System, device and medium for cardiovascular disease recognition based on electrocardiogram fusion
By preprocessing, segmenting, and dynamically modeling ECG signals through an ECG fusion recognition system, a non-threshold recursive graph is generated. Combined with neural networks, this enables automated diagnosis of cardiovascular diseases, solving the problems of subjective dependence and insufficient information in existing methods and improving diagnostic accuracy and comprehensiveness.
Patent Information
- Application Number
- CN202511086074.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-05
- Publication Date
- 2025-11-18
- Estimated Expiration
- 2045-08-05
AI Technical Summary
Existing ECG analysis methods rely on manual interpretation, are easily affected by subjective factors, and are difficult to fully capture subtle abnormalities in ECG signals. They also ignore the intrinsic connection between the time domain and the dynamic domain, which limits the accuracy and comprehensiveness of cardiovascular disease diagnosis.
A cardiovascular disease identification system based on electrocardiogram fusion is adopted. The system performs preprocessing through an acquisition module, signal segmentation through a segmentation module, dynamic analysis through a dynamic modeling module, and non-threshold recursive graph generation through a fusion recursion module. Finally, a neural network is used for disease identification.
It enables automated diagnosis of cardiovascular diseases, improves the ability to characterize subtle abnormalities in electrocardiogram signals, enhances diagnostic performance, and avoids the problem of inappropriate threshold selection.
Smart Images
Figure CN120570618B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of cardiovascular disease identification technology, and in particular to cardiovascular disease identification systems, devices and media based on electrocardiogram fusion. Background Technology
[0002] Cardiovascular disease has become one of the leading causes of death worldwide, and its early and accurate diagnosis is crucial for improving patient outcomes. Electrocardiography (ECG), as a non-invasive and readily available diagnostic tool, plays a vital role in the diagnosis of cardiovascular disease. However, clinical ECG analysis methods rely on manual interpretation by physicians, which is susceptible to subjective factors and has low efficiency, limiting the accuracy and consistency of diagnosis.
[0003] In recent years, with the rapid development of artificial intelligence technology, ECG-based automated diagnostic methods have gradually become a research hotspot. Existing automated diagnostic methods mainly focus on traditional time-domain signal processing of raw ECG signals. However, in the early stages of disease, ECG may only show subtle changes, or even appear as a seemingly normal ECG, making it difficult for methods relying solely on traditional ECG feature extraction to comprehensively capture disease-related minute abnormalities. Furthermore, existing methods heavily rely on human experience in feature engineering, and the quality of feature selection directly affects diagnostic performance. More importantly, existing technologies often neglect the intrinsic connection between the time and dynamic domains of ECG signals, as well as the interaction between depolarization and repolarization processes, resulting in limitations in the comprehensiveness and depth of diagnosis. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention provides a cardiovascular disease identification system, device, and medium based on electrocardiogram (ECG) fusion; this invention provides a more comprehensive and in-depth analysis of ECG signals, offering an effective new paradigm for cardiovascular disease diagnosis.
[0005] On the one hand, a cardiovascular disease identification system based on electrocardiogram fusion is provided, including:
[0006] The acquisition module is configured to: acquire the raw ECG time-domain signal, preprocess the raw ECG time-domain signal, and obtain the preprocessed ECG time-domain signal;
[0007] The segmentation module is configured to segment the preprocessed ECG time-domain signal to obtain the ECG depolarization time-domain signal and the ECG repolarization time-domain signal.
[0008] The dynamic modeling module is configured to: use deterministic learning theory to perform dynamic modeling on the ECG depolarization time-domain signal to obtain the ECG depolarization dynamic domain signal; and perform dynamic modeling on the ECG repolarization time-domain signal to obtain the ECG repolarization dynamic domain signal.
[0009] The fusion recursive module is configured to: determine the non-threshold recursive graph in the depolarization time domain based on the Euclidean distance between any two states in the ECG depolarization time domain signal; determine the non-threshold recursive graph in the depolarization dynamic domain based on the Euclidean distance between any two states in the ECG depolarization dynamic domain signal; determine the non-threshold recursive graph in the repolarization time domain based on the Euclidean distance between any two states in the ECG repolarization time domain signal; and determine the non-threshold recursive graph in the repolarization dynamic domain based on the Euclidean distance between any two states in the ECG repolarization dynamic domain signal.
[0010] Based on the non-threshold recursive graphs in the depolarization time domain, dynamic domain, complex polarization time domain, and complex polarization dynamic domain, a joint non-threshold recursive graph is determined; based on the upper triangular matrix and lower triangular matrix of the joint non-threshold recursive graph, a fused recursive graph is determined.
[0011] The disease identification module is configured to input the fused recursive graph into the trained neural network model to obtain the identification results of cardiovascular disease categories.
[0012] Furthermore, an electronic device is also provided, comprising: a memory for non-transitory storage of computer-readable instructions; and a processor for executing the computer-readable instructions, wherein, when executed by the processor, the computer-readable instructions perform the following steps: acquiring a raw electrocardiogram (ECG) time-domain signal; preprocessing the raw ECG time-domain signal to obtain a preprocessed ECG time-domain signal; segmenting the preprocessed ECG time-domain signal to obtain an ECG depolarization time-domain signal and an ECG repolarization time-domain signal; using deterministic learning theory to perform dynamic modeling on the ECG depolarization time-domain signal to obtain an ECG depolarization dynamic domain signal; and performing dynamic modeling on the ECG repolarization time-domain signal to obtain an ECG repolarization dynamic domain signal; and determining, based on the Euclidean distance between any two time-states of the ECG depolarization time-domain signal, the following steps are performed: The depolarization time-domain non-threshold recursive graph is determined; based on the Euclidean distance between any two states in the ECG depolarization dynamic domain signal, the repolarization time-domain non-threshold recursive graph is determined; based on the Euclidean distance between any two states in the ECG repolarization time-domain signal, the repolarization dynamic domain non-threshold recursive graph is determined; based on the depolarization time-domain non-threshold recursive graph, the depolarization dynamic domain non-threshold recursive graph, the repolarization time-domain non-threshold recursive graph, and the repolarization dynamic domain non-threshold recursive graph, a non-threshold joint recursive graph is determined; based on the upper triangular matrix and lower triangular matrix of the non-threshold joint recursive graph, a fusion recursive graph is determined; the fusion recursive graph is input into the trained neural network model to obtain the identification results of cardiovascular disease categories.
[0013] Furthermore, a storage medium is also provided for non-transitory storage of computer-readable instructions, wherein when the non-transitory computer-readable instructions are executed by a computer, the following steps are performed:
[0014] The raw ECG time-domain signal is acquired, and the raw ECG time-domain signal is preprocessed to obtain the preprocessed ECG time-domain signal.
[0015] The preprocessed ECG time-domain signal is segmented to obtain the ECG depolarization time-domain signal and the ECG repolarization time-domain signal;
[0016] Using deterministic learning theory, dynamic modeling is performed on the time-domain signal of ECG depolarization to obtain the dynamic domain signal of ECG depolarization; and dynamic modeling is also performed on the time-domain signal of ECG repolarization to obtain the dynamic domain signal of ECG repolarization.
[0017] Based on the Euclidean distance between any two states in the ECG depolarization time domain signal, a non-threshold recursive graph in the depolarization time domain is determined; based on the Euclidean distance between any two states in the ECG depolarization dynamic domain signal, a non-threshold recursive graph in the depolarization dynamic domain is determined; based on the Euclidean distance between any two states in the ECG repolarization time domain signal, a non-threshold recursive graph in the repolarization time domain is determined; based on the Euclidean distance between any two states in the ECG repolarization dynamic domain signal, a non-threshold recursive graph in the repolarization dynamic domain is determined.
[0018] Based on the non-threshold recursive graphs in the depolarization time domain, dynamic domain, complex polarization time domain, and complex polarization dynamic domain, a joint non-threshold recursive graph is determined; based on the upper triangular matrix and lower triangular matrix of the joint non-threshold recursive graph, a fused recursive graph is determined.
[0019] The fused recursive graph is input into the trained neural network model to obtain the identification results of cardiovascular disease categories.
[0020] The above technical solution has the following advantages or beneficial effects:
[0021] 1. Clinical knowledge-driven automatic decomposition of ECG signals into depolarization and repolarization time-domain signals. Combining the physiological and pathological mechanisms of cardiovascular diseases, and based on the independence and complementarity of the QRS wave and ST-T segment, the ECG signal is automatically segmented into depolarization and repolarization time-domain signals using a multi-scale decomposition method, avoiding the shortcomings of inaccurate localization of key points in traditional QRS wave and ST-T segment segmentation.
[0022] 2. Time-domain signal dynamics information mining. Dynamic modeling is performed on ECG depolarization and repolarization time-domain signals to obtain more comprehensive and interpretable dynamic domain signals, which can be used to reveal the deep patterns of ECG signals and expand the information dimensions of ECG signals.
[0023] 3. Recursive Visualization and Fusion. A threshold-free joint recursive analysis method is used to transform the time series into a two-dimensional image, fusing ECG signals and dynamic domain signals, as well as ECG depolarization and repolarization time domain signals. This enhances the ability to represent subtle abnormalities in ECG signals and improves the performance of cardiovascular disease diagnosis. The threshold-free joint recursive analysis method avoids the problems of over- or under-representation of the system due to inappropriate threshold selection.
[0024] 4. Automated diagnosis of cardiovascular diseases. Features are automatically extracted and fused from the recurrence graph using neural networks to achieve automatic identification of cardiovascular diseases. Attached Figure Description
[0025] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0026] Figure 1 This is a system functional block diagram in an embodiment of the present invention. Detailed Implementation
[0027] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0028] Example 1
[0029] This embodiment provides a cardiovascular disease identification system based on electrocardiogram fusion;
[0030] like Figure 1 As shown, the cardiovascular disease identification system based on electrocardiogram fusion includes:
[0031] The acquisition module is configured to: acquire the raw ECG time-domain signal, preprocess the raw ECG time-domain signal, and obtain the preprocessed ECG time-domain signal;
[0032] The segmentation module is configured to segment the preprocessed ECG time-domain signal to obtain the ECG depolarization time-domain signal and the ECG repolarization time-domain signal.
[0033] The dynamic modeling module is configured to: use deterministic learning theory to perform dynamic modeling on the ECG depolarization time-domain signal to obtain the ECG depolarization dynamic domain signal; and perform dynamic modeling on the ECG repolarization time-domain signal to obtain the ECG repolarization dynamic domain signal.
[0034] The fusion recursive module is configured to: determine the non-threshold recursive graph in the depolarization time domain based on the Euclidean distance between any two states in the ECG depolarization time domain signal; determine the non-threshold recursive graph in the depolarization dynamic domain based on the Euclidean distance between any two states in the ECG depolarization dynamic domain signal; determine the non-threshold recursive graph in the repolarization time domain based on the Euclidean distance between any two states in the ECG repolarization time domain signal; and determine the non-threshold recursive graph in the repolarization dynamic domain based on the Euclidean distance between any two states in the ECG repolarization dynamic domain signal.
[0035] Based on the non-threshold recursive graphs in the depolarization time domain, dynamic domain, complex polarization time domain, and complex polarization dynamic domain, a joint non-threshold recursive graph is determined; based on the upper triangular matrix and lower triangular matrix of the joint non-threshold recursive graph, a fused recursive graph is determined.
[0036] The disease identification module is configured to input the fused recursive graph into the trained neural network model to obtain the identification results of cardiovascular disease categories.
[0037] Further, the original ECG time-domain signal is acquired, and the original ECG time-domain signal is preprocessed to obtain the preprocessed ECG time-domain signal. The preprocessing includes filtering and baseline correction.
[0038] For example, the preprocessing includes: using a bandpass filter to select a cutoff frequency of 0.5~40Hz to remove power frequency interference and high-frequency noise. Simultaneously, a 0.5Hz Butterworth filter is used to eliminate baseline drift caused by breathing or other factors.
[0039] Furthermore, the preprocessed ECG time-domain signal is segmented to obtain the ECG depolarization time-domain signal and the ECG repolarization time-domain signal, specifically including:
[0040] First, the preprocessed ECG time-domain signal is linearly transformed using the Kross matrix, which is as follows:
[0041] ;
[0042] ;
[0043] in, Let Kross be the matrix. Preprocessed ECG time-domain signal, where This indicates the first and second leads of the bipolar lead system. Represents 6 chest leads, Refers to the reduced-dimensional electrocardiogram signal. These represent the three leads of the ECG signal after dimensionality reduction.
[0044] The dimensionality-reduced three-lead ECG signal is decomposed using wavelet packets: wavelet basis functions (such as orthogonal wavelet bases like the Daubechies series) and the number of decomposition levels are selected. Through an iterative downsampling process, the signal is decomposed into sub-bands with different frequencies.
[0045] For any node wavelet packet coefficients (or signal) Through a low-pass filter and high-pass filter Decompose to obtain the first Wavelet packet coefficients of two child nodes of the layer and :
[0046] ;
[0047] ;
[0048] in, Indicates the number of decomposition layers. Indicates the node index in the current layer. This indicates the index of the sampling point of the current layer signal. This serves as the index for the sampling points of the next decomposition layer. and Defined by wavelet basis functions.
[0049] For sampling rate The time-domain signal of the electrocardiogram has a wavelet packet decomposition level of . , produced Sub-band. Arbitrary node. The corresponding frequency range is ,Right now:
[0050]
[0051] The set of wavelet packet coefficients for STT and QRS is defined as the set of node coefficients that satisfy the following conditions:
[0052]
[0053]
[0054] in, .
[0055] Collected wavelet packet coefficients and Reconstruction is performed using inverse wavelet packet transform, and ST-T band signals are obtained by iteratively applying a reconstruction filter and upsampling. and QRS group signal The formula for the inverse wavelet packet transform is as follows:
[0056] ;
[0057] in, and These are the low-pass and high-pass filters for wavelet reconstruction, respectively. The wavelet reconstruction filter... and With wavelet decomposition filter and The relationship depends on the specific wavelet basis function. Taking orthogonal wavelet basis as an example, the wavelet reconstruction filter is the time reversal of the wavelet decomposition filter, that is... , .
[0058] The QRS complex signal is a time-domain signal of electrocardiographic depolarization; the ST-T band signal is a time-domain signal of electrocardiographic repolarization.
[0059] The beneficial effect of the above technical solution is that it segments the two signals based on the independence and complementarity between the QRS complex signal and the ST-T band signal in the pathological mechanism of cardiovascular diseases.
[0060] Furthermore, deterministic learning theory is employed to dynamically model the ECG depolarization time-domain signal, obtaining the ECG depolarization dynamic domain signal; and dynamic modeling is also performed on the ECG repolarization time-domain signal, obtaining the ECG repolarization dynamic domain signal, specifically including:
[0061] Both the depolarization and repolarization time-domain signals are considered as three-dimensional nonlinear dynamic systems. The generated signal is obtained by sampling the depolarization and repolarization time-domain signals. Based on deterministic learning theory, a neural network identifier is used to perform local accurate neural network modeling of the intrinsic dynamics of depolarization and repolarization.
[0062]
[0063] in, It is an RBF (Radial Basis Function) neural network; It is an electrocardiogram signal. The estimated quantity, For estimating the weights of an RBF neural network, For the regression vector of the RBF neural network, For learning rate, The sampling time for the electrocardiogram (ECG) signal;
[0064] Applying the weight update algorithm to :
[0065] ;
[0066] in, For state estimation error, Represents the learning gain, where, for Least norm 2 The sampling time for the electrocardiogram (ECG) signal.
[0067] Furthermore, the inherent nonlinear dynamics of depolarization and repolarization are obtained. :
[0068]
[0069] in, , , This indicates that intrinsic electrocardiographic dynamics information is obtained through deterministic learning, when At that time, the dynamic domain information of ECG depolarization was obtained. ,when At that time, the dynamic domain information of ECG repolarization was obtained. .
[0070] The beneficial effect of the above technical solution is that by determining the learning theory for dynamic modeling, a more comprehensive and interpretable dynamic signal can be obtained, which is the dynamic domain signal.
[0071] Furthermore, based on the Euclidean distance between any two states in the ECG depolarization time-domain signal, a non-threshold recursive graph in the depolarization time-domain is determined; based on the Euclidean distance between any two states in the ECG depolarization dynamic-domain signal, a non-threshold recursive graph in the depolarization dynamic-domain is determined; based on the Euclidean distance between any two states in the ECG repolarization time-domain signal, a non-threshold recursive graph in the repolarization time-domain is determined; based on the Euclidean distance between any two states in the ECG repolarization dynamic-domain signal, a non-threshold recursive graph in the repolarization dynamic-domain is determined; specifically including:
[0072] Based on ECG depolarization time domain signal calculate and The Euclidean distance between two time-domain states determines the non-threshold recursive graph in the divider time domain. Based on ECG depolarization dynamic domain signal calculate and The Euclidean distance between two time-space states determines the non-threshold recursive graph of the divider dynamic domain. Based on ECG repolarization time-domain signals calculate and The Euclidean distance between two time-domain states determines the complex polar time-domain non-threshold recursive graph. Based on ECG repolarization dynamic domain signals calculate and The Euclidean distance between two time-space states determines the non-threshold recursive graph of the complex polar dynamic domain. Specifically, it includes:
[0073] ;
[0074] ;
[0075] in, The time series of ECG depolarization and repolarization time-domain signals after dimensionality reduction. These are depolarized dynamic domain signals and complex polarized dynamic domain signals. This represents the Euclidean norm, which in turn represents the Euclidean distance. This represents a non-threshold recursive graph of the ECG time-domain signal; Represents a non-threshold recursive graph of a dynamic domain signal;
[0076] when The depolarization time-domain non-threshold recursive graph is obtained. ;when The depolarization time-domain non-threshold recursive graph is obtained. ;when The depolarization time-domain non-threshold recursive graph is obtained. ;when The depolarization time-domain non-threshold recursive graph is obtained. .
[0077] The beneficial effect of the above technical solution is that it can convert time series into two-dimensional images to enhance the characterization of subtle abnormalities in electrocardiogram signals.
[0078] Furthermore, based on the non-threshold recursive graphs in the depolarization time domain, dynamic domain, complex polarization time domain, and complex polarization dynamic domain, a joint non-threshold recursive graph is determined; specifically including:
[0079] To enhance the similarity between the two time series while reducing their correlation, Hadamard product is calculated to obtain a non-threshold joint recurrence graph. :
[0080] ;
[0081] in, Represents the time-domain signal. Time and the Euclidean distance of time Represents the dynamic domain signal. Time and the Euclidean distance of time ;
[0082] when , The non-threshold joint recurrence graph of the divider is obtained. ;
[0083] when , The complex non-threshold joint recurrence graph is obtained. .
[0084] Furthermore, based on the upper and lower triangular matrices of the non-threshold joint recursive graph, the fused recursive graph is determined, specifically including:
[0085] The non-threshold joint recurrence plot of depolarization and repolarization is generated using Matlab software. functions and The function transforms the upper triangular matrix into the lower triangular matrix and adds them together to obtain the fused recursive graph. :
[0086] ;
[0087] in, The function is used to extract the upper triangular portion of a matrix. The function is used to extract the lower triangular portion of a matrix.
[0088] The beneficial effects of the above technical solution are: in order to effectively reflect the correlation between the depolarization and repolarization processes, and at the same time solve the problem that the recursion graph contains general repetitive information, the thresholdless joint recursion graphs of different subsystems are transformed into upper triangular matrices and lower triangular matrices, and then added together to obtain a fused recursion graph containing complete time-domain and dynamic-domain information of the depolarization and repolarization subsystems.
[0089] Furthermore, the fused recursive graph is input into the trained neural network model to obtain the identification results of cardiovascular disease categories, specifically including:
[0090] The fused recurrent graph is input into the trained convolutional neural network to obtain the identification results of cardiovascular disease categories.
[0091] The training process of the trained convolutional neural network includes:
[0092] Construct a training set, which is a fusion recursive graph of known real clinical labels of cardiovascular diseases (such as myocardial ischemia);
[0093] The training set is input into the convolutional neural network to train the network, resulting in the trained convolutional neural network.
[0094] Specifically, the fused recurrent graph is transformed into a format suitable for input to a convolutional neural network. Since the fused recurrent graph is color-coded, 3 channels are chosen.
[0095] The convolutional neural network (CNN) model structure is designed, and its parameters are determined based on the size of the fused recurrent graph. Features from the fused recurrent graph are automatically extracted. Finally, either the Softmax or Sigmoid function is selected to output probability values for different categories of cardiovascular diseases, achieving automatic identification of cardiovascular diseases.
[0096] The beneficial effects of the above technical solution are: using the fused recurrent graph as input, a neural network model (optionally, a convolutional neural network) is trained. The convolutional neural network utilizes its inherent ability to automatically extract image features to extract discriminative features from the fused recurrent graph. Finally, the trained model outputs the corresponding cardiovascular disease identification result.
[0097] This invention relates to the field of cardiovascular disease diagnostic technology, specifically including the following steps: preprocessing the acquired electrocardiogram (ECG) signals to obtain standardized, high-quality ECG signals; segmenting the preprocessed ECG signals into a depolarization subsystem (corresponding to the QRS complex) and a repolarization subsystem (corresponding to the ST-T segment) using a multi-scale decomposition method; applying deterministic learning theory to perform dynamic modeling for each of these subsystems to obtain dynamic domain signals; performing recursive analysis on the time-domain and dynamic-domain signals of the subsystems to generate a threshold-free joint recursive graph, and obtaining a fused recursive graph containing complete information of both the depolarization and repolarization subsystems through upper / lower triangular matrix transformation and addition; finally, inputting the fused recursive graph into a trained neural network model (e.g., a convolutional neural network) to automatically extract features and identify diseases, thereby achieving automatic diagnosis of cardiovascular diseases. This invention provides an effective new paradigm for automatic identification of cardiovascular diseases and has significant clinical application value.
[0098] Example 2
[0099] This embodiment also provides an electronic device, including: one or more processors, one or more memories, and one or more computer programs; wherein, the processor is connected to the memory, and the one or more computer programs are stored in the memory. When the electronic device is running, the processor executes the one or more computer programs stored in the memory to cause the electronic device to perform the following steps:
[0100] The raw ECG time-domain signal is acquired, and the raw ECG time-domain signal is preprocessed to obtain the preprocessed ECG time-domain signal.
[0101] The preprocessed ECG time-domain signal is segmented to obtain the ECG depolarization time-domain signal and the ECG repolarization time-domain signal;
[0102] Using deterministic learning theory, dynamic modeling is performed on the time-domain signal of ECG depolarization to obtain the dynamic domain signal of ECG depolarization; and dynamic modeling is also performed on the time-domain signal of ECG repolarization to obtain the dynamic domain signal of ECG repolarization.
[0103] Based on the Euclidean distance between any two states in the ECG depolarization time domain signal, a non-threshold recursive graph in the depolarization time domain is determined; based on the Euclidean distance between any two states in the ECG depolarization dynamic domain signal, a non-threshold recursive graph in the depolarization dynamic domain is determined; based on the Euclidean distance between any two states in the ECG repolarization time domain signal, a non-threshold recursive graph in the repolarization time domain is determined; based on the Euclidean distance between any two states in the ECG repolarization dynamic domain signal, a non-threshold recursive graph in the repolarization dynamic domain is determined.
[0104] Based on the non-threshold recursive graphs in the depolarization time domain, dynamic domain, complex polarization time domain, and complex polarization dynamic domain, a joint non-threshold recursive graph is determined; based on the upper triangular matrix and lower triangular matrix of the joint non-threshold recursive graph, a fused recursive graph is determined.
[0105] The fused recursive graph is input into the trained neural network model to obtain the identification results of cardiovascular disease categories.
[0106] Example 3
[0107] This embodiment also provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the following steps:
[0108] The raw ECG time-domain signal is acquired, and the raw ECG time-domain signal is preprocessed to obtain the preprocessed ECG time-domain signal; the preprocessed ECG time-domain signal is segmented to obtain the ECG depolarization time-domain signal and the ECG repolarization time-domain signal.
[0109] Using deterministic learning theory, dynamic modeling is performed on the time-domain signal of ECG depolarization to obtain the dynamic domain signal of ECG depolarization; and dynamic modeling is also performed on the time-domain signal of ECG repolarization to obtain the dynamic domain signal of ECG repolarization.
[0110] Based on the Euclidean distance between any two states in the ECG depolarization time domain signal, a non-threshold recursive graph in the depolarization time domain is determined; based on the Euclidean distance between any two states in the ECG depolarization dynamic domain signal, a non-threshold recursive graph in the depolarization dynamic domain is determined; based on the Euclidean distance between any two states in the ECG repolarization time domain signal, a non-threshold recursive graph in the repolarization time domain is determined; based on the Euclidean distance between any two states in the ECG repolarization dynamic domain signal, a non-threshold recursive graph in the repolarization dynamic domain is determined.
[0111] Based on the non-threshold recursive graphs in the depolarization time domain, dynamic domain, complex polarization time domain, and complex polarization dynamic domain, a joint non-threshold recursive graph is determined; based on the upper triangular matrix and lower triangular matrix of the joint non-threshold recursive graph, a fused recursive graph is determined.
[0112] The fused recursive graph is input into the trained neural network model to obtain the identification results of cardiovascular disease categories.
[0113] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A cardiovascular disease identification system based on electrocardiogram fusion, characterized in that, include: The acquisition module is configured to: acquire the raw ECG time-domain signal, preprocess the raw ECG time-domain signal, and obtain the preprocessed ECG time-domain signal. The segmentation module is configured to segment the preprocessed ECG time-domain signal to obtain the ECG depolarization time-domain signal and the ECG repolarization time-domain signal. The dynamic modeling module is configured to: use deterministic learning theory to perform dynamic modeling on the ECG depolarization time-domain signal to obtain the ECG depolarization dynamic domain signal; and perform dynamic modeling on the ECG repolarization time-domain signal to obtain the ECG repolarization dynamic domain signal. The fusion recursive module is configured to: determine the non-threshold recursive graph in the depolarization time domain based on the Euclidean distance between any two states in the ECG depolarization time domain signal; determine the non-threshold recursive graph in the depolarization dynamic domain based on the Euclidean distance between any two states in the ECG depolarization dynamic domain signal; determine the non-threshold recursive graph in the repolarization time domain based on the Euclidean distance between any two states in the ECG repolarization time domain signal; and determine the non-threshold recursive graph in the repolarization dynamic domain based on the Euclidean distance between any two states in the ECG repolarization dynamic domain signal. Specifically, it includes: Based on ECG depolarization time domain signal calculate and The Euclidean distance between two time-domain states determines the non-threshold recursive graph in the divider time domain. Based on ECG depolarization dynamic domain signal calculate and The Euclidean distance between two time-space states determines the non-threshold recursive graph of the divider dynamic domain. Based on ECG repolarization time-domain signals calculate and The Euclidean distance between two time-domain states determines the complex polar time-domain non-threshold recursive graph. Based on ECG repolarization dynamic domain signals calculate and The Euclidean distance between two time-space states determines the non-threshold recursive graph of the complex polar dynamic domain. Specifically, this includes: ； ； in, The time series of ECG depolarization and repolarization time-domain signals after dimensionality reduction. These are depolarized dynamic domain signals and complex polarized dynamic domain signals. Denotes the Euclidean norm; This represents a non-threshold recursive graph of the ECG time-domain signal; Represents a non-threshold recursive graph of a dynamic domain signal; when The depolarization time-domain non-threshold recursive graph is obtained. ;when The depolarization time-domain non-threshold recursive graph is obtained. ;when The depolarization time-domain non-threshold recursive graph is obtained. ;when The depolarization time-domain non-threshold recursive graph is obtained. ; Based on the non-threshold recursion graphs in the depolarization time domain, dynamic domain, complex polarity time domain, and complex polarity dynamic domain, a joint non-threshold recursion graph is determined; specifically, this includes: To enhance the similarity between the two time series while reducing their correlation, Hadamard product calculation is performed to obtain a non-threshold joint recurrence graph. : ； in, Represents the time-domain signal. Time and the Euclidean distance of time Represents the dynamic domain signal. Time and the Euclidean distance of time ; when , The non-threshold joint recurrence graph of the divider is obtained. ; when , The complex non-threshold joint recurrence graph is obtained. ; The fused recursive graph is determined based on the upper and lower triangular matrices of the non-threshold joint recursive graph; specifically, it includes: The non-threshold joint recurrence graph of depolarization and repolarization is used... functions and The function transforms the upper triangular matrix into the lower triangular matrix and adds them together to obtain the fused recursive graph. : ； in, The function is used to extract the upper triangular portion of a matrix. The function is used to extract the lower triangular portion of a matrix; The disease identification module is configured to input the fused recursive graph into the trained neural network model to obtain the identification results of cardiovascular disease categories.
2. The cardiovascular disease identification system based on electrocardiogram fusion as described in claim 1, characterized in that, The preprocessed ECG time-domain signal is segmented to obtain the ECG depolarization time-domain signal and the ECG repolarization time-domain signal, specifically including: First, the preprocessed ECG time-domain signal is linearly transformed using the Kross matrix, which is as follows: ； ； in, Let Kross be the matrix. Preprocessed ECG time-domain signal, where This indicates the first and second leads of the bipolar lead system. Represents 6 chest leads, Refers to the reduced-dimensional electrocardiogram signal. These represent the three leads of the ECG signal after dimensionality reduction; The dimensionality-reduced three-lead ECG signal was subjected to wavelet packet decomposition: wavelet basis functions and the number of decomposition levels were selected. Through an iterative downsampling process, the signal is decomposed into sub-bands with different frequencies; For any node wavelet packet coefficient Through a low-pass filter and high-pass filter Decompose to obtain the first Wavelet packet coefficients of two child nodes of the layer and : ； ； in, Indicates the number of decomposition layers. Indicates the node index in the current layer. This indicates the sampling point index of the current layer signal. This serves as the index for the sampling points of the next decomposition layer; For sampling rate The time-domain signal of the electrocardiogram has a wavelet packet decomposition level of . , produced Sub-band; arbitrary node The corresponding frequency range is : ； The set of wavelet packet coefficients for STT and QRS is defined as the set of node coefficients that satisfy the following conditions: ； ； in, ; Collected wavelet packet coefficients and Reconstruction is performed using inverse wavelet packet transform, and ST-T band signals are obtained by iteratively applying a reconstruction filter and upsampling. and QRS group signal The inverse wavelet packet transform formula is as follows: ； in, and These are the low-pass and high-pass filters for wavelet reconstruction, respectively. , ; The QRS complex signal is a time-domain signal of electrocardiographic depolarization; the ST-T band signal is a time-domain signal of electrocardiographic repolarization.
3. The cardiovascular disease identification system based on electrocardiogram fusion as described in claim 1, characterized in that, The fused recursive graph is input into the trained neural network model to obtain the identification results of cardiovascular disease categories, specifically including: The fused recursive graph is input into the trained convolutional neural network to obtain the identification result of cardiovascular disease category; wherein, the training process of the trained convolutional neural network includes: constructing a training set, which is a fused recursive graph of known real clinical labels of cardiovascular diseases; inputting the training set into the convolutional neural network to train the network to obtain the trained convolutional neural network.
4. The cardiovascular disease identification system based on electrocardiogram fusion as described in claim 1, characterized in that, The raw ECG time-domain signal is acquired, and the raw ECG time-domain signal is preprocessed to obtain the preprocessed ECG time-domain signal. The preprocessing includes filtering and baseline correction.
5. The cardiovascular disease identification system based on electrocardiogram fusion as described in claim 1, characterized in that, The preprocessing includes: using a bandpass filter with a cutoff frequency of 0.5~40Hz to remove power frequency interference and high-frequency noise; and using a Butterworth filter of 0.5Hz to eliminate baseline drift caused by breathing.
6. An electronic device, characterized in that, include: A memory for non-transitory storage of computer-readable instructions; and a processor for executing the computer-readable instructions, wherein, when executed by the processor, the computer-readable instructions perform the following steps: acquiring a raw electrocardiogram (ECG) time-domain signal; preprocessing the raw ECG time-domain signal to obtain a preprocessed ECG time-domain signal; segmenting the preprocessed ECG time-domain signal to obtain an ECG depolarization time-domain signal and an ECG repolarization time-domain signal; using deterministic learning theory to perform dynamic modeling of the ECG depolarization time-domain signal to obtain an ECG depolarization dynamic domain signal; and performing dynamic modeling of the ECG repolarization time-domain signal... Dynamic modeling is performed to obtain the ECG repolarization dynamic domain signal; based on the Euclidean distance between any two states in the ECG depolarization time domain signal, a non-threshold recursive graph in the depolarization time domain is determined; based on the Euclidean distance between any two states in the ECG depolarization dynamic domain signal, a non-threshold recursive graph in the depolarization dynamic domain is determined; based on the Euclidean distance between any two states in the ECG repolarization time domain signal, a non-threshold recursive graph in the repolarization time domain is determined; based on the Euclidean distance between any two states in the ECG repolarization dynamic domain signal, a non-threshold recursive graph in the repolarization dynamic domain is determined; specifically including: Based on ECG depolarization time domain signal calculate and The Euclidean distance between two time-domain states determines the non-threshold recursive graph in the divider time domain. Based on ECG depolarization dynamic domain signal calculate and The Euclidean distance between two time-space states determines the non-threshold recursive graph of the divider dynamic domain. Based on ECG repolarization time-domain signals calculate and The Euclidean distance between two time-domain states determines the complex polar time-domain non-threshold recursive graph. Based on ECG repolarization dynamic domain signals calculate and The Euclidean distance between two time-space states determines the non-threshold recursive graph of the complex polar dynamic domain. Specifically, this includes: ； ； in, The time series of ECG depolarization and repolarization time-domain signals after dimensionality reduction. These are depolarized dynamic domain signals and complex polarized dynamic domain signals. Denotes the Euclidean norm; This represents a non-threshold recursive graph of the ECG time-domain signal; Represents a non-threshold recursive graph of a dynamic domain signal; when The depolarization time-domain non-threshold recursive graph is obtained. ;when The depolarization time-domain non-threshold recursive graph is obtained. ;when The depolarization time-domain non-threshold recursive graph is obtained. ;when The depolarization time-domain non-threshold recursive graph is obtained. ; Based on the non-threshold recursion graphs in the depolarization time domain, dynamic domain, complex polarity time domain, and complex polarity dynamic domain, a joint non-threshold recursion graph is determined; specifically, this includes: To enhance the similarity between the two time series while reducing their correlation, Hadamard product calculation is performed to obtain a non-threshold joint recurrence graph. : ； in, Represents the time-domain signal. Time and the Euclidean distance of time Represents the dynamic domain signal. Time and the Euclidean distance of time ; when , The non-threshold joint recurrence graph of the divider is obtained. ; when , The complex non-threshold joint recurrence graph is obtained. ; The fused recursive graph is determined based on the upper and lower triangular matrices of the non-threshold joint recursive graph; specifically, it includes: The non-threshold joint recurrence graph of depolarization and repolarization is used... functions and The function transforms the upper triangular matrix into the lower triangular matrix and adds them together to obtain the fused recursive graph. : ； in, The function is used to extract the upper triangular portion of a matrix. The function is used to extract the lower triangular portion of a matrix; The fused recursive graph is input into the trained neural network model to obtain the identification results of cardiovascular disease categories.
7. A storage medium characterized by being non-transitory. The computer-readable instructions are stored, wherein when the non-transitory computer-readable instructions are executed by a computer, the following steps are performed: The raw ECG time-domain signal is acquired, and the raw ECG time-domain signal is preprocessed to obtain the preprocessed ECG time-domain signal. The preprocessed ECG time-domain signal is segmented to obtain the ECG depolarization time-domain signal and the ECG repolarization time-domain signal; Using deterministic learning theory, dynamic modeling is performed on the time-domain signal of ECG depolarization to obtain the dynamic domain signal of ECG depolarization; and dynamic modeling is also performed on the time-domain signal of ECG repolarization to obtain the dynamic domain signal of ECG repolarization. Based on the Euclidean distance between any two states in the ECG depolarization time-domain signal, a non-threshold recursive graph in the depolarization time domain is determined; based on the Euclidean distance between any two states in the ECG depolarization dynamic domain signal, a non-threshold recursive graph in the depolarization dynamic domain is determined; based on the Euclidean distance between any two states in the ECG repolarization time-domain signal, a non-threshold recursive graph in the repolarization time-domain signal is determined; based on the Euclidean distance between any two states in the ECG repolarization dynamic domain signal, a non-threshold recursive graph in the repolarization dynamic domain is determined; specifically including: Based on ECG depolarization time domain signal calculate and The Euclidean distance between two time-domain states determines the non-threshold recursive graph in the divider time domain. Based on ECG depolarization dynamic domain signal calculate and The Euclidean distance between two time-space states determines the non-threshold recursive graph of the divider dynamic domain. Based on ECG repolarization time-domain signals calculate and The Euclidean distance between two time-domain states determines the complex polar time-domain non-threshold recursive graph. Based on ECG repolarization dynamic domain signals calculate and The Euclidean distance between two time-space states determines the non-threshold recursive graph of the complex polar dynamic domain. Specifically, this includes: ； ； in, The time series of ECG depolarization and repolarization time-domain signals after dimensionality reduction. These are depolarized dynamic domain signals and complex polarized dynamic domain signals. Denotes the Euclidean norm; This represents a non-threshold recursive graph of the ECG time-domain signal; Represents a non-threshold recursive graph of a dynamic domain signal; when The depolarization time-domain non-threshold recursive graph is obtained. ;when The depolarization time-domain non-threshold recursive graph is obtained. ;when The depolarization time-domain non-threshold recursive graph is obtained. ;when The depolarization time-domain non-threshold recursive graph is obtained. ; Based on the non-threshold recursion graphs in the depolarization time domain, dynamic domain, complex polarity time domain, and complex polarity dynamic domain, a joint non-threshold recursion graph is determined; specifically, this includes: To enhance the similarity between the two time series while reducing their correlation, Hadamard product calculation is performed to obtain a non-threshold joint recurrence graph. : ； in, Represents the time-domain signal. Time and the Euclidean distance of time Represents the dynamic domain signal. Time and the Euclidean distance of time ; when , The non-threshold joint recurrence graph of the divider is obtained. ; when , The complex non-threshold joint recurrence graph is obtained. ; The fused recursive graph is determined based on the upper and lower triangular matrices of the non-threshold joint recursive graph; specifically, it includes: The non-threshold joint recurrence graph of depolarization and repolarization is used... functions and The function transforms the upper triangular matrix into the lower triangular matrix and adds them together to obtain the fused recursive graph. : ； in, The function is used to extract the upper triangular portion of a matrix. The function is used to extract the lower triangular portion of a matrix; The fused recursive graph is input into the trained neural network model to obtain the identification results of cardiovascular disease categories.