Electrocardiogram data processing device and method capable of achieving myocardial ischemia screening

Abstract

The invention relates to the technical field of electrocardiogram data processing, and provides an electrocardiogram data processing device and method capable of achieving myocardial ischemia screening, abnormal waveforms in an electrocardiogram are accurately judged through cooperative work of a plurality of modules of electrocardiogram signal preprocessing, waveform division, heart beat screening, alignment fitting, feature extraction and the like; the method comprises the following steps of: performing signal alignment and fitting, performing alignment and fitting superposition on a re-sampled wave band, eliminating the influence of difference between individuals and signal drift, performing alignment superposition on a plurality of pieces of heart beat data, then realizing denoising, realizing numerical value averaging of signal points, accurately extracting electrocardiogram characteristics related to myocardial ischemia, and improving the accuracy of myocardial ischemia. Therefore, automatic myocardial ischemia screening can be achieved through the extracted features, personal errors are greatly reduced, and powerful technical support is provided for clinical electrocardiogram analysis.

Description

Technical Field

[0001] The present disclosure relates to the technical field of electrocardiogram data processing, and more specifically, to an electrocardiogram data processing device and method capable of realizing myocardial ischemia screening. Background Art

[0002] The statements in this section merely provide background technical information related to the present disclosure and do not necessarily constitute prior art.

[0003] Electrocardiogram (ECG) is one of the basic tools for evaluating cardiac function and can accurately reflect the electrical activity of the heart. Electrophysiological information of the heart can be obtained through electrocardiogram, providing an important basis for diagnosing various heart diseases. Electrocardiogram records the process of electrical excitation propagation of cardiac muscles, forming different waveforms (P wave, QRS complex, T wave, etc.). The changes in these waveforms are directly related to the electrical activity of the heart and can help doctors evaluate the health status of the heart. With the increase in the amount and complexity of electrocardiogram data, traditional manual electrocardiogram analysis methods are becoming inadequate, so computer-aided diagnosis (CAD) analysis has become the key to solving this problem.

[0004] The method of judging myocardial ischemia through electrocardiogram mainly relies on the analysis of specific waveforms and segments. When ST segment depression, ST segment elevation, T wave inversion, T wave flattening or bidirectionality occur, it can be considered that myocardial ischemia characteristics appear. The method of using a computer to extract the representation of myocardial ischemia from electrocardiogram has made significant progress in many aspects, combining machine learning, deep learning and other advanced computing technologies. With the development of machine learning, numerous electrocardiogram algorithms have been proposed and applied to assist clinical electrocardiogram interpretation, saving a large amount of medical resources. In addition, the stability of the computer also reduces the influence of human factors;

[0005] Electrocardiogram electrical signals are usually affected by noise from various sources, including muscle activity, poor electrode contact, power supply interference, etc. Although there are now various mature noise processing methods in existing computer-aided diagnosis, noise influence is still a major obstacle in computer-aided ischemia diagnosis. Existing methods perform feature analysis on the entire waveband of electrocardiogram and cannot effectively eliminate the influence of noise, so the judgment feature information of myocardial ischemia cannot be effectively extracted from electrocardiogram data. Summary of the Invention

[0006] To solve the above problems, the present disclosure proposes an electrocardiogram data processing device and method capable of realizing myocardial ischemia screening, which can effectively reduce the influence of interference information in electrocardiogram and extract accurate electrocardiogram features for identifying myocardial ischemia.

[0007] To achieve the above object, the present disclosure adopts the following technical solutions:

[0008] The first aspect of the present disclosure provides an electrocardiogram data processing device capable of realizing myocardial ischemia screening, including:

[0009] A data preprocessing module configured to preprocess the acquired electrocardiogram to obtain electrocardiogram waveform data;

[0010] A waveform division module configured to divide the obtained electrocardiogram waveform to generate a sub-waveform mask and a heart beat classification result;

[0011] A heart beat screening module configured to count the number of heart beats of each type according to the heart beat category, determine the main category of heart beats, and screen out the heart beat curves of the main category;

[0012] An alignment and fitting module configured to perform equal-length resampling on the PR segment, ST segment, QRS wave, and T wave of the screened heart beat curves, and perform alignment and superposition fitting on the resampled wave bands to obtain a new heart beat curve;

[0013] A feature extraction module configured to extract curve features from the new heart beat curve to obtain electrocardiogram features.

[0014] The second aspect of the present disclosure provides an electrocardiogram data processing method capable of realizing myocardial ischemia screening, including the following steps:

[0015] Preprocess the acquired electrocardiogram to obtain electrocardiogram waveform data;

[0016] Divide the obtained electrocardiogram waveform to generate a sub-waveform mask and a heart beat classification result;

[0017] Count the number of heart beats of each type according to the heart beat category, determine the main category of heart beats, and screen out the heart beat curves of the main category;

[0018] Perform equal-length resampling on the PR segment, ST segment, QRS wave, and T wave of the screened heart beat curves, and perform alignment and superposition fitting based on kernel density on the resampled wave bands to obtain a new heart beat curve;

[0019] Extract curve features from the new heart beat curve to obtain electrocardiogram features.

[0020] The third aspect of the present disclosure provides an electronic device including a memory, a processor, and computer instructions stored on the memory and running on the processor. When the computer instructions are run by the processor, the following steps are completed:

[0021] Preprocess the acquired electrocardiogram to obtain electrocardiogram waveform data;

[0022] Divide the obtained electrocardiogram waveform to generate a sub-waveform mask and a heart beat classification result;

[0023] Count the number of heartbeats of each type according to the heartbeat category, determine the main category of the heartbeat, and filter out the heartbeat curves of the main category;

[0024] Perform resampling of equal length on the PR segment, ST segment, QRS wave, and T wave of the filtered heartbeat curve, and perform alignment superposition fitting based on kernel density on the resampled waveband to obtain a new heartbeat curve;

[0025] Extract curve features from the new heartbeat curve to obtain electrocardiogram features.

[0026] A fourth aspect of the present disclosure is an electrocardiogram data processing device capable of realizing myocardial ischemia screening, including an electrocardiogram acquisition device and a processor;

[0027] The electrocardiogram acquisition device is used to acquire electrocardiogram data;

[0028] The processor is configured to execute the following steps:

[0029] Preprocess the acquired electrocardiogram to obtain electrocardiogram waveform data;

[0030] Perform waveform division on the obtained electrocardiogram waveform to generate sub-waveform masks and heartbeat classification results;

[0031] Count the number of heartbeats of each type according to the heartbeat category, determine the main category of the heartbeat, and filter out the heartbeat curves of the main category;

[0032] Perform resampling of equal length on the PR segment, ST segment, QRS wave, and T wave of the filtered heartbeat curve, and perform alignment superposition fitting based on kernel density on the resampled waveband to obtain a new heartbeat curve;

[0033] Extract curve features from the new heartbeat curve to obtain electrocardiogram features.

[0034] Compared with the prior art, the beneficial effects of the present disclosure are as follows:

[0035] The present disclosure can efficiently extract valuable electrocardiogram features from electrocardiograms. Through the coordinated work of multiple modules such as preprocessing, waveform division, heartbeat screening, alignment fitting, and feature extraction of electrocardiogram signals, it can accurately judge abnormal waveforms in electrocardiograms; perform signal alignment and fitting. After resampling, the wavebands are aligned and superposed by fitting, eliminating the influence of inter-individual differences and signal drift. At the same time, after aligning and superposing multiple heartbeat data, denoising is achieved, and the numerical averaging of signal points is performed, enabling accurate extraction of electrocardiogram features related to myocardial ischemia. Thus, the extracted features can realize automated myocardial ischemia screening, greatly reducing human errors and providing strong technical support for clinical electrocardiogram analysis.

[0036] The advantages of the present disclosure and the advantages of additional aspects will be described in detail in the following specific embodiments. Brief Description of the Drawings

[0037] The accompanying drawings forming a part of this disclosure are used to provide a further understanding of the present disclosure. The schematic embodiments and descriptions thereof of the present disclosure are used to explain the present disclosure and do not constitute a limitation to the present disclosure.

[0038] Figure 1 is a block diagram of the electrocardiogram data processing device according to Embodiment 1 of the present disclosure;

[0039] Figure 2 is a waveform schematic diagram of the electrocardiogram sub - waveform after waveform division according to Embodiment 1 of the present disclosure;

[0040] Figure 3 is a schematic diagram of the specific output result of the waveform division module according to Embodiment 1 of the present disclosure;

[0041] Figure 4 is a schematic diagram of the superposition of heartbeat waveforms according to Embodiment 1 of the present disclosure;

[0042] Figure 5 is a front - view superposition of the heartbeat curves after alignment and superposition according to Embodiment 1 of the present disclosure;

[0043] Figure 6 is a scatter plot of the voltage values of each heartbeat at a time point after alignment and superposition according to Embodiment 1 of the present disclosure;

[0044] Figure 7 is a histogram of the voltage values of each heartbeat at a time point after alignment and superposition according to Embodiment 1 of the present disclosure;

[0045] Figure 8 is a new heartbeat curve obtained after alignment, superposition and fitting according to Embodiment 1 of the present disclosure;

[0046] Figure 9 is a flowchart of the electrocardiogram data processing method according to Embodiment 2 of the present disclosure. Detailed Description of the Embodiments

[0047] The present disclosure will be further described below in conjunction with the drawings and embodiments.

[0048] It should be noted that the following detailed descriptions are all exemplary and are intended to provide a further description of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present disclosure belongs.

[0049] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. It should be noted that, without conflict, the various embodiments and features in the present disclosure may be combined with each other. The embodiments will be described in detail below with reference to the drawings.

[0050] Embodiment 1

[0051] In the technical solutions disclosed in one or more embodiments, as Figures 1 to 8 shown, an electrocardiogram data processing device capable of realizing myocardial ischemia screening includes:

[0052] A data preprocessing module configured to preprocess the acquired electrocardiogram to obtain electrocardiogram waveform data;

[0053] A waveform partitioning module configured to partition the obtained electrocardiogram waveform to generate sub-waveform masks and heart beat classification results;

[0054] A heart beat screening module configured to count the number of heart beats of each type according to the heart beat category, determine the main category of heart beats, and screen out the heart beat curves of the main category;

[0055] An alignment and fitting module configured to perform equal-length resampling on the PR segment, ST segment, QRS wave, and T wave of the screened heart beat curves, and perform alignment superposition fitting based on kernel density on the resampled wave bands to obtain new heart beat curves;

[0056] A feature extraction module configured to extract curve features from the new heart beat curves to obtain electrocardiogram features;

[0057] This device can efficiently extract valuable electrocardiogram features from the electrocardiogram. Through the collaborative work of multiple modules such as preprocessing, waveform partitioning, heart beat screening, alignment fitting, and feature extraction of the electrocardiogram signal, it can accurately judge the abnormal waveforms in the electrocardiogram; at the same time, signal alignment and fitting are performed, and the resampled wave bands are aligned and superposed by fitting, eliminating the influence of individual differences and signal drift. At the same time, after aligning and superposing multiple heart beat data, denoising is achieved, and the numerical averaging of signal points is achieved. It can accurately extract electrocardiogram features related to myocardial ischemia, so that the extracted features can realize automated myocardial ischemia screening, and also greatly reduce human error, providing strong technical support for clinical electrocardiogram analysis.

[0058] Based on the obtained electrocardiogram features, classification or identification by a deep learning model can be performed for the auxiliary diagnosis of myocardial ischemia, providing auxiliary judgment data for doctors.

[0059] In the data preprocessing module, a 12-lead electrocardiogram can be used. The preprocessing includes unifying the data sampling rate to the same standard and performing denoising processing, including removing noises such as baseline drift, power interference, and electromyogram artifacts. The preprocessing module performs the following processes:

[0060] Step 11: Data resampling. According to the sampling of the electrocardiogram data, linear interpolation resampling is performed on the data, and sampling is carried out at the same set sampling frequency.

[0061] For example, if its sampling rate is unified to 500Hz, the dimension of the processed data is [12, 500*T]. Among them, 12 is the number of leads, 500 is the sampling frequency, and T is the time length in seconds.

[0062] Step 12: Remove baseline drift. Specifically, high-pass filtering is used to remove low-frequency components.

[0063] Step 13: Remove power interference. Adaptive filtering technology can be used to remove 50Hz - 60Hz power interference.

[0064] In some embodiments, the waveform division module uses a division method that performs image recognition and division through a convolutional neural network (CNN), including the following steps;

[0065] Step 21: Use a convolutional neural network to perform image recognition of the position masks corresponding to the P wave, QRS wave, and T wave. The corresponding waveform positions are 1, and other positions are 0.

[0066] Step 22: Divide the waveforms according to the obtained position masks to obtain multiple sub-waveforms. For each sub-waveform, classification is performed, and the maximum value of the classification probability is judged as the heart beat type corresponding to the sub-waveform.

[0067] As Figure 3 shown, color marking is used. The blue - yellow - red covered areas are the P wave, QRS complex, and T wave sub-waveforms respectively. The PR segment is between the end position of the P wave and the start position of the QRS wave, and the ST segment is between the end position of the QRS wave and the start position of the T wave. After division, one heart beat waveform is used as one sub-waveform.

[0068] Among them, the heart beat types can include N sinus, S atrial, V ventricular, J junctional, and aF atrial flutter and atrial fibrillation. Classification probabilities are calculated for each type of heart beat type to determine the heart beat type corresponding to each sub-waveform.

[0069] Each heartbeat waveform is the signal waveform generated by one electrical activity of the heart in an electrocardiogram, including a P wave, a QRS complex, and a T wave; usually, the signal between two adjacent R waves (RR interval) is regarded as a complete heartbeat.

[0070] In this embodiment, the format of data output is: [8, 500*T], where 8 represents the probability results of the mask with 3 sub-waveform divisions and 5 types of heartbeat type divisions.

[0071] In some embodiments, the heartbeat screening module is configured to remove ventricular (V) arrhythmias that cause paroxysmal changes, and screen out the main category of heartbeat curves according to the frequencies of various heartbeat waveforms, which can be used for the heartbeat of auxiliary judgment of myocardial ischemia.

[0072] Specifically, the method for screening out the main category of heartbeat curves includes the following steps:

[0073] Step 31: Remove the abnormal ventricular heartbeat waveform.

[0074] Step 32: Count the occurrence numbers of various heartbeat waveforms.

[0075] Step 33: Determine the main heartbeat waveform according to the occurrence numbers and the proportion of sinusoidal (N sinusoidal) heartbeat waveforms.

[0076] Specifically, if the proportion of sinus rhythm exceeds 30%, the sinus heartbeat is the main heartbeat; if it is less than 30%, the one with the highest occurrence number is the main heartbeat, and the heartbeat category to be analyzed next is determined.

[0077] In some embodiments, the alignment and fitting module is configured to perform the following steps:

[0078] Step 41: Count the widths of the PR segment, ST segment, QRS complex, and T wave of all heartbeats on the time axis, and calculate the maximum value of the width of each waveband respectively, which is expressed as follows:

[0079] M_pr = MAX(pr), M_st = MAX(st), M_t = MAX(t), M_qrs = MAX(qrs);

[0080] As Figure 3 shown, it includes the waveforms of all leads. Each lead waveform includes multiple heartbeat waveforms in sequence on the time axis. Identify the widths on the time axis of the heartbeat waveforms of all leads, find the maximum value of the width for the next alignment, which can make the lengths of all heartbeat waveforms consistent. For a 12-lead electrocardiogram, that is, identify the widths on the time axis of the heartbeat waveforms of 12 curves, so that after the subsequent alignment, the widths of all heartbeat waveforms are consistent;

[0081] Step 42: According to the maximum width, horizontally stretch the wave band of each heartbeat based on the endpoint values of the wave band until the wave band reaches the corresponding width, and then connect them in the chronological order of the wave bands to form an elongated waveform curve, achieving length alignment;

[0082] Optionally, horizontally stretch based on the value of the end point of the wave band until the corresponding length of the wave band is reached. For example, as Figure 2 shown, for a complete heartbeat curve, the first P wave band remains unchanged. For the PR segment, horizontally stretch it horizontally backward from the intersection of the PR segment and the QRS wave to obtain a PR segment with a length of MAX(pr); for the QRS wave, horizontally stretch it horizontally backward from the intersection of the QRS wave and the ST segment to obtain a QRS wave band with a length of M_qrs; for the ST segment, horizontally stretch it horizontally backward from the intersection of the ST segment and the T wave band to obtain an ST segment with a length of M_st; for the T wave band, horizontally stretch it horizontally backward based on the end value of the T wave band to obtain a T wave band with a length of M_t; the aligned heartbeat waveform curve is as Figure 3 shown, and the corresponding segments have equal widths on the time axis;

[0083] Alternatively, it can be horizontally stretched forward based on the initial value of each wave band, which can also uniformly increase the width of the heartbeat curve.

[0084] The actual length of each aligned heartbeat waveform curve is:

[0085] BEAT_len = p + M_pr + M_qrs + M_st + M_t

[0086] After the above alignment operation, resampling with equal length is achieved, making the waveforms of each heartbeat signal more standardized and providing uniform input data for subsequent feature extraction;

[0087] Step 43: Using the starting point of the QRS wave as a reference, superimpose the aligned heartbeat waveform curves of the same lead in the spatial coordinate system;

[0088] Specifically, the aligned heartbeat waveform curves are stacked in the spatial coordinate system and arranged in a front-to-back order aligned with the starting point of the QRS wave;

[0089] As Figure 4 shows, for multiple superimposed heartbeat curves, the heartbeat curves sampled from the same lead are aligned and set in the spatial coordinate system in a front-to-back manner. The three orthogonal coordinate axes are respectively:

[0090] X-axis (horizontal axis): represents time;

[0091] Y-axis (vertical axis): represents potential;

[0092] Z-axis (depth axis): Represents beats and indicates which cardiac cycle curve it is.

[0093] Step 44: Perform kernel density estimation of the potential on the superimposed signals along the time axis, select the point with the highest kernel density as the fitting value of the waveform corresponding to the previous time point, and connect the obtained fitting values to obtain a fitted cardiac cycle waveform curve.

[0094] For a 12-lead electrocardiogram, 12 cardiac cycle curves can be obtained, and each cardiac cycle curve includes Figure 5 the five wavebands shown.

[0095] In the already aligned data, sequentially obtain the potential data of each lead, each cardiac cycle, and each time point. Perform density estimation on the potential value (Potential) of each obtained time point, and take the value with the maximum density as the voltage value of that point; connect the maximum density values of each time point to form a new cardiac cycle curve.

[0096] Optionally, the kernel function used can be a Gaussian kernel function or a mean function.

[0097] In this embodiment, specifically, the formula of the kernel density estimation method is:

[0098]

[0099] where is the density estimation value at the time point x; n is the number of observations, that is, the number of superimposed cardiac cycle curves; x i is an independently and identically distributed observation, i = 1, 2,..., n;

[0100] K is the kernel function, which is a symmetric and usually non-negative function that satisfies:

[0101]

[0102] The Gaussian kernel function is used here:

[0103]

[0104] where h > 0 is the bandwidth (or window size), a parameter that controls the smoothness; a larger h results in a smoother estimate.

[0105] In this embodiment, by performing equal-length resampling on key wavebands such as the PR segment, ST segment, QRS complex, and T wave, the problem of inconsistent timing of electrocardiogram signals is solved, making the waveform lengths of each heartbeat signal more consistent and standardized, and providing uniform input data for subsequent feature extraction. At the same time, signal alignment and fitting are performed: the resampled wavebands are superimposed through alignment and fitting, eliminating the influence of inter-individual differences and signal drift. At the same time, after aligning and superimposing multiple heartbeat data, abnormal signal filtering and denoising are achieved, and the numerical averaging of signal points is performed, enabling accurate extraction of electrocardiogram features related to myocardial ischemia. This alignment method ensures the unity of the signal structure of each heartbeat, making the data more suitable for precise analysis, and achieving the effect of analyzing the features of a single heartbeat to the overall analysis. Especially in myocardial ischemia screening, it can effectively extract the features of pathological changes.

[0106] A feature extraction module, configured to resample the fitted heartbeat waveform curve for the PR segment, ST segment, and T waveband according to a set number of sampling points to obtain the collected electrocardiogram features;

[0107] Optionally, set the sampling data point ratios for different wavebands. Preferably, in this embodiment, the sampling data point ratios for the PR segment, ST segment, and T waveband are set to 2:3:5;

[0108] The specific sampling method is to extract the features of the PR segment, ST segment, and T waveband of 12 curves, resample the PR segment to 40 data points (80 milliseconds), the ST segment to 60 data points (120 milliseconds), and the T waveband to 100 data points (200 milliseconds). After data splicing, there are a total of 200 data points.

[0109] Embodiment 2

[0110] Based on Embodiment 1, this embodiment provides an electrocardiogram data processing method capable of realizing myocardial ischemia screening, as Figure 9 shown, including the following steps:

[0111] Step 1: Preprocess the acquired electrocardiogram to obtain electrocardiogram waveform data;

[0112] Step 2: Divide the obtained electrocardiogram waveform to generate sub-waveform masks and heartbeat classification results;

[0113] Step 3: Count the number of each type of heartbeat according to the heartbeat category, determine the main category of the heartbeat, and screen out the heartbeat curves of the main category;

[0114] Step 4: Perform equal-length resampling on the PR segment, ST segment, QRS complex, and T wave of the screened heartbeat curves, and perform alignment, superimposition, and fitting based on kernel density on the resampled wavebands to obtain new heartbeat curves;

[0115] Step 5: Extract curve features from the new heartbeat curve to obtain ECG features;

[0116] This method significantly improves the accuracy and robustness of electrocardiogram data processing through efficient preprocessing, waveform division of convolutional neural networks, heartbeat screening, and alignment fitting. Especially during the screening process for myocardial ischemia, it can accurately identify key features such as ST-segment changes and T-wave inversion, providing important support for clinical diagnosis. At the same time, the alignment fitting method based on kernel density effectively eliminates the time deviation between signals, ensures the accuracy of feature extraction, and greatly improves the performance of myocardial ischemia screening.

[0117] In Step 1, the preprocessing includes the following processes:

[0118] Step 11: Data resampling. According to the sampling of the electrocardiogram data, perform linear interpolation resampling on the data and sample it at the same set sampling frequency;

[0119] For example, unify its sampling rate to 500Hz, and the dimension of the processed data is [12, 500*T]. Among them, 12 is the number of leads. 500 is the sampling frequency; T is the time length in seconds.

[0120] Step 12: Remove baseline drift. Specifically, use high-pass filtering to remove low-frequency components.

[0121] Step 13: Remove power interference. You can use adaptive filtering technology to remove 50Hz - 60Hz power interference.

[0122] In Step 2, perform waveform division on the obtained ECG waveforms. Use a convolutional neural network (CNN) for image recognition and division, including the following steps;

[0123] Step 21: Use a convolutional neural network for image recognition to obtain the position masks corresponding to the P wave, QRS complex, and T wave. The corresponding waveform positions are 1, and other positions are 0.

[0124] Step 22: Divide the waveforms according to the obtained position masks to get multiple sub-waveforms. Classify each sub-waveform, and judge the maximum value of the classification probability as the heartbeat type corresponding to the sub-waveform;

[0125] As Figure 3 shown, use colors for marking. The blue-yellow-red covered areas are the P wave, QRS complex, and T wave respectively; the PR segment is between the end position of the P wave and the start position of the QRS complex, and the ST segment is between the end position of the QRS complex and the start position of the T wave. After division, one heartbeat waveform is used as one sub-waveform.

[0126] Among them, the heartbeat types can include N (sinus), S (atrial), V (ventricular), J (junctional), and aF (atrial flutter and atrial fibrillation). Classification probabilities are calculated for each type of heartbeat, so as to determine the heartbeat type corresponding to each sub-waveform.

[0127] Each heartbeat waveform is a signal waveform generated by one electrical activity of the heart in an electrocardiogram, including a P wave, a QRS complex, and a T wave; usually, the signal between two adjacent R waves (RR interval) is taken as a complete heartbeat.

[0128] In step 3, the number of heartbeats of each type is counted according to the heartbeat category, the main category of the heartbeat is determined, and the heartbeat curves of the main category are selected.

[0129] The method for selecting the heartbeat curves of the main category includes the following steps:

[0130] Step 31: Remove the waveforms of abnormal ventricular heartbeats (V).

[0131] Step 32: Count the number of occurrences of each type of heartbeat waveform.

[0132] Step 33: Determine the main heartbeat waveform based on the number of occurrences and the proportion of sinus (N sinus) heartbeat waveforms.

[0133] Specifically, if the proportion of sinus rhythm exceeds 30%, the sinus heartbeat is the main heartbeat. If it is less than 30%, the one with the highest number of occurrences is the main heartbeat, and the heartbeat category to be analyzed next is determined.

[0134] In step 4, the method for resampling the PR segment, ST segment, QRS complex, and T wave of the selected heartbeat curves at equal lengths and performing alignment superposition fitting based on kernel density on the resampled bands to obtain new heartbeat curves includes the following steps:

[0135] Step 41: Count the widths of the PR segment, ST segment, QRS complex, and T wave of all heartbeats on the time axis, and calculate the maximum value of the width of each band respectively, as shown below:

[0136] M_pr = MAX(pr), M_st = MAX(st), M_t = MAX(t), M_qrs = MAX(qrs);

[0137] As Figure 3 shown, it includes the waveforms of all leads. Each lead waveform includes multiple heartbeat waveforms in sequence along the time axis. The widths of the heartbeat waveforms of all leads are identified on the time axis, and the maximum value of the width is found for the next alignment, which can make the lengths of all heartbeat waveforms consistent. For a 12-lead electrocardiogram, that is, the widths of the heartbeat waveforms of 12 curves are identified on the time axis, so that after the subsequent alignment, the widths of all heartbeat waveforms are the same.

[0138] Step 42: According to the maximum width, stretch the waveband of each heartbeat horizontally based on the value of the end point until it reaches the width of the corresponding waveband, and then connect them in the chronological order of the wavebands to form an elongated waveform curve, achieving length alignment;

[0139] Optionally, stretch horizontally based on the value of the end point of the waveband until it reaches the length of the corresponding waveband. For example, as Figure 2 shown, for a complete heartbeat curve, the first P waveband remains unchanged. For the PR segment, stretch horizontally backward from the intersection of the PR segment and the QRS wave to obtain a PR segment with a length of MAX(pr); for the QRS wave, stretch horizontally backward from the intersection of the QRS wave and the ST segment to obtain a QRS waveband with a length of M_qrs; for the ST segment, stretch horizontally backward from the intersection of the ST segment and the T waveband to obtain an ST segment with a length of M_st; for the T waveband, stretch horizontally backward based on the end value of the T waveband to obtain a T waveband with a length of M_t; the aligned heartbeat waveform curve is as Figure 3 shown, and the corresponding segments have equal widths on the time axis; as Figure 3 shown is the effect of stretching and aligning an actual electrocardiogram. The symbols on the left represent the marks of 12 leads; the blue mark P above represents that the blue area is the P wave, the yellow mark QRS represents that the yellow area is the QRS wave; the red mark T represents that the red area is the T wave; the N mark position represents the starting point of a heartbeat waveform;

[0140] Alternatively, it can be stretched forward based on the initial value of each waveband, which can also uniformly increase the width of the heartbeat curve.

[0141] The actual length of each aligned heartbeat waveform curve is:

[0142] BEAT_len = p + M_pr + M_qrs + M_st + M_t

[0143] After the above alignment operation, resampling with equal length is achieved, making the waveforms of each heartbeat signal more standardized and providing uniform input data for subsequent feature extraction;

[0144] Step 43: Taking the starting point of the QRS wave as a reference, superimpose the aligned heartbeat waveform curves of the same lead in the spatial coordinate system;

[0145] Specifically, the aligned heartbeat waveform curves are stacked in the spatial coordinate system and arranged in a sequential order front to back with the starting point of the QRS wave aligned;

[0146] As Figure 4Among them, it is multiple superimposed heartbeat curves. The heartbeat curves sampled from the same lead are aligned in the spatial coordinate system in the front-back manner. The three orthogonal coordinate axes are respectively:

[0147] X-axis (horizontal axis): represents Time;

[0148] Z-axis (vertical axis): represents Potential;

[0149] Y-axis (depth axis): represents Beats, indicating which heartbeat curve;

[0150] Step 44: Perform kernel density estimation of the potential on the superimposed signal along the time axis, select the point with the highest kernel density as the fitting value of the waveform corresponding to the previous time point, and connect the obtained fitting values to obtain a fitted heartbeat waveform curve;

[0151] For a 12-lead electrocardiogram, 12 heartbeat curves can be obtained. Each heartbeat curve includes Figure 5 the five bands shown;

[0152] In the already aligned data, obtain the potential data of each lead, each heartbeat, and each time point in sequence. Perform density estimation on the potential value (Potential) of each obtained time point, and take the value with the maximum density as the voltage value of this point; connect the maximum density values of each time point to form a new heartbeat curve;

[0153] Optionally, the kernel function used can be a Gaussian kernel function or a mean function;

[0154] In this embodiment, specifically, the formula of the kernel density estimation method is:

[0155]

[0156] Among them, is the density estimation value at the time point x; n is the number of observations, that is, the number of superimposed heartbeat curves; x i is an independently and identically distributed observation value, i = 1, 2,..., n;

[0157] K is the kernel function, which is a symmetric and usually non-negative function, and it satisfies:

[0158]

[0159] The Gaussian kernel function is used here:

[0160]

[0161] Among them, h>0 is the bandwidth (or window size), a parameter that controls the degree of smoothing; a larger h results in a smoother estimate.

[0162] Figure 5 is the superimposed image, which contains multiple heartbeat curves, and the density after superposition is marked with colors. The darker the color, the higher the density, which is equivalent to Figure 4 seen from the front (along the Y-axis), and the density is marked with colors. The redder the color, the higher the density; Figures 5 to 7 is for visually showing Figure 8 the intermediate image drawn for the formation process of Figure 6 and Figure 7 are the general principle and schematic diagram of density estimation. Figure 6 is when there is Figure 5 Based on the superimposed image, the heartbeat data of each point is obtained to form a scatter plot; Figure 7 is at Figure 6 The histogram formed on the voltage scatter plot can visually display the density change. Converting the Figure 7 scatter plot of Figure 7 into the Figure 8 histogram of Figure 8 is at Figure 5 After density estimation is performed on the basis of

[0163] In this embodiment, by performing equal-length resampling on key bands such as the PR segment, ST segment, QRS complex, and T wave of the electrocardiogram signal, the problem of inconsistent time series of the electrocardiogram signal is solved, making the waveform lengths of each heartbeat signal more consistent and standardized, providing uniform input data for subsequent feature extraction. At the same time, signal alignment and fitting are performed: the resampled bands are superimposed through alignment and fitting, eliminating the influence of individual differences and signal drift. At the same time, after aligning and superimposing multiple heartbeat data, abnormal signal filtering, denoising are achieved, and the numerical averaging of signal points is done, enabling accurate extraction of electrocardiogram features related to myocardial ischemia. This alignment method ensures the unity of the signal structure of each heartbeat, making the data more suitable for precise analysis, and achieving the effect of analyzing the features of a single heartbeat to the overall analysis. Especially in myocardial ischemia screening, it can effectively extract the features of pathological changes.

[0164] In step 5, the feature extraction method is: for the fitted heartbeat waveform curve, resampling is performed on the PR segment, ST segment, and T wave bands according to the set number of sampling points to obtain the collected electrocardiogram features;

[0165] Optionally, set the sampling data point ratios for different bands. Preferably, in this embodiment, it is set for the PR segment, and the sampling data point ratios of the ST segment and the T band are 2:3:5;

[0166] The specific sampling method is as follows: For the PR segment, ST segment, and T band characteristics of 12 curves, resample the PR segment to 40 data points (80 milliseconds), the ST segment to 60 data points (120 milliseconds), and the T band to 100 data points (200 milliseconds). After data splicing, there are a total of 200 data points.

[0167] Embodiment 3

[0168] This embodiment provides an electronic device, including a memory, a processor, and computer instructions stored on the memory and running on the processor. When the computer instructions are run by the processor, the following steps are completed:

[0169] Step 1: Preprocess the acquired electrocardiogram to obtain electrocardiogram waveform data;

[0170] Step 2: Divide the obtained electrocardiogram waveform to generate sub-waveform masks and heart beat classification results;

[0171] Step 3: Count the number of each type of heart beat according to the heart beat category, determine the main category of the heart beat, and screen out the heart beat curves of the main category;

[0172] Step 4: Resample the PR segment, ST segment, QRS wave, and T wave of the screened heart beat curves to the same length, and perform kernel density-based alignment, superposition, and fitting on the resampled bands to obtain new heart beat curves;

[0173] Step 5: Extract curve features from the new heart beat curves to obtain electrocardiogram features;

[0174] It should be noted here that each module in this embodiment corresponds one by one to each step in Embodiment 1, and the specific implementation process is the same, so it will not be repeated here.

[0175] Embodiment 4

[0176] This embodiment provides an electrocardiogram data processing device capable of realizing myocardial ischemia screening, including an electrocardiogram acquisition device and a processor;

[0177] The electrocardiogram acquisition device is used to acquire electrocardiogram data;

[0178] The processor is configured to execute the following steps:

[0179] Step 1: Preprocess the acquired electrocardiogram to obtain electrocardiogram waveform data;

[0180] Step 2: Divide the obtained electrocardiogram waveform to generate sub-waveform masks and heart beat classification results;

[0181] Step 3: Count the number of heart beats of each type according to the heart beat category, determine the main category of heart beats, and filter out the heart beat curves of the main category;

[0182] Step 4: Resample the PR segment, ST segment, QRS wave, and T wave of the filtered heart beat curve at equal lengths, and perform kernel density-based alignment and superposition fitting on the resampled wave bands to obtain a new heart beat curve;

[0183] Step 5: Extract curve features from the new heart beat curve to obtain electrocardiogram features;

[0184] Optionally, the electrocardiogram acquisition device can be any device capable of realizing electrocardiogram acquisition, such as: 12-lead electrocardiograph, Holter monitor, wearable ECG device, electrode chest strap (ECG chest strap), or wireless electrocardiogram acquisition device (wireless ECG), etc.

[0185] The above are only the preferred embodiments of the present disclosure and are not used to limit the present disclosure. For those skilled in the art, the present disclosure can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

[0186] Although the specific implementation manners of the present disclosure have been described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that, based on the technical solutions of the present disclosure, various modifications or deformations that can be made without creative efforts by those skilled in the art are still within the protection scope of the present disclosure.

Claims

1. An electrocardiogram data processing device capable of implementing myocardial ischemia screening, characterized in that: include: A data preprocessing module is configured to preprocess the acquired electrocardiogram to obtain electrocardiogram waveform data; A waveform division module is configured to perform waveform division on the obtained ECG waveform to generate sub-waveform masks and heart beat classification results; A heartbeat screening module is configured to count the number of each type of heartbeat according to the heartbeat category, determine the main category of the heartbeat, and screen out the heartbeat curve of the main category; The alignment and fitting module is configured to resample the PR segment, ST segment, QRS wave and T wave of the screened heartbeat curve to equal length, align and superimpose the resampled bands to obtain a new heartbeat curve; The feature extraction module is configured to extract curve features from the new heartbeat curve to obtain electrocardiogram features.

2. An electrocardiogram data processing device capable of implementing myocardial ischemia screening as claimed in claim 1, characterized in that: The waveform segmentation module uses a convolutional neural network to perform image recognition segmentation, including the following steps: Image recognition is performed through a convolutional neural network to obtain the position masks corresponding to the P wave, QRS wave, and T wave; The obtained position mask is used to identify the band of each heartbeat waveform, and a heartbeat waveform is used as a sub-waveform to obtain a waveform division result; Each sub-waveform is classified, and the maximum value of the classification probability is determined as the heartbeat type of the corresponding sub-waveform.

3. The electrocardiogram data processing device capable of implementing myocardial ischemia screening as claimed in claim 1, characterized in that: The alignment fitting module is configured to perform the following steps: Count the widths of the PR segment, ST segment, QRS wave, and T wave of all heart beats on the time axis, and calculate the maximum value of the width of each band; According to the maximum value of the width, the band of each heartbeat is horizontally stretched based on the endpoint value of the band so that the band reaches the width of the corresponding band, and then connected in the order of the time of the band to form a stretched waveform curve to achieve length alignment; Taking the starting point of the QRS wave as a reference, the aligned heartbeat waveform curves of the same lead are superimposed and set in the spatial coordinate system; The kernel density of the potential of the superimposed signal is estimated along the time axis, and the point with the highest kernel density is selected as the fitting value of the waveform corresponding to the previous time point. The obtained fitting values ​​are connected to obtain a fitted heartbeat waveform curve.

4. An electrocardiogram data processing device capable of implementing myocardial ischemia screening as claimed in claim 3, characterized in that: Stretch horizontally based on the endpoint value of the band, and stretch horizontally backward based on the value of the end point of the band to reach the length of the corresponding band; or stretch forward based on the initial value of each band to reach the length of the corresponding band.

5. The electrocardiogram data processing device capable of implementing myocardial ischemia screening as claimed in claim 1, characterized in that: It also includes a feature extraction module, which is configured to sample the PR segment, ST segment and T band of the fitted heartbeat waveform curve according to a set number of sampling points to obtain the collected electrocardiographic features.

6. A method for processing electrocardiogram data capable of screening myocardial ischemia, characterized in that: The steps include: Preprocessing the acquired electrocardiogram to obtain electrocardiogram waveform data; Performing waveform division on the obtained ECG waveform to generate sub-waveform masks and heart beat classification results; According to the heartbeat category, the number of each type of heartbeat is counted, the main category of the heartbeat is determined, and the heartbeat curve of the main category is screened out; The PR segment, ST segment, QRS wave and T wave of the screened heart rate curve are resampled to equal length, and the resampled bands are aligned and superimposed based on kernel density to obtain a new heart rate curve; The curve features of the new heartbeat curve are extracted to obtain the electrocardiogram features.

7. The electrocardiogram data processing method capable of implementing myocardial ischemia screening as claimed in claim 6, characterized in that: The obtained ECG waveform is segmented and image recognition and segmentation are performed using a convolutional neural network, including the following steps: The convolutional neural network is used to identify the position masks corresponding to the P wave, QRS wave and T wave in the image; The obtained position mask is used to identify the band of each heartbeat waveform, and a heartbeat waveform is used as a sub-waveform to obtain a waveform division result; Each sub-waveform is classified, and the maximum value of the classification probability is determined as the heartbeat type of the corresponding sub-waveform.

8. The electrocardiogram data processing method capable of implementing myocardial ischemia screening as claimed in claim 6, characterized in that: The method of resampling the PR segment, ST segment, QRS wave and T wave of the screened heartbeat curve to equal length, performing alignment superposition fitting based on kernel density on the resampled bands to obtain a new heartbeat curve comprises the following steps: Count the widths of the PR segment, ST segment, QRS wave, and T wave of all heart beats on the time axis, and calculate the maximum value of the width of each band; According to the maximum value of the width, the band of each heartbeat is horizontally stretched based on the endpoint value of the band so that the band reaches the width of the corresponding band, and then connected in the order of the time of the band to form a stretched waveform curve to achieve length alignment; Taking the starting point of the QRS wave as a reference, the aligned heartbeat waveform curves of the same lead are superimposed and set in the spatial coordinate system; The kernel density of the potential of the superimposed signal is estimated along the time axis, and the point with the highest kernel density is selected as the fitting value of the waveform corresponding to the previous time point, and the obtained fitting values ​​are connected to obtain a fitted heartbeat waveform curve; Alternatively, the feature extraction method is: for the fitted heartbeat waveform curve, the PR segment, ST segment and T wave segment are sampled again according to a set number of sampling points to obtain the collected ECG features.

9. An electronic device, characterized in that: The system comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein the computer instructions are executed by the processor to complete the following steps: Preprocessing the acquired electrocardiogram to obtain electrocardiogram waveform data; Performing waveform division on the obtained ECG waveform to generate sub-waveform masks and heart beat classification results; According to the heartbeat category, the number of each type of heartbeat is counted, the main category of the heartbeat is determined, and the heartbeat curve of the main category is screened out; The PR segment, ST segment, QRS wave and T wave of the screened heart rate curve are resampled to equal length, and the resampled bands are aligned and superimposed based on kernel density to obtain a new heart rate curve; The curve features of the new heartbeat curve are extracted to obtain the electrocardiogram features.

10. An electrocardiogram data processing device capable of implementing myocardial ischemia screening, characterized in that: It includes an electrocardiogram acquisition device and a processor; An electrocardiogram acquisition device, used for acquiring electrocardiogram data; The processor is configured to perform the following steps: Preprocessing the acquired electrocardiogram to obtain electrocardiogram waveform data; Performing waveform division on the obtained ECG waveform to generate sub-waveform masks and heart beat classification results; According to the heartbeat category, the number of each type of heartbeat is counted, the main category of the heartbeat is determined, and the heartbeat curve of the main category is screened out; The PR segment, ST segment, QRS wave and T wave of the screened heart rate curve are resampled to equal length, and the resampled bands are aligned and superimposed based on kernel density to obtain a new heart rate curve; The curve features of the new heartbeat curve are extracted to obtain the electrocardiogram features.

Images