JP6831944B1 - How to extend ECG data and its devices, electronic devices and media - Google Patents

Abstract

The embodiments of the present disclosure disclose a method for expanding ECG data and its devices, electronic devices and media. This electrocardiogram data expansion method is based on a step of acquiring electrocardiogram data, a step of processing the electrocardiogram data to acquire a plurality of heartbeat data, and at least two heartbeat data among the plurality of heartbeat data. Includes steps to generate. It allows data expansion on ECG data, increases the amount of data in the training dataset, and fully trains the machine learning model to enhance the representation of the trained model. be able to. [Selection diagram] None

Description

The present disclosure relates to the field of medical technology, and specifically to a method for expanding electrocardiogram data and its device, electronic device and medium.

Data expansion is one of the most commonly used techniques for deep learning, primarily to increase the training dataset, make the dataset as diverse as possible, and increase the generalization ability of the trained model. used. In practice, not all extended methods are suitable for real-world learning data. Depending on the characteristics of the dataset, it is necessary to decide which data expansion method to use. Current data extensions primarily include horizontal / vertical inversion, rotation, scaling, cropping, cutting, translation, contrast, color jitter, noise and more.

An electrocardiogram shows a pattern in which changes in electrical activity that occur in each cycle of heart movement from the body surface are recorded using an electrocardiogram measuring device, and the signals in each segment have specific medical significance. There is. However, the use of traditional data expansion methods such as inversion, rotation, and cropping destroys the medical significance of electrocardiograms and cannot be effective in training machine learning models. That is, general image processing data expansion means cannot be applied to data expansion of electrocardiographic signals.

The embodiments of the present disclosure provide ECG data expansion methods, devices, electronic devices and media for solving problems in related techniques.

In a first aspect, the embodiments of the present disclosure provide a method of expanding ECG data. This method
Steps to get ECG data and
A step of processing the electrocardiogram data to acquire a plurality of heart rate data, and
It includes a step of generating extended data of learning data in machine learning based on at least two heartbeat data among the plurality of heartbeat data.

With reference to the first aspect, in the first embodiment of the first aspect of the present disclosure, the step of processing the electrocardiogram data to acquire a plurality of heart rate data is
The step of preprocessing the electrocardiogram data and
It includes a step of performing heartbeat identification on the preprocessed electrocardiogram data and acquiring a plurality of heartbeat data.

With reference to the first embodiment of the first aspect, in the second embodiment of the first aspect of the present disclosure, the pretreatment is
It includes a step of performing a multi-level wavelet decomposition on the electrocardiogram data by a discrete wavelet transform, setting the lowest level approximation to zero, and acquiring the electrocardiogram data after baseline calibration by discrete wavelet reconstruction.

In the third embodiment of the first aspect of the present disclosure, with reference to the first embodiment or the second embodiment of the first aspect, the pretreatment is performed.
It includes a step of performing multi-level wavelet decomposition on the electrocardiogram data by stationary wavelet transform, performing threshold filtering on detailed values, and acquiring electrocardiogram data after noise reduction by stationary wavelet reconstruction.

With reference to the first embodiment of the first aspect, in the fourth embodiment of the first aspect of the present disclosure, the step of performing heartbeat identification on the preprocessed electrocardiogram data and acquiring a plurality of heartbeat data is ,
A step of performing multi-level wavelet decomposition by stationary wavelet transform on the lead data of the first lead in the preprocessed electrocardiogram data and acquiring the decomposition result, and
A step of performing feature identification on the decomposition result and determining the position of the wave peak of the R wave in the guidance data of the first lead, and
It includes a step of dividing the lead data of the first lead and / or other leads into a plurality of heart rate data based on the position of the wave peak of the R wave in the lead data of the first lead.

With reference to the first aspect, in the fifth embodiment of the first aspect of the present disclosure, the step of generating extended data based on at least two heartbeat data among the plurality of heartbeat data is
A step of determining the first heartbeat data and the second heartbeat data from the plurality of heartbeat data, and
When at least one of the guidance data of the first heartbeat data and the second heartbeat data can be joined, the first heartbeat data and the second heartbeat data are joined so as to generate the third heartbeat data. Steps and
The step of acquiring the label of the first heart rate data and / or the second heart rate data, and
It includes a step of determining the label of the third heart rate data based on the label of the first heart rate data and / or the second heart rate data, and determining the labeled third heart rate data as extended data.

In the sixth embodiment of the first aspect of the present disclosure with reference to the fifth embodiment of the first aspect, at least one of the guidance data of the first heartbeat data and the second heartbeat data can be joined. In the case, the step of joining the first heart rate data and the second heart rate data so as to generate the third heart rate data is
For each type of guidance data, the guidance data of the type in the first heartbeat data and the second heartbeat data have the same R wave direction, and a zero cross point occurs within a preset time after the R wave. When a certain condition is satisfied, the step of joining the kind of guidance data in the first heartbeat data and the second heartbeat data as the guidance data of the kind of false heartbeat data based on the position of the zero cross point, and the step If the conditions are not satisfied, the step of determining the guidance data of the type of the second heartbeat data as the guidance data of the type of the false heartbeat data, and
When the false heartbeat data and the second heartbeat data are not exactly the same, the step of determining the false heartbeat data as the third heartbeat data is included.

In a second aspect, the embodiments of the present disclosure provide an electrocardiogram data expansion device. This device
An acquisition module that is configured to acquire ECG data, and
A processing module configured to process the electrocardiogram data and acquire a plurality of heart rate data,
It includes a generation module configured to generate extended data of learning data in machine learning based on at least two heartbeat data out of the plurality of heartbeat data.

In a third aspect, the embodiments of the present disclosure provide an electronic device. This electronic device includes a memory and a processor. The memory is used to store one or more computer instructions, wherein the one or more computer instructions are described in any of the first and first embodiments, first to sixth embodiments. It is performed by the processor to realize the method.

In a fourth aspect, the embodiments of the present disclosure provide a computer-readable storage medium in which computer instructions are stored. This computer instruction is executed by the processor so as to realize the method according to any one of the first aspect and the first to sixth embodiments of the first aspect.

According to the technical solution provided by the embodiments of the present disclosure, the electrocardiogram data is acquired, the electrocardiogram data is processed to obtain a plurality of heartbeat data, and at least two heartbeat data out of the plurality of heartbeat data are obtained. By generating extended data for training data in machine learning based on, data expansion can be performed on the electrocardiogram data, the amount of data in the training dataset can be increased, and machine learning. You can train your model well to improve the performance of your trained model.

It should be understood that the general description above and the detailed description below are merely exemplary and descriptive and do not limit this disclosure.

Other features, objectives and advantages of the present disclosure will become more apparent by elaborating on the following non-limiting embodiments with reference to the accompanying drawings.
The flowchart of the extension method of the electrocardiogram data which concerns on embodiment of this disclosure is shown. A schematic diagram of the electrocardiogram data according to the embodiment of the present disclosure is shown. A schematic diagram of the electrocardiogram data after baseline calibration according to the embodiment of the present disclosure is shown. A schematic diagram of the electrocardiogram data after the noise reduction processing according to the embodiment of the present disclosure is shown. The flowchart of the heartbeat identification which concerns on the Example of this disclosure is shown. A schematic diagram of an R wave in the reverse direction according to the embodiment of the present disclosure is shown. A schematic diagram of an R wave in the reverse direction according to the embodiment of the present disclosure is shown. A schematic diagram of heartbeat identification according to an embodiment of the present disclosure is shown. A flowchart for generating extended data according to an embodiment of the present disclosure is shown. A flowchart for generating extended data according to another embodiment of the present disclosure is shown. A schematic diagram of a heartbeat C generated based on a heartbeat A and a heartbeat B according to an embodiment of the present disclosure is shown. The control figure of the heartbeat B and the heartbeat C which concerns on the Example of this disclosure is shown. A control chart of heartbeat A and heartbeat C according to the embodiment of the present disclosure is shown. A block diagram of an electrocardiogram data expansion device according to an embodiment of the present disclosure is shown. A block diagram of an electronic device according to an embodiment of the present disclosure is shown. A schematic structure diagram of a computer system suitable for realizing a method for expanding ECG data according to an embodiment of the present disclosure is shown.

Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can readily implement them. Also, for clarity, parts unrelated to the description of the exemplary embodiments are omitted in the accompanying drawings.

In the present disclosure, terms such as "including" and "having" are intended to indicate the existence of features, numbers, steps, acts, components, parts or combinations thereof disclosed herein. It should be understood that it is not intended to rule out the possibility of the existence of multiple other features, numbers, steps, acts, members, parts or combinations thereof.

It should also be noted that the embodiments of the present disclosure and the features of the embodiments can be combined with each other if there is no contradiction. Hereinafter, the present disclosure will be described in detail together with examples with reference to the accompanying drawings.

Building a labeled dataset requires a huge amount of work. In addition, the incidence of different diseases among people is different, and there are many problems that the amount of categorical data is imbalanced in the data sets collected directly in clinical medicine. Insufficient data volumes in a dataset or imbalanced data volumes in different categories can reduce the performance of machine learning algorithms, for example, poor classification accuracy and low recall.

An electrocardiogram shows a pattern in which changes in electrical activity that occur in each cycle of heart movement from the body surface are recorded using an electrocardiogram measuring device, and the signals in each segment have specific medical significance. There is. However, the use of traditional data expansion methods such as inversion, rotation, and cropping destroys the medical significance of electrocardiograms and cannot be effective in learning machine learning models. That is, general image processing data expansion means cannot be applied to data expansion of electrocardiographic signals.

According to the technical solution provided by the embodiments of the present disclosure, the electrocardiogram data is acquired, the electrocardiogram data is processed to obtain a plurality of heartbeat data, and at least two heartbeat data of the plurality of heartbeat data are obtained. Based on this, by generating extended data for training data in machine learning, data expansion can be performed on the electrocardiogram data, the amount of data in the training dataset can be increased, and machine learning You can train your model well to improve the performance of your trained model.

FIG. 1 shows a flowchart of an electrocardiogram data expansion method according to an embodiment of the present disclosure.
As shown in FIG. 1, this method includes steps S101 to S103.

In step S101, the electrocardiogram data is acquired.
In step S102, the electrocardiogram data is processed to acquire a plurality of heartbeat data.
In step S103, extended data of learning data in machine learning is generated based on at least two heartbeat data of the plurality of heartbeat data.

According to the technical solution provided by the embodiments of the present disclosure, the electrocardiogram data is acquired, the electrocardiogram data is processed to obtain a plurality of heartbeat data, and at least two heartbeat data of the plurality of heartbeat data are obtained. Based on this, by generating extended data for training data in machine learning, data expansion can be performed on the electrocardiogram data, the amount of data in the training dataset can be increased, and machine learning You can train your model well to improve the performance of your trained model.

FIG. 2 shows a schematic diagram of electrocardiogram data according to the embodiment of the present disclosure. As shown in FIG. 2, the electrocardiogram data is continuously collected and generally contains a plurality of heartbeats, and thus has a characteristic of constant periodicity. The largest wave peak in each heartbeat is called the R wave. The R wave has distinctive features and can be used as a reference for dividing the heartbeat.

In ECG terminology, the placement of electrodes attached to the surface of the human body when recording an electrocardiogram, and the connection method between the electrodes and the amplifier are called ECG induction. There are 12 types of commonly used leads, which are lead I, lead II, lead III, lead aVR, lead aVL, lead aVF, lead V1, lead V2, lead V3, lead V4, lead V5 and lead V6, respectively. It is an induction. The electrocardiographic signals of all leads are detected at about the same time. They are almost the same in terms of distribution density and position of R waves. The electrocardiogram data exemplified in FIG. 2 is only one lead data, and other lead data may be similar thereto.

According to the embodiment of the present disclosure, the step of processing the electrocardiogram data and acquiring a plurality of heartbeat data is a plurality of steps of preprocessing the electrocardiogram data and performing heartbeat identification on the preprocessed electrocardiogram data. Includes steps to acquire heart rate data for.

According to the technical solution provided by the embodiments of the present disclosure, the electrocardiogram is obtained by preprocessing the electrocardiogram data, performing heartbeat identification on the preprocessed electrocardiogram data, and acquiring a plurality of heartbeat data. Data can be expanded on the data, the amount of data for training can be increased, and the machine learning model can be sufficiently trained to enhance the representation of the model.

According to the embodiments of the present disclosure, the preprocessing performs multi-level wavelet decomposition on the electrocardiogram data by discrete wavelet transform, sets the lowest level approximation to zero, and baselines by discrete wavelet reconstruction. Includes steps to acquire electrocardiogram data after calibration.

The wavelet transform is an ideal tool for performing time-frequency analysis and processing of signals. The wavelet transform algorithm commonly used in computers is the mallat algorithm. The central idea of this algorithm is to use a filter to perform the 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. A is an approximate value of the signal (value acquired by the low frequency filter), and D is a detailed value of the signal (value acquired by the high frequency filter). Decomposition of the original signal by such a pair of filters is called one-level decomposition. The signal decomposition process can be repeated. That is, multi-level decomposition can be performed. By continuously decomposing the low frequency components without further decomposing the high frequency components, lower frequency components can be obtained and a wavelet decomposition tree can be formed. If desired, the number of decomposition levels can be determined.

In the embodiments of the present disclosure, the discrete wavelet transform (DWT) is adopted to downsample the data after each level of decomposition, so that the amount of data is relatively small and the calculation is relatively fast. As the number of levels continues to increase, the resolution of the resulting low frequency components gradually decreases. In contrast, the Discrete Wavelet Reconstruction Algorithm (IDWT) involves two processes: upsampling and filtering. Upsampling means inserting 0 between the downsampling data.

The wavelet used during baseline variation elimination is the db5 wavelet. As shown in FIG. 3, the baseline variation was removed by performing a 9-level wavelet decomposition on the signal using the DWT, setting the lowest level approximation to zero, and then performing the IDWT. Later electrocardiographic signals can be obtained.

According to the technical solution provided by the embodiments of the present disclosure, the discrete wavelet transform is used to perform a multi-level wavelet decomposition on the electrocardiogram data to set the lowest level approximation to zero and discrete wavelet reconstruction. By acquiring the electrocardiogram data after the baseline calibration, the baseline calibration can be performed better on the electrocardiogram data.

According to the embodiments of the present disclosure, noise reduction processing can also be performed in addition to baseline calibration in the pretreatment process. The electromyographic data set includes EMG interference and 50/60 Hz utility frequency noise. For example, the wavelet transform threshold method can be used to remove high frequency noise in the electrocardiographic signal.

According to the embodiment of the present disclosure, in the preprocessing, multi-level wavelet decomposition is performed on the electrocardiogram data by stationary wavelet transform, threshold filtering is performed on detailed values, and noise is reduced by stationary wavelet reconstruction. Includes steps to acquire ECG data for.

According to the embodiments of the present disclosure, stationary wavelet transform (SWT) is used to remove noise. The biggest difference between SWT and DWT is that SWT does not perform downsampling after each level decomposition. Therefore, although the amount of data is relatively large, the resolution of each level remains unchanged. Resolution is especially important because the removal of high frequency noise is not simply the complete removal of a level of signal, but the threshold removal. For the same reason, there is no need to perform upsampling during the steady wavelet reconstruction (ISWT) process.

ｄは詳細値、Ｎはその長さ、ｍｉｄは中央値である。使用される閾値方法は、ハーフソフト閾値法であり、ハード閾値法のようにウェーブレット係数の突然変化を生成することも、ソフト閾値法のように偏差を生成することもない。ハーフソフト閾値法の式は、以下のとおりである。

式では、λ１＝λｔｈ、λ２＝１．２５λｔｈとする。閾値フィルタリングの後にＩＳＷＴを行うと、高周波ノイズを除去することができる。ノイズ低減処理後の心電図データは図４に示される。 The wavelet used during high frequency noise removal is the bio2.6 wavelet. After performing a 6-level wavelet decomposition on the signal using SWT, first set the highest detail value of the frequency at the first two levels to zero. The frequencies represented by these detailed values are too high to contain most electrocardiographic information. Next, threshold filtering is performed on the detailed values at the 3rd to 6th levels. The threshold is selected using the following equation.

d is the detailed value, N is the length thereof, and mid is the median value. The threshold method used is the half-soft threshold method, which does not generate a sudden change in the wavelet coefficient as in the hard threshold method, nor does it generate a deviation as in the soft threshold method. The formula of the half-soft threshold method is as follows.

In the formula, λ1 = λth and λ2 = 1.25λth. High frequency noise can be removed by performing ISWT after threshold filtering. The electrocardiogram data after the noise reduction processing is shown in FIG.

According to the technical solution provided by the embodiments of the present disclosure, the stationary wavelet transform performs multi-level wavelet decomposition on the electrocardiogram data, threshold filtering on the detailed values, and noise due to stationary wavelet reconstruction. By acquiring the reduced electrocardiogram data, the noise reduction effect of the electrocardiogram data can be improved.

FIG. 5 shows a flowchart of heartbeat identification according to the embodiment of the present disclosure.

As shown in FIG. 5, this method comprises steps S501-S503.

In step S501, the induction data of the first lead in the preprocessed electrocardiogram data is subjected to multi-level wavelet decomposition by stationary wavelet transform, and the decomposition result is acquired.
In step S502, feature identification is performed on the decomposition result, and the position of the wave peak of the R wave in the guidance data of the first lead is determined.
In step S503, the guidance data of the first guidance and / or other guidance is divided into a plurality of heart rate data based on the position of the wave peak of the R wave in the guidance data of the first guidance.

であってもよい。 According to the embodiments of the present disclosure, in step S501, the SWT can be used to perform wavelet transform on the signal. The wavelet used is, for example,

It may be.

According to the embodiments of the present disclosure, in step S502, the converted fifth level detail value can be selected. The characteristics of this level of R wave are most obvious. The upward R wave appears at this level in the form of a negative-positive maxima pair. The downward R wave appears in the form of positive-negative maxima pairs at this level. Find these maxima pairs and find the location of their zero crossing points. The corresponding R wave peak is a maximum / minimum value within ± 0.05 s of the zero crossing point.

Even in the electrocardiogram of a healthy person, the peak direction of the R wave of each type of induction is not always constant. In the electrocardiogram after the lesion, the peak direction of the R wave may be opposite to the normal direction. 6A and 6B show schematic views of the reverse R wave according to the embodiment of the present disclosure. FIG. 6A shows a schematic diagram of the pathologically caused R-wave inversion situation. FIG. 6B shows a schematic diagram of R wave inversion by the electrodes arranged in the opposite direction.

In the actual operation, 12 types of leads of the same person's simultaneously detected electrocardiographic signals (these 12 types of leads are lead I, lead II, lead III, lead aVR, lead aVL, lead aVF, respectively, V1 lead, V2 lead, V3 lead, V4 lead, V5 lead, and V6 lead. Hereinafter, each kind of lead can be referred to based on the reference order of the above 12 kinds of leads. For example, the first lead can be referred to as the first lead. Lead I and lead 7 can refer to lead V1), each performing a single baseline removal and filtering operation. After that, the R wave identification is performed only for the first lead, and the result of the R wave identification of the first lead is used to divide all the other leads. The reason for this is that since the electrocardiographic signals of all leads are detected almost simultaneously, they are almost the same with respect to the distribution density and position of the R wave.

After the division, it is necessary to calibrate the R wave positions of 11 types of leads other than the first lead. Since it takes a certain amount of time for potential conduction between adjacent cells in the human body, the positions of the R wave peaks of these 12 types of leads may differ by up to ± 0.05 s. The following operations are performed for each of the other 11 types of guidance. Find the maximum and minimum values within ± 0.05 s near the position of the R wave peak of the first lead. If the maximum value is less than one-third of the absolute value of the minimum value, the position with the minimum value is recognized as the R wave peak of this lead, otherwise the position with the maximum value is the R wave of this lead. Admit that it is the peak. Thus, the methods of the embodiments of the present disclosure may further include calibrating the position of the R wave peak in the data of the other leads based on the position of the R wave peak in the data of the first lead. ..

According to the embodiments of the present disclosure, the first and last heartbeat signals may be incomplete at the time of division, so the first and last heartbeat signals are rejected and the midpoint between the R peaks of two adjacent heartbeats is set. It can be used as a dividing point. By this method, the complete electrocardiographic signal can be divided into individual heartbeats. The result after the division is shown in FIG.

According to the technical solution provided by the embodiments of the present disclosure, the first lead data in the preprocessed electrocardiogram data is subjected to multi-level wavelet decomposition by steady wavelet conversion to obtain the decomposition result. In the step, the step of performing feature identification on the decomposition result and determining the position of the R wave peak in the data of the first lead, and the position of the wave peak of the R wave in the data of the first lead. Based on this, by executing the step of dividing the data of the first lead and / or the other leads into a plurality of heartbeat data, the heartbeat data can be accurately identified, and the data expansion for the electrocardiogram data can be performed. It can be done, the amount of data in the training dataset can be increased, and the machine learning model can be well trained to improve the performance of the trained model.

FIG. 8 shows a flowchart for generating extended data according to the embodiment of the present disclosure.

As shown in FIG. 8, this method comprises steps S801-S804.

In step S801, the first heartbeat data and the second heartbeat data are determined from the plurality of heartbeat data.
In step S802, when at least one of the first heartbeat data and the second heartbeat data can be joined, the first heartbeat data and the second heartbeat data are joined to generate the third heartbeat data. To do.
In step S803, the label of the first heart rate data and / or the second heart rate data is acquired.
In step S804, the label of the third heartbeat data is determined based on the label of the first heartbeat data and / or the second heartbeat data, and the labeled third heartbeat data is determined as extended data.

According to the embodiments of the present disclosure, two divided heartbeat signals (each lead) can be randomly selected as heartbeat A and heartbeat B, respectively. This random selection is based on one of the following three methods.
(1) Two heartbeat signals of the same patient at different times are randomly selected as A and B.
(2) Two heartbeat signals of two different patients suffering from the same type of disease are randomly selected as A and B.
(3) The heartbeat signal of one healthy person is randomly selected as A, and the heartbeat signal of one patient is B.

According to the examples of the present disclosure, the degree of generalization of data can be increased by using the three methods in combination at the same time.

It is also possible to detect in advance whether or not the two selected signals can be joined. For example, the peak directions of the R waves of various leads in the heartbeat A and the heartbeat B are compared, and when the peak directions of the R waves of a certain type of lead data in A and B are the same, the lead data are joined. Mark as possible, and if the peak directions are not the same, mark as unjoinable. If you mark all guidance data as unjoinable, this join fails and the heartbeat is reselected. Joining can be performed if there is at least one type of guidance data marked as joinable.

According to the examples of the present disclosure, for the above three methods, the position of the main feature data for detecting the lesion is determined, and the data at this position of the third heartbeat data becomes heartbeat A or heartbeat B. Determine if it comes from. If it is derived from heart rate A, a label for the third heart rate data is placed based on the label for heart rate A. If it is derived from heart rate B, the label of the third heart rate data is placed based on the label of heart rate B. For example, since the main detection position of myocardial infarction lesion is the ST segment (the segment after the R wave), if the ST segment of the expanded heartbeat C is derived from heartbeat B, the label of heartbeat C is heartbeat B. Admit that they are the same. For example, when heartbeat B is derived from electrocardiogram data of a myocardial infarction patient, the label of heartbeat C is set as suffering from myocardial infarction disease. If the heart rate B is derived from the electrocardiogram data of a healthy person, the label of the heart rate C is set as healthy.

According to the technical solution provided by the embodiment of the present disclosure, a step of determining the first heartbeat data and the second heartbeat data from the plurality of heartbeat data, and at least in the first heartbeat data and the second heartbeat data. When one type of guidance data can be joined, the step of joining the first heartbeat data and the second heartbeat data to generate the third heartbeat data, and the first heartbeat data and / or the second heartbeat data. A step of acquiring a label and a step of determining the label of the third heartbeat data based on the label of the first heartbeat data and / or the second heartbeat data, and determining the labeled third heartbeat data as extended data. By performing data expansion on the electrocardiogram data, the amount of data in the training dataset can be increased, and the machine learning model is well trained and of the trained model. You can enhance the expression.

According to the embodiment of the present disclosure, when at least one kind of induction of the first heartbeat data and the second heartbeat data can be joined, the first heartbeat data and the second heartbeat data are joined together. 3 The step of generating heart rate data is
For each type of guidance data, the guidance data of the type in the first heartbeat data and the second heartbeat data have the same R wave direction, and a zero cross point occurs within a preset time after the R wave. When a certain condition is satisfied, the step of joining the kind of guidance data in the first heartbeat data and the second heartbeat data as the guidance data of the kind of false heartbeat data based on the position of the zero cross point, and the step If the conditions are not satisfied, the step of determining the guidance data of the type of the second heartbeat data as the guidance data of the type of the false heartbeat data, and
When the false heartbeat data and the second heartbeat data are not exactly the same, the step of determining the false heartbeat data as the third heartbeat data is included.

Hereinafter, the method of the embodiment of the present disclosure will be described as an example with reference to FIGS. 9, 10, 11A and 11B.

FIG. 9 shows a flowchart for generating extended data according to another embodiment of the present disclosure.

As shown in FIG. 9, this method comprises steps S901-S910.

In step S901, heartbeat A and heartbeat B are selected. Since the above description of FIG. 8 can be referred to, it is not repeated here.

In step S902, one untreated lead is selected.

In step S903, it is determined whether or not the R wave directions in the heartbeat A and the heartbeat B are the same for this guidance, and if they are the same, step S904 is continuously executed, and if they are not the same, step S907 is performed. Execute.

In step S904, the position of the first zero crossing point within 100 ms after the R wave peak of the type of guidance data in heartbeat A and heartbeat B is found.

In step S905, it is determined whether or not a zero cross point has been found in both A and B for this guidance, and if found, step S906 is continuously executed, and if not found, step S907 is executed.

In practice, there are occasional lesions with special waveforms for which no zero crossing point can be found (eg, ST segment elevation). Avoiding such waveforms before joining is a relatively good process, as it is not even possible to find other waveforms that can be joined to such waveforms.

In step S906, the first half of the guidance data of the type of heartbeat A is resampled and assigned to the first half of heartbeat C, the second half of the guidance data of the type of heartbeat B is assigned to the second half of heartbeat C, and the heartbeat C. Acquire the guidance data of the relevant type.

According to the embodiments of the present disclosure, the first half of the induction data of the type of heart rate A is resampled and assigned to the first half of heart rate C, for example, by horizontal compression or expansion, the first half of heart rate A. It can be performed by matching the length of the heartbeat B with the length of the first half of the heartbeat B. The first half portion refers to the data from the start position of the heartbeat to the position of the zero cross point, and the second half portion refers to the data from the position of the zero cross point to the end position.

FIG. 10 shows a schematic diagram of a heartbeat C generated based on a heartbeat A and a heartbeat B according to an embodiment of the present disclosure. As shown in FIG. 10, the heartbeat A and the heartbeat B of the two heartbeat data can generate the heartbeat C of one new heartbeat data. Heart rate C is different from both heart rate A and heart rate B. FIG. 11A shows a control diagram of heartbeat B and heartbeat C according to the embodiment of the present disclosure. As shown in FIG. 11A, the latter half of the heartbeat C coincides with the heartbeat B. FIG. 11B shows a control diagram of heartbeat A and heartbeat C according to the embodiment of the present disclosure. As shown in FIG. 11B, the first half of the heartbeat C is the result of resample (stretching) the first half of the heartbeat A. 10, 11A and 11B show only one type of guidance data. Data for other leads processed by the methods of this example may have similar characteristics.

Please refer back to FIG. In step S907, the guidance data of the type of heart rate B is directly assigned to heart rate C.

In step S908, it is determined whether or not all the guidance processes have been completed, and if it is completed, step S909 is executed. If not, the process returns to step S902 to continue one type of unprocessed guidance. To select.

In step S909, it is determined whether or not each induction of heartbeat C is the same as heartbeat B, and if they are the same, heartbeat C is rejected, and the process returns to S901 to reselect the heartbeat. If not, step S910 is performed. Execute.

In step S910, the heartbeat C is determined as one extended data.

After executing step S910, it is possible to return to step S901 to continuously generate new extended data.

According to the technical solution provided by the embodiment of the present disclosure, with respect to the guidance data for each type, the guidance data of the type in the first heartbeat data and the second heartbeat data have the same R wave direction. If the condition is that there is a zero cross point within a preset time after the R wave, the guidance data of the type in the first heartbeat data and the second heartbeat data is obtained based on the position of the zero cross point. , A step of joining the false heartbeat data as the guidance data of the type, and a step of determining the guidance data of the type of the second heartbeat data as the guidance data of the type of the false heartbeat data when the conditions are not satisfied. When the false heartbeat data and the second heartbeat data are not exactly the same, the data can be expanded on the electrocardiogram data by executing the step of determining the false heartbeat data as the third heartbeat data. , The amount of data in the training dataset can be increased, and the machine learning model can be well trained to improve the performance of the trained model.

The method of the embodiment of the present disclosure can improve the identification accuracy based on the existing deep learning method. After expanding the electrocardiographic signal, the problem of inadequate or imbalanced data in the training data set is solved, and the identification accuracy caused by the inadequate or imbalanced data in the training data set is low. In order to solve the problem, the identification accuracy of such data is greatly improved. In order to increase the overall amount of valid data, when solving the problem of imbalanced data, it improves the identification accuracy of a small number of individual data as well as the overall identification accuracy of all data.

FIG. 12 shows a block diagram of the electrocardiogram data expansion device according to the embodiment of the present disclosure. The device can be part or all of an electronic device, depending on software, hardware, or a combination of both.

As shown in FIG. 12, the electrocardiogram data expansion device 1200 includes an acquisition module 1210, a processing module 1220, and a generation module 1230.

The acquisition module 1210 is configured to acquire electrocardiogram data.
The processing module 1220 is configured to process the electrocardiogram data and acquire a plurality of heartbeat data.
The generation module 1230 is configured to generate extended data of learning data in machine learning based on at least two heartbeat data among the plurality of heartbeat data.

According to the technical solution provided by the embodiments of the present disclosure, an acquisition module configured to acquire electrocardiogram data, configured to process the electrocardiogram data to acquire multiple heart rate data. Data expansion for electrocardiogram data by a processing module and a generation module configured to generate extended data for training data in machine learning based on at least two heart rate data of the plurality of heart rate data. It can be done, the amount of data in the training dataset can be increased, and the machine learning model can be well trained to improve the performance of the trained model.

The present disclosure further discloses electronic devices. FIG. 13 shows a block diagram of an electronic device according to an embodiment of the present disclosure.

As shown in FIG. 13, the electronic device 1300 includes a memory 1301 and a processor 1302. The memory 1301 is used to store a program that assists the electronic device in performing any of the ECG data expansion methods or code generation methods of the above embodiments. The processor 1302 is configured to execute a program stored in the memory 1301.

According to an embodiment of the present disclosure, the memory 1301 is used to store one or more computer instructions, and when the one or more computer instructions are executed by the processor 1302:
Steps to get ECG data and
A step of processing the electrocardiogram data to acquire a plurality of heart rate data, and
A step of generating extended data of learning data in machine learning is realized based on at least two heartbeat data among the plurality of heartbeat data.

According to the embodiments of the present disclosure, the step of processing the electrocardiogram data to acquire a plurality of heart rate data is
The step of preprocessing the electrocardiogram data and
It includes a step of performing heartbeat identification on the preprocessed electrocardiogram data and acquiring a plurality of heartbeat data.
According to the embodiments of the present disclosure, the pretreatment
It includes a step of performing a multi-level wavelet decomposition on the electrocardiogram data by a discrete wavelet transform, setting the lowest level approximation to zero, and acquiring the electrocardiogram data after baseline calibration by discrete wavelet reconstruction.

According to the embodiments of the present disclosure, the pretreatment
It includes a step of performing multi-level wavelet decomposition on the electrocardiogram data by stationary wavelet transform, performing threshold filtering on detailed values, and acquiring electrocardiogram data after noise reduction by stationary wavelet reconstruction.

According to the embodiments of the present disclosure, the step of performing heart rate identification on preprocessed electrocardiogram data and acquiring a plurality of heart rate data is
A step of performing multi-level wavelet decomposition by stationary wavelet transform on the lead data of the first lead in the preprocessed electrocardiogram data and acquiring the decomposition result, and
A step of performing feature identification on the decomposition result and determining the position of the wave peak of the R wave in the guidance data of the first lead, and
It includes a step of dividing the lead data of the first lead and / or other leads into a plurality of heart rate data based on the position of the wave peak of the R wave in the lead data of the first lead.

According to the embodiments of the present disclosure, the step of generating extended data based on at least two heart rate data of the plurality of heart rate data is
A step of determining the first heartbeat data and the second heartbeat data from the plurality of heartbeat data, and
When at least one kind of guidance data in the first heartbeat data and the second heartbeat data can be joined, the step of joining the first heartbeat data and the second heartbeat data to generate the third heartbeat data,
The step of acquiring the label of the first heart rate data and / or the second heart rate data, and
It includes a step of determining the label of the third heart rate data based on the label of the first heart rate data and / or the second heart rate data, and determining the labeled third heart rate data as extended data.

According to the embodiment of the present disclosure, when at least one kind of guidance data of the first heartbeat data and the second heartbeat data can be joined, the first heartbeat data and the second heartbeat data are joined to form a third heartbeat. The step of generating data is
With respect to the guidance data for each type, the guidance data of the type in the first heartbeat data and the second heartbeat data have the same R wave direction, and a zero cross point occurs within a preset time after the R wave. When a certain condition is satisfied, the step of joining the kind of guidance data in the first heartbeat data and the second heartbeat data as the guidance data of the kind of false heartbeat data based on the position of the zero cross point, and the step If the conditions are not satisfied, the step of determining the guidance data of the type of the second heartbeat data as the guidance data of the type of the false heartbeat data, and
When the false heartbeat data and the second heartbeat data are not exactly the same, the step of determining the false heartbeat data as the third heartbeat data is included.

FIG. 14 shows a schematic structural diagram of a computer system suitable for realizing a method for expanding electrocardiogram data according to an embodiment of the present disclosure.

As shown in FIG. 14, the computer system 1400 includes a processing unit 1401. The processing unit 1401 may execute various processes in the above embodiment according to the program stored in the read-only memory (ROM) 1402 or the program loaded from the storage portion 1408 into the random access memory (RAM) 1403. it can. The RAM 1403 stores various programs and data necessary for operating the system 1400. The processing units 1401, ROM 1402, and RAM 1403 are connected to each other via the bus 1404. The input / output (I / O) interface 1405 is also connected to bus 1404.

The I / O interface 1405 includes an input portion 1406 including a keyboard and a mouse, an output portion 1407 including a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker, and a storage portion 1408 including a hard disk and the like. , The communication portion 1409 of the network interface card including the LAN card, the modem and the like is connected. The communication portion 1409 executes communication processing via a network such as the Internet. The driver 1410 is also connected to the I / O interface 1405 as needed. Removable media 1411 such as magnetic disks, optical disks, magneto-optical disks, and semiconductor memories are attached to the driver 1410 as needed. As a result, the computer program read from the removable media 1411 is installed in the storage portion 1408 as needed. The processing unit 1401 can be realized as a processing unit such as a CPU, GPU, TPU, FPGA, or NPU.

In particular, according to the embodiments of the present disclosure, the methods described above can be implemented as computer software programs. For example, the embodiments of the present disclosure include computer program products. This computer program product includes a computer program that is physically contained in a readable medium. The computer program includes program code for executing the above method. In such an embodiment, the computer program can be downloaded and installed from the network via communication portion 1409 and / or installed from removable media 1411.

The flowcharts and block diagrams in the accompanying drawings illustrate the systems, methods and feasible system architectures, functions and operations of the various embodiments of the present disclosure by computer program products. In this regard, each block in the flowchart or block diagram can represent a module, program segment or part of the code. The module, program segment, or part of the code includes one or more executable instructions for implementing a given logical function. Also note that in some alternative implementations, the functions marked on the blocks may occur in a different order than those marked on the attached drawing. For example, two blocks displayed one after the other can actually be executed basically in parallel, or in reverse order depending on the related functions. In addition, each block in the block diagram and / or flowchart, and a combination of blocks in the block diagram and / or flowchart may be implemented by a system based on dedicated hardware for performing a predetermined function or operation. Also note that it can be done, or it can be implemented with a combination of dedicated hardware and computer instructions.

The units or modules referred to in the embodiments of the present disclosure can be implemented in the form of software and can also be implemented in the form of programmable hardware. The units or modules described can also be placed on the processor. The names of these units or modules do not constitute restrictions on the unit or module itself in certain circumstances.

In another aspect, the present disclosure further 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 according to the above embodiment, or may be a computer-readable storage medium that is not installed in the device and exists alone. It may be a storage medium. One or more programs are stored in the computer-readable storage medium storage. The program is used by one or more processors to perform the methods described in the present disclosure.

The above description is merely a description of the preferred embodiments of the present disclosure and the technical principles used. Those skilled in the art will not be limited to technical solutions based on a specific combination of the above-mentioned technical features, and the above-mentioned techniques will be provided on the premise that the invention does not deviate from the concept of the above-mentioned invention. It should be understood that other technical solutions with any combination of features or their equivalent features are also included. For example, technical solutions formed by the mutual exchange of the above features with technical features having similar functions disclosed in the present disclosure (but not limited to these).

Claims (10)

Steps to get ECG data and
A step of processing the electrocardiogram data to acquire a plurality of heart rate data, and
A method for expanding electrocardiogram data, which comprises a step of generating extended data for learning data in machine learning based on at least two heartbeat data among the plurality of heartbeat data. The step of processing the electrocardiogram data to acquire a plurality of heart rate data is
The step of preprocessing the electrocardiogram data and
The method according to claim 1, further comprising a step of performing heartbeat identification on the preprocessed electrocardiogram data and acquiring a plurality of heartbeat data. The pretreatment is
It includes the steps of performing a multi-level wavelet decomposition on the electrocardiogram data by the discrete wavelet transform, setting the lowest level approximation to zero, and acquiring the electrocardiogram data after baseline calibration by the discrete wavelet reconstruction. The method according to claim 2, which is characterized. The pretreatment is
It is characterized by including a step of performing multi-level wavelet decomposition on the electrocardiogram data by stationary wavelet transform, performing threshold filtering on detailed values, and acquiring electrocardiogram data after noise reduction by stationary wavelet reconstruction. The method according to claim 2 or 3. The step of performing heartbeat identification on preprocessed electrocardiogram data and acquiring multiple heartbeat data is
A step of performing multi-level wavelet decomposition by stationary wavelet transform on the lead data of the first lead in the preprocessed electrocardiogram data and acquiring the decomposition result, and
A step of performing feature identification on the decomposition result and determining the position of the wave peak of the R wave in the guidance data of the first lead, and
It is characterized by including a step of dividing the lead data of the first lead and / or other leads into a plurality of heart rate data based on the position of the wave peak of the R wave in the lead data of the first lead. The method according to claim 2. The step of generating extended data of learning data in machine learning based on at least two heart rate data among the plurality of heart rate data is
A step of determining the first heartbeat data and the second heartbeat data from the plurality of heartbeat data, and
When at least one kind of guidance data of the first heartbeat data and the second heartbeat data can be joined, the step of joining the first heartbeat data and the second heartbeat data so that the third heartbeat data is generated. When,
The step of acquiring the label of the first heart rate data and / or the second heart rate data, and
It is characterized by including a step of determining the label of the third heart rate data based on the label of the first heart rate data and / or the second heart rate data, and determining the labeled third heart rate data as extended data. The method according to claim 1. When at least one kind of guidance data of the first heartbeat data and the second heartbeat data can be joined, the first heartbeat data and the second heartbeat data are joined so that the third heartbeat data is generated. The step is
For each type of guidance data, the guidance data of the type in the first heartbeat data and the second heartbeat data have the same R wave direction, and a zero cross point occurs within a preset time after the R wave. When a certain condition is satisfied, the step of joining the kind of guidance data in the first heartbeat data and the second heartbeat data as the guidance data of the kind of false heartbeat data based on the position of the zero cross point, and the step If the conditions are not satisfied, the step of determining the guidance data of the type of the second heartbeat data as the guidance data of the type of the false heartbeat data, and
The method according to claim 6, wherein when the false heartbeat data and the second heartbeat data are not completely the same, the step of determining the false heartbeat data as the third heartbeat data is included. An acquisition module that is configured to acquire ECG data, and
A processing module configured to process the electrocardiogram data and acquire a plurality of heart rate data,
An ECG data expansion device comprising a generation module configured to generate expansion data for learning data in machine learning based on at least two heart rate data among the plurality of heart rate data. .. An electronic device that includes a memory and a processor, the memory being used to store one or more computer instructions, the one or more computer instructions being any one of claims 1-7. An electronic device, characterized in that it is performed by said processor to implement the method described in the section. A readable storage medium in which computer instructions are stored, characterized in that the computer instructions are executed by a processor to implement the method according to any one of claims 1-7. A readable storage medium.

Images