Method for predicting heart failure based on left ventricular ejection fraction

Abstract

The invention discloses a method for predicting heart failure based on left ventricular ejection fraction, which develops and trains a primitive model to automatically provide help and evaluate the level of clinical decision based on LVEF heart failure. The training model is prevented from using any or biased clinical variables, and the following two steps are ensured: investigating the statistical significance of each variable to distinguish the three categories; secondly, a novel dimension reduction technology is adopted to visually observe the characterization of the optimal variable in the radial direction, and each LVEF-based HF category is separated; and on the basis, the performance of the developed model is trained, the importance of the most important clinical variables is discussed, and the significance of the clinical variables is explained in detail based on the application of LVEF-based deep learning in HF analysis, so that patient data of three heart failure types are distinguished.

Description

Technical Field

[0001] This invention belongs to the field of pathological assessment of heart failure, and specifically relates to a method for predicting heart failure based on left ventricular ejection fraction. Background Technology

[0002] Heart failure (HF) is a chronic, progressive pathological condition characterized by the heart's inability to pump enough blood through the systemic circulation to nourish the tissues. Several conditions, such as coronary artery disease (CAD) and hypertension, are considered major causes of HF progression. According to the European Society of Cardiology (ESC), more than 26 million people worldwide suffer from HF caused by coronary artery disease. Furthermore, the World Health Organization (WHO) estimates that more than 7.2 million people worldwide die from HF each year.

[0003] Based on left ventricular ejection fraction (LVEF), heart failure patients exhibit significantly reduced cardiac systolic function. LVEF refers to the amount of oxygenated blood pumped from the left ventricle with each heart contraction. It is considered an important diagnostic indicator for assessing the progression of heart failure, especially in its early stages. Based on LVEF, heart failure can be classified into three main categories according to American echocardiography and the European Society for Cardiovascular Imaging (ASE / EACVI): heart failure with preserved ejection fraction (HFpEF) exceeding 55%, heart failure with intermediate ejection fraction (HFmEF) between 50% and 55%, and heart failure with reduced ejection fraction (HFrEF) below 50%. The narrower range of the HFmEF category is considered a variable criterion depending on its etiology. Other guidelines, including the ESC, recommend different cutoff values ​​for HF classification, with the cutoff value for HFrEF as low as 40%. Literature indicates that there are no strict rules, and treatment is loosely correlated with LVEF and clinical presentation.

[0004] Therefore, accurate assessment of heart failure based on LVEF presents a significant challenge for clinicians. Although HFpEF covers half of HF patients, it is not well understood due to its similarity in symptoms and adverse outcomes to HFrEF and HFmEF, and is often difficult to detect. Furthermore, HFmEF represents one-fifth of the HF population and remains unclear because its pathogenesis has been observed to be more similar to HFrEF than to HFpEF, which differs from the guidelines applied. This raises the question of whether it should be considered a transient entity between HFpEF and HFrEF, or a distinct entity. Therefore, further research is needed to investigate the effectiveness of LVEF-based classification of heart failure patients. According to recently published studies, patient clinical profiles allow for differentiation among the three HF categories, particularly based on the presence of comorbidities and quality of life according to ESC guidelines. Based on these clinical results, we found that HFmEF patients fall between HFpEF and HFrEF, but are closer to HFpEF. Additionally, they are younger than those with HFpEF and are more likely to have diabetes and hypertension than those with HFrEF. Therefore, further research on larger patient cohorts is still needed to understand how demographic and clinical characteristics relate to each heart failure category defined by clinically measured ejection fraction, particularly in optimizing treatment regimens to improve patient stratification and risk management.

[0005] In the present technology, echocardiographic LVEF assessment tools require highly specialized knowledge and expensive equipment. Furthermore, knowledge regarding the complex relationship between demographics, clinical information, and the three LVEF-based heart failure categories remains limited. Summary of the Invention

[0006] This invention aims to provide a method for predicting heart failure based on left ventricular ejection fraction (LVEF). It develops a well-trained model that can automatically assist in assessing the level of clinical decision-making in LVEF-based heart failure. To prevent the training model from using arbitrary or biased clinical variables, the following two steps are ensured: first, investigating the statistical significance of each variable to distinguish the three categories; second, employing a novel dimensionality reduction technique to radially visualize the representation of the optimal variables and separate each LVEF-based heart failure (HF) category. Based on this, the developed model is trained to improve performance, exploring the importance of the most clinically significant variables and the application of LVEF-based deep learning in HF analysis, detailing the significance of these clinical variables in distinguishing patient data from the three types of heart failure.

[0007] To solve the above technical problems, the following technical solution is adopted:

[0008] A method for predicting heart failure based on left ventricular ejection fraction, characterized by the following steps:

[0009] (1) Input existing patient case samples and clinical information;

[0010] (2) Statistical analysis results and feature selection: determine the most important markers for maximizing performance in the proposed deep learning model;

[0011] (3) Construct deep learning network models with convolutional neural networks for regression and classification;

[0012] (4) Develop deep learning network models.

[0013] After optimization, the number of existing patients in step (1) is 150-300.

[0014] After optimization, step (3) involves using two convolutional layers, each of which is batch normalized and corrected linearly combined to extract contaminated features from the clinical markers of patients in each LVEF category.

[0015] After optimization, the kernel sizes of the two convolutional layers are (1, 3) and (1, 2), respectively.

[0016] After optimization, the two convolutional layers have 32 filters and 64 filters, respectively.

[0017] After optimization, step (4) includes:

[0018] a. During the training phase, the model was trained for 300 epochs with a mini-batch size of 64. The adaptive moment estimate was selected as the optimizer, with a learning rate of 0.001, a regularization rate of 0.0001, and a decay rate of 0.90.

[0019] b. In the prediction phase, a leave-two scheme was adopted, in which each subject was retained as a test subject in each training iteration.

[0020] The above technical solution has the following beneficial effects.

[0021] This invention presents a method for predicting heart failure based on left ventricular ejection fraction (LVEF). A well-trained model can be developed to automatically assist in assessing the effectiveness of clinical decision-making in LVEF-based heart failure. To prevent the training model from using arbitrary or biased clinical variables, the following two steps are ensured: first, investigating the statistical significance of each variable to distinguish the three categories; second, employing a novel dimensionality reduction technique to radially visualize the representation of the optimal variables and separate each LVEF-based heart failure (HF) category. Based on this, the developed model is trained to assess its performance, exploring the importance of the most clinically significant variables and the application of LVEF-based deep learning in HF analysis, detailing the significance of these clinical variables in patient data that distinguishes the three types of heart failure.

[0022] This invention provides a prediction for each topic in the dataset while maximizing the amount of data included in the training model. It allows each patient to be treated as a fully hidden test set during model training, resolving any issues regarding generality in both the training and prediction phases. Attached Figure Description

[0023] The present invention will be further described below with reference to the accompanying drawings:

[0024] Figure 1 This is a flowchart of the prediction method of the present invention. Detailed Implementation

[0025] This invention aims to provide a method for predicting heart failure based on left ventricular ejection fraction (LVEF). It develops a well-trained model that can automatically assist in assessing the level of clinical decision-making in LVEF-based heart failure. To prevent the training model from using arbitrary or biased clinical variables, the following two steps are ensured: first, investigating the statistical significance of each variable to distinguish the three categories; second, employing a novel dimensionality reduction technique to radially visualize the representation of the optimal variables and separate each LVEF-based heart failure (HF) category. Based on this, the developed model is trained to improve performance, exploring the importance of the most clinically significant variables and the application of LVEF-based deep learning in HF analysis, detailing the significance of these clinical variables in distinguishing patient data from the three types of heart failure.

[0026] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments:

[0027] A method for predicting heart failure based on left ventricular ejection fraction includes the following steps:

[0028] (1) Input case samples from 200 existing patients ( Figure 1 A) and clinical information ( Figure 1 B);

[0029] (2) Statistical analysis results and feature selection 以帮助 Identify the most important markers for maximizing performance in the proposed deep learning model. Figure 1 C);

[0030] (3) A deep learning network model with a convolutional neural network (CNN) was constructed for regression and classification. Two convolutional layers were used, each of which was subjected to batch normalization (BN) and calibrated linear joint (ReLU) to extract contaminated features from patient clinical markers for each LVEF category. The kernel sizes of the two convolutional layers were (1, 3) and (1, 2), respectively, with 32 and 64 filters.

[0031] (4) Develop deep learning network models, including the training and prediction phases. Figure 1E)

[0032] a. During the training phase, the model was trained for 300 epochs with a mini-batch size of 64. Adaptive Moment Estimation (ADAM) was selected as the optimizer with a learning rate of 0.001, a regularization rate of 0.0001, and a decay rate of 0.90.

[0033] b. In the prediction phase, a leave-two scheme was adopted, that is, each subject was retained as a test subject in each training iteration.

[0034] The above are merely specific embodiments of the present invention, but the technical features of the present invention are not limited thereto. Any simple changes, equivalent substitutions, or modifications made based on the present invention to solve essentially the same technical problems and achieve essentially the same technical effects are all covered within the protection scope of the present invention.

Claims

1. A method for predicting heart failure based on left ventricular ejection fraction, characterized in that... Includes the following steps: (1) Input existing patient case samples and clinical information; (2) Statistical analysis results and feature selection: determine the most important markers for maximizing performance in the proposed deep learning model; (3) Construct deep learning network models with convolutional neural networks for regression and classification; (4) Develop deep learning network models.

2. The method for predicting heart failure based on left ventricular ejection fraction according to claim 1, characterized in that: In step (1), the number of existing patients is 150-300.

3. The method for predicting heart failure based on left ventricular ejection fraction according to claim 1, characterized in that: Step (3): Two convolutional layers are used, and batch normalization and corrected linear combination are performed on each layer to extract contaminated features from the clinical markers of patients in each LVEF category.

4. The method for predicting heart failure based on left ventricular ejection fraction according to claim 3, characterized in that: The kernel sizes of the two convolutional layers are (1, 3) and (1, 2), respectively.

5. The method for predicting heart failure based on left ventricular ejection fraction according to claim 4, characterized in that: The two convolutional layers have 32 filters and 64 filters, respectively.

6. The method for predicting heart failure based on left ventricular ejection fraction according to claim 1, characterized in that: Step (4) includes: a. During the training phase, the model was trained for 300 epochs with a mini-batch size of 64. The adaptive moment estimate was selected as the optimizer, with a learning rate of 0.001, a regularization rate of 0.0001, and a decay rate of 0.

90. b. In the prediction phase, a leave-two scheme was adopted, in which each subject was retained as a test subject in each training iteration.

Images