A cardiovascular health risk prediction method, apparatus, device and medium
By simultaneously acquiring electrocardiogram (ECG) and heart sound signals, constructing phase sequences, and performing feature fusion, the problem of low accuracy in single-modal signal detection is solved, and high-accuracy prediction of cardiovascular health risks is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHANGCHUN UNIV OF SCI & TECH
- Filing Date
- 2026-03-25
- Publication Date
- 2026-07-24
AI Technical Summary
Existing cardiovascular disease detection methods based on single-modal signals have limitations in accuracy, resulting in low accuracy in cardiovascular disease detection.
By synchronously acquiring electrocardiogram (ECG) signal segments and heart sound signal segments, a phase sequence is constructed, and feature encoding and feature fusion are performed. The phase-encoded features are embedded into the ECG-encoded features and heart sound-encoded features to enhance the temporal correlation between ECG features and heart sound features. The complementary information of multimodal signals is used to predict cardiovascular health risks.
It significantly improves the accuracy of cardiovascular health risk prediction, makes full use of the complementary information of electrocardiogram signals and heart sound signals, provides rich multimodal characterization evidence, and improves the accuracy of detection.
Smart Images
Figure CN122440144A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and medium for predicting cardiovascular health risks. Background Technology
[0002] Cardiovascular disease is one of the leading threats to human health worldwide, consistently ranking as the cause of death. With the accelerated pace of modern life, the prevalence of unhealthy lifestyles, and the ongoing aging population, the incidence of cardiovascular disease is increasing year by year, posing a significant challenge to public health systems. Although electrocardiography (ECG) and phonocardiography (PHC) are widely used in clinical practice and have achieved some success in screening and preliminary diagnosis of heart disease, these single-modal signal-based detection methods still have limitations in the precise identification of specific diseases, resulting in low accuracy. Therefore, improving the accuracy of cardiovascular disease detection is an urgent problem to be solved. Summary of the Invention
[0003] In view of this, embodiments of this application provide a method, apparatus, device, and medium for predicting cardiovascular health risks, in order to solve the problem of low detection accuracy in the detection of cardiovascular diseases.
[0004] In a first aspect, embodiments of this application provide a method for predicting cardiovascular health risks, the method comprising: Acquire electrocardiogram (ECG) signal segments and heart sound signal segments synchronously collected from the person to be tested within the corresponding period, and construct a phase sequence of sampling times within the corresponding period based on the ECG signal segments and the heart sound signal segments; The electrocardiogram (ECG) signal segment is feature-encoded to obtain ECG-encoded features; the heart sound signal segment is feature-encoded to obtain heart sound-encoded features; and the phase sequence is feature-encoded to obtain phase-encoded features. The phase encoding features are embedded into the electrocardiogram encoding features and the heart sound encoding features to obtain enhanced electrocardiogram features and enhanced heart sound features; The enhanced electrocardiogram features and the enhanced heart sound features are fused to obtain the fused features; Based on the fused features, cardiovascular health risk is predicted for the individuals to be tested, and cardiovascular health risk prediction results are obtained.
[0005] Secondly, embodiments of this application provide a cardiovascular health risk prediction device, the cardiovascular health risk prediction device comprising: The acquisition module is used to acquire electrocardiogram (ECG) signal segments and heart sound signal segments synchronously collected by the person to be tested within a corresponding period, and to construct a phase sequence of sampling times within the corresponding period based on the ECG signal segments and the heart sound signal segments. The feature encoding module is used to perform feature encoding on the electrocardiogram signal segment to obtain electrocardiogram encoded features, to perform feature encoding on the heart sound signal segment to obtain heart sound encoded features, and to perform feature encoding on the phase sequence to obtain phase encoded features; An embedding module is used to embed the phase coding feature into the electrocardiogram coding feature and the heart sound coding feature to obtain enhanced electrocardiogram features and enhanced heart sound features; The fusion module is used to fuse the enhanced electrocardiogram features and the enhanced heart sound features to obtain the fused features; The prediction module is used to predict the cardiovascular health risk of the person to be tested based on the fused features, and obtain the cardiovascular health risk prediction result.
[0006] Thirdly, embodiments of this application provide a computer device, the computer device including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the cardiovascular health risk prediction method as described in any of the preceding claims.
[0007] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the cardiovascular health risk prediction method as described in any of the preceding claims.
[0008] The advantages of this application compared to the prior art are: In this application, the phase-encoded features are embedded into the electrocardiogram (ECG) encoding features and the heart sound encoding features to obtain enhanced ECG and heart sound features. This achieves temporal alignment enhancement of the phase information for ECG and heart sound signals. This embedding mechanism enables the model to explicitly perceive the relative temporal position relationship between ECG and heart sound signals within the cardiac cycle, strengthening the correlation between different modal signals at the same phase moment. The enhanced ECG and heart sound features are fused along the feature dimension, providing rich multimodal representation data for accurate prediction of cardiovascular health risks. Cardiovascular health risk prediction based on the fused features can fully utilize the complementary information of ECG and heart sound signals, significantly improving the accuracy of cardiovascular health risk prediction. Attached Figure Description
[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0010] Figure 1 This is a flowchart illustrating a cardiovascular health risk prediction method provided in one embodiment of this application. Figure 2 This is a schematic diagram of the structure of a cardiovascular health risk prediction device provided in one embodiment of this application; Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0012] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.
[0013] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.
[0014] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.
[0015] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."
[0016] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0017] References to "one embodiment" or "some embodiments" in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.
[0018] It should be understood that the sequence number of each step in the following embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.
[0019] To illustrate the technical solution of this application, specific embodiments are described below.
[0020] As shown Figure 1 , Figure 1 This is a flowchart illustrating a cardiovascular health risk prediction method provided in one embodiment of this application, as shown below. Figure 1 As shown, the cardiovascular health risk prediction method may include the following steps.
[0021] S101: Obtain the electrocardiogram (ECG) signal segments and heart sound signal segments synchronously collected by the person to be tested within the corresponding period, and construct the phase sequence of each sampling time within the corresponding period based on the ECG signal segments and heart sound signal segments.
[0022] In step S101, the ECG signal segments and heart sound signal segments synchronously acquired within the corresponding period are ECG signals and heart sound signals continuously recorded within the same time period. The phase sequence of the sampling time is a representation of the phase relationship between the ECG signal and the heart sound signal at each sampling time within the corresponding period.
[0023] In this embodiment, electrocardiogram (ECG) signal segments and heart sound signal segments synchronously acquired by the subject within a corresponding period are obtained. This synchronous acquisition can be achieved using a multi-lead ECG acquisition device and a heart sound sensor array. The ECG signal segments contain ECG waveform data for at least one complete cardiac cycle, and the heart sound signal segments contain at least one complete heart sound cycle data aligned with the time of the ECG signal segments. The timing accuracy of the synchronous acquisition should be controlled within milliseconds to ensure a strict correspondence between the two physiological signals in the time dimension. During acquisition, the ECG signal sampling frequency can be set to 500Hz to 1000Hz, and the heart sound signal sampling frequency can be set to 2000Hz to 4000Hz to meet the signal resolution requirements of subsequent phase analysis.
[0024] Based on ECG and heart sound signal segments, a phase sequence of sampling times within a corresponding period is constructed. For any sampling time within a corresponding period, a linear interpolation method is used to calculate its phase value, and the linear length of the corresponding period on the time axis is normalized and mapped to a mathematically periodic manifold [0, 2]. (Space). The formula for calculating the phase value is: in, Let t be the phase value at sampling time t. This represents the end time of the corresponding period. This represents the start time of the corresponding period.
[0025] Optionally, acquire electrocardiogram (ECG) signal segments and heart sound signal segments synchronously collected from the person to be tested within the corresponding period, including: The original electrocardiogram (ECG) signal and original heart sound signal were acquired synchronously from the person to be tested. The original ECG signal and original heart sound signal were divided into segments with the adjacent first heart sound peak as the period, and ECG signal segments and heart sound signal segments within the corresponding period after segmentation were obtained.
[0026] In this embodiment, the original electrocardiogram (ECG) signal and original heart sound signal are acquired synchronously from the person being tested. The ECG and heart sound signals are then segmented using the adjacent first heart sound peaks as the period, resulting in ECG and heart sound signal segments within the corresponding period after segmentation. The first heart sound peak corresponds to the start of cardiac systole, and the time interval between two adjacent first heart sound peaks constitutes a complete cardiac cycle. By detecting the peak position of the first heart sound in the original heart sound signal, the start and end boundaries of each cycle can be determined. This allows the synchronously acquired original ECG and original heart sound signals to be segmented according to the same time window, ensuring that the ECG and heart sound signal segments within each cycle are strictly aligned in time.
[0027] First, when denoising the original heart sound signal, a high-pass filter with a cutoff frequency of 20Hz is used to remove baseline drift and low-frequency noise. Then, multi-scale denoising is performed using wavelet transform to decompose the original heart sound signal into several detail components. By analyzing and reconstructing a specific wavelet sub-band signal containing the main energies of the first and second heart sounds, breath sounds and environmental noise can be effectively filtered out, significantly improving the signal-to-noise ratio of the transient characteristics. A nonlinear energy operator is applied to this sub-band signal to generate an instantaneous energy envelope, which can sensitively reflect the instantaneous rate of change of signal energy. The specific formula is: in, Let t be the energy envelope at sampling time t. For nonlinear Teager energy operators, The subband signal at sampling time t. for Subband signal at the sampling time for The sub-band signal at the sampling time, where t is the sampling time t. for Sampling time, for Sampling time.
[0028] The obtained energy envelope After median filtering and smoothing, an adaptive dynamic threshold algorithm is used for preliminary peak candidate region identification. This threshold can be dynamically adjusted based on the moving mean and standard deviation of the envelope. To make the final determination of the first heart sound, a physiological refractory period constraint is introduced to ensure that no repeated detection is performed within a short period after a peak is detected, thereby eliminating waveform splitting. By analyzing the temporal intervals between heart sound peaks and combining the rule that the systolic period is usually shorter than the diastolic period, the peak before the first short interval is finally determined as the first heart sound peak, realizing automated anchoring of the beat-by-beat cardiac cycle. The times L and R of two adjacent first heart sound peaks are defined as a single heart beat cycle interval [L,R].
[0029] In this embodiment, the adjacent first heart sound peaks of the original heart sound signal are used as period segmentation points to divide the electrocardiogram (ECG) signal and the heart sound signal into several cardiac cycle segments. Signal segmentation is achieved through heart sound cycle anchoring, which effectively overcomes the cycle segmentation error that may occur in the detection of the R-peak of the ECG signal itself under noise interference or arrhythmia, and improves the robustness of cycle positioning.
[0030] S102: Perform feature encoding on the electrocardiogram signal segment to obtain electrocardiogram encoded features, perform feature encoding on the heart sound signal segment to obtain heart sound encoded features, and perform feature encoding on the phase sequence to obtain phase encoded features.
[0031] In step S102, the ECG coding feature characterizes the waveform morphology information and temporal dynamic change law in the ECG signal segment, the heart sound coding feature characterizes the heart sound component structure and hemodynamic information in the heart sound signal segment, and the phase coding feature characterizes the relative position, temporal relationship and periodic structure of each sampling point in the normalized periodic space within the cardiac cycle.
[0032] In this embodiment, the encoder of the cardiovascular health risk prediction model performs feature encoding on electrocardiogram (ECG) signal segments to obtain ECG-coded features, on heart sound signal segments to obtain heart sound-coded features, and on phase sequences to obtain phase-coded features. The cardiovascular health risk prediction model is used to predict cardiovascular health risks. The network structure of the cardiovascular health risk prediction model includes an ECG feature encoder, a heart sound feature encoder, a phase-aware position encoder, a conflict-aware fusion module, and a fully connected layer. The ECG feature encoder encodes features from ECG signal segments to obtain ECG-coded features, while the heart sound feature encoder encodes features from heart sound signal segments to obtain heart sound-coded features. The phase-aware position encoder encodes phase sequences to obtain phase-coded features. The conflict-aware fusion module fuses the ECG-coded features and heart sound-coded features. The fully connected layer performs nonlinear transformations and dimensionality mappings on the fused features, converting high-dimensional abstract features into risk probability outputs.
[0033] It should be noted that the ECG feature encoder, heart sound feature encoder, and phase-aware position encoder are pre-trained encoders. These encoders can be multi-layer self-attention networks based on the Transformer architecture. Specifically, the ECG feature encoder employs a combination of temporal convolutional networks and bidirectional long short-term memory networks to capture subtle waveform variations and temporal dependencies in the ECG signal. The heart sound feature encoder uses a one-dimensional residual convolutional network combined with gated recurrent units. The phase-aware position encoder embeds phase information into a continuous vector space by combining learnable sinusoidal position encoding with relative position bias.
[0034] Optionally, the phase sequence is feature-encoded to obtain phase-encoded features, including: Calculate the sinusoidal and cosine position feature components at each sampling time in the phase sequence; The sine position feature component and the cosine position feature component are concatenated to obtain the phase-coded feature.
[0035] In this embodiment, the sinusoidal position feature component and the cosine position feature component at each sampling time in the phase sequence are calculated using the following formula: Where t is the sampling time, In frequency The sinusoidal positional characteristic components below, In frequency The cosine position feature component below, Let t be the phase value at sampling time t. is the frequency number of the Fourier expansion.
[0036] The sine position feature component and the cosine position feature component are concatenated to obtain the phase-coded feature.
[0037] In this embodiment, by leveraging the complementary properties of sine and cosine functions, the one-dimensional phase value is mapped to a high-dimensional continuous space, thereby enhancing the model's ability to perceive subtle phase differences.
[0038] S103: Embed the phase coding features into the ECG coding features and the heart sound coding features to obtain the enhanced ECG features and the enhanced heart sound features.
[0039] In step S103, the enhanced electrocardiogram (ECG) features and the enhanced heart sound features characterize the temporal correlation between the ECG signal and the heart sound signal under a specific cardiac phase.
[0040] In this embodiment, phase-encoded features are embedded into ECG and heart sound encoding features to obtain enhanced ECG and heart sound features. An adaptive gating mechanism can be used to embed the phase-encoded features into the ECG and heart sound encoding features, or other methods can be used; this embodiment does not impose any limitations.
[0041] Optionally, phase-encoded features are embedded into ECG-encoded features and heart sound-encoded features to obtain enhanced ECG features and enhanced heart sound features, including: The ECG coding features are added to the phase coding features to obtain the enhanced ECG features; The enhanced heart sound features are obtained by adding the heart sound coding features to the phase coding features.
[0042] In this embodiment, the phase-encoded features are linearly projected, and the corresponding dimensions are adjusted to obtain the projected phase-encoded features, making the dimensions of the projected phase-encoded features equal to the dimensions of the ECG-encoded features and the heart-encoded features. The ECG-encoded features are added to the projected phase-encoded features to obtain the enhanced ECG features; the heart-voice-encoded features are added to the projected phase-encoded features to obtain the enhanced heart-voice-encoded features.
[0043] S104: Perform feature fusion on the enhanced electrocardiogram features and the enhanced heart sound features to obtain the fused features.
[0044] In step S104, the fused features characterize the deep correlation between the electrocardiogram signal and the heart sound signal in terms of temporal phase relationship.
[0045] In this embodiment, an adaptive fusion strategy based on an attention mechanism is adopted. The enhanced ECG features and the enhanced heart sound features are concatenated along the feature dimension to form a joint feature vector. The attention weights of each modality feature are calculated by a multilayer perceptron. After normalization, the attention weights are used to perform weighted fusion of the enhanced ECG features and the enhanced heart sound features to obtain the fused features.
[0046] Optionally, the enhanced electrocardiogram features and the enhanced heart sound features are fused to obtain fused features, including: Determine the conflict-complementary gating coefficients between electrocardiogram coding features and heart sound coding features; Based on the conflict complementarity gating coefficient, the enhanced electrocardiogram features and the enhanced heart sound features are weighted and fused to obtain the fused features.
[0047] In this embodiment, conflict-complementary gating coefficients are determined for ECG coding features and heart sound coding features. These coefficients characterize the dynamic interaction between ECG and heart sound features at the information level. Based on these coefficients, the enhanced ECG and heart sound features are weighted and fused to obtain the fused features. The calculation formula is as follows: in, The characteristics after fusion For conflict-complementary gating coefficients, To enhance the electrocardiogram characteristics, This is an enhanced representation of heart sounds.
[0048] It should be noted that, in order to extract multi-level temporal features of ECG and heart sounds, the cardiovascular health risk prediction model includes a multi-stage ECG feature encoder, a heart sound feature encoder, a phase-aware position encoder, and a conflict-aware fusion module, such as four stages. Each stage contains an ECG feature encoder, a heart sound feature encoder, a phase-aware position encoder, and a conflict-aware fusion module. The ECG feature encoder and the heart sound feature encoder operate in parallel. The ECG feature encoder is used to encode features in ECG signal segments to obtain ECG-coded features, and the heart sound feature encoder is used to encode features in heart sound signal segments to obtain heart sound-coded features. In the first stage, to adapt to the long-term waveform features of ECG and heart sounds, both the ECG feature encoder and the heart sound feature encoder uniformly use a receptive field convolution kernel of size 7 and set the basic number of channels to 64. During feature extraction, the first stage maintains the original temporal resolution to capture microscopic morphology. In the subsequent three stages, both the ECG feature encoder and the heart sound feature encoder used a stride convolution with a stride of 5 for progressive downsampling, compressing the time dimension from 2000 to 16. At the same time, the number of feature channels increased progressively in a linear fashion according to 64, 128, 192, and 256. In order to compensate for the information loss caused by downsampling, a multi-scale original signal injection mechanism was introduced in the third and fourth stages. ECG signal segments and heart sound signal segments that had been averaged by 5 times and 25 times, respectively, were concatenated with the current layer feature map, achieving complete feature coverage from waveform details to global semantics.
[0049] After obtaining the fused features at each stage, the fused features are downsampled and aligned, and then used as historical prior information to be input into the conflict-aware fusion module of the next stage, realizing cross-stage feature fusion until the last stage, to obtain the final fused features.
[0050] Optionally, determining the conflict-complementary gating coefficients between ECG coding features and heart sound coding features includes: Calculate the conflict modulation factor between electrocardiogram coding features and heart sound coding features; The phase sequence is convolved to obtain the convolved phase sequence; The conflict complementarity gating coefficients are calculated based on the electrocardiogram coding features, heart sound coding features, conflict regulation factor, and the phase sequence after convolution.
[0051] In this embodiment, when determining the conflict complementarity gating coefficient between ECG-coded features and heart sound-coded features, a conflict adjustment factor between the two features is calculated. This conflict adjustment factor represents the semantic consistency between the ECG-coded features and the heart sound-coded features. The calculation formula is as follows: in, As a conflict moderating factor, These are ECG coding features. Encoding features for heart sounds, To encode ECG features The result after projecting onto the common subspace and performing L2 normalization. To encode heart sound features The result after projecting onto the common subspace and performing L2 normalization. This is a temperature coefficient used to scale similarity values.
[0052] The temporal gate is calculated by convolution from the concatenated bimodal features, and the channel gate is generated by passing the global average pooling vector of the bimodal features through a fully connected layer. The calculation formula is as follows: in, For time gates, The representation is the result of concatenating ECG-coded features with heart sound-coded features. for Convolution operation, i.e., on conduct Convolution operation, For passageway doors, ECG coding features The global average pooling vector, Encoding features for heart sounds The global average pooling vector, It is a fully connected layer, that is, a layer after splicing. Perform nonlinear transformation.
[0053] The phase sequence is convolved to obtain the convolved phase sequence. Based on the ECG coding features, heart sound coding features, conflict modulation factor, and the convolved phase sequence, the conflict complementarity gating coefficient is calculated using the following formula: in, For conflict-complementary gating coefficients, For the Sigmoid function, For passageway doors, For time gates, As a conflict moderating factor, This is the conflict adjustment coefficient. The phase sequence after convolution. This is the phase offset coefficient.
[0054] S105: Based on the fused features, perform cardiovascular health risk prediction on the individuals to be tested, and obtain the cardiovascular health risk prediction results.
[0055] In step S105, the cardiovascular health risk prediction results include healthy and unhealthy.
[0056] In this embodiment, the fused features are input into the fully connected layer of the cardiovascular health risk prediction model. The fully connected layer performs a nonlinear transformation on the fused features to extract higher-order semantic feature representations. A Softmax classification layer is then used to perform probability normalization on these higher-order semantic feature representations, calculating the probability values of whether the person being tested belongs to the healthy or unhealthy category. The probability values of the healthy and unhealthy categories are compared, and the category with the higher probability value is used as the cardiovascular health risk prediction result for the person being tested.
[0057] It should be noted that the cardiovascular health risk prediction model needs to be trained before use to obtain a well-trained model. The network structure of the cardiovascular health risk prediction model includes an electrocardiogram (ECG) feature encoder, a heart sound feature encoder, a phase-aware position encoder, a conflict-aware fusion module, and a fully connected layer. The ECG and heart sound feature encoders operate in parallel. The ECG feature encoder encodes ECG signal segments to obtain ECG-coded features, and the heart sound feature encoder encodes heart sound signal segments to obtain heart sound-coded features. The phase-aware position encoder encodes phase sequences to obtain phase-coded features. The conflict-aware fusion module fuses the ECG and heart sound-coded features. The fully connected layer performs nonlinear transformations and dimensional mappings on the fused features, converting high-dimensional abstract features into risk probability outputs.
[0058] During training, synchronous raw electrocardiogram (ECG) signals and raw heart sound signals were collected from patients with early-stage and diagnosed cardiovascular diseases, as well as from healthy subjects, to construct a multimodal dataset. A diverse age-distributed subject group was selected, including a normal control group, a high-risk cardiovascular disease group, and a group diagnosed with cardiovascular disease, to ensure the diversity and balance of the dataset. A uniform sampling rate of 500Hz was used throughout the data collection process. The normal control group consisted of individuals with no history of cardiovascular disease, no clinical symptoms, and no abnormalities found in basic clinical examinations. The high-risk cardiovascular disease group consisted of individuals who had not yet experienced organic cardiovascular events but possessed clear cardiovascular risk factors, or whose clinical examinations showed early changes in cardiac function but did not meet the diagnostic criteria. The group diagnosed with cardiovascular disease consisted of patients diagnosed with mid-to-late stage coronary heart disease, arrhythmia, or heart failure according to comprehensive clinical diagnostic criteria.
[0059] In this embodiment, a five-fold cross-validation strategy is used to conduct end-to-end training of the cardiovascular health risk prediction model. According to the preset configuration, the dataset is randomly divided into five mutually exclusive subsets, and the experiment is executed in five rounds. In each round of the experiment, one subset is selected as the validation set, and the other four are selected as the training set. This ensures that every sample participates in model validation, eliminates the randomness bias of a single partition, and more objectively evaluates the generalization ability of the cardiovascular health risk prediction model to capture cardiovascular disease characteristics.
[0060] During training, the Adam optimizer is used to update parameters. The initial learning rate is set to 0.001, the maximum number of iterations is 100, and the batch size of the training set is set to 16 to ensure the stability of gradient descent. The batch size of the test set is set to 64.
[0061] During training, the original ECG and original heart sound signals from each dataset are segmented with the peak value of the adjacent first heart sound as the period. This results in ECG and heart sound signal segments within the corresponding period after segmentation. A phase sequence within the corresponding period is then constructed. The ECG, heart sound, and phase sequences are combined into a triplet input to the model for supervised training of the cardiovascular health risk prediction model, resulting in a well-trained cardiovascular health risk prediction model.
[0062] In this application, phase-encoded features are embedded into ECG and heart sound encoding features to obtain enhanced ECG and heart sound features. This achieves temporal alignment enhancement of phase information for ECG and heart sound signals. This embedding mechanism enables the model to explicitly perceive the relative temporal position relationship between ECG and heart sound signals within the cardiac cycle, strengthening the correlation between different modal signals at the same phase moment. The enhanced ECG and heart sound features are fused along the feature dimension, providing rich multimodal representation for accurate prediction of cardiovascular health risks. Cardiovascular health risk prediction based on the fused features can fully utilize the complementary information of ECG and heart sound signals, significantly improving the accuracy of cardiovascular health risk prediction.
[0063] Please see Figure 2 , Figure 2 This is a schematic diagram of a cardiovascular health risk prediction device according to an embodiment of this application. This cardiovascular health risk prediction device corresponds one-to-one with the cardiovascular health risk prediction method described in the above embodiments. Please refer to [link / reference] for details. Figure 2 as well as Figure 2 The relevant descriptions in the corresponding embodiments are shown below. For ease of explanation, only the parts relevant to this embodiment are shown. See also... Figure 2 The cardiovascular health risk prediction device 200 includes: an acquisition module 201, a feature encoding module 202, an embedding module 203, a fusion module 204, and a prediction module 205.
[0064] The acquisition module 201 is used to acquire the electrocardiogram signal segments and heart sound signal segments synchronously collected by the person to be tested within the corresponding period, and to construct the phase sequence of the sampling time within the corresponding period based on the electrocardiogram signal segments and heart sound signal segments.
[0065] The feature encoding module 202 is used to perform feature encoding on electrocardiogram signal segments to obtain electrocardiogram encoded features, to perform feature encoding on heart sound signal segments to obtain heart sound encoded features, and to perform feature encoding on phase sequences to obtain phase encoded features.
[0066] The embedding module 203 is used to embed the phase coding features into the electrocardiogram coding features and the heart sound coding features to obtain enhanced electrocardiogram features and enhanced heart sound features.
[0067] The fusion module 204 is used to fuse the enhanced electrocardiogram features and the enhanced heart sound features to obtain the fused features.
[0068] The prediction module 205 is used to predict the cardiovascular health risk of the person to be tested based on the fused features, and obtain the cardiovascular health risk prediction result.
[0069] Optionally, the acquisition module 201 includes: The segmentation unit is used to acquire the original electrocardiogram (ECG) signal and original heart sound signal synchronously collected from the person to be tested. The original ECG signal and original heart sound signal are segmented with the adjacent first heart sound peak as the period to obtain ECG signal segments and heart sound signal segments within the corresponding period after segmentation.
[0070] Optionally, the feature encoding module 202 includes: The calculation unit is used to calculate the sinusoidal position feature components and cosine position feature components at each sampling time in the phase sequence.
[0071] The splicing unit is used to splice the sinusoidal position feature components with the cosine position feature components to obtain the phase-coded features.
[0072] Optionally, the above-mentioned embedded module 203 includes: The first enhancement unit is used to add the ECG coding features and the phase coding features to obtain the enhanced ECG features.
[0073] The second enhancement unit is used to add the heart sound coding features and the phase coding features to obtain the enhanced heart sound features.
[0074] Optionally, the fusion module 204 includes: The determination unit is used to determine the conflict-complementary gating coefficients between electrocardiogram coding features and heart sound coding features.
[0075] The fusion unit is used to perform weighted fusion of the enhanced electrocardiogram features and the enhanced heart sound features according to the conflict complementarity gating coefficient, so as to obtain the fused features.
[0076] Optionally, the determining unit mentioned above includes: The first computational subunit is used to calculate the conflict regulation factor between electrocardiogram coding features and heart sound coding features.
[0077] The convolution sub-unit is used to perform convolution processing on the phase sequence to obtain the convolved phase sequence.
[0078] The second computational subunit is used to calculate the conflict complementary gating coefficient based on the electrocardiogram coding features, heart sound coding features, conflict regulation factor and convolutional phase sequence.
[0079] It should be noted that the information interaction and execution process between the above-mentioned units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.
[0080] Figure 3 This is a schematic diagram of the structure of a computer device provided in one embodiment of this application. For example... Figure 3As shown, the computer device of this embodiment includes: at least one processor ( Figure 3 Only one is shown in the diagram), a memory, and a computer program stored in the memory and executable on at least one processor, which, when executed by the processor, implements the steps in any of the above embodiments of the cardiovascular health risk prediction methods.
[0081] This computer device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that... Figure 3 The examples of computer devices are merely examples and do not constitute a limitation on computer devices. Computer devices may include more or fewer components than shown in the illustration, or combinations of certain components, or different components, such as network interfaces, displays, and input devices.
[0082] The processor referred to can be a CPU, but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.
[0083] Memory includes readable storage media, internal memory, etc., wherein internal memory can be the RAM of a computer device, providing an environment for the operation of the operating system and computer-readable instructions stored in the readable storage media. The readable storage media can be the hard drive of a computer device, or in other embodiments, it can be an external storage device of the computer device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, memory can include both internal storage units and external storage devices of the computer device. Memory is used to store the operating system, applications, bootloader, data, and other programs, such as program code for computer programs. Memory can also be used to temporarily store data that has been output or will be output.
[0084] Those skilled in the art will understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the functions described above can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above device can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here. If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the above method embodiments. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable medium can include at least: any entity or device capable of carrying computer program code, a recording medium, a computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.
[0085] The implementation of all or part of the processes in the methods of the above embodiments can also be accomplished by a computer program product. When the computer program product is run on a computer device, it causes the computer device to execute the steps in the above method embodiments.
[0086] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0087] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0088] In the embodiments provided in this application, it should be understood that the disclosed apparatus / computer devices and methods can be implemented in other ways. For example, the apparatus / computer device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.
[0089] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0090] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.
Claims
1. A method for predicting cardiovascular health risks, characterized in that, The cardiovascular health risk prediction method includes: Acquire electrocardiogram (ECG) signal segments and heart sound signal segments synchronously collected from the person to be tested within the corresponding period, and construct a phase sequence of sampling times within the corresponding period based on the ECG signal segments and the heart sound signal segments; The electrocardiogram (ECG) signal segment is feature-encoded to obtain ECG-encoded features; the heart sound signal segment is feature-encoded to obtain heart sound-encoded features; and the phase sequence is feature-encoded to obtain phase-encoded features. The phase encoding features are embedded into the electrocardiogram encoding features and the heart sound encoding features to obtain enhanced electrocardiogram features and enhanced heart sound features; The enhanced electrocardiogram features and the enhanced heart sound features are fused to obtain the fused features; Based on the fused features, cardiovascular health risk is predicted for the individuals to be tested, and cardiovascular health risk prediction results are obtained.
2. The cardiovascular health risk prediction method as described in claim 1, characterized in that, The acquisition of electrocardiogram (ECG) signal segments and heart sound signal segments synchronously collected from the person to be tested within the corresponding period includes: The original electrocardiogram (ECG) signal and original heart sound signal of the person to be tested are acquired synchronously. The original ECG signal and the original heart sound signal are divided into segments with the adjacent first heart sound peak as the period, so as to obtain ECG signal segments and heart sound signal segments within the corresponding period after segmentation.
3. The cardiovascular health risk prediction method as described in claim 1, characterized in that, The step of feature encoding the phase sequence to obtain phase-encoded features includes: Calculate the sinusoidal and cosine position feature components at each sampling moment in the phase sequence; The sinusoidal position feature component and the cosine position feature component are concatenated to obtain the phase-coded feature.
4. The cardiovascular health risk prediction method as described in claim 1, characterized in that, The step of embedding the phase-encoded features into the electrocardiogram (ECG)-encoded features and the heart sound-encoded features to obtain enhanced ECG features and enhanced heart sound features includes: The ECG coding features are added to the phase coding features to obtain the enhanced ECG features; The enhanced heart sound features are obtained by adding the heart sound coding features to the phase coding features.
5. The cardiovascular health risk prediction method as described in claim 1, characterized in that, The feature fusion of the enhanced electrocardiogram features and the enhanced heart sound features to obtain the fused features includes: Determine the conflict-complementary gating coefficients between the electrocardiogram coding features and the heart sound coding features; Based on the conflict complementarity gating coefficient, the enhanced electrocardiogram features and the enhanced heart sound features are weighted and fused to obtain the fused features.
6. The cardiovascular health risk prediction method as described in claim 5, characterized in that, The determination of the conflict-complementary gating coefficients between the electrocardiogram coding features and the heart sound coding features includes: Calculate the conflict modulation factor between the electrocardiogram coding features and the heart sound coding features; The phase sequence is convolved to obtain the convolved phase sequence; The conflict complementary gating coefficients are calculated based on the electrocardiogram coding features, the heart sound coding features, the conflict modulation factor, and the convolutional phase sequence.
7. A cardiovascular health risk prediction device, characterized in that, The cardiovascular health risk prediction device includes: The acquisition module is used to acquire electrocardiogram (ECG) signal segments and heart sound signal segments synchronously collected by the person to be tested within a corresponding period, and to construct a phase sequence of sampling times within the corresponding period based on the ECG signal segments and the heart sound signal segments. The feature encoding module is used to perform feature encoding on the electrocardiogram signal segment to obtain electrocardiogram encoded features, to perform feature encoding on the heart sound signal segment to obtain heart sound encoded features, and to perform feature encoding on the phase sequence to obtain phase encoded features; An embedding module is used to embed the phase coding feature into the electrocardiogram coding feature and the heart sound coding feature to obtain enhanced electrocardiogram features and enhanced heart sound features; The fusion module is used to fuse the enhanced electrocardiogram features and the enhanced heart sound features to obtain the fused features; The prediction module is used to predict the cardiovascular health risk of the person to be tested based on the fused features, and obtain the cardiovascular health risk prediction result.
8. The cardiovascular health risk prediction device as described in claim 7, characterized in that, The acquisition module includes: The segmentation unit is used to acquire the original electrocardiogram (ECG) signal and the original heart sound signal synchronously collected by the person to be tested, and to segment the ECG signal and the heart sound signal with the adjacent first heart sound peak as the period to obtain ECG signal segments and heart sound signal segments within the corresponding period after segmentation.
9. A computer device, characterized in that, The computer device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the cardiovascular health risk prediction method as described in any one of claims 1 to 6.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the cardiovascular health risk prediction method as described in any one of claims 1 to 6.