Method for constructing sudden cardiac arrest risk prediction model
Through multimodal data fusion and machine learning algorithms, a cardiac arrest risk prediction model is constructed, which solves the problem of insufficient LVEF assessment in the existing technology, realizes accurate identification and personalized treatment decisions for high-risk patients, and optimizes cardiac arrest Stop risk stratification to improve prediction performance.
Patent Information
- Application Number
- CN202411163658.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-23
- Publication Date
- 2024-11-12
AI Technical Summary
Existing cardiac arrest risk assessment methods rely on left ventricular ejaculation fraction (LVEF), with problems of under-evaluation and over-treatment, making it difficult to accurately identify high-risk patients, and the limitations of traditional methods lead to high lethality rates of cardiac arrest and disability rate.
By acquiring clinical text data and image images, preprocessing, segmentation, feature extraction and feature selection, a cardiac arrest risk prediction model is constructed, combined with machine learning algorithms, imagingomics features and clinical data are used , develop efficient cardiac arrest risk assessment tools.
It has achieved accurate identification of high-risk cardiac arrest patients, reduced the incidence of cardiac arrest, provided personalized treatment decisions, optimized cardiac arrest risk stratification, improved predictive performance, and has good clinical application value.
Smart Images

Figure CN118942715A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method for constructing a cardiac arrest risk prediction model. Background Art
[0002] Cardiac arrest is a serious disease, usually occurring without warning. About 60% to 80% of cardiac arrests occur outside hospitals, which greatly increases the difficulty of implementing high-quality cardiopulmonary resuscitation in the early stage. Globally, the overall survival rate of cardiac arrest is less than 10%, and this proportion is as low as 1.2% in China. Even after successful resuscitation, less than 20% of survivors can retain good cranial nerve function. Given the high mortality and disability rates of cardiac arrest, resuscitation and prevention have become key links in reducing the burden of disease. However, relying on improving the success rate of resuscitation has limited effect, because this requires the public to master high-quality first aid skills. Although left ventricular ejection fraction (LVEF) has long been used as the main basis for cardiac arrest risk stratification, its limitations in predictive effectiveness are becoming increasingly prominent.
[0003] Currently, in the guidelines for SCD (sudden cardiac death) and ventricular arrhythmias developed by the European Society of Cardiology, risk stratification and primary prevention are mainly based on left ventricular ejection fraction (LVEF), patients with LVEF ≤ 35% are classified as high-risk groups, and implantable cardioverter defibrillator (ICD) is recommended for prevention. However, this standard faces the dilemma of insufficient evaluation and overtreatment in actual application. On the one hand, among patients who receive primary prevention under the guidance of this standard, only 5% receive appropriate ICD treatment each year, indicating that a considerable number of patients receive unnecessary ICD implantation. On the other hand, 70%-80% of SCD patients have LVEF> 35%, indicating that there is a lack of effective risk assessment tools for this group. As research evidence continues to reveal the limitations of LVEF in cardiac arrest risk stratification, this poses a huge challenge to the traditional method of relying on LVEF to identify high-risk patients.
[0004] Therefore, how to break through the limitations of existing risk stratification methods is crucial to reducing the incidence of cardiac arrest in heart failure patients and improving their prognosis. Accurately identifying high-risk patients and taking preventive measures in advance has become a more effective way to reduce the global burden of cardiac arrest. Summary of the invention
[0005] The purpose of the present invention is to provide a method for constructing a cardiac arrest risk prediction model for accurately identifying high-risk cardiac arrest patients and taking preventive measures in advance to reduce the incidence of cardiac arrest.
[0006] In order to achieve the above objectives, the present invention is implemented by the following technical solutions:
[0007] A method for constructing a cardiac arrest risk prediction model comprises: step S1, obtaining data, wherein the data comprises clinical text data and image data. step S2, preprocessing the image data. step S3, segmenting the preprocessed image data. step S4, extracting features from the segmented image data to obtain radiomic features. step S5, constructing a cardiac arrest risk prediction model based on the radiomic features and the clinical text data.
[0008] Optionally, the step S2 includes: step S21, dividing the image into a training image and an automatically segmented image. step S22, annotating the training image.
[0009] Optionally, the step S3 includes: step S31, using nnU-Net to segment the labeled training image. Step S32, automatically segmenting the automatically segmented image. Step S33, using the PyRadiomics library to extract the radiomics features from the segmented training image and the automatically segmented image. Step S34, avoiding overfitting processing on the radiomics features.
[0010] Optionally, the step S34 includes: step S341, using a feature selection method to perform feature selection on the imaging omics features to obtain a plurality of features. Step S342, using a heat map correlation matrix to analyze the correlation between the plurality of features and cardiac arrest, and the correlation between the plurality of features. Step S343, selecting the plurality of features related to the cardiac arrest as an imaging omics feature subset.
[0011] Optionally, the feature selection method includes: embedding method, packaging method or filtering method.
[0012] Optionally, the step S5 comprises: step S51, using a machine learning algorithm to develop a model to obtain a plurality of models. step S52, determining an optimal model of the plurality of models. step S53, adding the imaging genomics feature subset to the optimal model, and combining the clinical text data to obtain the cardiac arrest risk prediction model.
[0013] Optionally, the step S52 includes: step S521, using grid search and cross validation to determine the best hyperparameters of several of the models. Step S522, using Voting Classifier to perform machine learning on several of the models, and using a "soft voting" strategy to determine the weight of each of the models. Step S523, determining the best threshold of the ROC curve based on the modified Youden index, and calculating the performance indicators of several of the models. Step S524, determining the optimal model based on the best hyperparameters, the weight of each of the models and the performance indicators of several of the models.
[0014] Optionally, the model includes: logistic regression, naive Bayes, support vector classifier, random forest and category boosting classifier.
[0015] Optionally, the method for constructing a cardiac arrest risk prediction model further includes: step S6, performing a performance evaluation on the cardiac arrest risk prediction model. The step S6 includes: step S61, using the area under the curve, accuracy, precision, recall rate and F1 score as model performance evaluation indicators of the cardiac arrest risk prediction model. Step S62, using the mean and standard deviation of the model performance evaluation indicators to evaluate the performance of the cardiac arrest risk prediction model.
[0016] On the other hand, the present invention further provides an electronic device, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the method described above is implemented.
[0017] The present invention has at least one of the following technical effects:
[0018] The present invention uses multimodal data, including text and image information, to provide a more comprehensive perspective for cardiac arrest risk assessment and enhance the accuracy and generalization of the model. The cardiac arrest risk prediction model constructed by the present invention can accurately identify high-risk cardiac arrest patients, thereby guiding personalized treatment decisions and reducing the incidence of cardiac arrest.
[0019] The present invention can optimize and improve the accurate risk stratification of cardiac arrest by introducing these high-dimensional imaging features. The method of optimizing cardiac arrest risk assessment using imaging omics features means a shift from traditional risk markers to more detailed exploration of microscopic features, and through these microscopic features of myocardial scars, we can further deepen our understanding of the pathogenesis of cardiac arrest.
[0020] The present invention combines cutting-edge machine learning and deep learning technologies, and its application scope is not limited by LVEF, providing a more extensive scientific basis for risk stratification for patients with heart failure. Moreover, the integrated model constructed by the present invention exhibits superior predictive performance and has good clinical application value. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 A flowchart of a method for constructing a cardiac arrest risk prediction model provided by one embodiment of the present invention;
[0022] Figure 2 A detailed flow chart of a method for constructing a cardiac arrest risk prediction model provided by one embodiment of the present invention;
[0023] Figure 3 A flowchart of image segmentation provided by an embodiment of the present invention;
[0024] Figure 4 A flowchart of feature screening and cardiac arrest risk prediction model construction provided in one embodiment of the present invention. DETAILED DESCRIPTION
[0025] The following is a further detailed description of a method for constructing a cardiac arrest risk prediction model proposed by the present invention in conjunction with the accompanying drawings and specific embodiments. According to the following description, the advantages and features of the present invention will be clearer. It should be noted that the drawings are in a very simplified form and use non-precise proportions, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention. In order to make the purposes, features and advantages of the present invention more obvious and easy to understand, please refer to the accompanying drawings. It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the limiting conditions for the implementation of the present invention, so they have no technical substantive significance. Any structural modification, change in proportional relationship or adjustment of size, without affecting the effects that can be produced by the present invention and the purposes that can be achieved, should still fall within the scope of the technical content disclosed by the present invention.
[0026] As described in the background technology, cardiac arrest is a serious public health challenge worldwide. Cardiac arrest is a complex pathophysiological state involving multiple pathogenic factors and substrates. Traditional single linear models are difficult to reveal the potential associations between its variables. The rapid progress of artificial intelligence has opened up new ways to achieve more accurate cardiac arrest risk assessment. Machine learning and deep learning, as powerful analytical tools, can effectively mine complex patterns and associations in large amounts of clinical data, thereby improving the accuracy and efficiency of cardiac arrest prediction. These methods are particularly suitable for processing high-dimensional, nonlinear and large-scale medical data, covering multimodal data such as clinical text materials, electrocardiogram electrophysiological signals and medical images.
[0027] At present, the application of artificial intelligence in the field of cardiac arrest is still in its infancy, especially the lack of multimodal data fusion mode. Multimodal data fusion refers to the technology of integrating data types from different sources for analysis and processing. In patients with chronic heart failure, the risk of cardiac arrest is much higher than that of the general population, and its pathogenesis is usually related to the reentrant pathway formed by myocardial scarring. Rapid ventricular arrhythmias are the main direct cause of cardiac arrest, and myocardial scars are the key substrates of ventricular arrhythmias. Myocardial scars are closely related to the risk of cardiac arrest. At present, three-dimensional image segmentation technology is mostly used in the field of oncology, and is rarely used in myocardial scars to explore their relationship with clinical prognosis, especially in the study of serious adverse outcomes such as cardiac arrest.
[0028] like Figure 2 As shown, in order to solve the above problems, the present invention uses artificial intelligence technology to deeply explore myocardial scars, and applies three-dimensional image segmentation and imaging omics technology to extract imaging omics features of myocardial scars from delayed gadolinium-enhanced cardiac images. Through feature selection, imaging omics markers closely related to cardiac arrest are screened out, and clinical data are integrated with multimodal data such as cardiac images to develop efficient cardiac arrest risk assessment tools to guide personalized treatment decisions. The present invention is divided into three aspects: first, three-dimensional image segmentation of myocardial scars is performed based on delayed gadolinium-enhanced cardiac magnetic resonance images, and then feature extraction technology is used to capture its imaging omics features, and then risk markers significantly associated with cardiac arrest are screened out through feature selection. Finally, these relevant features are combined with clinical data, and a machine learning algorithm is used to construct a cardiac arrest prediction model.
[0029] Figure 1 A flowchart of a method for constructing a cardiac arrest risk prediction model provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the present embodiment provides a method for constructing a cardiac arrest risk prediction model, comprising: step S1, obtaining data, wherein the data includes clinical text data and image data. Step S2, preprocessing the image data. Step S3, segmenting the preprocessed image data. Step S4, extracting features from the segmented image data to obtain radiomic features. Step S5, constructing a cardiac arrest risk prediction model based on the radiomic features and the clinical text data.
[0030] This embodiment uses multimodal data, including text and image information, to provide a more comprehensive perspective for cardiac arrest risk assessment, and enhances the accuracy and generalization of the cardiac arrest risk prediction model. Moreover, the cardiac arrest risk prediction model constructed in this embodiment can accurately identify high-risk cardiac arrest patients, thereby guiding personalized treatment decisions and reducing the incidence of cardiac arrest.
[0031] In this embodiment or some other embodiments, step S1 includes: collecting data of patients diagnosed with symptomatic heart failure, cardiac magnetic resonance examination and delayed gadolinium enhancement positive between January 2018 and April 2022. Collect clinical data of patients, including general demographic data, laboratory tests (blood routine, biochemistry, etc.), electrocardiogram, echocardiogram, cardiac magnetic resonance image. Follow-up for 2.3 years (1.6, 3.4 years, i.e. median and quartile, non-normal distribution data), the endpoint event is cardiac arrest event, including sudden cardiac death, ventricular fibrillation and hemodynamically unstable ventricular tachycardia.
[0032] like Figure 3 As shown, the step S2 includes: step S21, dividing the image into a training image and an automatically segmented image. Step S21, annotating the training image.
[0033] For details, please continue to refer to Figure 3 Image preprocessing includes: downloading left ventricular short-axis cardiac magnetic resonance images in DICOM (Data Imaging and Communications in Medicine) format, 70% of the data set is used for training, and 30% is used for automatic segmentation. The image data used for training is imported into ITK-SNAP (Image Toolkit-Segmentation and Navigation Platform, a medical image processing software) software for manual annotation of myocardial scars. It should be noted that ITK-SNAP software is used to train the myocardial scar image segmentation framework and realize the remaining 30% automatic segmentation.
[0034] like Figure 3 and Figure 4 As shown, the step S31 includes: using nnU-Net (no new U-NET, which is a medical image segmentation model) to segment the labeled training image.
[0035] Step S32: automatically segment the automatically segmented image.
[0036] Step S33: extract the radiomics features from the segmented training image and the automatically segmented image using the PyRadiomics library (a software library).
[0037] Step S34: performing overfitting prevention processing on the radiomics features.
[0038] Step S341: Use a feature selection method to perform feature selection on the imaging omics features to obtain a number of features.
[0039] Step S342: Analyze the correlation between some of the features and cardiac arrest, and the correlation between some of the features using a heat map correlation matrix.
[0040] Step S343: Select several features related to the cardiac arrest as a radiomics feature subset.
[0041] For details, please continue to refer to Figure 3 and Figure 4 For the training set images, the nnU-Net framework is used to segment the myocardial scar 3D images, and the remaining images are automatically segmented. For the segmented myocardial scar 3D images, the PyRadiomics library is used to extract radiomics features. Preferably, 833 radiomics features are extracted, but this embodiment is not limited thereto.
[0042] In order to avoid overfitting, feature selection is first performed on the imaging omics features. Three methods, namely, embedding, packaging and filtering, are used for feature selection. The correlation between features and cardiac arrest, as well as the correlation between features, are analyzed using a heat map correlation matrix. The three features with the highest correlation with cardiac arrest are selected as the optimal imaging omics feature subset. The feature selection methods include: embedding, packaging or filtering. The details are as follows:
[0043] (1) Embedding method: The random forest algorithm combined with 5-fold cross validation is used to evaluate the importance of features. The number of trees is set to 100, and the top 10 features in importance are selected.
[0044] (2) Wrapping method: Recursive feature elimination is used, and logistic regression is selected as the base model. In each step of recursive feature elimination, the feature that contributes the least to the model is removed until the 10 most important features are selected.
[0045] (3) Unsupervised learning: K-means clustering algorithm is used to determine the optimal number of clusters based on the elbow rule, and features with similar characteristics are grouped. The feature closest to the center of each cluster is selected as the representative feature.
[0046] (4) Filtering method: Summarize the features filtered by the above method and use the heat map correlation matrix to analyze the correlation between the features and cardiac arrest, as well as the correlation between the features.
[0047] For cases where the correlation coefficient between features is >0.7, only features with a higher correlation with cardiac arrest are retained to avoid introducing highly correlated features into the model. Finally, the three features with the highest correlation with cardiac arrest were selected as the optimal imaging genomics feature subset. It should be noted that in this embodiment, the sample size of the training set = 87 cases, and the sample size of automatic segmentation is 35 cases. That is, as mentioned above, "70% of the data set is used for training and 30% for automatic segmentation."
[0048] In this embodiment or some other embodiments, as Figure 4As shown, the step S5 includes: step S51, using a machine learning algorithm to develop a model to obtain several models.
[0049] Step S52: Determine the optimal model of the models. (That is, evaluate the performance of each model and select the model with the best performance.)
[0050] Step S521: Use grid search and cross validation to determine the best hyperparameters of several models.
[0051] Step S522: Use Voting Classifier to perform machine learning on the models, and use a "soft voting" strategy to determine the weight of each model.
[0052] Step S523: determine the optimal threshold of the ROC curve (receiver operating characteristic curve) according to the modified Youden index, and calculate several performance indicators of the model.
[0053] Step S524: determine the optimal model according to the optimal hyperparameters, the weight of each model and the performance indicators of several models.
[0054] Step S53: adding the imaging genomics feature subset into the optimal model, and combining it with the clinical text data to obtain the cardiac arrest risk prediction model.
[0055] For details, please continue to refer to Figure 4 , the step S5 constructs a cardiac arrest prediction model including: using 5 supervised machine learning algorithms for model development, including logistic regression, naive Bayes, support vector classifier (SVC), random forest, and category boosting classifier (CatBoost). Grid search and 3-fold cross validation are used to determine the best hyperparameters of the 5 models to achieve the best model performance. The Voting Classifier is used to integrate the above 5 machine learning algorithms, and the "soft voting" strategy is used to determine the weight of each classifier to achieve the weight distribution of the best model performance. The derived data set is randomly divided into a training set and a test set in a ratio of 8:2. In order to better balance the impact of false negatives and false positives on clinical decision-making, the optimal threshold of the ROC curve is determined based on the modified Youden index (sensitivity + specificity + 1 / 3 × accuracy - 1). The performance indicators of the following models are calculated under the optimal threshold. In order to facilitate clinical use, the forward search method is used for feature screening to retain the feature subset of clinical data with the best model performance. Finally, a subset of radiomics features was added to the model to construct a cardiac arrest prediction model, which was named PentaVote classifier.
[0056] It should be noted that the model prediction performance is different under different weight distributions. The purpose of adjusting the weight distribution is to achieve the area under the curve (AUC) of the best model. The derivation data set is randomly divided into a training set and a test set in a ratio of 8:2, which means that all samples are randomly divided into a training set and a validation set in a ratio of 8:2, the training set is used to train the model, and the test set is used to evaluate the model performance. The model performance mentioned below is the result on the test set.
[0057] In this embodiment or some other embodiments, the method for constructing a cardiac arrest risk prediction model further includes: step S6, performing a performance evaluation on the cardiac arrest risk prediction model.
[0058] The step S6 comprises: step S61, using the area under the curve, accuracy, precision, recall and F1 score as model performance evaluation indicators of the cardiac arrest risk prediction model. It should be noted that the F1 score is also called the balanced F score, which is defined as the harmonic mean of precision and recall.
[0059] Step S62: Evaluate the performance of the cardiac arrest risk prediction model using the mean value and standard deviation of the model performance evaluation index.
[0060] It should be noted that 10 consecutive numbers (1 to 10) were selected as random seeds during the model development process, and the mean and standard deviation of each performance index were used to evaluate the model performance. The model performance is shown in Table 1.
[0061] Table 1 Performance indicators of the PentaVote model
[0062]
[0063] In summary, this embodiment adopts a multimodal data fusion mode, that is, it integrates multi-source data, including text and image information, to provide a more comprehensive perspective for cardiac arrest risk assessment and enhance the accuracy and generalization ability of the model. This multimodal data fusion method is relatively rare in previous studies, especially in the field of cardiac arrest.
[0064] This example explores the imaging features of myocardial scars. The method of optimizing cardiac arrest risk assessment using imaging features means a shift from traditional risk markers to more detailed exploration of microscopic features. Through these microscopic features of myocardial scars, we can further deepen our understanding of the pathogenesis of cardiac arrest.
[0065] This embodiment uses advanced AI technology to construct an integrated model that shows superior predictive performance and has good clinical application value. In addition, this study combines cutting-edge 3D myocardial scar image segmentation technology with clinical prognosis. This type of comprehensive application is relatively rare in previous studies, reflecting the high innovation of this study in both technological innovation and promoting its transformation into clinical practice.
[0066] This embodiment can optimize and improve the accurate risk stratification of cardiac arrest by introducing these high-dimensional imaging features. It also combines cutting-edge machine learning and deep learning technologies, and its application scope is not limited by LVEF, providing a more extensive scientific basis for risk stratification for patients with heart failure. Moreover, the integrated model constructed by the present invention shows excellent predictive performance and has good clinical application value.
[0067] This embodiment can accurately identify high-risk patients and take preventive measures in advance, becoming a more effective way to reduce the global burden of cardiac arrest. In the field of cardiac arrest research, landmark large-scale clinical trials all use traditional linear models with a single data modality.
[0068] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0069] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may include that the first and second features are in direct contact, or may include that the first and second features are not in direct contact but are in contact through another feature between them. Moreover, a first feature being "above", "above" and "above" a second feature includes that the first feature is directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below", "below" and "below" a second feature includes that the first feature is directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature.
[0070] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.
Claims
1. A method for constructing a cardiac arrest risk prediction model, It is characterized in that include: Step S1, obtaining data, wherein the data includes clinical text data and image images; Step S2, preprocessing the image; Step S3, segmenting the pre-processed image; Step S4, extracting features from the segmented image to obtain radiomics features; Step S5: constructing a cardiac arrest risk prediction model based on the radiomics features and the clinical text data.
2. The method for constructing a cardiac arrest risk prediction model according to claim 1, It is characterized in that The step S2 comprises: Step S21, dividing the image into a training image and an automatic segmentation image; Step S22: labeling the training image.
3. The method for constructing a cardiac arrest risk prediction model according to claim 2, It is characterized in that The step S3 comprises: Step S31, using nnU-Net to segment the labeled training image; Step S32, automatically segmenting the automatically segmented image; Step S33, extracting the radiomics features from the segmented training image and the automatically segmented image using the PyRadiomics library; Step S34: performing overfitting prevention processing on the radiomics features.
4. The method for constructing a cardiac arrest risk prediction model according to claim 3, It is characterized in that The step S34 comprises: Step S341, using a feature selection method to perform feature selection on the radiomics features to obtain a number of features; Step S342, analyzing the correlation between the plurality of features and cardiac arrest, and the correlation between the plurality of features using a heat map correlation matrix; Step S343: Select several features related to the cardiac arrest as a radiomics feature subset.
5. The method for constructing a cardiac arrest risk prediction model according to claim 4, It is characterized in that The feature selection method includes: embedding method, packaging method or filtering method.
6. The method for constructing a cardiac arrest risk prediction model according to claim 4, It is characterized in that The step S5 comprises: Step S51: using a machine learning algorithm to develop a model to obtain several models; Step S52, determining the optimal model of the plurality of models; Step S53: adding the imaging genomics feature subset into the optimal model, and combining it with the clinical text data to obtain the cardiac arrest risk prediction model.
7. The method for constructing a cardiac arrest risk prediction model according to claim 6, It is characterized in that The step S52 comprises: Step S521, using grid search and cross validation to determine the best hyperparameters of several models; Step S522: using Voting Classifier to perform machine learning on the models, and using a "soft voting" strategy to determine the weight of each model; Step S523, determining the optimal threshold of the ROC curve according to the modified Youden index, and calculating several performance indicators of the model; Step S524: determine the optimal model according to the optimal hyperparameters, the weight of each model and the performance indicators of several models.
8. The method for constructing a cardiac arrest risk prediction model according to claim 6, It is characterized in that The models include: logistic regression, naive Bayes, support vector classifier, random forest, and class boosting classifier.
9. The method for constructing a cardiac arrest risk prediction model according to claim 1, It is characterized in that The method also includes: step S6, performing a performance evaluation on the cardiac arrest risk prediction model; The step S6 comprises: Step S61, using the area under the curve, accuracy, precision, recall rate and F1 score as model performance evaluation indicators of the cardiac arrest risk prediction model; Step S62: Evaluate the performance of the cardiac arrest risk prediction model using the mean value and standard deviation of the model performance evaluation index.
10. An electronic device, It is characterized in that The method comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the method according to any one of claims 1 to 9 is implemented.
Citation Information
Cited By
Diagnostic model for sudden cardiac death and construction method thereof
CN119581041A
Processing method and device for hypertrophic cardiomyopathy risk prediction model
CN120565056A
