WO2022036968A1 - 心电图数据增广方法、装置、电子设备及介质 - Google Patents

心电图数据增广方法、装置、电子设备及介质 Download PDF

Info

Publication number
WO2022036968A1
WO2022036968A1 PCT/CN2020/137330 CN2020137330W WO2022036968A1 WO 2022036968 A1 WO2022036968 A1 WO 2022036968A1 CN 2020137330 W CN2020137330 W CN 2020137330W WO 2022036968 A1 WO2022036968 A1 WO 2022036968A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
heartbeat
lead
beat
electrocardiogram
Prior art date
Application number
PCT/CN2020/137330
Other languages
English (en)
French (fr)
Inventor
王景峰
黄凯
陈样新
张玉玲
郭思璐
宋日辉
赵宝泽
张恒非
Original Assignee
生物岛实验室
中山大学孙逸仙纪念医院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 生物岛实验室, 中山大学孙逸仙纪念医院 filed Critical 生物岛实验室
Publication of WO2022036968A1 publication Critical patent/WO2022036968A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis

Definitions

  • the present disclosure relates to the field of medical technology, and in particular, to an electrocardiogram data augmentation method, device, electronic device and medium.
  • Data augmentation is one of the commonly used techniques in deep learning. It is mainly used to increase the training data set to make the data set as diverse as possible, so that the trained model has stronger generalization ability. In practical applications, not all augmentation methods are applicable to the current training data, and it is necessary to determine which data augmentation methods should be used according to the characteristics of the dataset. At present, data augmentation mainly includes: horizontal/vertical flip, rotation, zoom, crop, cut, translation, contrast, color jitter, noise, etc.
  • An electrocardiogram is the use of an electrocardiograph to record from the body surface the graph of electrical activity changes generated by each cardiac cycle of the heart, and each small segment of the signal has a certain medical significance.
  • conventional data augmentation methods such as flipping, rotating, and cropping will destroy the medical significance of ECG, and cannot play a positive role in the training of machine learning models.
  • Ordinary image processing data augmentation methods cannot be applied to ECG. Data augmentation of the signal.
  • embodiments of the present disclosure provide an electrocardiogram data augmentation method, apparatus, electronic device, and medium.
  • an embodiment of the present disclosure provides an ECG data augmentation method, including:
  • Augmented data is generated based on at least two of the plurality of beat data.
  • the processing of the electrocardiogram data to obtain a plurality of cardiac beat data includes:
  • the preprocessing includes:
  • Multi-layer wavelet decomposition is performed on the electrocardiogram data through discrete wavelet transform, the approximate value of the lowest layer is set to zero, and the baseline-calibrated electrocardiogram data is obtained through discrete wavelet reconstruction.
  • the preprocessing includes:
  • Multi-layer wavelet decomposition is performed on the electrocardiogram data through stationary wavelet transformation, threshold filtering is performed on detail values, and denoised electrocardiogram data are obtained through stationary wavelet reconstruction.
  • performing heartbeat recognition on the preprocessed electrocardiogram data to obtain multiple heartbeat data including:
  • the data of the first lead and/or the other leads are divided into a plurality of heart beat data according to the position of the peak of the R wave in the data of the first lead.
  • the generating augmented data based on at least two heartbeat data in the plurality of heartbeat data includes:
  • the label of the third cardiac data is determined based on the labels of the first cardiac data and/or the second cardiac data, and the third cardiac data with the label is determined as augmented data.
  • the case where at least one lead of the first cardiac beat data and the second cardiac beat data can be spliced splicing the first heartbeat data and the second heartbeat data to generate third heartbeat data, including:
  • the following operations are performed for each lead, in the case that the R-wave directions of the first and second heartbeat data in the current lead are the same and there is a zero-crossing point within a predetermined time after the R-wave, based on the The position of the zero-crossing point is described, and the data of the first heart beat data and the second heart beat data under the current lead are spliced into the data of the temporary heart beat data under the current lead; otherwise, the data of the second heart beat data under the current lead It is determined as the temporary cardiac data in the current lead;
  • the temporary heart beat data is determined to be the third heart beat data.
  • an electrocardiogram data augmentation device including:
  • an acquisition module configured to acquire ECG data
  • a processing module configured to process the electrocardiogram data to obtain a plurality of heartbeat data
  • a generating module configured to generate augmented data based on at least two of the plurality of beat data.
  • embodiments of the present disclosure provide an electronic device, including a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are processed by the The device executes to implement the method according to any one of the first aspect and the first to sixth implementation manners of the first aspect.
  • an embodiment of the present disclosure provides a computer-readable storage medium on which computer instructions are stored, and when the computer instructions are executed by a processor, implement the first aspect and the first to sixth aspects of the first aspect The method described in any one of the implementation manners.
  • the present disclosure by acquiring electrocardiogram data; processing the electrocardiogram data to acquire multiple heartbeat data; and generating augmented data based on at least two heartbeat data in the multiple heartbeat data, so that the The data augmentation of the ECG data can increase the data volume of the training set, which in turn can fully train the machine learning model and improve the performance of the model.
  • FIG. 1 shows a flowchart of an electrocardiogram data augmentation method according to an embodiment of the present disclosure
  • FIG. 2 shows a schematic diagram of electrocardiogram data according to an embodiment of the present disclosure
  • FIG. 3 shows a schematic diagram of ECG data after baseline calibration according to an embodiment of the present disclosure
  • FIG. 4 shows a schematic diagram of electrocardiogram data after noise reduction processing according to an embodiment of the present disclosure
  • FIG. 5 shows a flowchart of heart beat recognition according to an embodiment of the present disclosure
  • FIGS. 6A and 6B illustrate schematic diagrams of R-wave inversions according to embodiments of the present disclosure
  • FIG. 7 shows a schematic diagram of heart beat recognition according to an embodiment of the present disclosure
  • FIG. 8 illustrates a flow chart of generating augmented data according to an embodiment of the present disclosure
  • FIG. 9 shows a flowchart of generating augmented data according to another embodiment of the present disclosure.
  • FIG. 10 shows a schematic diagram of a heartbeat C generated from a heartbeat A and a heartbeat B according to an embodiment of the present disclosure
  • 11A shows a comparison diagram of heartbeat B and heartbeat C according to an embodiment of the present disclosure
  • FIG. 11B shows a comparison diagram of heartbeat A and heartbeat C according to an embodiment of the present disclosure
  • FIG. 12 shows a block diagram of an electrocardiogram data augmentation apparatus according to an embodiment of the present disclosure
  • FIG. 13 shows a block diagram of an electronic device according to an embodiment of the present disclosure.
  • FIG. 14 shows a schematic structural diagram of a computer system suitable for implementing an electrocardiogram data augmentation method according to an embodiment of the present disclosure.
  • An electrocardiogram is the use of an electrocardiograph to record from the body surface the graph of electrical activity changes generated by each cardiac cycle of the heart, and each small segment of the signal has a certain medical significance.
  • conventional data augmentation methods such as flipping, rotating, and cropping will destroy the medical significance of ECG, and cannot play a positive role in the training of machine learning models.
  • Ordinary image processing data augmentation methods cannot be applied to ECG. Data augmentation of the signal.
  • the present disclosure by acquiring electrocardiogram data; processing the electrocardiogram data to acquire multiple heartbeat data; and generating augmented data based on at least two heartbeat data in the multiple heartbeat data, so that the The data augmentation of the ECG data can increase the data volume of the training set, which in turn can fully train the machine learning model and improve the performance of the model.
  • FIG. 1 shows a flowchart of an electrocardiogram data augmentation method according to an embodiment of the present disclosure.
  • the method includes steps S101-S103.
  • step S101 obtain electrocardiogram data
  • step S102 processing the electrocardiogram data to obtain a plurality of heart beat data
  • step S103 augmented data is generated based on at least two heartbeat data among the plurality of heartbeat data.
  • the present disclosure by acquiring electrocardiogram data; processing the electrocardiogram data to acquire multiple heartbeat data; and generating augmented data based on at least two heartbeat data in the multiple heartbeat data, so that the The data augmentation of the ECG data can increase the data volume of the training set, which in turn can fully train the machine learning model and improve the performance of the model.
  • FIG. 2 shows a schematic diagram of electrocardiogram data according to an embodiment of the present disclosure.
  • the ECG data is acquired continuously, usually including multiple heartbeats, and thus has a certain periodicity.
  • ECG leads In ECG terminology, the placement of electrodes on the human body surface and the connection between electrodes and amplifiers during ECG recording are called ECG leads. There are 12 commonly used leads, namely lead I, lead II, lead III, lead aVR, lead aVL, lead aVF, lead V1, lead V2, lead V3, lead V4 lead, lead V5, and lead V6.
  • the ECG signals of all leads are measured almost at the same time, and they are almost the same in terms of R wave distribution density and location.
  • the electrocardiogram data shown in FIG. 2 is only the data of one lead, and the data of other leads can be similar.
  • the processing of the electrocardiogram data to obtain multiple heartbeat data includes preprocessing the electrocardiogram data, and performing heartbeat recognition on the preprocessed electrocardiogram data to obtain multiple heartbeat data.
  • the preprocessing includes performing multi-layer wavelet decomposition on the ECG data through discrete wavelet transform, setting the approximate value of the lowest layer to zero, and obtaining baseline-calibrated ECG data through discrete wavelet reconstruction.
  • Wavelet transform is an ideal tool for signal time-frequency analysis and processing.
  • the commonly used wavelet transform algorithm in computers is the mallat algorithm.
  • the core idea of this algorithm is to use filters to perform wavelet transform.
  • the original input signal S passes through two complementary filters (low-pass filter and high-pass filter) to generate two signals, A and D, where A is the approximate value of the signal (the value obtained by the low-frequency filter), and D is the detail value of the signal (value obtained by the high frequency filter).
  • the decomposition of the original signal by such a pair of filters is called first-order decomposition.
  • the signal decomposition process can be iterative, that is, multi-level decomposition can be performed. If the high-frequency components are no longer decomposed, but the low-frequency components are continuously decomposed, the lower-frequency components can be obtained, and a wavelet decomposition tree can be formed, and the decomposition level can be determined according to the needs.
  • discrete wavelet transform DWT
  • IDLT discrete wavelet reconstruction algorithm
  • the wavelet used in the process of eliminating baseline drift is db5 wavelet. After using DWT to decompose the signal with nine layers of wavelet, set the approximate value of the lowest layer to zero, and then perform IDWT to obtain the ECG signal after eliminating the baseline drift, as shown in Figure 3.
  • multi-layer wavelet decomposition is performed on the electrocardiogram data through discrete wavelet transform, the approximate value of the lowest layer is set to zero, and the baseline-calibrated electrocardiogram data is obtained through discrete wavelet reconstruction, so that it is possible to Better baseline calibration of ECG data.
  • the preprocessing process can also perform noise reduction processing.
  • the ECG data set includes EMG interference and 50/60Hz power frequency noise.
  • the wavelet transform threshold method can be used to eliminate the high frequency noise of the ECG signal.
  • the preprocessing includes performing multi-layer wavelet decomposition on the electrocardiogram data through stationary wavelet transformation, performing threshold filtering on detail values, and obtaining denoised electrocardiogram data through stationary wavelet reconstruction.
  • stationary wavelet transform is employed to remove noise.
  • SWT stationary wavelet transform
  • the biggest difference between SWT and DWT is that it does not perform downsampling after each layer is decomposed, so the amount of data is large and the amount of calculation is large, but the resolution of each layer remains unchanged.
  • For removing high-frequency noise it is not simply to remove the signal of a certain layer completely, but to remove the threshold value, so the resolution is particularly important.
  • ISWT stationary wavelet reconstruction
  • the wavelet used in the process of eliminating high frequency noise is bio2.6 wavelet. After using SWT to decompose the signal with 6 layers of wavelet, first set the highest frequency detail value of the first two layers to 0, the frequency represented by these detail values is too high and hardly contains any ECG signal information. Then threshold filtering is performed on the detail values of layers 3 to 6, and the threshold is selected using the following formula:
  • the threshold method used is the semi-soft threshold method, which does not produce wavelet coefficient mutation like the hard threshold method, nor does it produce deviations like the soft threshold method.
  • the formula of the semi-soft threshold method is as follows:
  • multi-layer wavelet decomposition is performed on the ECG data through stationary wavelet transform, threshold filtering is performed on the detail values, and denoised ECG data is obtained through stationary wavelet reconstruction, so that the ECG data can be improved.
  • the noise reduction effect on the data is performed on the ECG data through stationary wavelet transform, threshold filtering is performed on the detail values, and denoised ECG data is obtained through stationary wavelet reconstruction, so that the ECG data can be improved.
  • FIG. 5 shows a flowchart of heart beat recognition according to an embodiment of the present disclosure.
  • the method includes steps S501-S503.
  • step S501 perform multi-layer wavelet decomposition on the data of the first lead in the preprocessed electrocardiogram data through stationary wavelet transform to obtain a decomposition result;
  • step S502 feature identification is performed on the decomposition result, and the position of the peak of the R wave in the data of the first lead is determined;
  • step S503 the data of the first lead and/or other leads are divided into a plurality of heart beat data according to the position of the peak of the R wave in the data of the first lead.
  • SWT may be used to perform wavelet transform on the signal, and the used wavelet may be, for example:
  • the transformed detail value of the fifth layer may be selected, and the characteristic of the R wave of this layer is the most obvious.
  • the characteristic of the R wave of this layer is the most obvious.
  • the R wave in the upward direction it will appear in the form of a negative-positive maximum value pair in this layer; for the R wave in the downward direction, it will appear in the form of a positive-negative maximum value pair in this layer.
  • Find these pairs of maxima and find out the positions of their zero-crossing points, and the corresponding R-wave peaks are the maximum/minimum values within ⁇ 0.05s of the zero-crossing point.
  • FIG. 6A and 6B show schematic diagrams of R-wave inversion according to an embodiment of the present disclosure, wherein FIG. 6A shows a schematic diagram of a pathologically-induced R-wave inversion situation, and FIG. 6B shows a schematic diagram of an R-wave inversion caused by an electrode Schematic diagram of R wave inversion caused by reverse connection.
  • the 12 leads of the ECG signal measured by the same person at the same time are: lead I, lead II, lead III, lead aVR, lead aVL).
  • the first lead refers to lead I
  • the seventh lead refers to lead V1 to perform separate baseline removal and filtering operations, and then only the first lead is identified for R waves, and the first lead is used.
  • the R wave identification results to cut all other leads.
  • the reason for this is that the ECG signals of all leads are measured almost at the same time, so they are almost the same in terms of R-wave distribution density and location.
  • the R wave position calibration should be performed on 11 leads except the first lead, because the potential conduction between adjacent cells in the human body takes a certain time, so the R wave peak position of these 12 leads may have the largest position. ⁇ 0.05s difference.
  • the first and last heart beat signals may be incomplete, the first and last heart beat signals can be discarded, and the midpoint of the R peaks of two adjacent heart beats can be taken as the cutting point . In this way, a complete ECG signal can be accurately cut into one heartbeat.
  • the result after segmentation is shown in Figure 7.
  • a decomposition result is obtained by performing multi-layer wavelet decomposition on the data of the first lead in the preprocessed electrocardiogram data through stationary wavelet transform; and performing feature identification on the decomposition result, determining the position of the peak of the R wave in the data of the first lead; according to the position of the peak of the R wave in the data of the first lead,
  • the data is divided into multiple heartbeat data, so that the heartbeat data can be accurately identified, which is conducive to data augmentation of the electrocardiogram data, and can increase the data volume of the training set, thereby fully training the machine learning model and improving the performance of the model.
  • FIG. 8 illustrates a flow diagram of generating augmented data according to an embodiment of the present disclosure.
  • the method includes steps S801 to S804.
  • step S801 determine the first heartbeat data and the second heartbeat data from the plurality of heartbeat data
  • step S802 in the case where at least one lead of the first cardiac beat data and the second cardiac beat data can be spliced, splicing the first cardiac beat data and the second cardiac beat data to generate third cardiac beat data;
  • step S803 obtain the label of the first heartbeat data and/or the second heartbeat data
  • a label of the third heartbeat data is determined based on the labels of the first heartbeat data and/or the second heartbeat data, and the third heartbeat data with the label is determined as augmented data.
  • two divided heart beat signals may be randomly selected as heart beat A and heart beat B, respectively. This random selection is based on one of three methods:
  • a heart beat signal of a normal person is randomly selected as A, and a heart beat signal of a patient is selected as B.
  • the generalization degree of the data can be increased by combining the three methods and using them simultaneously.
  • the two selected signals can be spliced. For example, the directions of the R wave peaks of each lead of the heartbeat A and the heartbeat B are compared respectively. If the R wave peak directions of a certain lead of A and B are the same, the lead can be marked for splicing, otherwise the mark cannot be spliced. If all leads are marked for splicing, the splicing fails, and the heartbeat is reselected; if there is at least one lead that is marked for splicing, splicing can be performed.
  • the label of the third heart beat data can be set according to the label of the heart beat A; otherwise, the label of the third heart beat data can be set according to the label of the heart beat B.
  • the main detection position of myocardial infarction is in the ST segment (the segment after the R wave)
  • the ST segment of the augmented beat C is derived from the beat B, it is considered that the label of the beat C is the same as that of the beat B.
  • the label of the heartbeat C is set as suffering from myocardial infarction disease; if the heartbeat B is derived from the ECG data of a healthy person, the label of the heartbeat C is set as healthy.
  • the first cardiac beat data and the second cardiac beat data from the plurality of cardiac beat data; between at least one lead of the first cardiac beat data and the second cardiac beat data
  • splicing splicing the first heartbeat data and the second heartbeat data to generate third heartbeat data
  • acquiring the labels of the first heartbeat data and/or the second heartbeat data based on the first heartbeat data and/or Or the label of the second heartbeat data determines the label of the third heartbeat data, and determines that the third heartbeat data with the label is augmented data, so that data augmentation can be performed on the electrocardiogram data, and the data volume of the training set can be increased, and then
  • the machine learning model can be fully trained and the performance of the model can be improved.
  • the first heart beat data and the second heart beat data are spliced to generate a third heart beat data, including:
  • the following operations are performed for each lead, in the case that the R-wave directions of the first and second heartbeat data in the current lead are the same and there is a zero-crossing point within a predetermined time after the R-wave, based on the The position of the zero-crossing point is described, and the data of the first heart beat data and the second heart beat data under the current lead are spliced into the data of the temporary heart beat data under the current lead; otherwise, the data of the second heart beat data under the current lead It is determined as the temporary cardiac data in the current lead;
  • the temporary heart beat data is determined to be the third heart beat data.
  • FIG. 9 shows a flowchart of generating augmented data according to another embodiment of the present disclosure.
  • the method includes steps S901 to S910.
  • step S901 heart beat A and heart beat B are selected. Reference may be made to the above description about FIG. 8 , which will not be repeated here.
  • step S902 an unprocessed lead is selected.
  • step S903 in this lead, check whether the R-wave directions of the heartbeat A and the heartbeat B are the same, if so, go to step S904, otherwise go to step S907.
  • step S904 find the first zero-crossing point position within 100 ms after the R wave peak of the lead of the heartbeat A and the heartbeat B.
  • step S905 check whether the zero-crossing point is found for both A and B for the lead, if so, go to step S906, otherwise go to step S907.
  • step S906 the first half of the lead of the beat A is resampled and assigned to the first half of the beat C, and the second half of the lead of the beat B is assigned to the second half of the beat C, so as to obtain the lead data.
  • the first half of the lead of the heartbeat A is resampled and assigned to the first half of the heartbeat C, for example, the length of the first half of the heartbeat A may be the same as the length of the heartbeat B by horizontal compression or stretching.
  • the first half is the same length.
  • the first half segment refers to the data from the start position of the heart beat to the zero-crossing position mentioned above, and the second half refers to the data from the zero-crossing position to the ending position.
  • FIG. 10 shows a schematic diagram of a heartbeat C generated from a heartbeat A and a heartbeat B according to an embodiment of the present disclosure.
  • two heartbeat data, heartbeat A and heartbeat B can generate a new heartbeat data, heartbeat C.
  • Heartbeat C is different from Heartbeat A and Heartbeat B.
  • FIG. 11A shows a comparison diagram of beats B and C according to an embodiment of the present disclosure. As shown in FIG. 11A , the second half of the heartbeat C coincides with the heartbeat B.
  • FIG. 11B shows a comparison diagram of beats A and C according to an embodiment of the present disclosure. As shown in FIG.
  • the first half of the heartbeat C is the result of resampling (stretching) the first half of the heartbeat A.
  • FIG. FIG. 10 , FIG. 11A and FIG. 11B only show the data of one lead, and the data of other leads processed according to the method of this embodiment may have similar characteristics.
  • step S907 the lead of the beat B is directly assigned to the beat C.
  • step S908 check whether the processing of all leads is completed, if yes, execute step S909, otherwise return to step S902 to continue to select an unprocessed lead.
  • step S909 check whether each lead of the heartbeat C is the same as the heartbeat B, if so, discard the heartbeat C, and return to S901 to reselect the heartbeat, otherwise, go to step S910.
  • step S910 it is determined that the heart beat C is an augmented data.
  • step S910 After step S910 is executed, it is possible to return to step S901 to continue generating new augmented data.
  • the R waves of the first cardiac beat data and the second cardiac beat data in the current lead are in the same direction and at a predetermined time after the R wave.
  • the data of the first heartbeat data and the second heartbeat data under the current lead are spliced into the temporary heartbeat data under the current lead; otherwise, the The data of the second heartbeat data in the current lead is determined as the data of the temporary heartbeat data in the current lead; if the temporary heartbeat data is not exactly the same as the second heartbeat data, it is determined that the temporary heartbeat data is the first heartbeat data.
  • Three heart beat data so that the data of the electrocardiogram data can be augmented, the data volume of the training set can be increased, and the machine learning model can be fully trained, and the performance of the model can be improved.
  • the method of the embodiment of the present disclosure can improve the recognition accuracy on the basis of the existing deep learning method.
  • the problem of insufficient or unbalanced data in the training set is solved, and the problem of low recognition accuracy caused by insufficient or unbalanced data in the training set is also solved. the recognition accuracy of such data. Since the overall effective data volume is increased, when solving the problem of data imbalance, it not only improves the recognition accuracy of individual few data, but also improves the overall recognition accuracy of all data.
  • FIG. 12 shows a block diagram of an electrocardiogram data augmentation apparatus according to an embodiment of the present disclosure.
  • the apparatus may be realized by software, hardware or a combination of the two to become part or all of the electronic device.
  • the ECG data augmentation apparatus 1200 includes an acquisition module 1210 , a processing module 1220 and a generation module 1230 .
  • an acquisition module 1210 configured to acquire electrocardiogram data
  • a processing module 1220 configured to process the electrocardiogram data to obtain a plurality of heartbeat data
  • the generating module 1230 is configured to generate augmented data based on at least two heartbeat data in the plurality of heartbeat data.
  • an acquisition module is configured to acquire electrocardiogram data;
  • a processing module is configured to process the electrocardiogram data to acquire multiple heartbeat data;
  • At least two heartbeat data in the heartbeat data generate augmented data, which can perform data augmentation on the electrocardiogram data, increase the data volume of the training set, and then fully train the machine learning model and improve the performance of the model.
  • FIG. 13 shows a block diagram of the electronic device according to an embodiment of the present disclosure.
  • the electronic device 1300 includes a memory 1301 and a processor 1302, wherein the memory 1301 is used to store a program that supports the electronic device to perform the ECG data augmentation method or the code generation method in any of the above embodiments , the processor 1302 is configured to execute the program stored in the memory 1301 .
  • the memory 1301 is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor 1302 to implement the following steps:
  • Augmented data is generated based on at least two of the plurality of beat data.
  • the processing of the electrocardiogram data to obtain a plurality of heart beat data includes:
  • the preprocessing includes:
  • Multi-layer wavelet decomposition is performed on the electrocardiogram data through discrete wavelet transform, the approximate value of the lowest layer is set to zero, and the baseline-calibrated electrocardiogram data is obtained through discrete wavelet reconstruction.
  • the preprocessing includes:
  • Multi-layer wavelet decomposition is performed on the electrocardiogram data through stationary wavelet transformation, threshold filtering is performed on detail values, and denoised electrocardiogram data are obtained through stationary wavelet reconstruction.
  • performing heartbeat recognition on the preprocessed electrocardiogram data to obtain a plurality of heartbeat data includes:
  • the data of the first lead and/or the other leads are divided into a plurality of heart beat data according to the position of the peak of the R wave in the data of the first lead.
  • the generating augmented data based on at least two heartbeat data in the plurality of heartbeat data includes:
  • the label of the third cardiac data is determined based on the labels of the first cardiac data and/or the second cardiac data, and the third cardiac data with the label is determined as augmented data.
  • the first heart beat data and the second heart beat data are spliced to generate a third heart beat data, including:
  • the following operations are performed for each lead, in the case that the R-wave directions of the first and second heartbeat data in the current lead are the same and there is a zero-crossing point within a predetermined time after the R-wave, based on the The position of the zero-crossing point is described, and the data of the first heart beat data and the second heart beat data under the current lead are spliced into the data of the temporary heart beat data under the current lead; otherwise, the data of the second heart beat data under the current lead It is determined as the temporary cardiac data in the current lead;
  • the temporary heart beat data is determined to be the third heart beat data.
  • FIG. 14 shows a schematic structural diagram of a computer system suitable for implementing the electrocardiogram data augmentation method according to an embodiment of the present disclosure.
  • a computer system 1400 includes a processing unit 1401 that can execute the above-described implementation according to a program stored in a read only memory (ROM) 1402 or a program loaded from a storage section 1408 into a random access memory (RAM) 1403 various treatments in the example.
  • ROM read only memory
  • RAM random access memory
  • various programs and data required for the operation of the system 1400 are also stored.
  • the processing unit 1401, the ROM 1402, and the RAM 1403 are connected to each other through a bus 1404.
  • An input/output (I/O) interface 1405 is also connected to bus 1404 .
  • the following components are connected to the I/O interface 1405: an input section 1406 including a keyboard, a mouse, etc.; an output section 1407 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1408 including a hard disk, etc. ; and a communication section 1409 including a network interface card such as a LAN card, a modem, and the like.
  • the communication section 1409 performs communication processing via a network such as the Internet.
  • Drivers 1410 are also connected to I/O interface 1405 as needed.
  • a removable medium 1411 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is mounted on the drive 1410 as needed so that a computer program read therefrom is installed into the storage section 1408 as needed.
  • the processing unit 1401 may be implemented as a processing unit such as a CPU, a GPU, a TPU, an FPGA, and an NPU.
  • embodiments of the present disclosure include a computer program product comprising a computer program tangibly embodied on a readable medium thereof, the computer program containing program code for performing the above-described method.
  • the computer program may be downloaded and installed from the network via the communication portion 1409, and/or installed from the removable medium 1411.
  • each block in the flowchart or block diagrams may represent a module, segment, or portion of code that contains one or more functions for implementing the specified logical function(s) executable instructions.
  • the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved.
  • each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations can be implemented in dedicated hardware-based systems that perform the specified functions or operations , or can be implemented in a combination of dedicated hardware and computer instructions.
  • the units or modules involved in the embodiments of the present disclosure may be implemented in a software manner, or may be implemented in a programmable hardware manner.
  • the described units or modules may also be provided in the processor, and the names of these units or modules do not constitute a limitation on the units or modules themselves in certain circumstances.
  • the present disclosure also provides a computer-readable storage medium
  • the computer-readable storage medium may be a computer-readable storage medium included in the electronic device or computer system in the above-mentioned embodiments; it may also exist independently , a computer-readable storage medium that does not fit into a device.
  • the computer-readable storage medium stores one or more programs used by one or more processors to perform the methods described in the present disclosure.

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Surgery (AREA)
  • Pathology (AREA)
  • Veterinary Medicine (AREA)
  • Molecular Biology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Public Health (AREA)
  • General Health & Medical Sciences (AREA)
  • Animal Behavior & Ethology (AREA)
  • Physiology (AREA)
  • Psychiatry (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Power Engineering (AREA)
  • General Physics & Mathematics (AREA)
  • Fuzzy Systems (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

一种心电图数据增广方法、装置（1200）、电子设备（1300）及介质，该心电图数据增广方法，包括：获取心电图数据（S101）；处理心电图数据以获取多个心拍数据（S102）；以及基于多个心拍数据中的至少两个心拍数据生成增广数据（S103），从而能够对心电图数据进行数据增广，可增加训练集的数据量，进而可充分地训练机器学习的模型，提高模型的表现。

Description

心电图数据增广方法、装置、电子设备及介质
相关申请的交叉引用
本公开要求2020年8月21日提交的中国专利申请号为“CN202010850174.2”的优先权，其全部内容作为整体并入本申请中。
技术领域
本公开涉及医疗技术领域，具体涉及一种心电图数据增广方法、装置、电子设备及介质。
背景技术
数据增广是深度学习中常用的技巧之一，主要用于增加训练数据集，让数据集尽可能的多样化，使得训练的模型具有更强的泛化能力。在实际应用中，并非所有的增广方式都适用当前的训练数据，需要根据数据集特征来确定应该使用哪几种数据增广方式。目前数据增广主要包括：水平/垂直翻转、旋转、缩放、裁剪、剪切、平移、对比度、色彩抖动、噪声等。
心电图是利用心电图机从体表记录心脏每一心动周期所产生的电活动变化图形，其中每一小段信号都具有一定的医学意义。而使用常规的数据增广方法诸如翻转、旋转、裁剪等方法，则会破坏心电图的医学意义，无法对机器学习模型的训练起到积极作用，普通图像处理的数据增广手段无法适用于心电信号的数据增广。
发明内容
为了解决相关技术中的问题，本公开实施例提供一种心电图数据增广方法、装置、电子设备及介质。
第一方面，本公开实施例中提供了一种心电图数据增广方法，包括：
获取心电图数据；
处理所述心电图数据以获取多个心拍数据；以及
基于所述多个心拍数据中的至少两个心拍数据生成增广数据。
结合第一方面，本公开在第一方面的第一种实现方式中，所述处理所述 心电图数据以获取多个心拍数据，包括：
对所述心电图数据进行预处理；
对经过预处理的心电图数据进行心拍识别以获取多个心拍数据。
结合第一方面的第一种实现方式，本公开在第一方面的第二种实现方式中，所述预处理包括：
通过离散小波变换对所述心电图数据进行多层小波分解，将最低一层的近似值置零，通过离散小波重构以获取基线校准后的心电图数据。
结合第一方面的第一种或第二种实现方式，本公开在第一方面的第三种实现方式中，所述预处理包括：
通过平稳小波变换对所述心电图数据进行多层小波分解，对细节值进行阈值滤波，通过平稳小波重构以获取降噪后的心电图数据。
结合第一方面的第一种实现方式，本公开在第一方面的第四种实现方式中，所述对经过预处理的心电图数据进行心拍识别以获取多个心拍数据，包括：
对经过预处理的心电图数据中的第一导联的数据，通过平稳小波变换进行多层小波分解，获得分解结果；
对所述分解结果进行特征识别，确定所述第一导联的数据中的R波的波峰的位置；
根据所述第一导联的数据中的R波的波峰的位置，将所述第一导联和/或其他导联的数据分割为多个心拍数据。
结合第一方面，本公开在第一方面的第五种实现方式中，所述基于所述多个心拍数据中的至少两个心拍数据生成增广数据包括：
从所述多个心拍数据中确定第一心拍数据和第二心拍数据；
在所述第一心拍数据和第二心拍数据的至少一个导联之间可拼接的情况下，拼接所述第一心拍数据和第二心拍数据以生成第三心拍数据；
获取所述第一心拍数据和/或第二心拍数据的标签；
基于所述第一心拍数据和/或第二心拍数据的标签确定所述第三心拍数据的标签，确定带有标签的第三心拍数据为增广数据。
结合第一方面的第五种实现方式，本公开在第一方面的第六种实现方式中，所述在所述第一心拍数据和第二心拍数据的至少一个导联之间可拼接的 情况下，拼接所述第一心拍数据和第二心拍数据以生成第三心拍数据，包括：
对于每个导联执行以下操作，在所述第一心拍数据和第二心拍数据在当前导联下的R波方向相同且在R波后的预定时间内都存在过零点的情况下，基于所述过零点的位置，将第一心拍数据和第二心拍数据在当前导联下的数据拼接为临时心拍数据在当前导联下的数据；否则，将第二心拍数据在当前导联下的数据确定为临时心拍数据在当前导联下的数据；
在所述临时心拍数据与第二心拍数据不完全相同的情况下，确定所述临时心拍数据为第三心拍数据。
第二方面，本公开实施例提供了一种心电图数据增广装置，包括：
获取模块，被配置为获取心电图数据；
处理模块，被配置为处理所述心电图数据以获取多个心拍数据；以及
生成模块，被配置为基于所述多个心拍数据中的至少两个心拍数据生成增广数据。
第三方面，本公开实施例提供了一种电子设备，包括存储器和处理器，其中，所述存储器用于存储一条或多条计算机指令，其中，所述一条或多条计算机指令被所述处理器执行以实现如第一方面、第一方面的第一种至第六种实现方式任一项所述的方法。
第四方面，本公开实施例中提供了一种计算机可读存储介质，其上存储有计算机指令，该计算机指令被处理器执行时实现如第一方面、第一方面的第一种至第六种实现方式任一项所述的方法。
根据本公开实施例提供的技术方案，通过获取心电图数据；处理所述心电图数据以获取多个心拍数据；以及基于所述多个心拍数据中的至少两个心拍数据生成增广数据，从而能够对心电图数据进行数据增广，可增加训练集的数据量，进而可充分地训练机器学习的模型，提高模型的表现。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，并不能限制本公开。
附图说明
结合附图，通过以下非限制性实施方式的详细描述，本公开的其它特征、目的和优点将变得更加明显。在附图中：
图1示出根据本公开实施例的心电图数据增广方法的流程图；
图2示出根据本公开实施例的心电图数据的示意图；
图3示出根据本公开实施例的基线校准后的心电图数据的示意图；
图4示出根据本公开实施例的降噪处理后的心电图数据的示意图；
图5示出根据本公开实施例的心拍识别的流程图；
图6A和图6B示出根据本公开实施例的R波反向的示意图；
图7示出根据本公开实施例的心拍识别的示意图；
图8示出根据本公开实施例的生成增广数据的流程图；
图9示出根据本公开另一实施例的生成增广数据的流程图；
图10示出根据本公开实施例的根据心拍A和心拍B生成的心拍C的示意图；
图11A示出根据本公开实施例的心拍B和心拍C的对照图；
图11B示出根据本公开实施例的心拍A和心拍C的对照图；
图12示出根据本公开实施例的心电图数据增广装置的框图；
图13示出根据本公开实施例的电子设备的框图；以及
图14示出根据本公开实施例的适于实现心电图数据增广方法的计算机系统的结构示意图。
具体实施方式
下文中，将参考附图详细描述本公开的示例性实施例，以使本领域技术人员可容易地实现它们。此外，为了清楚起见，在附图中省略了与描述示例性实施例无关的部分。
在本公开中，应理解，诸如“包括”或“具有”等的术语旨在指示本说明书中所公开的特征、数字、步骤、行为、部件、部分或其组合的存在，并且不欲排除一个或多个其他特征、数字、步骤、行为、部件、部分或其组合存在或被添加的可能性。
另外还需要说明的是，在不冲突的情况下，本公开中的实施例及实施例中的特征可以相互组合。下面将参考附图并结合实施例来详细说明本公开。
一个带有标签的数据集的构建需要付出巨大的工作量。同时，由于不同疾病在人群中的发病率是不一样的，医学临床上直接采集的数据集，常常有 类别数据量不均衡的问题。数据集的数据量不足或者不同类别的数据量不均衡都会机器学习算法的表现欠佳，比如分类的准确率低，召回率低等。
心电图是利用心电图机从体表记录心脏每一心动周期所产生的电活动变化图形，其中每一小段信号都具有一定的医学意义。而使用常规的数据增广方法诸如翻转、旋转、裁剪等方法，则会破坏心电图的医学意义，无法对机器学习模型的训练起到积极作用，普通图像处理的数据增广手段无法适用于心电信号的数据增广。
根据本公开实施例提供的技术方案，通过获取心电图数据；处理所述心电图数据以获取多个心拍数据；以及基于所述多个心拍数据中的至少两个心拍数据生成增广数据，从而能够对心电图数据进行数据增广，可增加训练集的数据量，进而可充分地训练机器学习的模型，提高模型的表现。
图1示出根据本公开实施例的心电图数据增广方法的流程图。
如图1所示，该方法包括步骤S101～S103。
在步骤S101中，获取心电图数据；
在步骤S102中，处理所述心电图数据以获取多个心拍数据；以及
在步骤S103中，基于所述多个心拍数据中的至少两个心拍数据生成增广数据。
根据本公开实施例提供的技术方案，通过获取心电图数据；处理所述心电图数据以获取多个心拍数据；以及基于所述多个心拍数据中的至少两个心拍数据生成增广数据，从而能够对心电图数据进行数据增广，可增加训练集的数据量，进而可充分地训练机器学习的模型，提高模型的表现。
图2示出根据本公开实施例的心电图数据的示意图。如图2所示，心电图数据被连续采集，通常包括多个心拍，从而具有一定的周期性特点。其中，在每个心拍中存在一个最大的波峰称为R波，R波特征突出，因而可以作为分割心拍的基础。
在心电图的专业术语中，将记录心电图时电极在人体体表的放置位置及电极与放大器的连接方式称为心电图的导联。常用的导联为12个，分别为第Ⅰ导联、第Ⅱ导联、第Ⅲ导联、aVR导联、aVL导联、aVF导联、V1导联、V2导联、V3导联、V4导联、V5导联以及V6导联。所有导联的心电信号都是几乎同时测得的，它们在R波分布密集程度、位置等方面都几乎相差无几。 图2所示意的心电图数据仅为一个导联的数据，其他导联的数据可以与之类似。
根据本公开实施例，所述处理所述心电图数据以获取多个心拍数据，包括，对所述心电图数据进行预处理，以及对经过预处理的心电图数据进行心拍识别以获取多个心拍数据。
根据本公开实施例提供的技术方案，通过对所述心电图数据进行预处理；对经过预处理的心电图数据进行心拍识别以获取多个心拍数据，从而能够对心电图数据进行数据增广，可增加训练集的数据量，进而可充分地训练机器学习的模型，提高模型的表现。
根据本公开实施例，所述预处理包括通过离散小波变换对所述心电图数据进行多层小波分解，将最低一层的近似值置零，通过离散小波重构以获取基线校准后的心电图数据。
小波变换是一种进行信号时频分析和处理的理想工具，计算机中常用的小波变换算法为mallat算法，这种算法的核心思想是使用滤波器来执行小波变换。
原始输入信号S通过两个互补的滤波器(低通滤波器和高通滤波器)产生A和D两个信号，A为信号的近似值(低频滤波器所获得的值)，D为信号的细节值(高频滤波器所获得的值)。原始信号通过这样的一对滤波器进行的分解叫做一级分解。信号分解过程可以迭代，也就是说可以进行多级分解。若对高频分量不再分解，而对低频分量连续进行分解，就可以得到更低频的分量，形成小波分解树，可以根据需要确定分解级数。
本公开实施例采用离散小波变换(DWT)，在每一级分解后对数据进行降采样，因此数据量较小，计算较快。随着层数的不断增高，得到的低频分量分辨率逐渐降低。与之相对的，离散小波重构算法(IDWT)则包含了升采样和滤波两个过程，其中升采样即为在降采样数据之间插入0。
在消除基线漂移的过程中所使用的小波为db5小波。在使用DWT对信号进行九层小波分解后，将最低一层的近似值置零，再进行IDWT，即可获得消除基线漂移后的心电信号，如图3所示。
根据本公开实施例提供的技术方案，通过离散小波变换对所述心电图数据进行多层小波分解，将最低一层的近似值置零，通过离散小波重构以获取 基线校准后的心电图数据，从而能够较好地对心电图数据进行基线校准。
根据本公开实施例，预处理过程除了基线校准外，还可以进行降噪处理。心电图数据集中包括肌电干扰和50/60Hz工频噪声，例如可以使用小波变换阈值法消除心电信号的高频噪声。
根据本公开实施例，所述预处理包括通过平稳小波变换对所述心电图数据进行多层小波分解，对细节值进行阈值滤波，通过平稳小波重构以获取降噪后的心电图数据。
根据本公开实施例，采用平稳小波变换(SWT)消除噪声。SWT与DWT最大的区别在于其在每一层分解后不进行降采样，因此数据量较大，计算量较大，但每一层的分辨率都保持不变。对于去除高频噪声而言，并不是简单地将某一层的信号完全删除，而是进行阈值删除，因此分辨率就显得尤为重要。同理，在平稳小波重构(ISWT)的过程中，也不需要进行升采样。
在消除高频噪声的过程中所使用的小波为bior2.6小波。在使用SWT对信号进行6层小波分解后，首先将前两层频率最高的细节值置0，这些细节值所代表的频率过高，几乎不包含任何心电信号信息。然后对3到6层细节值进行阈值滤波，阈值的选取使用以下公式：
Figure PCTCN2020137330-appb-000001
其中d为细节值，N为其长度，mid为中值，使用的阈值方法为半软阈值法，既不会像硬阈值法一样产生小波系数突变，也不会像软阈值法一样产生偏差。半软阈值法的公式如下：
Figure PCTCN2020137330-appb-000002
其中取λ 1＝λ th，λ 2＝1.25λ th。在进行阈值滤波后再进行ISWT，即可去除高频噪声，降噪处理后的心电图数据如图4所示。
根据本公开实施例提供的技术方案，通过平稳小波变换对所述心电图数 据进行多层小波分解，对细节值进行阈值滤波，通过平稳小波重构以获取降噪后的心电图数据，从而能够提高心电图数据的降噪效果。
图5示出根据本公开实施例的心拍识别的流程图。
如图5所示，该方法包括步骤S501～S503。
在步骤S501中，对经过预处理的心电图数据中的第一导联的数据，通过平稳小波变换进行多层小波分解，获得分解结果；
在步骤S502中，对所述分解结果进行特征识别，确定所述第一导联的数据中的R波的波峰的位置；
在步骤S503中，根据所述第一导联的数据中的R波的波峰的位置，将所述第一导联和/或其他导联的数据分割为多个心拍数据。
根据本公开实施例，在步骤S501，可以使用SWT对信号进行小波变换，使用的小波例如可以是：
Figure PCTCN2020137330-appb-000003
根据本公开实施例，在步骤S502，可以选取变换后的第五层细节值，该层的R波的特征最明显。对于方向向上的R波，在该层将会以负-正极大值对的形态呈现；对于方向向下的R波，在该层将会以正-负极大值对的形态呈现。找出这些极大值对并找出它们过零点的位置，对应的R波峰就为过零点±0.05s内的极大/极小值。
即使是在正常人的心电图中，每个导联R波峰的方向都不固定。而病变后的心电图更是会出现R波峰方向与正常相反的情况。图6A和图6B示出根据本公开实施例的R波反向的示意图，其中，图6A示出了一种病理性导致的R波倒置的情况的示意图，图6B示出了一种由于电极接反导致的R波倒置的示意图。
实际操作中，对同一个人同时测得的心电信号的12个导联(这12个导联分别为：第Ⅰ导联、第Ⅱ导联、第Ⅲ导联、aVR导联、aVL导联、aVF导联、V1导联、V2导联、V3导联、V4导联、V5导联、V6导联；下文中将使用上面列举的12个导联的提及顺序来指代每一个导联，例如第一导联指第Ⅰ导联，第七导联指V1导联)分别进行单独的去除基线和滤波操作，然后 只对第一导联进行R波识别，并使用第一导联的R波识别结果来对其他所有导联进行切割。这么做的原因是所有导联的心电信号都是几乎同时测得的，因此它们在R波分布密集程度、位置等方面都几乎相差无几。
分割后还要对除第一导联以外的11个导联进行R波位置校准，因为人体相邻细胞之间的电位传导需要一定时间，因此这12个导联的R波峰位置可能会有最大±0.05s的差异。对其他的11个导联分别进行如下操作：在第一导联的R波峰的位置附近的±0.05s的范围内寻找最大值和最小值，若最大值小于最小值的绝对值的三分之一，则认为最小值所在的位置为该导联的R波峰，否则认为最大值所在的位置为该导联的R波峰。因此，本公开实施例的方法还可以包括，根据所述第一导联的数据中的R波的波峰的位置，对其他导联的数据中的R波的波峰位置进行校准。
根据本公开实施例，在切割时，由于第一个和最后一个心拍信号可能不完整，可以抛弃第一个和最后一个心拍信号，并取相邻两个心拍的R峰的中点作为切割点。用这种方法可以准确地将一个完整的心电信号切成一个一个的心拍。分割后的结果如图7所示。
根据本公开实施例提供的技术方案，通过对经过预处理的心电图数据中的第一导联的数据，通过平稳小波变换进行多层小波分解，获得分解结果；对所述分解结果进行特征识别，确定所述第一导联的数据中的R波的波峰的位置；根据所述第一导联的数据中的R波的波峰的位置，将所述第一导联和/或其他导联的数据分割为多个心拍数据，从而能够准确识别心拍数据，有利于对心电图数据进行数据增广，可增加训练集的数据量，进而可充分地训练机器学习的模型，提高模型的表现。
图8示出根据本公开实施例的生成增广数据的流程图。
如图8所示，该方法包括步骤S801～S804。
在步骤S801中，从所述多个心拍数据中确定第一心拍数据和第二心拍数据；
在步骤S802中，在所述第一心拍数据和第二心拍数据的至少一个导联之间可拼接的情况下，拼接所述第一心拍数据和第二心拍数据以生成第三心拍数据；
在步骤S803中，获取所述第一心拍数据和/或第二心拍数据的标签；
在步骤S804中，基于所述第一心拍数据和/或第二心拍数据的标签确定所述第三心拍数据的标签，确定带有标签的第三心拍数据为增广数据。
根据本公开实施例，可以随机选择两个分割后的心拍信号(各个导联)，分别作为心拍A和心拍B。该随机选择基于以下三种方法中的一种：
(1)随机选择同一个病人不同时间的两个心拍信号作为A和B；
(2)随机选择同一类疾病的两个不同病人的两个心拍信号作为A和B；
(3)随机选择一个正常人的心拍信号作为A，再选择一个病人的心拍信号作为B。
根据本公开实施例，可以通过将三种方法结合起来同时使用来增加数据的泛化程度。
此外，还可以初步检测选取的两个信号能否进行拼接。例如，分别对比心拍A和心拍B的各个导联的R波峰方向，若A和B的某一导联R波峰方向相同则标记该导联可拼接，否则标记不可拼接。若所有导联都被标记不可拼接，则此次拼接失败，重新选择心拍；若存在至少一个导联被标记可拼接，则可以进行拼接。
根据本公开实施例，对于上述三种方法，确定用于检测病变的主要特征数据所处的位置，判断第三心拍数据上该位置的数据来自于心拍A或是心拍B，若来自于心拍A，则根据心拍A的标签设置第三心拍数据的标签；反之，可以根据心拍B的标签设置第三心拍数据的标签。例如，因为心梗病变的主要检测位置在ST段(R波之后的片段)，而若增广出的心拍C的ST段来源于心拍B，因此认为心拍C的标签与心拍B相同。例如，如果心拍B来源于心梗病人的心电图数据，则设置心拍C的标签为患有心梗疾病，如果心拍B来源于健康的人的心电图数据，则设置心拍C的标签为健康。
根据本公开实施例提供的技术方案，通过从所述多个心拍数据中确定第一心拍数据和第二心拍数据；在所述第一心拍数据和第二心拍数据的至少一个导联之间可拼接的情况下，拼接所述第一心拍数据和第二心拍数据以生成第三心拍数据；获取所述第一心拍数据和/或第二心拍数据的标签；基于所述第一心拍数据和/或第二心拍数据的标签确定所述第三心拍数据的标签，确定带有标签的第三心拍数据为增广数据，从而能够对心电图数据进行数据增广， 可增加训练集的数据量，进而可充分地训练机器学习的模型，提高模型的表现。
根据本公开实施例，所述在所述第一心拍数据和第二心拍数据的至少一个导联之间可拼接的情况下，拼接所述第一心拍数据和第二心拍数据以生成第三心拍数据，包括：
对于每个导联执行以下操作，在所述第一心拍数据和第二心拍数据在当前导联下的R波方向相同且在R波后的预定时间内都存在过零点的情况下，基于所述过零点的位置，将第一心拍数据和第二心拍数据在当前导联下的数据拼接为临时心拍数据在当前导联下的数据；否则，将第二心拍数据在当前导联下的数据确定为临时心拍数据在当前导联下的数据；
在所述临时心拍数据与第二心拍数据不完全相同的情况下，确定所述临时心拍数据为第三心拍数据。
下面参照图9、图10、图11A和图11B对本公开实施例的方法进行示例性说明。
图9示出根据本公开另一实施例的生成增广数据的流程图。
如图9所示，该方法包括步骤S901～S910。
在步骤S901，选择心拍A和心拍B。可以参照上文关于图8的描述，此处不再赘述。
在步骤S902，选择一个未处理的导联。
在步骤S903，在该导联下，心拍A和心拍B的R波方向是否相同，若是，则继续执行步骤S904，否则执行步骤S907。
在步骤S904，寻找心拍A和心拍B该导联的R波峰后100ms内的第一个过零点位置。
在步骤S905，对该导联，A和B是否都找到过零点，若是，则继续执行步骤S906，否则执行步骤S907。
实际中偶尔会有特殊的波形病变无法找到过零点(如ST段上移)，对于此类波形，甚至无法找到能够与之能拼接的其他波形，因此，在拼接前避开此类波形不失为一种较好的处理方式。
在步骤S906，将心拍A的该导联的前半段重采样并赋值给心拍C的前半段，将心拍B的该导联的后半段赋值给心拍C的后半段，得到心拍C在该导 联下的数据。
根据本公开实施例，将心拍A的该导联的前半段重采样并赋值给心拍C的前半段，例如可以是通过水平压缩或拉伸的方式使得心拍A的前半段的长度与心拍B的前半段的长度一致。其中，前半段是指由心拍起始位置至上文所述过零点位置的数据，后半段是指由过零点位置至结束位置的数据。
图10示出根据本公开实施例的根据心拍A和心拍B生成的心拍C的示意图。如图10所示，两个心拍数据心拍A和心拍B可以生成一个新的心拍数据心拍C。心拍C与心拍A和心拍B都不同。图11A示出根据本公开实施例的心拍B和心拍C的对照图。如图11A所示，心拍C的后半段与心拍B一致。图11B示出根据本公开实施例的心拍A和心拍C的对照图。如图11B所示，心拍C的前半段为心拍A的前半段重采样(拉伸)之后的结果。图10、图11A和图11B所示仅为一个导联的数据，根据本实施例的方法处理的其他导联的数据可以具有类似的特点。
返回参考图9。在步骤S907，将心拍B的该导联直接赋值给心拍C。
在步骤S908，是否所有导联处理完成，若是，则执行步骤S909，否则返回步骤S902继续选择一个未处理的导联。
在步骤S909，是否心拍C的每一个导联都与心拍B相同，若是，则丢弃心拍C，返回S901重新选择心拍，否则，执行步骤S910。
在步骤S910，确定心拍C为一个增广数据。
在步骤S910执行之后，可以返回步骤S901继续生成新的增广数据。
根据本公开实施例提供的技术方案，通过对于每个导联执行以下操作，在所述第一心拍数据和第二心拍数据在当前导联下的R波方向相同且在R波后的预定时间内都存在过零点的情况下，基于所述过零点的位置，将第一心拍数据和第二心拍数据在当前导联下的数据拼接为临时心拍数据在当前导联下的数据；否则，将第二心拍数据在当前导联下的数据确定为临时心拍数据在当前导联下的数据；在所述临时心拍数据与第二心拍数据不完全相同的情况下，确定所述临时心拍数据为第三心拍数据，从而能够对心电图数据进行数据增广，可增加训练集的数据量，进而可充分地训练机器学习的模型，提高模型的表现。
本公开实施例的方法可以在现有的深度学习方法基础上提升识别准确率。 在将心电信号进行增广后，由于解决了训练集数据不足或数据不均衡的问题，同时解决了由于训练集数据不足或数据不均衡而导致的识别准确率不高的问题，大幅度提升了这类数据的识别准确率。由于提升了总体的有效数据量，因此在解决数据不均衡问题时，其在提升个别较少的数据识别准确率的同时，也提升了对所有数据的总体识别准确率。
图12示出根据本公开实施例的心电图数据增广装置的框图。其中，该装置可以通过软件、硬件或者两者的结合实现成为电子设备的部分或者全部。
如图12所示，所述心电图数据增广装置1200包括获取模块1210、处理模块1220和生成模块1230。
获取模块1210，被配置为获取心电图数据；
处理模块1220，被配置为处理所述心电图数据以获取多个心拍数据；以及
生成模块1230，被配置为基于所述多个心拍数据中的至少两个心拍数据生成增广数据。
根据本公开实施例提供的技术方案，通过获取模块，被配置为获取心电图数据；处理模块，被配置为处理所述心电图数据以获取多个心拍数据；生成模块，被配置为基于所述多个心拍数据中的至少两个心拍数据生成增广数据，能够对心电图数据进行数据增广，可增加训练集的数据量，进而可充分地训练机器学习的模型，提高模型的表现。
本公开还公开了一种电子设备，图13示出根据本公开的实施例的电子设备的框图。
如图13所示，所述电子设备1300包括存储器1301和处理器1302，其中，所述存储器1301用于存储支持电子设备执行上述任一实施例中的心电图数据增广方法或代码生成方法的程序，所述处理器1302被配置为用于执行所述存储器1301中存储的程序。
根据本公开实施例，所述存储器1301用于存储一条或多条计算机指令，其中，所述一条或多条计算机指令被所述处理器1302执行以实现以下步骤：
获取心电图数据；
处理所述心电图数据以获取多个心拍数据；以及
基于所述多个心拍数据中的至少两个心拍数据生成增广数据。
根据本公开实施例，所述处理所述心电图数据以获取多个心拍数据，包括：
对所述心电图数据进行预处理；
对经过预处理的心电图数据进行心拍识别以获取多个心拍数据。
根据本公开实施例，所述预处理包括：
通过离散小波变换对所述心电图数据进行多层小波分解，将最低一层的近似值置零，通过离散小波重构以获取基线校准后的心电图数据。
根据本公开实施例，所述预处理包括：
通过平稳小波变换对所述心电图数据进行多层小波分解，对细节值进行阈值滤波，通过平稳小波重构以获取降噪后的心电图数据。
根据本公开实施例，所述对经过预处理的心电图数据进行心拍识别以获取多个心拍数据，包括：
对经过预处理的心电图数据中的第一导联的数据，通过平稳小波变换进行多层小波分解，获得分解结果；
对所述分解结果进行特征识别，确定所述第一导联的数据中的R波的波峰的位置；
根据所述第一导联的数据中的R波的波峰的位置，将所述第一导联和/或其他导联的数据分割为多个心拍数据。
根据本公开实施例，所述基于所述多个心拍数据中的至少两个心拍数据生成增广数据包括：
从所述多个心拍数据中确定第一心拍数据和第二心拍数据；
在所述第一心拍数据和第二心拍数据的至少一个导联之间可拼接的情况下，拼接所述第一心拍数据和第二心拍数据以生成第三心拍数据；
获取所述第一心拍数据和/或第二心拍数据的标签；
基于所述第一心拍数据和/或第二心拍数据的标签确定所述第三心拍数据的标签，确定带有标签的第三心拍数据为增广数据。
根据本公开实施例，所述在所述第一心拍数据和第二心拍数据的至少一个导联之间可拼接的情况下，拼接所述第一心拍数据和第二心拍数据以生成第三心拍数据，包括：
对于每个导联执行以下操作，在所述第一心拍数据和第二心拍数据在当 前导联下的R波方向相同且在R波后的预定时间内都存在过零点的情况下，基于所述过零点的位置，将第一心拍数据和第二心拍数据在当前导联下的数据拼接为临时心拍数据在当前导联下的数据；否则，将第二心拍数据在当前导联下的数据确定为临时心拍数据在当前导联下的数据；
在所述临时心拍数据与第二心拍数据不完全相同的情况下，确定所述临时心拍数据为第三心拍数据。
图14示出适于用来实现根据本公开实施例的心电图数据增广方法的计算机系统的结构示意图。
如图14所示，计算机系统1400包括处理单元1401，其可以根据存储在只读存储器(ROM)1402中的程序或者从存储部分1408加载到随机访问存储器(RAM)1403中的程序而执行上述实施例中的各种处理。在RAM 1403中，还存储有系统1400操作所需的各种程序和数据。处理单元1401、ROM 1402以及RAM 1403通过总线1404彼此相连。输入/输出(I/O)接口1405也连接至总线1404。
以下部件连接至I/O接口1405：包括键盘、鼠标等的输入部分1406；包括诸如阴极射线管(CRT)、液晶显示器(LCD)等以及扬声器等的输出部分1407；包括硬盘等的存储部分1408；以及包括诸如LAN卡、调制解调器等的网络接口卡的通信部分1409。通信部分1409经由诸如因特网的网络执行通信处理。驱动器1410也根据需要连接至I/O接口1405。可拆卸介质1411，诸如磁盘、光盘、磁光盘、半导体存储器等等，根据需要安装在驱动器1410上，以便于从其上读出的计算机程序根据需要被安装入存储部分1408。其中，所述处理单元1401可实现为CPU、GPU、TPU、FPGA、NPU等处理单元。
特别地，根据本公开的实施例，上文描述的方法可以被实现为计算机软件程序。例如，本公开的实施例包括一种计算机程序产品，其包括有形地包含在及其可读介质上的计算机程序，所述计算机程序包含用于执行上述方法的程序代码。在这样的实施例中，该计算机程序可以通过通信部分1409从网络上被下载和安装，和/或从可拆卸介质1411被安装。
附图中的流程图和框图，图示了按照本公开各种实施例的系统、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段或代码的一部分，所述模块、 程序段或代码的一部分包含一个或多个用于实现规定的逻辑功能的可执行指令。也应当注意，在有些作为替换的实现中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个接连地表示的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图和/或流程图中的每个方框、以及框图和/或流程图中的方框的组合，可以用执行规定的功能或操作的专用的基于硬件的系统来实现，或者可以用专用硬件与计算机指令的组合来实现。
描述于本公开实施例中所涉及到的单元或模块可以通过软件的方式实现，也可以通过可编程硬件的方式来实现。所描述的单元或模块也可以设置在处理器中，这些单元或模块的名称在某种情况下并不构成对该单元或模块本身的限定。
作为另一方面，本公开还提供了一种计算机可读存储介质，该计算机可读存储介质可以是上述实施例中电子设备或计算机系统中所包含的计算机可读存储介质；也可以是单独存在，未装配入设备中的计算机可读存储介质。计算机可读存储介质存储有一个或者一个以上程序，所述程序被一个或者一个以上的处理器用来执行描述于本公开的方法。
以上描述仅为本公开的较佳实施例以及对所运用技术原理的说明。本领域技术人员应当理解，本公开中所涉及的发明范围，并不限于上述技术特征的特定组合而成的技术方案，同时也应涵盖在不脱离所述发明构思的情况下，由上述技术特征或其等同特征进行任意组合而形成的其它技术方案。例如上述特征与本公开中公开的(但不限于)具有类似功能的技术特征进行互相替换而形成的技术方案。

Claims (10)

  1. 一种心电图数据增广方法，包括：
    获取心电图数据；
    处理所述心电图数据以获取多个心拍数据；以及
    基于所述多个心拍数据中的至少两个心拍数据生成增广数据。
  2. 根据权利要求1所述的方法，其特征在于，所述处理所述心电图数据以获取多个心拍数据，包括：
    对所述心电图数据进行预处理；
    对经过预处理的心电图数据进行心拍识别以获取多个心拍数据。
  3. 根据权利要求2所述的方法，其特征在于，所述预处理包括：
    通过离散小波变换对所述心电图数据进行多层小波分解，将最低一层的近似值置零，通过离散小波重构以获取基线校准后的心电图数据。
  4. 根据权利要求2或3所述的方法，其特征在于，所述预处理包括：
    通过平稳小波变换对所述心电图数据进行多层小波分解，对细节值进行阈值滤波，通过平稳小波重构以获取降噪后的心电图数据。
  5. 根据权利要求2所述的方法，其特征在于，所述对经过预处理的心电图数据进行心拍识别以获取多个心拍数据，包括：
    对经过预处理的心电图数据中的第一导联的数据，通过平稳小波变换进行多层小波分解，获得分解结果；
    对所述分解结果进行特征识别，确定所述第一导联的数据中的R波的波峰的位置；
    根据所述第一导联的数据中的R波的波峰的位置，将所述第一导联和/或其他导联的数据分割为多个心拍数据。
  6. 根据权利要求1所述的方法，其特征在于，所述基于所述多个心拍数据中的至少两个心拍数据生成增广数据包括：
    从所述多个心拍数据中确定第一心拍数据和第二心拍数据；
    在所述第一心拍数据和第二心拍数据的至少一个导联之间可拼接的情况下，拼接所述第一心拍数据和第二心拍数据以生成第三心拍数据；
    获取所述第一心拍数据和/或第二心拍数据的标签；
    基于所述第一心拍数据和/或第二心拍数据的标签确定所述第三心拍数 据的标签，确定带有标签的第三心拍数据为增广数据。
  7. 根据权利要求6所述的方法，其特征在于，所述在所述第一心拍数据和第二心拍数据的至少一个导联之间可拼接的情况下，拼接所述第一心拍数据和第二心拍数据以生成第三心拍数据，包括：
    对于每个导联执行以下操作，在所述第一心拍数据和第二心拍数据在当前导联下的R波方向相同且在R波后的预定时间内都存在过零点的情况下，基于所述过零点的位置，将第一心拍数据和第二心拍数据在当前导联下的数据拼接为临时心拍数据在当前导联下的数据；否则，将第二心拍数据在当前导联下的数据确定为临时心拍数据在当前导联下的数据；
    在所述临时心拍数据与第二心拍数据不完全相同的情况下，确定所述临时心拍数据为第三心拍数据。
  8. 一种心电图数据增广装置，包括：
    获取模块，被配置为获取心电图数据；
    处理模块，被配置为处理所述心电图数据以获取多个心拍数据；以及
    生成模块，被配置为基于所述多个心拍数据中的至少两个心拍数据生成增广数据。
  9. 一种电子设备，其特征在于，包括存储器和处理器；其中，所述存储器用于存储一条或多条计算机指令，其中，所述一条或多条计算机指令被所述处理器执行以实现如权利要求1～7中任一项所述的方法。
  10. 一种可读存储介质，其上存储有计算机指令，其特征在于，该计算机指令被处理器执行时实现如权利要求1～7中任一项所述的方法。
PCT/CN2020/137330 2020-08-21 2020-12-17 心电图数据增广方法、装置、电子设备及介质 WO2022036968A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010850174.2 2020-08-21
CN202010850174.2A CN111772625B (zh) 2020-08-21 2020-08-21 心电图数据增广方法、装置、电子设备及介质

Publications (1)

Publication Number Publication Date
WO2022036968A1 true WO2022036968A1 (zh) 2022-02-24

Family

ID=72763033

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/137330 WO2022036968A1 (zh) 2020-08-21 2020-12-17 心电图数据增广方法、装置、电子设备及介质

Country Status (3)

Country Link
JP (1) JP6831944B1 (zh)
CN (1) CN111772625B (zh)
WO (1) WO2022036968A1 (zh)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111772625B (zh) * 2020-08-21 2021-08-10 生物岛实验室 心电图数据增广方法、装置、电子设备及介质
CN113812959B (zh) * 2021-07-26 2023-10-13 北京海思敏医疗技术有限公司 用于小型数据处理设备的心电数据处理方法、电子设备和存储介质
CN113974644B (zh) * 2021-11-26 2023-03-24 苏州大学 一种房颤信号生成方法及系统
WO2023101526A1 (ko) * 2021-12-03 2023-06-08 서울대학교병원 다채널 1차원 시그널로부터 얻어진 이미지를 분석하는 이미지 기반 인공지능 모델을 훈련하는데 사용되는 훈련 이미지 생성 방법 및 이를 수행하는 장치
KR102589471B1 (ko) * 2021-12-15 2023-10-16 한양대학교 에리카산학협력단 데이터 증강 장치 및 방법
KR102560516B1 (ko) * 2021-12-27 2023-07-26 강원대학교산학협력단 심전도 데이터 및 신체 정보를 이용하는 딥러닝 모델 기반 심장 질환 발병 예측 장치 및 방법

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060079797A1 (en) * 2001-05-29 2006-04-13 Bischoff Edward T Cardiac rhythm monitoring device
CN101843480A (zh) * 2009-03-27 2010-09-29 华为技术有限公司 处理生物电信号的方法及装置
CN108090509A (zh) * 2017-12-13 2018-05-29 四川大学 一种数据长度自适应的心电图分类方法
CN110393519A (zh) * 2019-08-19 2019-11-01 广州视源电子科技股份有限公司 心电信号的分析方法、装置、存储介质和处理器
CN111460953A (zh) * 2020-03-26 2020-07-28 山东省人工智能研究院 一种基于对抗域自适应学习的心电信号分类方法
CN111772625A (zh) * 2020-08-21 2020-10-16 生物岛实验室 心电图数据增广方法、装置、电子设备及介质

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10426364B2 (en) * 2015-10-27 2019-10-01 Cardiologs Technologies Sas Automatic method to delineate or categorize an electrocardiogram
JP7336755B2 (ja) * 2017-07-28 2023-09-01 パナソニックＩｐマネジメント株式会社 データ生成装置、生体データ計測システム、識別器生成装置、データ生成方法、識別器生成方法及びプログラム
CN110495872B (zh) * 2019-08-27 2022-03-15 中科麦迪人工智能研究院(苏州)有限公司 基于图片及心搏信息的心电分析方法、装置、设备及介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060079797A1 (en) * 2001-05-29 2006-04-13 Bischoff Edward T Cardiac rhythm monitoring device
CN101843480A (zh) * 2009-03-27 2010-09-29 华为技术有限公司 处理生物电信号的方法及装置
CN108090509A (zh) * 2017-12-13 2018-05-29 四川大学 一种数据长度自适应的心电图分类方法
CN110393519A (zh) * 2019-08-19 2019-11-01 广州视源电子科技股份有限公司 心电信号的分析方法、装置、存储介质和处理器
CN111460953A (zh) * 2020-03-26 2020-07-28 山东省人工智能研究院 一种基于对抗域自适应学习的心电信号分类方法
CN111772625A (zh) * 2020-08-21 2020-10-16 生物岛实验室 心电图数据增广方法、装置、电子设备及介质

Also Published As

Publication number Publication date
CN111772625A (zh) 2020-10-16
CN111772625B (zh) 2021-08-10
JP6831944B1 (ja) 2021-02-17
JP2022035929A (ja) 2022-03-04

Similar Documents

Publication Publication Date Title
WO2022036968A1 (zh) 心电图数据增广方法、装置、电子设备及介质
Butun et al. 1D-CADCapsNet: One dimensional deep capsule networks for coronary artery disease detection using ECG signals
Karpagachelvi et al. ECG feature extraction techniques-a survey approach
Meng et al. Enhancing dynamic ECG heartbeat classification with lightweight transformer model
WO2020136569A1 (en) Method and system to characterize disease using parametric features of a volumetric object and machine learning
WO2012106729A1 (en) System and method for evaluating an electrophysiological signal
JP2021528148A (ja) 生体物理信号における非同期ノイズを定量化し、かつ除去するための方法およびシステム
Kumar et al. Fuzz-ClustNet: Coupled fuzzy clustering and deep neural networks for Arrhythmia detection from ECG signals
Lokesh et al. CNN based deep learning methods for precise analysis of cardiac arrhythmias
Varanini et al. A new method for QRS complex detection in multichannel ECG: Application to self-monitoring of fetal health
CN110767293B (zh) 一种脑部辅助诊断系统
Sun et al. The application of deep learning in electrocardiogram: Where we came from and where we should go?
Tseng et al. Sliding large kernel of deep learning algorithm for mobile electrocardiogram diagnosis
Deevi et al. HeartNetEC: a deep representation learning approach for ECG beat classification
Sulas et al. A non-invasive multimodal foetal ECG–Doppler dataset for antenatal cardiology research
Tseng et al. Clustering analysis of aging diseases and chronic habits with multivariate time series electrocardiogram and medical records
WO2021189900A1 (zh) 医学图像分析方法、装置、电子设备及可读存储介质
Tutuko et al. DAE-ConvBiLSTM: End-to-end learning single-lead electrocardiogram signal for heart abnormalities detection
CN113180688A (zh) 基于残差神经网络的冠心病心电图筛查系统及方法
Ayushi et al. A survey of ECG classification for arrhythmia diagnoses using SVM
CN114569116A (zh) 基于三通道图像及迁移学习的心冲击图室颤辅助诊断系统
Sabherwal Wavelet Transform As Method for ECG Signal Analysis
Lwin et al. Applications of wavelet transform
Wu et al. ECG Characteristic Detection Using DenseNet based on Attention Mechanism and Feature Pyramid
Luo et al. AMCNet: attention-based multiscale convolutional network for DCM MRI segmentation

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20950169

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 29.06.2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20950169

Country of ref document: EP

Kind code of ref document: A1