A method for establishing a prediction model of endpoint events in HCM patients based on LGE radiomics and clinical characteristics, a method for predicting SCD risk, and a device.

Abstract

This disclosure relates to the field of medical big data technology, specifically to a method for establishing a prediction model of endpoint events in HCM patients based on LGE radiomics features and clinical predictive features, a method for predicting SCD risk based on the HCM patient endpoint event prediction model, and an apparatus. The method for establishing the model innovatively combines LGE radiomics with clinical features and employs multiple machine learning methods to establish multiple radiomics-clinical feature joint prediction models corresponding to each patient group in the cardiovascular death patient group, heart failure hospitalized patient group, and secondary endpoint event patient group. Then, the optimal radiomics-clinical feature joint prediction model for each patient group is selected as the HCM patient endpoint event prediction model for that patient group. This disclosure achieves the prediction of multiple cardiovascular endpoint events in HCM patients, not limited to sudden cardiac death, and demonstrates good predictive efficacy in various predictions.

Description

Technical Field

[0001] This disclosure relates to the field of medical big data technology, specifically to a method for establishing an endpoint event prediction model for patients with hypertrophic cardiomyopathy (HCM) based on delayed gadolinium enhancement (LGE) radiomics features and clinical prediction features, a method for predicting the risk of sudden cardiac death (SCD) based on the HCM patient endpoint event prediction model, and an apparatus. Background Technology

[0002] Hemorrhage-induced heart disease (HCM) is a common hereditary heart disease, inherited in an autosomal dominant pattern. It is characterized by significant myocardial hypertrophy (often involving the left ventricular wall) accompanied by diastolic dysfunction, and is typically accompanied by symptoms such as chest pain, dyspnea, syncope, and sudden death. Studies indicate that the prevalence of HCM in the general adult population is 0.16%–0.23%, with an annual mortality rate of 1.5%–4%, or even higher. In reality, the clinical course of HCM is highly variable. 47%–50% of HCM patients can maintain a long-term asymptomatic or mildly symptomatic benign course throughout their lives. On the other hand, 40%–50% of HCM patients will experience at least one adverse cardiovascular event, including sudden cardiac death, heart failure, atrial fibrillation / stroke, etc., resulting in a significant social health burden.

[0003] Because HCM patients have a high risk of developing sudden cardiac death (SCD) and a very poor prognosis due to SCD, current research on the prognosis of HCM patients mainly focuses on the early prediction and prevention of SCD, and has achieved good results. However, heart failure, as the leading cause of morbidity and mortality in HCM and the greatest unmet treatment need in adult HCM patients, is gradually becoming the focus of research.

[0004] However, there is still a lack of relevant risk prediction studies for heart failure and cardiovascular death in HCM patients. Therefore, it is an urgent problem to find predictive factors and establish predictive models for HCM patients to predict the risk of cardiovascular endpoint events (cardiovascular death, heart failure, readmission rate, etc.) and guide the formulation of treatment plans for HCM patients. Summary of the Invention

[0005] To address the problems in the related technologies, this disclosure provides a method for establishing a prediction model of HCM patient endpoint events based on LGE radiomics features and clinical prediction features, a method for predicting SCD risk based on the prediction model of HCM patient endpoint events, and an apparatus.

[0006] In a first aspect, this disclosure provides a method for establishing a prediction model for endpoint events in HCM patients based on LGE radiomics features and clinical prediction features. The method is implemented by a computer and includes:

[0007] The HCM patient endpoint event group and the HCM patient control group were obtained. The HCM patient endpoint event group included the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group. The HCM patient control group included the first HCM patient control group, the second HCM patient control group, and the third HCM patient control group, which corresponded to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively.

[0008] Based on the baseline clinical predictive characteristics of the cardiovascular death patient group and the first HCM patient control group, a first clinical characteristic predictor is obtained for the cardiovascular death patient group; based on the baseline clinical predictive characteristics of the heart failure hospitalized patient group and the second HCM patient control group, a second clinical characteristic predictor is obtained for the heart failure hospitalized patient group; based on the baseline clinical predictive characteristics of the secondary endpoint event patient group and the third HCM patient control group, a third clinical characteristic predictor is obtained for the secondary endpoint event patient group; wherein, the first clinical characteristic predictor includes: left atrial diameter, family history of sudden cardiac death; the second clinical characteristic predictor includes: NYHA functional class, non-sustained ventricular tachycardia; and the third clinical characteristic predictor includes: age, NYHA functional class.

[0009] Based on the LGE images of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, the screened radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are obtained; and the radiomics feature scores of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are calculated based on the corresponding screened radiomics features.

[0010] For each patient group among the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: based on the radiomics feature scores and corresponding clinical feature predictors of each patient group, multiple radiomics-clinical feature joint prediction models are established using various machine learning methods; the multiple radiomics-clinical feature joint prediction models are validated, and the optimal radiomics-clinical feature joint prediction model is selected as the HCM patient endpoint event prediction model for each patient group.

[0011] According to an embodiment of this disclosure, obtaining a first clinical feature predictor for the cardiovascular death patient group based on the baseline clinical prediction characteristics of the cardiovascular death patient group and the first HCM patient control group includes: inputting the baseline clinical prediction characteristics of the cardiovascular death patient group and the first HCM patient control group into a multivariate survival analysis model using a stepwise regression method to obtain the first clinical feature predictor.

[0012] Based on the baseline clinical prediction characteristics of the hospitalized heart failure patient group and the second HCM patient control group, a second clinical feature predictor is obtained for the hospitalized heart failure patient group, including: inputting the baseline clinical prediction characteristics of the hospitalized heart failure patient group and the second HCM patient control group into a multivariate survival analysis model using stepwise regression to obtain the second clinical feature predictor;

[0013] Based on the baseline clinical predictive characteristics of the secondary endpoint event patient group and the third HCM patient control group, the third clinical feature predictor for the secondary endpoint event patient group is obtained, including: inputting the baseline clinical predictive characteristics of the secondary endpoint event patient group and the third HCM patient control group into a multivariate survival analysis model using stepwise regression to obtain the third clinical feature predictor.

[0014] The baseline clinical predictive features include: patient basic information, echocardiographic information, electrocardiogram information, and basic parameters of cardiac magnetic resonance imaging.

[0015] According to embodiments of this disclosure, the multifactor survival analysis model is any one or more of the following:

[0016] Cox proportional hazards model, Weibull distribution model, and competitive risks model;

[0017] The stepwise regression method includes any one or more of the following:

[0018] Forward stepwise regression, backward stepwise regression, and bidirectional stepwise regression.

[0019] According to embodiments of this disclosure, the HCM endpoint events for HCM patients in the cardiovascular death patient group include sudden cardiac death and / or death from heart failure;

[0020] The HCM endpoint event for HCM patients in the heart failure hospitalization group included heart failure hospitalization;

[0021] The secondary endpoint events for HCM patients in the patient group include any one or more of the following: unplanned readmission due to cardiovascular disease, unplanned readmission due to other causes, and death from non-cardiovascular disease.

[0022] No sudden cardiac death or heart failure occurred among the HCM patients in the first HCM patient control group;

[0023] No HCM patients in the second HCM patient control group were hospitalized for heart failure;

[0024] No unplanned readmissions due to cardiovascular disease, unplanned readmissions for other reasons, or non-cardiovascular disease deaths occurred in the third HCM patient control group.

[0025] According to embodiments of this disclosure, the step of obtaining screened radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group based on LGE image groups includes:

[0026] For each patient group in the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, an LGE image set for each patient group is obtained, the LGE image set including LGE images of HCM patients in each patient group; the left ventricular myocardial short-axis section of the LGE image in the LGE image set is taken as the region of interest, and the pre-screening radiomics features of each patient group are extracted from the region of interest;

[0027] The pre-screening radiomics features of each patient group were input into the LASSO (Least Absolute Contraction and Selection) model for correlation analysis to obtain the post-screening radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group.

[0028] According to embodiments of this disclosure, calculating radiomics feature scores for the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group based on the corresponding screened radiomics features includes:

[0029] Based on the feature distribution and LASSO coefficient of the corresponding screened radiomics features, the radiomics feature scores of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are calculated respectively.

[0030] According to embodiments of this disclosure, the machine learning method includes any one of the following:

[0031] Logistic regression, K-nearest neighbor method, elastic network method, and Naive Bayes method.

[0032] According to embodiments of this disclosure, the step of validating the plurality of radiomics-clinical feature joint prediction models and selecting the optimal radiomics-clinical feature joint prediction model includes: calculating the area under the ROC curve of the plurality of radiomics-clinical feature joint prediction models, and selecting the radiomics-clinical feature joint prediction model with the largest area under the ROC curve as the optimal radiomics-clinical feature joint prediction model.

[0033] Secondly, this disclosure provides a method for predicting the risk of sudden cardiac death (SCD) based on an endpoint event prediction model for HCM patients. The method is implemented by a computer and includes:

[0034] The HCM patient endpoint event group and the HCM patient control group were obtained. The HCM patient endpoint event group included the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group. The HCM patient control group included the first HCM patient control group, the second HCM patient control group, and the third HCM patient control group, which corresponded to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively.

[0035] Based on the baseline clinical predictive characteristics of the cardiovascular death patient group and the first HCM patient control group, a first clinical characteristic predictor is obtained for the cardiovascular death patient group; based on the baseline clinical predictive characteristics of the heart failure hospitalized patient group and the second HCM patient control group, a second clinical characteristic predictor is obtained for the heart failure hospitalized patient group; based on the baseline clinical predictive characteristics of the secondary endpoint event patient group and the third HCM patient control group, a third clinical characteristic predictor is obtained for the secondary endpoint event patient group; wherein, the first clinical characteristic predictor includes: left atrial diameter, family history of sudden cardiac death; the second clinical characteristic predictor includes: NYHA functional class, non-sustained ventricular tachycardia; and the third clinical characteristic predictor includes: age, NYHA functional class.

[0036] Based on the LGE images of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, the screened radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are obtained; and the radiomics feature scores of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are calculated based on the corresponding screened radiomics features.

[0037] For each patient group among the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: based on the radiomics feature scores and corresponding clinical feature predictors of each patient group, multiple radiomics-clinical feature joint prediction models are established using various machine learning methods; the multiple radiomics-clinical feature joint prediction models are validated, and the optimal radiomics-clinical feature joint prediction model is selected as the HCM patient endpoint event prediction model for each patient group;

[0038] Acquire a dataset of HCM patients to be tested, wherein the dataset of HCM patients to be tested includes data from multiple HCM patients to be tested;

[0039] The dataset of HCM patients to be tested is input into the HCM patient endpoint event prediction model of the cardiovascular death patient group to obtain the corresponding SCD risk prediction results.

[0040] According to embodiments of this disclosure, the machine learning method includes any one of the following:

[0041] Logistic regression, K-nearest neighbor method, elastic network method, and Naive Bayes method.

[0042] Thirdly, this disclosure provides an apparatus for establishing a prediction model for endpoint events in HCM patients based on LGE radiomics features and clinical prediction features. The apparatus is installed in a computer and includes:

[0043] The first acquisition module is configured to acquire HCM patient endpoint event groups and HCM patient control groups. The HCM patient endpoint event groups include cardiovascular death patient groups, heart failure hospitalized patient groups, and secondary endpoint event patient groups. The HCM patient control groups include a first HCM patient control group, a second HCM patient control group, and a third HCM patient control group, which correspond to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively.

[0044] The second acquisition module is configured to acquire a first clinical feature predictor for the cardiovascular death patient group based on the baseline clinical predictor characteristics of the cardiovascular death patient group and the first HCM patient control group; acquire a second clinical feature predictor for the heart failure hospitalized patient group based on the baseline clinical predictor characteristics of the heart failure hospitalized patient group and the second HCM patient control group; and acquire a third clinical feature predictor for the secondary endpoint event patient group based on the baseline clinical predictor characteristics of the secondary endpoint event patient group and the third HCM patient control group; wherein the first clinical feature predictor includes: left atrial diameter, family history of sudden cardiac death; the second clinical feature predictor includes: NYHA functional class, non-sustained ventricular tachycardia; and the third clinical feature predictor includes: age, NYHA functional class.

[0045] The first calculation module is configured to obtain, based on the LGE images of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, the selected radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group; and to calculate the radiomics feature scores of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group based on the corresponding selected radiomics features.

[0046] The first model selection module is configured to, for each patient group among the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: establish multiple radiomics-clinical feature joint prediction models using various machine learning methods based on the radiomics feature scores and corresponding clinical feature predictors of each patient group; validate the multiple radiomics-clinical feature joint prediction models, and select the optimal radiomics-clinical feature joint prediction model as the HCM patient endpoint event prediction model for each patient group.

[0047] Fourthly, this disclosure provides an apparatus for predicting the risk of sudden cardiac death (SCD) based on an endpoint event prediction model for HCM patients, comprising:

[0048] The third acquisition module is configured to acquire HCM patient endpoint event groups and HCM patient control groups. The HCM patient endpoint event groups include cardiovascular death patient groups, heart failure hospitalized patient groups, and secondary endpoint event patient groups. The HCM patient control groups include a first HCM patient control group, a second HCM patient control group, and a third HCM patient control group, which correspond to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively.

[0049] The fourth acquisition module is configured to acquire, based on the baseline clinical predictive characteristics of the cardiovascular death patient group and the first HCM patient control group, a first clinical feature predictor factor for the cardiovascular death patient group; based on the baseline clinical predictive characteristics of the heart failure hospitalized patient group and the second HCM patient control group, a second clinical feature predictor factor for the heart failure hospitalized patient group; and based on the baseline clinical predictive characteristics of the secondary endpoint event patient group and the third HCM patient control group, a third clinical feature predictor factor for the secondary endpoint event patient group; wherein the first clinical feature predictor factor includes: left atrial diameter, family history of sudden cardiac death; the second clinical feature predictor factor includes: NYHA functional class, non-sustained ventricular tachycardia; and the third clinical feature predictor factor includes: age, NYHA functional class.

[0050] The second calculation module is configured to obtain, based on the LGE images of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, the filtered radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group; and to calculate the radiomics feature scores of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group based on the corresponding filtered radiomics features.

[0051] The second model selection module is configured to, for each patient group among the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: establish multiple radiomics-clinical feature joint prediction models using various machine learning methods based on the radiomics feature scores and corresponding clinical feature predictors of each patient group; validate the multiple radiomics-clinical feature joint prediction models, and select the optimal radiomics-clinical feature joint prediction model as the HCM patient endpoint event prediction model for each patient group;

[0052] The data acquisition module is configured to acquire a set of HCM patient data to be tested, which includes data on multiple HCM patients to be tested.

[0053] The SCD prediction module is configured to input the HCM patient data set to be tested into the HCM patient endpoint event prediction model of the cardiovascular death patient group to obtain the corresponding SCD risk prediction results.

[0054] Fifthly, this disclosure provides an electronic device including a memory and a processor; wherein the memory is used to store computer instructions, which are executed by the processor to implement the method as described in either the first or second aspect.

[0055] In a sixth aspect, embodiments of this disclosure provide a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the method as described in either the first or second aspect.

[0056] In a seventh aspect, this disclosure provides a computer program product including computer instructions that, when executed by a processor, implement the method described in either the first or second aspect.

[0057] According to the technical solution provided in this disclosure, LGE radiomics is combined with clinical features. Based on the radiomics feature scores and corresponding clinical feature predictors of each patient group in the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, multiple radiomics-clinical feature joint prediction models are established using various machine learning methods. The multiple radiomics-clinical feature joint prediction models are then validated, and the optimal radiomics-clinical feature joint prediction model is selected as the HCM patient endpoint event prediction model for each patient group.

[0058] This invention innovatively combines LGE radiomics with clinical features, resulting in a predictive model with higher accuracy than models built solely using clinical predictive indicators. Furthermore, it can predict not only cardiovascular endpoints such as sudden cardiac death, but also other cardiovascular endpoints in HCM patients (heart failure death, heart failure hospitalization, unplanned readmission due to cardiovascular disease, unplanned readmission due to other reasons, and non-cardiovascular disease death). This enables the prediction of endpoints not given special attention in existing protocols, such as heart failure hospitalization, demonstrating excellent predictive power across all cardiovascular endpoints.

[0059] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0060] Other features, objects, and advantages of this disclosure will become more apparent from the following detailed description of non-limiting embodiments, taken in conjunction with the accompanying drawings. In the drawings:

[0061] Figure 1 A flowchart is shown illustrating a method for establishing a prediction model of endpoint events in HCM patients based on LGE radiomics features and clinical prediction features, according to an embodiment of the present disclosure.

[0062] Figure 2 This diagram illustrates the distribution of endpoint events in an HCM patient according to an embodiment of the present disclosure.

[0063] Figure 3 This diagram illustrates the area under the ROC curve of a combined radiomics-clinical feature prediction model for a cardiovascular death patient group in an embodiment of this disclosure.

[0064] Figure 4 This diagram illustrates the area under the ROC curve of a combined radiomics-clinical feature prediction model for a group of hospitalized patients with heart failure in an embodiment of this disclosure.

[0065] Figure 5 This diagram illustrates the area under the ROC curve of a combined radiomics-clinical feature prediction model for a patient group with secondary endpoint events, as shown in an embodiment of this disclosure.

[0066] Figure 6 A flowchart illustrating a method for predicting SCD risk based on an HCM patient endpoint event prediction model according to an embodiment of the present disclosure;

[0067] Figure 7 A structural block diagram of an apparatus for establishing a predictive model of HCM patient endpoint events based on LGE radiomics features and clinical predictive features, according to an embodiment of the present disclosure, is shown.

[0068] Figure 8 A structural block diagram of an apparatus for predicting SCD risk based on an HCM patient endpoint event prediction model is shown according to an embodiment of the present disclosure.

[0069] Figure 9 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown;

[0070] Figure 10 A schematic diagram of the structure of a computer system suitable for implementing the method according to embodiments of the present disclosure is shown. Detailed Implementation

[0071] In the following, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings to enable those skilled in the art to readily implement them. Furthermore, for clarity, portions unrelated to the description of exemplary embodiments have been omitted from the drawings.

[0072] In this disclosure, it should be understood that terms such as “comprising” or “having” are intended to indicate the presence of features, figures, steps, behaviors, components, parts or combinations thereof disclosed in this specification, and are not intended to exclude the possibility of the presence or addition of one or more other features, figures, steps, behaviors, components, parts or combinations thereof.

[0073] It should also be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other. This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0074] In this disclosure, any operation involving the acquisition of user information or user data, or the display of user information or user data to others, is an operation authorized or confirmed by the user, or actively selected by the user.

[0075] As mentioned earlier, early prediction and identification of the risk of cardiovascular endpoint events are of great significance for HCM patients. After investigation and in-depth analysis, the inventors found that although existing studies have revealed factors such as gender, brain natriuretic peptide levels, left atrial diameter, critical decrease in left ventricular ejection fraction, and left ventricular outflow tract pressure gradient as predictors of heart failure and adverse cardiovascular events in HCM patients, these indicators generally face limitations in practical application, such as insufficient sensitivity or lack of sufficient validation through large-scale clinical trials. Therefore, how to explore more precise and sensitive predictive factors and construct more reliable and accurate predictive models based on them, in order to achieve accurate assessment and prediction of disease progression and prognostic risk in HCM patients, is an urgent problem to be solved.

[0076] With the development of imaging technology, cardiac magnetic resonance imaging (CMR) has become the standard non-invasive imaging tool in the cardiovascular field. LGE, as a CMR technique, can indicate the presence and degree of myocardial fibrosis by observing the concentration of gadolinium contrast agent residues in scar areas. The inventors have found that although studies have explored the significant correlation between LGE and cardiovascular endpoints, there are still some significant issues when applying CMR-LGE results to predict cardiovascular endpoints in HCM patients:

[0077] First, the current threshold for LGE% is set at 15% when predicting the risk of cardiovascular endpoint events. However, some clinical studies have found that even HCM patients with LGE% below 15% have a higher risk of developing cardiovascular endpoint events. Therefore, relying solely on LGE% as a predictor of cardiovascular endpoint events may not be sensitive enough.

[0078] Furthermore, LGE enhancement is usually measured using the degree of enhancement in normal myocardium as a reference standard. In patients with extensive myocardial fibrosis, where a normal myocardial control is lacking, the measurement of LGE% may have significant errors. Therefore, more robust predictive factors are still needed in clinical practice.

[0079] Therefore, CMR-LGE suffers from poor sensitivity and large measurement errors in practical applications. The inventors noted that with the continuous development of artificial intelligence, radiomics has received considerable attention in the cardiovascular field. They innovatively combined CMR-LGE radiomics with clinical characteristics to realize a method for establishing a predictive model of endpoint events in HCM patients based on LGE radiomics features and clinical predictive features. This method is implemented by a computer and includes:

[0080] The study obtains HCM patient endpoint event groups and HCM patient control groups. The HCM patient endpoint event groups include a cardiovascular death patient group, a heart failure hospitalized patient group, and a secondary endpoint event patient group. The HCM patient control groups include a first HCM patient control group, a second HCM patient control group, and a third HCM patient control group, corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively. Based on the baseline clinical predictive characteristics of the cardiovascular death patient group and the first HCM patient control group, a first clinical feature predictor is obtained for the cardiovascular death patient group. Based on the baseline clinical predictive characteristics of the heart failure hospitalized patient group and the second HCM patient control group, a second clinical feature predictor is obtained for the heart failure hospitalized patient group. Based on the baseline clinical predictive characteristics of the secondary endpoint event patient group and the third HCM patient control group, a third clinical feature predictor is obtained for the secondary endpoint event patient group. The first clinical feature predictor includes: left atrial diameter and family history of sudden cardiac death. The second clinical feature predictor includes: N. YHA functional classification, non-sustained ventricular tachycardia, the third clinical feature predictor includes: age, NYHA functional classification; based on the LGE image sets of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, obtained the screened radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group; calculated the radiomics feature scores of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group based on the corresponding screened radiomics features; for each patient group in the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: based on the radiomics feature scores of each patient group and the corresponding clinical feature predictor, multiple radiomics-clinical feature joint prediction models were established using multiple machine learning methods; the multiple radiomics-clinical feature joint prediction models were validated, and the optimal radiomics-clinical feature joint prediction model was selected as the HCM patient endpoint event prediction model for each patient group.

[0081] This disclosure preliminarily explores the feasibility of using machine learning to combine LGE radiomics and clinical features to predict endpoint events in HCM patients. It can predict not only cardiovascular endpoint events such as sudden cardiac death, but also other endpoint events in HCM patients. The prediction accuracy is better than that of models built using only clinical predictive indicators, and it shows good predictive power in the prediction of various endpoint events, with good generalizability.

[0082] Figure 1 A flowchart illustrating a method for establishing a predictive model of endpoint events in HCM patients based on LGE radiomics features and clinical predictive features, according to an embodiment of this disclosure, is shown. The method is implemented by a computer, such as... Figure 1 As shown, the method includes the following steps S101 to S104:

[0083] In step S101, the HCM patient endpoint event group and the HCM patient control group are obtained. The HCM patient endpoint event group includes the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group. The HCM patient control group includes the first HCM patient control group, the second HCM patient control group, and the third HCM patient control group, which correspond to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively. Each of the cardiovascular death patient group, the heart failure hospitalized patient group, the secondary endpoint event patient group, the first HCM patient control group, the second HCM patient control group, and the third HCM patient control group includes multiple HCM patients.

[0084] According to embodiments of this disclosure, the HCM endpoint events for HCM patients in the cardiovascular death patient group include sudden cardiac death (SCD) and / or death from heart failure.

[0085] According to embodiments of this disclosure, the HCM endpoint event for HCM patients in the heart failure hospitalization group includes heart failure hospitalization.

[0086] According to embodiments of this disclosure, the HCM endpoint event for HCM patients in the secondary endpoint event patient group includes any one or more of the following: unplanned readmission due to cardiovascular disease, unplanned readmission due to other reasons, and death from non-cardiovascular disease.

[0087] According to embodiments of this disclosure, no sudden cardiac death or heart failure occurred among the HCM patients in the first HCM patient control group.

[0088] According to embodiments of this disclosure, no HCM patients in the second HCM patient control group were hospitalized for heart failure.

[0089] According to embodiments of this disclosure, no unplanned readmissions due to cardiovascular disease, unplanned readmissions for other reasons, or non-cardiovascular disease deaths occurred in the third HCM patient control group.

[0090] like Figure 2 As shown, in one specific implementation, 104 HCM patients were enrolled and followed up for an average duration of 2.78 years (range 1.16–5.28 years). Among them, 10 patients (9.6%) experienced cardiovascular death and were assigned to the cardiovascular death group, while the corresponding first HCM patient control group consisted of 94 patients. 15 patients (14.4%) experienced hospitalization for heart failure and were assigned to the heart failure hospitalization group, while the corresponding second HCM patient control group consisted of 89 patients. A total of 29 patients (27.9%) experienced secondary endpoint events, including 25 patients (24.0%) who were readmitted due to cardiovascular or cerebrovascular diseases and 4 patients (3.8%) who were readmitted for other reasons and were assigned to the secondary endpoint event group, while the corresponding third HCM patient control group consisted of 75 patients.

[0091] In step S102, a first clinical feature predictor for the cardiovascular death patient group is obtained based on the baseline clinical predictor characteristics of the cardiovascular death patient group and the first HCM patient control group; a second clinical feature predictor for the heart failure hospitalized patient group is obtained based on the baseline clinical predictor characteristics of the heart failure hospitalized patient group and the second HCM patient control group; a third clinical feature predictor for the secondary endpoint event patient group is obtained based on the baseline clinical predictor characteristics of the secondary endpoint event patient group and the third HCM patient control group; wherein, the first clinical feature predictor includes: left atrial diameter, family history of sudden cardiac death; the second clinical feature predictor includes: NYHA functional class, non-sustained ventricular tachycardia; and the third clinical feature predictor includes: age, NYHA functional class.

[0092] According to embodiments of this disclosure, the baseline clinical predictive features of the cardiovascular death patient group, the heart failure hospitalized patient group, the secondary endpoint event patient group, the first HCM patient control group, the second HCM patient control group, and the third HCM patient control group all include: basic patient information, echocardiographic information, electrocardiographic information, and basic cardiac magnetic resonance parameters.

[0093] Specifically, basic patient information includes: gender, age, height, weight, NYHA functional classification, family history of sudden cardiac death, etc. Echocardiographic information includes: ventricular wall thickness, left ventricular ejection fraction, left atrial diameter, left ventricular end-diastolic diameter, left ventricular posterior wall diastolic thickness, right ventricular end-diastolic diameter, etc. Electrocardiographic information includes: non-sustained ventricular tachycardia, atrial tachycardia, total number of premature ventricular contractions (PVCs) in 24 hours, total number of paired PVCs in 24 hours, etc. Basic cardiac magnetic resonance imaging parameters include: left ventricular myocardial mass, left ventricular end-diastolic volume, left ventricular end-systolic volume, ventricular wall thickness, LGE enhancement percentage, etc.

[0094] The NYHA functional classification divides cardiac function into four levels based on the degree of activity that induces symptoms of heart failure: Level I: No restriction on daily activities; Level II: Mild restriction on physical activity; Level III: Significant restriction on physical activity; Level IV: Unable to perform any physical activity.

[0095] According to embodiments of this disclosure, the baseline clinical predictive characteristics of the cardiovascular death patient group and the first HCM patient control group are input into a multivariate survival analysis model using stepwise regression to obtain the first clinical characteristic predictive factor.

[0096] According to embodiments of this disclosure, the baseline clinical predictive characteristics of the heart failure hospitalized patient group and the second HCM patient control group are input into a multivariate survival analysis model using stepwise regression to obtain the second clinical characteristic predictive factor.

[0097] According to embodiments of this disclosure, the baseline clinical predictive characteristics of the secondary endpoint event patient group and the third HCM patient control group are input into a multivariate survival analysis model using stepwise regression to obtain the third clinical characteristic predictor.

[0098] In detail, for baseline clinical predictive characteristics, these variables are input into a multivariate survival analysis model using stepwise regression to obtain clinical characteristic predictors corresponding to each patient group.

[0099] According to embodiments of this disclosure, the multifactor survival analysis model is any one or more of the following: Cox proportional hazards model, Weibull distribution model, and competitive risk model.

[0100] According to embodiments of this disclosure, the stepwise regression method includes any one or more of the following: forward stepwise regression, backward stepwise regression, and bidirectional stepwise regression.

[0101] In one specific embodiment, the baseline clinical predictive features were analyzed using a forward stepwise regression method for multivariate analysis (p-values ​​were in increments of 0.05 and out increments of 0.1) to screen out relevant clinical predictive factors in the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group.

[0102] according to Figure 2 The example shown further illustrates the predictors of clinical characteristics obtained using forward stepwise regression in multivariate Cox regression, as illustrated in Table 1:

[0103] Table 1

[0104] Predictors p-value Cardiovascular death left atrial diameter 0.004 Family history of sudden cardiac death 0.060 Hospitalization for heart failure NYHA functional classification 0.049 Non-sustained ventricular tachycardia 0.007 Secondary endpoint events age 0.023 NYHA functional classification 0.033

[0105] In the multivariate Cox regression, a forward stepwise regression method is used. By gradually adding baseline clinical predictive features to the Cox proportional hazards model and re-evaluating the model’s goodness of fit at each step, the optimal combination of clinical predictive features is found. Using p-values ​​as the criteria for inclusion (0.05) and exclusion (0.1), if the exclusion of a clinical predictive feature leads to a significant decrease in the model's goodness of fit (judged by a p-value less than 0.05), the clinical predictive feature is retained in the model; otherwise, it can be considered for removal from the model (judged by a p-value greater than 0.1). Thus, the clinical predictive factors corresponding to the cardiovascular death patient group include: left atrial diameter and family history of sudden cardiac death (although the variable "family history of sudden cardiac death" is not statistically significant, it is included in subsequent studies because it is a recognized risk factor for cardiovascular endpoint events in HCM patients and is only borderline uncorrelated). The clinical predictive factors corresponding to the hospitalized heart failure patient group include: NYHA functional class and non-sustained ventricular tachycardia. The clinical predictive factors corresponding to the secondary endpoint event patient group include: age and NYHA functional class.

[0106] In step S103, based on the LGE image groups of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, the screened radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are obtained; and the radiomics feature scores of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are calculated based on the corresponding screened radiomics features.

[0107] Specifically, for each patient group in the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, an LGE image set for each patient group is obtained, the LGE image set including LGE images of HCM patients in each patient group; the left ventricular myocardial short-axis section of the LGE image in the LGE image set is taken as the region of interest, and the pre-screening radiomics features of each patient group are extracted from the region of interest.

[0108] Specifically, the pre-screening radiomics features of each patient group are input into the Least Absolute Shrinkage and Selection Operator (LASSO) model for correlation analysis to obtain the post-screening radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group.

[0109] Furthermore, based on the feature distribution and LASSO coefficient of the corresponding screened radiomics features, radiomics feature scores are calculated for the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively.

[0110] Specifically, LGE images in each patient group were delineated, and the left ventricular myocardium was delineated as a Region of Interest (ROI) on the short-axis LGE image of the heart. Radiomic features of the left ventricular myocardium were then extracted based on the ROI. Morphological features, first-order statistical features, and second- and higher-order texture features were extracted from the original images and under multiple filters as pre-screening radiomic features. These pre-screening radiomic features were then input into the LASSO model for correlation analysis. Based on the λ value that minimizes the average cross-validation error, radiomic features with LASSO coefficients ≠ 0 were selected.

[0111] exist Figure 2 In the example shown, after screening, the cardiovascular death patient group included seven radiomics features, which are:

[0112] Logarithm-gldm-Small Dependence Emphasis; Wavelet Transform LHL-Strength-Strength (adjacent gray-level difference matrix); Wavelet Transform HLH-Sum Entropy (gray-level co-occurrence matrix feature); Wavelet Transform LHH-Kurtosis (first order feature); Wavelet Transform HHH-Difference Entropy (gray-level co-occurrence matrix feature); Wavelet Transform HLL-Small Area Emphasis (gray-level size region matrix feature); Wavelet Transform HHH-Mean (first order feature).

[0113] The screening included three radiomics features in the heart failure hospitalization group:

[0114] Wavelet transform HLH - First order Mean; Local binary mode (3D image) - Gray run matrix - Run Entropy (lbp-3D-k-glrlm); Wavelet transform LHH - Gray level size region matrix feature - Size Zone Non-Uniformity Normalized (wavelet-LHH-glszm).

[0115] The secondary endpoint event group included six radiomics features after screening:

[0116] Local binary mode (3D image) - first-order feature - minimum eigenvalue (lbp-3D-k-first-order-Minimum); wavelet transform HHH - first-order feature - mean (wavelet-HHH-first-order-Mean); wavelet transform HHH - gray-level size region matrix feature - zone variance (wavelet-HHH-glszm-Zone Variance); wavelet transform HLH - first-order feature - mean (wavelet-HLH-first-order-Mean); wavelet transform LHH - gray-level size region matrix feature - size zone non-uniformity normalized (wavelet-LHH glszm-Size Zone Non-UniformityNormalized); wavelet transform LHL - gray-level co-occurrence matrix feature - difference square (wavelet-LHL-glcm-DifferenceVariance).

[0117] According to embodiments of this disclosure, the Rad-score is calculated for each patient group by combining the distribution of each included screening radiomics feature and the corresponding feature coefficients of different radiomics features in LASSO regression. The Rad-score integrates multiple radiomics features into a single value to characterize the image features of interest, making it easier to apply complex radiomics information to clinical decision-making and research.

[0118] In step S104, for each patient group among the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: based on the radiomics feature scores and corresponding clinical feature predictors of each patient group, multiple radiomics-clinical feature joint prediction models are established using various machine learning methods; the multiple radiomics-clinical feature joint prediction models are validated, and the optimal radiomics-clinical feature joint prediction model is selected as the HCM patient endpoint event prediction model for each patient group.

[0119] According to embodiments of this disclosure, the machine learning method includes any one of the following: logistic regression, K-nearest neighbors, elastic network, and Naive Bayes.

[0120] According to embodiments of this disclosure, the area under the ROC curve of the plurality of radiomics-clinical feature joint prediction models is calculated, and the radiomics-clinical feature joint prediction model with the largest area under the ROC curve is taken as the optimal radiomics-clinical feature joint prediction model.

[0121] In this disclosure, clinical predictive indicators are combined with Rad-Scores of various endpoint events, and four machine learning methods—Logistic Regression (LR), K-Nearest Neighbors (KNN), Elastic Net (EN), and Naive Bayes—are used to establish corresponding joint predictive models of radiomics and clinical features. After each predictive model is established, the area under the ROC curve (AUC) of each model is calculated, and the joint predictive model of radiomics and clinical features with the largest AUC is selected as the optimal joint predictive model of radiomics and clinical features.

[0122] Figure 3 This diagram illustrates the area under the ROC curve of a combined radiomics-clinical feature prediction model for a cardiovascular death patient group in an embodiment of this disclosure. Figure 4 This diagram illustrates the area under the ROC curve of a combined radiomics-clinical feature prediction model for a group of hospitalized patients with heart failure in an embodiment of this disclosure. Figure 5 This diagram illustrates the area under the ROC curve of a combined radiomics-clinical feature prediction model for a patient group with secondary endpoint events, as shown in an embodiment of this disclosure.

[0123] like Figure 3-5 As shown, the optimal radiomics-clinical feature joint prediction model in the cardiovascular death patient group is a model trained based on logistic regression, the optimal radiomics-clinical feature joint prediction model in the heart failure hospitalized patient group is a model trained based on K-nearest neighbors, and the optimal radiomics-clinical feature joint prediction model in the secondary endpoint event patient group is a model trained based on Naive Bayes.

[0124] This disclosure innovatively combines machine learning with radiomics and clinical features for the prediction of endpoint events in HCM patients, improving the predictive efficacy for multiple endpoint events in HCM patients. The prediction accuracy of the established predictive model is superior to models built solely using clinical predictive indicators. Furthermore, this disclosure focuses on important endpoint events in HCM patients, such as cardiovascular death and hospitalization for heart failure, which have not received particular attention in previous studies, and discovers the potential value of LGE radiomic features in predicting these endpoint events, providing a more comprehensive approach to assessing the risk of endpoint events in HCM patients.

[0125] Figure 6 A flowchart illustrating a method for predicting SCD risk based on an endpoint event prediction model for HCM patients according to an embodiment of the present disclosure is shown. The method is implemented by a computer and includes steps S601 to S606.

[0126] In step S601, the HCM patient endpoint event group and the HCM patient control group are obtained. The HCM patient endpoint event group includes the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group. The HCM patient control group includes the first HCM patient control group, the second HCM patient control group, and the third HCM patient control group, which correspond to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively.

[0127] In step S602, a first clinical feature predictor for the cardiovascular death patient group is obtained based on the baseline clinical predictor characteristics of the cardiovascular death patient group and the first HCM patient control group; a second clinical feature predictor for the heart failure hospitalized patient group is obtained based on the baseline clinical predictor characteristics of the heart failure hospitalized patient group and the second HCM patient control group; a third clinical feature predictor for the secondary endpoint event patient group is obtained based on the baseline clinical predictor characteristics of the secondary endpoint event patient group and the third HCM patient control group; wherein, the first clinical feature predictor includes: left atrial diameter, family history of sudden cardiac death; the second clinical feature predictor includes: NYHA functional class, non-sustained ventricular tachycardia; and the third clinical feature predictor includes: age, NYHA functional class.

[0128] In step S603, based on the LGE images of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, the screened radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are obtained; and the radiomics feature scores of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are calculated based on the corresponding screened radiomics features.

[0129] In step S604, for each patient group among the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: based on the radiomics feature scores and corresponding clinical feature predictors of each patient group, multiple radiomics-clinical feature joint prediction models are established using various machine learning methods; the multiple radiomics-clinical feature joint prediction models are validated, and the optimal radiomics-clinical feature joint prediction model is selected as the HCM patient endpoint event prediction model for each patient group.

[0130] In step S605, a data set of HCM patients to be tested is obtained, which includes data from multiple HCM patients to be tested.

[0131] In step S606, the dataset of HCM patients to be tested is input into the HCM patient endpoint event prediction model of the cardiovascular death patient group to obtain the corresponding SCD risk prediction results.

[0132] In this disclosure, the endpoint event prediction model for HCM patients in the cardiovascular death patient group is used to predict the risk of SCD in HCM patients. The model can output a risk score or probability value for each patient to assist doctors in assessing the patient's risk status and developing corresponding prevention and treatment strategies.

[0133] Figure 7 The diagram illustrates a structural block diagram of an apparatus for establishing a predictive model of endpoint events in HCM patients based on LGE radiomics features and clinical predictive features, according to an embodiment of the present disclosure. This apparatus can be implemented as part or all of an electronic device through software, hardware, or a combination of both.

[0134] like Figure 7 As shown, the device 700 includes: a first acquisition module 701, a second acquisition module 702, a first calculation module 703, and a first model establishment and selection module 704.

[0135] The first acquisition module 701 is configured to acquire HCM patient endpoint event groups and HCM patient control groups. The HCM patient endpoint event groups include cardiovascular death patient groups, heart failure hospitalized patient groups, and secondary endpoint event patient groups. The HCM patient control groups include a first HCM patient control group, a second HCM patient control group, and a third HCM patient control group, which correspond to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively.

[0136] The second acquisition module 702 is configured to acquire a first clinical feature predictor for the cardiovascular death patient group based on the baseline clinical predictor characteristics of the cardiovascular death patient group and the first HCM patient control group; acquire a second clinical feature predictor for the heart failure hospitalized patient group based on the baseline clinical predictor characteristics of the heart failure hospitalized patient group and the second HCM patient control group; and acquire a third clinical feature predictor for the secondary endpoint event patient group based on the baseline clinical predictor characteristics of the secondary endpoint event patient group and the third HCM patient control group; wherein the first clinical feature predictor includes: left atrial diameter, family history of sudden cardiac death; the second clinical feature predictor includes: NYHA functional class, non-sustained ventricular tachycardia; and the third clinical feature predictor includes: age, NYHA functional class.

[0137] The first calculation module 703 is configured to obtain, based on the LGE images of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, the selected radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group; and to calculate the radiomics feature scores of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group based on the corresponding selected radiomics features.

[0138] The first model establishment and selection module 704 is configured to, for each patient group among the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: establish multiple radiomics-clinical feature joint prediction models using various machine learning methods based on the radiomics feature scores and corresponding clinical feature predictors of each patient group; validate the multiple radiomics-clinical feature joint prediction models, and select the optimal radiomics-clinical feature joint prediction model as the HCM patient endpoint event prediction model for each patient group.

[0139] Figure 8 This diagram illustrates a structural block diagram of an apparatus for predicting SCD risk based on an HCM patient endpoint event prediction model according to an embodiment of the present disclosure. The apparatus can be implemented as part or all of an electronic device through software, hardware, or a combination of both.

[0140] like Figure 8 As shown, the device 800 includes: a third acquisition module 801, a fourth acquisition module 802, a second calculation module 803, a second model establishment and selection module 804, a data acquisition module 805, and an SCD prediction module 806.

[0141] The third acquisition module 801 is configured to acquire HCM patient endpoint event groups and HCM patient control groups. The HCM patient endpoint event groups include cardiovascular death patient groups, heart failure hospitalized patient groups, and secondary endpoint event patient groups. The HCM patient control groups include a first HCM patient control group, a second HCM patient control group, and a third HCM patient control group, which correspond to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively.

[0142] The fourth acquisition module 802 is configured to acquire a first clinical feature predictor for the cardiovascular death patient group based on the baseline clinical prediction characteristics of the cardiovascular death patient group and the first HCM patient control group; acquire a second clinical feature predictor for the heart failure hospitalized patient group based on the baseline clinical prediction characteristics of the heart failure hospitalized patient group and the second HCM patient control group; and acquire a third clinical feature predictor for the secondary endpoint event patient group based on the baseline clinical prediction characteristics of the secondary endpoint event patient group and the third HCM patient control group; wherein the first clinical feature predictor includes: left atrial diameter, family history of sudden cardiac death; the second clinical feature predictor includes: NYHA functional class, non-sustained ventricular tachycardia; and the third clinical feature predictor includes: age, NYHA functional class.

[0143] The second calculation module 803 is configured to obtain, based on the LGE images of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, the selected radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group; and to calculate the radiomics feature scores of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group based on the corresponding selected radiomics features.

[0144] The second model selection module 804 is configured to, for each patient group among the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: establish multiple radiomics-clinical feature joint prediction models using various machine learning methods based on the radiomics feature scores and corresponding clinical feature predictors of each patient group; validate the multiple radiomics-clinical feature joint prediction models, and select the optimal radiomics-clinical feature joint prediction model as the HCM patient endpoint event prediction model for each patient group.

[0145] The data acquisition module 805 is configured to acquire a data set of HCM patients to be tested, which includes data from multiple HCM patients to be tested.

[0146] The SCD prediction module 806 is configured to input the HCM patient data set to be tested into the HCM patient endpoint event prediction model of the cardiovascular death patient group to obtain the corresponding SCD risk prediction result.

[0147] This disclosure also discloses an electronic device. Figure 9 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown.

[0148] like Figure 9 As shown, the electronic device includes a memory and a processor, wherein the memory is used to store computer instructions, which are executed by the processor to implement the method according to embodiments of the present disclosure.

[0149] The method is as follows:

[0150] The study obtained HCM patient endpoint event groups and HCM patient control groups. The HCM patient endpoint event groups included cardiovascular death patients, heart failure hospitalized patients, and secondary endpoint event patients. The HCM patient control groups included a first HCM patient control group, a second HCM patient control group, and a third HCM patient control group, respectively corresponding to the cardiovascular death patient group, the heart failure hospitalized patients group, and the secondary endpoint event patients group.

[0151] Based on the baseline clinical predictive characteristics of the cardiovascular death patient group and the first HCM patient control group, a first clinical characteristic predictor is obtained for the cardiovascular death patient group; based on the baseline clinical predictive characteristics of the heart failure hospitalized patient group and the second HCM patient control group, a second clinical characteristic predictor is obtained for the heart failure hospitalized patient group; based on the baseline clinical predictive characteristics of the secondary endpoint event patient group and the third HCM patient control group, a third clinical characteristic predictor is obtained for the secondary endpoint event patient group; wherein, the first clinical characteristic predictor includes: left atrial diameter, family history of sudden cardiac death; the second clinical characteristic predictor includes: NYHA functional class, non-sustained ventricular tachycardia; and the third clinical characteristic predictor includes: age, NYHA functional class.

[0152] Based on the LGE images of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, the screened radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are obtained; and the radiomics feature scores of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are calculated based on the corresponding screened radiomics features.

[0153] For each patient group among the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: based on the radiomics feature scores and corresponding clinical feature predictors of each patient group, multiple radiomics-clinical feature joint prediction models are established using various machine learning methods; the multiple radiomics-clinical feature joint prediction models are validated, and the optimal radiomics-clinical feature joint prediction model is selected as the HCM patient endpoint event prediction model for each patient group.

[0154] According to an embodiment of this disclosure, a dataset of HCM patients to be tested is obtained, the dataset of HCM patients to be tested includes multiple HCM patients to be tested; the dataset of HCM patients to be tested is input into the HCM patient endpoint event prediction model of the cardiovascular death patient group to obtain the corresponding SCD risk prediction result.

[0155] Figure 10 A schematic diagram of the structure of a computer system suitable for implementing the method according to embodiments of the present disclosure is shown.

[0156] like Figure 10 As shown, the computer system includes a processing unit that can execute various methods described above based on a program stored in a read-only memory (ROM) or a program loaded from a storage portion into a random access memory (RAM). The RAM also stores various programs and data required for the operation of the computer system. The processing unit, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0157] The following components are connected to the I / O interface: input sections including keyboards, mice, etc.; output sections including cathode ray tubes (CRTs), liquid crystal displays (LCDs), and speakers; storage sections including hard disks, etc.; and communication sections including network interface cards such as LAN cards and modems. The communication section performs communication processes via a network such as the Internet. Drives are also connected to the I / O interface as needed. Removable media, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on the drive as needed so that computer programs read from them can be installed into the storage section as needed. The processing unit can be implemented as a CPU, GPU, TPU, FPGA, NPU, etc.

[0158] In particular, according to embodiments of this disclosure, the methods described above can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program containing program code for performing the methods described above. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium.

[0159] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0160] The units or modules described in the embodiments of this disclosure can be implemented in software or programmable hardware. The described units or modules can also be located in a processor, and the names of these units or modules do not necessarily constitute a limitation on the unit or module itself.

[0161] In another aspect, this disclosure also provides a computer-readable storage medium, which may be a computer-readable storage medium included in the electronic device or computer system described above; or it may be a standalone computer-readable storage medium not assembled into a device. The computer-readable storage medium stores one or more programs, which are used by one or more processors to perform the methods described in this disclosure.

[0162] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

Claims

1. A method for establishing an endpoint event prediction model for patients with hypertrophic cardiomyopathy (HCM) based on delayed gadolinium-enhanced LGE radiomics features and clinical predictive features, characterized in that, The method is implemented by a computer and includes: The HCM patient endpoint event group and the HCM patient control group were obtained. The HCM patient endpoint event group included the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group. The HCM patient control group included the first HCM patient control group, the second HCM patient control group, and the third HCM patient control group, which corresponded to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively. Based on the baseline clinical predictive characteristics of the cardiovascular death patient group and the first HCM patient control group, a first clinical characteristic predictor is obtained for the cardiovascular death patient group; based on the baseline clinical predictive characteristics of the heart failure hospitalized patient group and the second HCM patient control group, a second clinical characteristic predictor is obtained for the heart failure hospitalized patient group; based on the baseline clinical predictive characteristics of the secondary endpoint event patient group and the third HCM patient control group, a third clinical characteristic predictor is obtained for the secondary endpoint event patient group; wherein, the first clinical characteristic predictor includes: left atrial diameter, family history of sudden cardiac death; the second clinical characteristic predictor includes: NYHA functional class, non-sustained ventricular tachycardia; the third clinical characteristic predictor includes: age, NYHA functional class; and the baseline clinical predictive characteristics include: patient basic information, echocardiographic information, electrocardiographic information, and basic parameters of cardiac magnetic resonance imaging; Based on the LGE images of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, screened radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are obtained, including: for each patient group in the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, obtaining an LGE image set for each patient group, the LGE image set including LGE images of HCM patients in each patient group; and sorting the LGE images in the LGE image set... The left ventricular myocardial short-axis section is used as the region of interest (ROI). Pre-screening radiomics features of each patient group are extracted from the ROI. The pre-screening radiomics features of each patient group are input into the LASSO (Least Absolute Contraction and Selection) model for correlation analysis to obtain post-screening radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group. Radiomics feature scores are calculated for the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group based on the corresponding post-screening radiomics features. For each patient group among the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: based on the radiomics feature scores and corresponding clinical feature predictors of each patient group, multiple radiomics-clinical feature joint prediction models are established using various machine learning methods; the multiple radiomics-clinical feature joint prediction models are validated, and the optimal radiomics-clinical feature joint prediction model is selected as the HCM patient endpoint event prediction model for each patient group.

2. The method according to claim 1, characterized in that: Based on the baseline clinical prediction characteristics of the cardiovascular death patient group and the first HCM patient control group, the first clinical feature predictor of the cardiovascular death patient group is obtained, including: inputting the baseline clinical prediction characteristics of the cardiovascular death patient group and the first HCM patient control group into a multivariate survival analysis model using stepwise regression to obtain the first clinical feature predictor. Based on the baseline clinical prediction characteristics of the hospitalized heart failure patient group and the second HCM patient control group, a second clinical feature predictor is obtained for the hospitalized heart failure patient group, including: inputting the baseline clinical prediction characteristics of the hospitalized heart failure patient group and the second HCM patient control group into a multivariate survival analysis model using stepwise regression to obtain the second clinical feature predictor; Based on the baseline clinical predictive characteristics of the secondary endpoint event patient group and the third HCM patient control group, the third clinical characteristic predictor is obtained, including: inputting the baseline clinical predictive characteristics of the secondary endpoint event patient group and the third HCM patient control group into a multivariate survival analysis model using stepwise regression to obtain the third clinical characteristic predictor.

3. The method according to claim 2, characterized in that: The multifactor survival analysis model is any one or more of the following: Cox proportional hazards model, Weibull distribution model, and competitive risks model; The stepwise regression method includes any one or more of the following: Forward stepwise regression, backward stepwise regression, and bidirectional stepwise regression.

4. The method according to claim 2, characterized in that: The endpoint events for HCM patients in the cardiovascular death patient group included sudden cardiac death and / or death from heart failure; The HCM endpoint event for HCM patients in the heart failure hospitalization group included heart failure hospitalization; The secondary endpoint events for HCM patients in the patient group include any one or more of the following: unplanned readmission due to cardiovascular disease, unplanned readmission due to other causes, and death from non-cardiovascular disease. No sudden cardiac death or heart failure occurred among the HCM patients in the first HCM patient control group; No HCM patients in the second HCM patient control group were hospitalized for heart failure; No unplanned readmissions due to cardiovascular disease, unplanned readmissions for other reasons, or non-cardiovascular disease deaths occurred in the third HCM patient control group.

5. The method according to claim 1, characterized in that, The calculation of radiomics feature scores for the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group based on the corresponding screened radiomics features includes: Based on the feature distribution and LASSO coefficient of the corresponding screened radiomics features, the radiomics feature scores of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are calculated respectively.

6. The method according to claim 1, characterized in that, The machine learning method includes any of the following: Logistic regression, K-nearest neighbor method, elastic network method, and Naive Bayes method.

7. The method according to claim 1, characterized in that, The validation of the multiple radiomics-clinical feature joint prediction models and the selection of the optimal radiomics-clinical feature joint prediction model include: Calculate the area under the ROC curve of the multiple radiomics-clinical feature joint prediction models, and select the radiomics-clinical feature joint prediction model with the largest area under the ROC curve as the optimal radiomics-clinical feature joint prediction model.

8. A method for predicting the risk of sudden cardiac death (SCD) based on an endpoint event prediction model for HCM patients, characterized in that, The method is implemented by a computer and includes: The HCM patient endpoint event group and the HCM patient control group were obtained. The HCM patient endpoint event group included the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group. The HCM patient control group included the first HCM patient control group, the second HCM patient control group, and the third HCM patient control group, which corresponded to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively. Based on the baseline clinical predictive characteristics of the cardiovascular death patient group and the first HCM patient control group, a first clinical characteristic predictor is obtained for the cardiovascular death patient group; based on the baseline clinical predictive characteristics of the heart failure hospitalized patient group and the second HCM patient control group, a second clinical characteristic predictor is obtained for the heart failure hospitalized patient group; based on the baseline clinical predictive characteristics of the secondary endpoint event patient group and the third HCM patient control group, a third clinical characteristic predictor is obtained for the secondary endpoint event patient group; wherein, the first clinical characteristic predictor includes: left atrial diameter, family history of sudden cardiac death; the second clinical characteristic predictor includes: NYHA functional class, non-sustained ventricular tachycardia; the third clinical characteristic predictor includes: age, NYHA functional class; and the baseline clinical predictive characteristics include: patient basic information, echocardiographic information, electrocardiographic information, and basic parameters of cardiac magnetic resonance imaging; Based on the LGE images of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, screened radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group are obtained, including: for each patient group in the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, obtaining an LGE image set for each patient group, the LGE image set including LGE images of HCM patients in each patient group; and sorting the LGE images in the LGE image set... The left ventricular myocardial short-axis section is used as the region of interest (ROI). Pre-screening radiomics features of each patient group are extracted from the ROI. The pre-screening radiomics features of each patient group are input into the LASSO (Least Absolute Contraction and Selection) model for correlation analysis to obtain post-screening radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group. Radiomics feature scores are calculated for the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group based on the corresponding post-screening radiomics features. For each patient group among the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: based on the radiomics feature scores and corresponding clinical feature predictors of each patient group, multiple radiomics-clinical feature joint prediction models are established using various machine learning methods; the multiple radiomics-clinical feature joint prediction models are validated, and the optimal radiomics-clinical feature joint prediction model is selected as the HCM patient endpoint event prediction model for each patient group; Acquire a dataset of HCM patients to be tested, wherein the dataset of HCM patients to be tested includes data from multiple HCM patients to be tested; The dataset of HCM patients to be tested is input into the HCM patient endpoint event prediction model of the cardiovascular death patient group to obtain the corresponding SCD risk prediction results.

9. The method according to claim 8, characterized in that, The machine learning method includes any of the following: Logistic regression, K-nearest neighbor method, elastic network method, and Naive Bayes method.

10. A device for establishing a predictive model for endpoint events in HCM patients based on LGE radiomics features and clinical predictive features, characterized in that, The device is installed in a computer and includes: The first acquisition module is configured to acquire HCM patient endpoint event groups and HCM patient control groups. The HCM patient endpoint event groups include cardiovascular death patient groups, heart failure hospitalized patient groups, and secondary endpoint event patient groups. The HCM patient control groups include a first HCM patient control group, a second HCM patient control group, and a third HCM patient control group, which correspond to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively. The second acquisition module is configured to acquire a first clinical feature predictor for the cardiovascular death patient group based on the baseline clinical predictor characteristics of the cardiovascular death patient group and the first HCM patient control group; acquire a second clinical feature predictor for the heart failure hospitalized patient group based on the baseline clinical predictor characteristics of the heart failure hospitalized patient group and the second HCM patient control group; and acquire a third clinical feature predictor for the secondary endpoint event patient group based on the baseline clinical predictor characteristics of the secondary endpoint event patient group and the third HCM patient control group. The first clinical feature predictor includes: left atrial diameter and family history of sudden cardiac death; the second clinical feature predictor includes: NYHA functional class and non-sustained ventricular tachycardia; the third clinical feature predictor includes: age and NYHA functional class; and the baseline clinical predictor includes: patient basic information, echocardiographic information, electrocardiographic information, and basic parameters of cardiac magnetic resonance imaging. The first calculation module is configured to acquire, based on the LGE images of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, filtered radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, including: for each patient group in the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, acquiring an LGE image set for each patient group, the LGE image set including LGE images of HCM patients in each patient group; and processing the LGE images in the LGE image set... The left ventricular myocardial short-axis section of the GE image is used as the region of interest. Pre-screening radiomics features of each patient group are extracted from the region of interest. The pre-screening radiomics features of each patient group are input into the LASSO (Least Absolute Contraction and Selection) model for correlation analysis to obtain post-screening radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group. Radiomics feature scores are calculated for the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group based on the corresponding post-screening radiomics features. The first model selection module is configured to, for each patient group among the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: establish multiple radiomics-clinical feature joint prediction models using various machine learning methods based on the radiomics feature scores and corresponding clinical feature predictors of each patient group; validate the multiple radiomics-clinical feature joint prediction models, and select the optimal radiomics-clinical feature joint prediction model as the HCM patient endpoint event prediction model for each patient group.

11. A device for predicting SCD risk based on an HCM patient endpoint event prediction model, characterized in that, The device is installed in a computer and includes: The third acquisition module is configured to acquire HCM patient endpoint event groups and HCM patient control groups. The HCM patient endpoint event groups include cardiovascular death patient groups, heart failure hospitalized patient groups, and secondary endpoint event patient groups. The HCM patient control groups include a first HCM patient control group, a second HCM patient control group, and a third HCM patient control group, which correspond to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, respectively. The fourth acquisition module is configured to acquire, based on the baseline clinical predictive characteristics of the cardiovascular death patient group and the first HCM patient control group, a first clinical feature predictor factor for the cardiovascular death patient group; based on the baseline clinical predictive characteristics of the heart failure hospitalized patient group and the second HCM patient control group, a second clinical feature predictor factor for the heart failure hospitalized patient group; and based on the baseline clinical predictive characteristics of the secondary endpoint event patient group and the third HCM patient control group, a third clinical feature predictor factor for the secondary endpoint event patient group; wherein the first clinical feature predictor factor includes: left atrial diameter, family history of sudden cardiac death; the second clinical feature predictor factor includes: NYHA functional class, non-sustained ventricular tachycardia; the third clinical feature predictor factor includes: age, NYHA functional class; and the baseline clinical predictive characteristics include: patient basic information, echocardiographic information, electrocardiographic information, and basic parameters of cardiac magnetic resonance imaging. The second calculation module is configured to acquire, based on the LGE images of the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, filtered radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, including: for each patient group in the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group, acquiring an LGE image set for each patient group, the LGE image set including LGE images of HCM patients in each patient group; and processing the LGE images in the LGE image set... The left ventricular myocardial short-axis section of the GE image is used as the region of interest. Pre-screening radiomics features of each patient group are extracted from the region of interest. The pre-screening radiomics features of each patient group are input into the LASSO (Least Absolute Contraction and Selection) model for correlation analysis to obtain post-screening radiomics features corresponding to the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group. Radiomics feature scores are calculated for the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group based on the corresponding post-screening radiomics features. The second model selection module is configured to, for each patient group among the cardiovascular death patient group, the heart failure hospitalized patient group, and the secondary endpoint event patient group: establish multiple radiomics-clinical feature joint prediction models using various machine learning methods based on the radiomics feature scores and corresponding clinical feature predictors of each patient group; validate the multiple radiomics-clinical feature joint prediction models, and select the optimal radiomics-clinical feature joint prediction model as the HCM patient endpoint event prediction model for each patient group; The data acquisition module is configured to acquire a set of HCM patient data to be tested, which includes data on multiple HCM patients to be tested. The SCD prediction module is configured to input the HCM patient data set to be tested into the HCM patient endpoint event prediction model of the cardiovascular death patient group to obtain the corresponding SCD risk prediction results.

12. An electronic device, characterized in that, It includes a memory and a processor; wherein the memory is used to store computer instructions, wherein the computer instructions are executed by the processor to implement the method according to any one of claims 1 to 9.

13. A computer-readable storage medium storing computer instructions thereon, characterized in that, When executed by a processor, the computer instructions implement the method described in any one of claims 1 to 9.

14. A computer program product comprising computer instructions that, when executed by a processor, implement the method of any one of claims 1 to 9.

Images