Granular computing-based method and device for grading and classifying prediction of new-onset myocardial infarction

Abstract

The application discloses a new-onset myocardial infarction grading and classification prediction method and device based on granular computing, which comprises the following steps: establishing a new-onset myocardial infarction patient sample from real clinical data; performing single-feature modeling analysis on a plurality of clinical features through an xgboost model to obtain an AUROC value, and retaining a plurality of first clinical features; performing modeling on the plurality of clinical features through a plurality of tree models to output feature importance score of each tree model, and retaining a plurality of second clinical features; performing intersection processing on the first clinical features and the second clinical features to obtain target clinical features; constructing an integrated model and performing training; outputting feature importance score of the trained integrated model, and performing remodeling analysis on clinical features arranged in the front of a set number to output feature importance score of a newly constructed model, and screening out the most important clinical features according to the feature importance score; and verifying the newly constructed model through internal and external data. The application can reduce the cost of collecting data and improve the robustness of the model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical artificial intelligence, in particular to a new-onset myocardial infarction grading and classification prediction method and device based on granular computing. BACKGROUND

[0002] Myocardial infarction (MI) is a serious acute cardiovascular disease that threatens human health, and its morbidity and mortality rates are high worldwide. Despite the continuous progress of medical technology in recent years, the prognosis of myocardial infarction patients is still not optimistic, especially the acute mortality and long-term disability rates are still high. Accurate and timely grading and classification of new-onset myocardial infarction is crucial for optimizing treatment plans and improving patient outcomes. Currently, clinicians mainly rely on electrocardiogram, myocardial enzyme spectrum and other traditional methods to diagnose and grade myocardial infarction, but these methods have certain limitations.

[0003] In recent years, with the rapid development of artificial intelligence technology, myocardial infarction prediction models based on machine learning, deep learning and other methods have gradually emerged. At present, although many machine learning-based methods have been used for risk prediction of myocardial infarction and have achieved good performance. However, they either require all patient admission test data or require additional electrocardiogram and other multi-modal data, which is both expensive and time-consuming for each patient. Granular computing, as a new computing model, can effectively handle complex, uncertain and incomplete information, providing a new way to solve the above problems.

[0004] Therefore, the grading and classification prediction of new-onset myocardial infarction still has challenges, and existing methods cannot effectively screen important features of new-onset myocardial infarction in clinical applications, and there is a lack of a simple, accurate and robust machine learning method. SUMMARY

[0005] Therefore, the present application provides a new-onset myocardial infarction grading and classification prediction method and device based on granular computing, which can effectively screen important features of new-onset myocardial infarction in clinical applications. At the same time, it reduces the cost of collecting data and improves the robustness of the model.

[0006] In order to achieve the above purpose, the present application provides the following technical scheme: a new-onset myocardial infarction grading and classification prediction method based on granular computing, comprising:

[0007] Collecting basic information and laboratory examination data of a plurality of patients from real clinical data; screening the collected basic information and laboratory examination data of the patients to establish a new-onset myocardial infarction patient sample cohort;

[0008] obtaining an AUROC value of each of the clinical features through single-feature modeling analysis of the clinical features by an xgboost model; and selecting a plurality of first clinical features to be retained according to a first selection strategy based on the AUROC values;

[0009] modeling the clinical features by a plurality of tree models, and outputting a feature importance score of each of the tree models; and selecting a plurality of second clinical features to be retained according to a second selection strategy based on the feature importance scores.

[0010] performing intersection processing on the first clinical features and the second clinical features to obtain a plurality of target clinical features.

[0011] constructing an ensemble model based on an automatic machine learning framework AutoGluon; and training the ensemble model based on the target clinical features to obtain a trained ensemble model.

[0012] analyzing the target clinical features by the trained ensemble model to output a feature importance score of the trained ensemble model; ranking the target clinical features according to the feature importance score of the ensemble model, and re-modeling a plurality of clinical features ranked in the front to output a feature importance score of a new model; and screening the most important clinical features when the feature importance score of the new model and the feature importance score of the ensemble model meet a set requirement.

[0013] verifying the new model by internal and external data to evaluate the robustness of the new model and the effectiveness of the most important clinical features.

[0014] As an optimal solution of the granular computing-based classification and prediction method for new-onset myocardial infarction, in the process of screening the basic information and the laboratory examination data of the collected patients and establishing the sample queue of the new-onset myocardial infarction patients, samples with new-onset myocardial infarction are screened from the basic information and the laboratory examination data of the patients; samples with a data missing rate greater than 20% are deleted from the screened samples, and the sample queue of the new-onset myocardial infarction patients is established.

[0015] As an optimal solution of the granular computing-based classification and prediction method for new-onset myocardial infarction, in the process of selecting a plurality of first clinical features to be retained according to the first selection strategy based on the AUROC values, the first selection strategy is to retain clinical features with an AUROC value greater than 0.5.

[0016] As a preferred scheme of the new-onset myocardial infarction grading classification prediction method based on granular computing, in the process of modeling a plurality of clinical characteristics through a plurality of tree models, the tree models include: xgboost, catboost and lightgbm tree models; the second selection strategy is: retaining the clinical characteristics whose feature importance score of xgboost, catboost and lightgbm three tree models is greater than 0.

[0017] As a preferred scheme of the new-onset myocardial infarction grading classification prediction method based on granular computing, in the process of training the ensemble model through the target clinical characteristics, the training step is:

[0018] The ensemble model applies a separate embedding layer to each of the target clinical characteristics, and connects the numerical characteristics into a large vector;

[0019] A group of base models are trained separately in a general way by using a multi-layer stack ensemble strategy to obtain an aggregated prediction result; the aggregated prediction result is used as a feature of the base model to train the ensemble model;

[0020] Based on the granular computing strategy, the target clinical characteristics are decomposed into a plurality of data sets; each of the data sets is further divided into a training subset and a validation subset; a plurality of training subsets or a plurality of validation subsets are combined into a final training subset or a final validation subset; the ensemble model is subjected to multi-granularity cross-validation through the final training subset and the final validation subset.

[0021] The application also provides a new-onset myocardial infarction grading classification prediction device based on granular computing, based on the above new-onset myocardial infarction grading classification prediction method based on granular computing, comprising:

[0022] A data screening module is used to collect basic information and laboratory examination data of a plurality of patients from real clinical data; the collected basic information and laboratory examination data of the patients are screened to establish a new-onset myocardial infarction patient sample queue;

[0023] A first clinical characteristic screening module is used to perform single-feature modeling analysis on a plurality of clinical characteristics through an xgboost model to obtain an AUROC value of each of the clinical characteristics; according to the AUROC value, a plurality of first clinical characteristics are selected according to a first selection strategy;

[0024] A second clinical characteristic screening module is used to model a plurality of clinical characteristics through a plurality of tree models, and output a feature importance score of each of the tree models; according to the feature importance score, a plurality of second clinical characteristics are selected according to a second selection strategy;

[0025] a target clinical feature acquisition module, configured to perform intersection processing on the first clinical features and the second clinical features to obtain target clinical features;

[0026] an integrated model construction and training module, configured to construct an integrated model based on an automatic machine learning framework AutoGluon, and train the integrated model based on the target clinical features to obtain a trained integrated model;

[0027] a most important clinical feature screening module, configured to analyze the target clinical features by the trained integrated model, output feature importance scores of the trained integrated model, sort the target clinical features according to the feature importance scores of the integrated model, re-model the clinical features in the front of the sorted list, output feature importance scores of a new model, and screen the most important clinical features when the feature importance scores of the new model and the feature importance scores of the integrated model meet a set requirement.

[0028] a new model verification module, configured to verify the new model by internal and external data, and evaluate robustness of the new model and effectiveness of the most important clinical features.

[0029] As an optimal scheme of the device for predicting classification and grading of new-onset myocardial infarction based on granular computing, in the process of screening the basic information of the patients and the laboratory examination data to establish the sample queue of the patients with new-onset myocardial infarction, samples with new-onset myocardial infarction are screened from the basic information of the patients and the laboratory examination data; samples with a data missing rate greater than 20% are deleted, and the sample queue of the patients with new-onset myocardial infarction is established.

[0030] As an optimal scheme of the device for predicting classification and grading of new-onset myocardial infarction based on granular computing, in the process of selecting the first clinical features according to the first selection strategy based on the AUROC values, the first selection strategy is to retain clinical features with an AUROC value greater than 0.5.

[0031] As an optimal scheme of the device for predicting classification and grading of new-onset myocardial infarction based on granular computing, in the process of modeling the clinical features by the tree models, the tree models include xgboost, catboost and lightgbm tree models; and the second selection strategy is to retain clinical features with feature importance scores greater than 0 of the three tree models.

[0032] As a preferred scheme of the new granular computing-based classification and prediction device for grading new myocardial infarction, in the integrated model construction and training module, the training submodule comprises:

[0033] A large vector generation submodule is configured to apply a separate embedding layer to each target clinical feature and connect the numerical features into a large vector.

[0034] A multi-layer stack integrated training submodule is configured to use a multi-layer stack integration strategy to separately train a group of base models in a general manner to obtain an aggregated prediction result, and use the aggregated prediction result as a feature of the base model to train the integrated model.

[0035] A granular computing training submodule is configured to decompose the target clinical features into a plurality of data sets based on a granular computing strategy, each of the data sets is further divided into a training subset and a validation subset, a plurality of the training subsets or a plurality of the validation subsets are combined into a final training subset or a final validation subset, and the integrated model is subjected to multi-granularity cross-validation through the final training subset and the final validation subset.

[0036] The present application has the following advantages: the present application collects basic information and laboratory examination data of a plurality of patients from real clinical data; the collected basic information and laboratory examination data of the patients are screened to establish a sample queue of new myocardial infarction patients; a single feature modeling analysis is performed on a plurality of clinical features by an xgboost model to obtain an AUROC value of each clinical feature; according to the AUROC value, a first selection strategy is used to select a plurality of first clinical features; a plurality of tree models are used to model a plurality of clinical features to respectively output feature importance scores of each tree model; according to the feature importance scores, a second selection strategy is used to select a plurality of second clinical features; an intersection processing is performed on the plurality of first clinical features and the plurality of second clinical features to obtain a plurality of target clinical features; an integrated model is constructed based on an automatic machine learning framework AutoGluon; the integrated model is trained by the target clinical features to obtain a trained integrated model; the target clinical features are analyzed by the trained integrated model to output feature importance scores of the trained integrated model; the target clinical features are sorted according to the feature importance scores of the integrated model, and a plurality of clinical features in the front are re-modeled and analyzed to output feature importance scores of a new model; when the feature importance scores of the new model and the feature importance scores of the integrated model meet a set requirement, the most important clinical features are screened out; the new model is verified by internal and external data to evaluate the robustness of the new model and the effectiveness of the most important clinical features. The present application solves the challenge of high cost of clinical data collection in new myocardial infarction classification and prediction. By modeling and analyzing a large number of clinical features of patients, the important features of new myocardial infarction occurrence in clinical application are accurately screened out based on the granular computing method; the new myocardial infarction classification and prediction model constructed by the present method only needs to input common clinical features (such as blood routine) of patients, reduces the cost of data collection, and improves the robustness of the model. The present application introduces automatic machine learning, multi-layer stacking integration strategy and granular computing data division method to further improve the accuracy and robustness of the model; the present application has stable performance in real clinical internal and external verification sets. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed in the following embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only exemplary, and those skilled in the art can also obtain other implementation drawings according to the provided drawings without creative labor.

[0038] The structures, proportions, sizes, etc. shown in the specification are only used to cooperate with the content disclosed in the specification, to be understood and read by those skilled in the art, and are not used to limit the conditions that can be implemented by the application, so they do not have technical significance. Any modification of the structure, change of the proportion relationship or adjustment of the size, without affecting the effect and purpose that can be achieved by the application, should still fall within the scope covered by the disclosed technology.

[0039] Figure 1 A flowchart of a new myocardial infarction grading classification prediction method based on granular computing provided in embodiment 1 of the application is shown in the figure.

[0040] Figure 2 A specific implementation flowchart of the new myocardial infarction grading classification prediction method based on granular computing provided in embodiment 1 of the application is shown in the figure.

[0041] Figure 3 A schematic diagram of the multi-layer stack integration strategy and the granular computing strategy in the new myocardial infarction grading classification prediction method based on granular computing provided in embodiment 1 of the application is shown in the figure.

[0042] Figure 4 A schematic diagram of the feature importance ranking of the integrated model in the new myocardial infarction grading classification prediction method based on granular computing provided in embodiment 1 of the application is shown in the figure.

[0043] Figure 5 A schematic diagram of the AUROC value of the newly created model in the test set in the new myocardial infarction grading classification prediction method based on granular computing provided in embodiment 1 of the application is shown in the figure.

[0044] Figure 6 A schematic diagram of the architecture of the new myocardial infarction grading classification prediction device based on granular computing provided in embodiment 2 of the application is shown in the figure. DETAILED DESCRIPTION

[0045] The embodiments of the application are described below by specific examples, and those skilled in the art can easily understand other advantages and effects of the application from the disclosed content. Obviously, the described embodiments are part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0046] Embodiment 1

[0047] Referring to Figure 1 and Figure 2 , the new myocardial infarction grading classification prediction method based on granular computing provided in embodiment 1 of the application includes the following steps:

[0048] S1, collecting basic information and laboratory examination data of a plurality of patients from real clinical data; screening the collected basic information and laboratory examination data of the patients to establish a sample queue of new-onset myocardial infarction patients;

[0049] S2, performing single-feature modeling analysis on a plurality of clinical features by an xgboost model to obtain an AUROC value of each of the clinical features; and selecting a plurality of first clinical features according to a first selection strategy according to the AUROC values.

[0050] S3, modeling a plurality of the clinical features by a plurality of tree models to respectively output a feature importance score of each of the tree models; and selecting a plurality of second clinical features according to a second selection strategy according to the feature importance scores.

[0051] S4, performing intersection processing on the plurality of first clinical features and the plurality of second clinical features to obtain a plurality of target clinical features.

[0052] S5, constructing an integrated model based on an automatic machine learning framework AutoGluon; training the integrated model based on the target clinical features to obtain a trained integrated model.

[0053] S6, analyzing the target clinical features by the trained integrated model to output a feature importance score of the trained integrated model; ranking the target clinical features according to the feature importance score of the integrated model, and re-modeling the clinical features ranked in the front to output a feature importance score of a new model; and screening the most important clinical features when the feature importance score of the new model and the feature importance score of the integrated model meet a set requirement.

[0054] S7, verifying the new model by internal and external data to evaluate the robustness of the new model and the effectiveness of the most important clinical features.

[0055] In the embodiment, in step S1, basic information and laboratory examination data of a plurality of patients are collected from real clinical data; the collected basic information and laboratory examination data of the patients are screened to establish a sample queue of new-onset myocardial infarction patients.

[0056] In the process of screening the collected basic information and laboratory examination data of the patients to establish the sample queue of new-onset myocardial infarction patients, samples suffering from new-onset myocardial infarction are screened from the basic information and laboratory examination data of the patients; samples with a data missing rate greater than 20% in the screened samples are deleted to establish the sample queue of new-onset myocardial infarction patients.

[0057] Specifically, 102 clinical characteristics of 3556 patients were collected from real clinics, 1018 (28.6%) of which had new myocardial infarction in hospital, and the rest of the patients did not have myocardial infarction; samples with a missing rate greater than 20% in the data were deleted, and finally 3419 patients were obtained.

[0058] In this embodiment, in step S2, the xgboost model is used for single-feature modeling analysis of a plurality of clinical characteristics, and the AUROC value of each clinical characteristic is obtained; according to the AUROC value, a first selected strategy is used to select a plurality of first clinical characteristics;

[0059] Specifically, the xgboost model is used for single-feature modeling analysis of 102 clinical characteristics, and each time only one feature is used for model training and testing. Based on the bootstrap method, each feature is randomly sampled 10 times, and then tested on the test set and the receiver operating characteristic curve is drawn and the AUROC value is calculated. The AUROC value of each clinical characteristic when modeling alone is obtained, and the clinical characteristics with an AUROC greater than 0.5 are retained.

[0060] Specifically, the AUROC performance of clinical characteristics such as B-type brain natriuretic peptide precursor, average platelet area, urine glucose, and lymphocyte percentage is high, and the AUROC of clinical characteristics such as total protein of liver function and electrolyte potassium is less than 0.5.

[0061] In this embodiment, in step S3, a plurality of tree models are used for modeling a plurality of clinical characteristics, and the feature importance score of each tree model is output respectively; according to the feature importance score, a second selected strategy is used to select a plurality of second clinical characteristics;

[0062] Specifically, three tree models of xgboost, catboost and lightgbm are used for modeling of 102 features, and then the feature importance scores of the models are output respectively;

[0063] Clinical characteristics with feature importance scores greater than 0 in the three tree models are retained, including clinical characteristics such as myocardial enzyme creatine kinase, stool routine red blood cells, and hemoglobin.

[0064] In this embodiment, in step S4, the intersection of a plurality of first clinical characteristics and a plurality of second clinical characteristics is processed to obtain a plurality of target clinical characteristics;

[0065] Specifically, the clinical characteristics with an AUROC greater than 0.5 and a feature importance score greater than 0 in steps S2 and S3 are retained, and the remaining features are deleted from the data set to obtain a plurality of target clinical characteristics;

[0066] Specifically, a total of 12 data that produced noise to the model were deleted, and the remaining 90 clinical features were further used for subsequent modeling.

[0067] In this embodiment, in step S5, an integrated model is constructed based on the automatic machine learning framework AutoGluon; the integrated model is trained using the target clinical features to obtain a trained integrated model;

[0068] Among them, such as Figure 3 As shown, in the process of training the integrated model by the target clinical features, the training steps are:

[0069] S51, the integrated model applies a separate embedding layer to each of the target clinical features and concatenates them with the numerical features into a large vector;

[0070] Specifically, based on the clinical features screened in step S4, the automatic machine learning framework AutoGluon is applied for further modeling to construct an integrated model; the integrated model applies a separate embedding layer to each categorical feature (such as gender), and then connects it with the numerical feature (such as age) into a large vector.

[0071] S52. Using a multi-layer stack integration strategy, a set of basic models are individually trained in a common manner to obtain aggregated prediction results; and the aggregated prediction results are used as features of the basic models to train the integrated model;

[0072] Specifically, a multi-layer stacking integration strategy is used to first train a set of separate base models in a general way, and then the aggregated predictions of the base models are used as their features to train the stacker model. The stacker model can improve the shortcomings of individual basic predictions and utilize the interactions between the basic models to provide enhanced predictive capabilities.

[0073] S53. Based on the granular computing strategy, the target clinical characteristics are decomposed into several data sets; each of the data sets is further divided into a training subset and a validation subset; several of the training subsets or several of the validation subsets are merged into a final training subset or a final validation subset; and the integrated model is subjected to multi-granularity cross-validation through the final training subset and the final validation subset.

[0074] Specifically, based on a granular computing strategy, a dataset is decomposed into multiple datasets, each of which is further divided into a training subset and a validation subset (the granulation process). Subsequently, the multiple training or validation subsets are merged into a final training or validation subset (the organization process). This method forms a multi-granularity hierarchical structure of the dataset, enabling multi-granularity cross-validation of the ensemble model.

[0075] In this embodiment, in step S6, the target clinical features are analyzed by the trained ensemble model to output feature importance scores of the trained ensemble model; the target clinical features are sorted according to the feature importance scores of the ensemble model, and the clinical features ranked in the front are re-modeled to output feature importance scores of the newly built model; when the feature importance scores of the newly built model and the feature importance scores of the ensemble model meet the set requirements, the most important clinical features are screened out.

[0076] Specifically, according to the trained ensemble model obtained in step S5, the feature importance scores thereof are output, and the features with high feature importance scores are sequentially modeled and tested; the target clinical features are sorted according to the feature importance scores of the ensemble model, and the clinical features ranked in the front are re-modeled to output feature importance scores of the newly built model; when the feature importance scores of the newly built model and the feature importance scores of the ensemble model meet the set requirements, the most important clinical features are screened out; as shown in FIG. 6, when the top 22 features in terms of feature importance are retained, the performance of the newly built model and the performance of the ensemble model are not much different, and finally, the 20 clinical features are selected as the input features for predicting the risk of new myocardial infarction, including “myocardial enzyme 0 lactate dehydrogenase, systolic blood pressure, myocardial enzyme 0 creatine kinase, hepatitis B virus surface antigen, coagulation complete set_ quantitative fibrinogen, large platelet percentage, age, liver function_ albumin / globulin ratio, liver function_ glutamic-oxaloacetic transaminase, myocardial enzyme 0 myoglobin, B-type brain natriuretic peptide precursor, hemoglobin, urine sediment urobilinogen, renal function_ uric acid, coagulation complete set_ PT international normalized ratio, fecal routine_ red blood cells, fecal routine_ appearance, urine sediment specific gravity, low-density lipoprotein cholesterol, liver function_ glutamic-pyruvic transaminase, urine sediment hyaline cast, and liver function_ globulin”. Figure 4

[0077] In this embodiment, in step S7, the newly built model is verified by internal and external data to evaluate the robustness of the newly built model and the effectiveness of the most important clinical features.

[0078] Specifically, the newly built model trained with 22 clinical features in step S6 is used to test the external verification to evaluate the robustness of the model and the effectiveness of the screened features, and the final evaluation result is shown in FIG. 7. Figure 5 As shown in FIG. 7, the area under the receiver operating characteristic curve (AUROC, or AUC) of the model in the external verification set, that is, the area of the cyan line in the figure, is 0.915, which indicates that the machine learning model constructed by the present application can also perform well in the external verification set and has good generalization performance.

[0079] In a possible embodiment, a verification example is provided as follows:

[0080] ​As shown in Table 1, the comparative results of the learning model constructed by the application and other classic machine learning models are shown. In the table, the area under the receiver operating characteristic curve (AUROC) commonly used in binary classification problems is used as an evaluation index:

[0081] Model Area under the receiver operating characteristic curve (AUROC) K-Nearest Neighbors method 0.685 Neural Network 0.693 Random Forest 0.893 LightGBM 0.881 CatBoost 0.890 XGBoost 0.885 The method presented in the invention 0.915

[0082] Table 1 Performance results of different models in new myocardial infarction prediction task

[0083] As can be seen from Table 1, the model proposed by the application is more accurate in prediction and has better generalization performance compared with other classic machine learning models.

[0084] To sum up, the present application collects basic information and laboratory examination data of a plurality of patients from real clinical data; the collected basic information and laboratory examination data of the patients are screened to establish a sample queue of new myocardial infarction patients; a single feature modeling analysis is performed on a plurality of clinical characteristics by an xgboost model to obtain an AUROC value of each clinical characteristic; according to the AUROC value, a first selected strategy is used to select a plurality of first clinical characteristics; a plurality of tree models are used to model a plurality of clinical characteristics, and the feature importance score of each tree model is output; according to the feature importance score, a second selected strategy is used to select a plurality of second clinical characteristics; the plurality of first clinical characteristics and the plurality of second clinical characteristics are intersected to obtain a plurality of target clinical characteristics; an integrated model is constructed based on an automatic machine learning framework AutoGluon; the integrated model is trained by the target clinical characteristics to obtain a trained integrated model; the target clinical characteristics are analyzed by the trained integrated model to output the feature importance score of the trained integrated model; the target clinical characteristics are sorted according to the feature importance score of the integrated model, and a plurality of clinical characteristics ranked in the front are re-modeled to output the feature importance score of the new model; when the feature importance score of the new model meets the set requirement, the most important clinical characteristics are screened out; the new model is verified by internal and external data to evaluate the robustness of the new model and the effectiveness of the most important clinical characteristics. The present application solves the challenge of high cost of clinical data collection in new myocardial infarction classification and prediction. By modeling and analyzing a large number of clinical characteristics of patients, the important characteristics of new myocardial infarction in clinical application are accurately screened out based on the granular computing method; the new myocardial infarction classification and prediction model constructed by the method only needs to input common clinical characteristics (such as blood routine), reduces the cost of data collection, and improves the robustness of the model. The present application introduces automatic machine learning, multi-layer stacking integration strategy and granular computing data division method to further improve the accuracy and robustness of the model; the present application has stable performance in real clinical internal and external verification sets.

[0085] It should be noted that the method of the present embodiment can be performed by a single device, such as a computer or a server. The method of the present embodiment can also be applied to a distributed scenario, and completed by multiple devices cooperating with each other. In this distributed scenario, one of the multiple devices can only perform one or more steps in the method of the present embodiment, and the multiple devices can interact with each other to complete the method.

[0086] It is to be understood that the foregoing description is directed to some embodiments of the disclosure. Other embodiments fall within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order and still achieve desirable results. Additionally, the processes depicted in the accompanying figures do not necessarily require the particular order shown or sequential order to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

[0087] Embodiment 2

[0088] Referring to Figure 6 Embodiment 2 of the present application also provides a new AMI classification and prediction device based on granular computing, comprising:

[0089] A data screening module 001 is configured to collect basic information and laboratory examination data of a plurality of patients from real clinical data, screen the collected basic information and laboratory examination data of the patients, and establish a sample queue of AMI patients.

[0090] A first clinical feature screening module 002 is configured to perform single-feature modeling analysis on a plurality of clinical features by an xgboost model to obtain an AUROC value of each clinical feature, and select a plurality of first clinical features according to a first selection strategy according to the AUROC value.

[0091] A second clinical feature screening module 003 is configured to model a plurality of clinical features by a plurality of tree models, respectively output a feature importance score of each tree model, and select a plurality of second clinical features according to a second selection strategy according to the feature importance score.

[0092] A target clinical feature acquisition module 004 is configured to perform intersection processing on the plurality of first clinical features and the plurality of second clinical features to obtain a plurality of target clinical features.

[0093] An integrated model construction and training module 005 is configured to construct an integrated model based on an automatic machine learning framework AutoGluon, train the integrated model based on the target clinical features, and obtain a trained integrated model.

[0094] The most important clinical feature screening module 006 is used for the most important clinical features to analyze the target clinical features by the trained ensemble model, output the feature importance score of the trained ensemble model; the target clinical features are sorted according to the feature importance score of the ensemble model, and the clinical features ranked in the front are re-modeled and analyzed, and the feature importance score of the new model is output; when the feature importance score of the new model and the feature importance score of the ensemble model meet the set requirements, the most important clinical features are screened out;

[0095] The new model verification module 007 is used for verifying the new model by internal and external data, and evaluating the robustness of the new model and the effectiveness of the most important clinical features.

[0096] In the data screening module 001 in the embodiment, in the process of screening the collected basic information of the patients and the laboratory examination data and establishing the sample queue of the new myocardial infarction patients, samples suffering from new myocardial infarction are screened from the basic information of the patients and the laboratory examination data; samples with a data missing rate greater than 20% in the screened samples are deleted, and the sample queue of the new myocardial infarction patients is established.

[0097] In the first clinical feature screening module 002 in the embodiment, in the process of selecting to retain a plurality of the first clinical features according to the first selection strategy according to the AUROC value, the first selection strategy is to retain the clinical features with an AUROC value greater than 0.5.

[0098] In the second clinical feature screening module 003 in the embodiment, in the process of modeling a plurality of the clinical features by a plurality of the tree models, the tree models include xgboost, catboost and lightgbm tree models; and the second selection strategy is to retain the clinical features with feature importance scores of the three tree models of xgboost, catboost and lightgbm all greater than 0.

[0099] In the ensemble model construction and training module 005 in the embodiment, the training submodule includes:

[0100] The large vector generation submodule 051 is used for the ensemble model to apply a separate embedding layer to each of the target clinical features, and connect the numerical features into a large vector;

[0101] The multi-layer stack ensemble training submodule 052 is used for using a multi-layer stack ensemble strategy to separately train a group of basic models in a general way to obtain an aggregated prediction result; and using the aggregated prediction result as a feature of the basic model to train the ensemble model.

[0102] The granular computing training submodule 053 is configured to decompose the target clinical feature into a plurality of data sets based on a granular computing strategy; each of the data sets is further divided into a training subset and a verification subset; a plurality of the training subsets or a plurality of the verification subsets are combined into a final training subset or a final verification subset; and the integrated model is subjected to multi-granularity cross-validation through the final training subset and the final verification subset.

[0103] It should be noted that the information interaction and execution process between the modules of the system described above are based on the same concept as the method embodiment of the present application in Embodiment 1, and the technical effects brought by the method embodiment are the same as those of the method embodiment of the present application. For specific content, please refer to the description of the method embodiment of the present application as described above. Here, no further description is given.

[0104] Embodiment 3

[0105] Embodiment 3 of the present application provides a non-transitory computer readable storage medium, which stores program codes of a new-onset myocardial infarction grading and classification prediction method based on granular computing, and the program codes include instructions for executing the new-onset myocardial infarction grading and classification prediction method based on granular computing of Embodiment 1 or any possible implementation manner thereof.

[0106] The computer readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. containing one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)), etc.

[0107] Embodiment 4

[0108] Embodiment 4 of the present application provides an electronic device, comprising a memory and a processor.

[0109] The processor and the memory complete mutual communication through a bus; the memory stores program instructions executable by the processor, and the processor calling the program instructions can execute the new-onset myocardial infarction grading and classification prediction method based on granular computing of Embodiment 1 or any possible implementation manner thereof.

[0110] Specifically, the processor can be implemented by hardware or software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented by software, the processor can be a general-purpose processor, which realizes by reading software codes stored in a memory. The memory can be integrated in the processor or exist independently outside the processor.

[0111] In the embodiments described above, all or some of the modules or steps can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or some of the modules or steps can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded into and executed by a computer, all or some of the procedures or functions described in the embodiments of the present application are generated. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable systems. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transferred from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode.

[0112] Obviously, those skilled in the art should understand that the modules or steps of the present application described above can be implemented by a general computing system, which can be concentrated on a single computing system or distributed on a network composed of multiple computing systems, and optionally, they can be implemented by program codes executable by a computing system, so that they can be stored in a storage system and executed by a computing system, and in some cases, the steps shown or described can be executed in different order, or they can be made into individual integrated circuit modules, or multiple modules or steps can be made into a single integrated circuit module. Thus, the present application is not limited to any particular combination of hardware and software.

[0113] Although the present application has been described in detail above with general description and specific embodiments, some modifications or improvements can be made on the basis of the present application, which is obvious to those skilled in the art. Therefore, these modifications or improvements made on the basis of not deviating from the spirit of the present application are within the scope of the present application.

Claims

1. A method for predicting the classification of newly diagnosed myocardial infarction based on granular computing, characterized in that: include: From real clinical data, basic information and laboratory test data of several patients were collected; Screening the collected basic information of the patients and the laboratory test data to establish a sample cohort of patients with new myocardial infarction; Performing single-feature modeling analysis on a plurality of clinical features using an xgboost model to obtain an AUROC value for each clinical feature; selecting and retaining a plurality of first clinical features according to a first selection strategy based on the AUROC value; Modeling the clinical features using a plurality of tree models, and outputting a feature importance score for each of the tree models; selecting and retaining a plurality of second clinical features according to a second selection strategy based on the feature importance scores; Performing intersection processing on a plurality of the first clinical features and a plurality of the second clinical features to obtain a plurality of target clinical features; Building an integrated model based on the automatic machine learning framework AutoGluon; training the integrated model using the target clinical features to obtain a trained integrated model; The target clinical features are analyzed by the trained integrated model, and the feature importance scores of the trained integrated model are output; the target clinical features are sorted according to the feature importance scores of the integrated model, and a set number of clinical features ranked at the top are remodeled and analyzed, and the feature importance scores of the newly created model are output; when the feature importance scores of the newly created model and the feature importance scores of the integrated model meet the set requirements, the most important clinical features are screened out; The newly created model was validated by internal and external data to evaluate the robustness of the newly created model and the effectiveness of the most important clinical features.

2. The method for predicting the grade and classification of newly-occurring myocardial infarction based on granular computing according to claim 1, characterized in that: In the process of screening the collected basic information of the patients and the laboratory test data to establish the sample cohort of patients with new myocardial infarction, samples with new myocardial infarction are screened from the basic information of the patients and the laboratory test data; samples with a data missing rate greater than 20% among the screened samples are deleted to establish the sample cohort of patients with new myocardial infarction.

3. The method for predicting the grade and classification of newly-occurring myocardial infarction based on granular computing according to claim 2, characterized in that: In the process of selecting and retaining a plurality of the first clinical features according to the AUROC value and the first selection strategy, the first selection strategy is: retaining clinical features with an AUROC value greater than 0.

5.

4. The method for predicting the grade and classification of newly-occurring myocardial infarction based on granular computing according to claim 3, characterized in that: In the process of modeling the clinical features using the tree models, the tree models include xgboost, catboost and lightgbm tree models; the second selection strategy is to retain the clinical features whose feature importance scores of the three tree models xgboost, catboost and lightgbm are all greater than 0.

5. The method for predicting the grade and classification of newly-occurring myocardial infarction based on granular computing according to claim 4, characterized in that: In the process of training the integrated model using the target clinical features, the training steps are: The ensemble model applies a separate embedding layer to each of the target clinical features and concatenates them with the numerical features into a large vector; Using a multi-layer stacking ensemble strategy, a set of base models are trained separately in a common way to obtain aggregated prediction results; Using the aggregated prediction results as features of the base model to train the ensemble model; Based on a granular computing strategy, the target clinical features are decomposed into several data sets; each of the data sets is further divided into a training subset and a validation subset; Merging the plurality of training subsets or the plurality of validation subsets into a final training subset or a final validation subset; The integrated model is subjected to multi-granularity cross-validation using the final training subset and the final validation subset.

6. A device for predicting the grading and classification of newly-occurring myocardial infarction based on granular computing, using the method for predicting the grading and classification of newly-occurring myocardial infarction based on granular computing according to any one of claims 1 to 5, characterized in that: include: The data screening module is used to collect basic information and laboratory test data of several patients from real clinical data; Screening the collected basic information of the patients and the laboratory test data to establish a sample cohort of patients with new myocardial infarction; A first clinical feature screening module is configured to perform single feature modeling analysis on a plurality of clinical features using an xgboost model to obtain an AUROC value for each clinical feature; and select and retain a plurality of first clinical features according to a first selection strategy based on the AUROC value; A second clinical feature screening module is configured to model the plurality of clinical features using a plurality of tree models, and output a feature importance score for each of the tree models; and select and retain a plurality of second clinical features according to a second selection strategy based on the feature importance scores; a target clinical feature acquisition module, configured to perform intersection processing on a plurality of the first clinical features and a plurality of the second clinical features to obtain a plurality of target clinical features; An integrated model construction and training module is used to build an integrated model based on the automatic machine learning framework AutoGluon; the integrated model is trained using the target clinical features to obtain a trained integrated model; The most important clinical feature screening module is used to analyze the target clinical features using the trained integrated model for the most important clinical features, and output the feature importance score of the trained integrated model; sort the target clinical features according to the feature importance score of the integrated model, and remodel and analyze the clinical features that are ranked at the top by a set number, and output the feature importance score of the newly created model; when the feature importance score of the newly created model and the feature importance score of the integrated model meet the set requirements, the most important clinical features are screened out; The newly created model verification module is used to verify the newly created model through internal and external data, and to evaluate the robustness of the newly created model and the effectiveness of the most important clinical features.

7. The device for predicting the grade and classification of newly-occurring myocardial infarction based on granular computing according to claim 6, characterized in that: In the data screening module, in the process of screening the collected basic information of the patient and the laboratory test data to establish the sample cohort of patients with new myocardial infarction, samples with new myocardial infarction are screened from the basic information of the patient and the laboratory test data; samples with a data missing rate greater than 20% among the screened samples are deleted to establish the sample cohort of patients with new myocardial infarction.

8. The device for predicting the grade and classification of newly-occurring myocardial infarction based on granular computing according to claim 7, characterized in that: In the first clinical feature screening module, in the process of selecting and retaining a number of the first clinical features according to the AUROC value and the first selection strategy, the first selection strategy is: retaining clinical features with an AUROC value greater than 0.

5.

9. The device for predicting the grade and classification of newly-occurring myocardial infarction based on granular computing according to claim 8, characterized in that: In the second clinical feature screening module, in the process of modeling the several clinical features through the several tree models, the tree models include: xgboost, catboost and lightgbm tree models; the second selection strategy is: retaining the clinical features whose feature importance scores of the three tree models of xgboost, catboost and lightgbm are all greater than 0.

10. The device for predicting the grade and classification of newly-occurring myocardial infarction based on granular computing according to claim 9, characterized in that: In the integrated model construction and training module, the training submodule includes: A large vector generation submodule, used for the integrated model to apply a separate embedding layer to each of the target clinical features and concatenate them with the numerical features into a large vector; The multi-layer stack ensemble training submodule is used to train a set of basic models in a common way using a multi-layer stack ensemble strategy to obtain aggregated prediction results. Using the aggregated prediction results as features of the base model to train the ensemble model; A granular computing training submodule, configured to decompose the target clinical features into a plurality of data sets based on a granular computing strategy; each of the data sets is further divided into a training subset and a validation subset; Merging the plurality of training subsets or the plurality of validation subsets into a final training subset or a final validation subset; The integrated model is subjected to multi-granularity cross-validation using the final training subset and the final validation subset.

Images