embodiment of the invention, a Stacking ensemble model is constructed based on the underlying model and a preset meta-learning model, and the Stacking ensemble model is tested using a test set to obtain the final recurrence risk probability ensemble model, including: The standard dataset is input into the underlying model, and the prediction results of each model are output sequentially. The prediction results are fused according to the preset top-level meta-learning model, and a loss function for optimizing the parameters of the top-level meta-learning model is constructed based on the fused prediction results. The loss function is minimized to obtain the top-level meta-learning model that optimizes the fusion prediction results; The Stacking ensemble model, constructed from the underlying model and the meta-learning model with optimal parameters, was tested using a test set to obtain the recurrence risk probability ensemble model.

[0059] In this embodiment of the invention, the meta-learning model is optimized using the following loss function: In the formula, L is the loss function; N is the number of data points in the standard dataset; This represents the true label of the i-th data point, where 0 indicates no recurrence and 1 indicates recurrence. This represents the fusion result of the top-level meta-learning model on the i-th data.

[0060] It should be noted that the ovarian tumor recurrence risk probability prediction device based on multi-model integration provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the ovarian tumor recurrence risk probability prediction device based on multi-model integration provided in the above embodiments and the ovarian tumor recurrence risk probability prediction method embodiment based on multi-model integration belong to the same concept. The specific implementation process is detailed in the method embodiment, and will not be repeated here.

[0061] Embodiments of this application also provide a computer device, please refer to... Figure 10 The computer device includes a processor and a memory, the memory storing at least one instruction, at least one program, code set, or instruction set, the at least one instruction, at least one program, code set, or instruction set being loaded and executed by the processor to implement the multi-model integration-based ovarian tumor recurrence risk probability prediction method provided in the above-described method embodiments.

[0062] Embodiments of this application also provide a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the multi-model integration-based ovarian tumor recurrence risk probability prediction method provided in the above-described method embodiments.

[0063] Embodiments of this application also provide a computer program product, which includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium and executes the computer program, causing the computer device to perform the ovarian tumor recurrence risk probability prediction method based on multi-model integration as described in any of the above embodiments.

[0064] For ease of description, the above systems or devices are described separately as various modules or units based on their functions. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware components.

[0065] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0066] Finally, it should be noted that in this document, relational terms such as first, second, third, and fourth are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0067] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for predicting the probability of ovarian tumor recurrence based on multi-model ensemble, characterized in that, The method includes: Acquire clinical testing data for the target to be predicted; wherein, the clinical testing data includes clinical testing data for ovarian cancer and clinical testing data for borderline ovarian tumors; The clinical test data is input into a pre-trained ensemble model for recurrence risk probability, and the output is a prediction result of the recurrence risk probability of the target ovarian tumor. The ensemble model for recurrence risk probability is obtained by training a basic model and a meta-learning model on a training set and a test set. The training set and the test set are obtained by dividing a standard dataset. The standard dataset is obtained by preprocessing multi-source clinical data.

2. The method as described in claim 1, characterized in that, Preprocessing of multi-source clinical data yields a standard dataset, including: Missing values ​​were imputed in multi-source clinical data to obtain complete data; The complete data is subjected to feature standardization processing to obtain standard data that eliminates dimensional differences; The standard data is subjected to sample balancing to obtain a standard dataset containing relapsed sample data that meets the proportion requirements.

3. The method as described in claim 1, characterized in that, The recurrence risk probability ensemble model was trained in the following manner: The core hyperparameters of several pre-defined basic learning models are optimized using a combination of grid search and three-fold cross-validation on the training set. Based on the preset scoring criteria, all optimized basic models are screened, and the basic models that meet the preset requirements are determined as the underlying models. A Stacking ensemble model is constructed based on the underlying model and the preset meta-learning model, and the Stacking ensemble model is tested using a test set to obtain the final recurrence risk probability ensemble model.

4. The method as described in claim 3, characterized in that, The basic model that meets the preset requirements is the underlying model, including: Based on the nonlinear feature expression and recall index in the validation results, the extremely random tree model and the random forest model were determined as the underlying models of the Stacking ensemble model for borderline ovarian tumors. Based on the class imbalance, nonlinear relationships, and high-dimensional feature indicators in the validation results, CatBoost, ExtraTrees, Random Forest, and XGBoost were identified as the underlying models of the ovarian cancer Stacking ensemble model.

5. The method as described in claim 3, characterized in that, A Stacking ensemble model is constructed based on the underlying model and the preset meta-learning model, and the Stacking ensemble model is tested using a test set to obtain the final recurrence risk probability ensemble model, including: The training set data is input into the underlying model, and the prediction results of each model are output sequentially. The prediction results are fused according to a preset meta-learning model, and a loss function for optimizing the parameters of the meta-learning model is constructed based on the fused prediction results. The loss function is minimized to obtain the meta-learning model that optimizes the fusion prediction results; The Stacking ensemble model, constructed from the underlying model and the meta-learning model with optimal parameters, was tested using a test set to obtain the recurrence risk probability ensemble model.

6. The method as described in claim 5, characterized in that, The meta-learning model is optimized using the following loss function: In the formula, L is the loss function; N is the number of data points in the standard dataset; This represents the true label of the i-th data point, where 0 indicates no recurrence and 1 indicates recurrence. This represents the fusion result of the top-level meta-learning model on the i-th data.

7. A device for predicting the probability of ovarian tumor recurrence risk based on multi-model ensemble, characterized in that, The device includes: The acquisition module is used to acquire clinical test data of the target to be predicted; wherein, the clinical test data includes clinical test data of ovarian cancer and clinical test data of borderline ovarian tumors; The prediction module is used to input the clinical test data into a pre-trained recurrence risk probability ensemble model and output the prediction result of the recurrence risk probability of the target ovarian tumor; wherein, the recurrence risk probability prediction ensemble model is obtained by training a basic model and a meta-learning model on a training set and a test set, the training set and the test set are obtained by dividing a standard dataset, and the standard dataset is obtained by preprocessing multi-source clinical data.

8. A computer device, characterized in that, The computer device includes a memory and a processor. The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to implement the steps of the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the method described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1-6.

Images