Atrial fibrillation monitoring method based on ECG (ECG) signals
阅读说明：本技术 一种基于ecg信号的房颤监测方法 (Atrial fibrillation monitoring method based on ECG (ECG) signals ) 是由 代超 何帆 周振 于 2019-11-19 设计创作，主要内容包括：本发明公开了一种基于ECG信号的房颤监测方法,涉及心电图分析技术领域,本发明包括将心电图12导联ECG原始信号描绘在同一张画布上,保存为ECG信号图片；标记ECG信号图片中的波形以及R波高峰,利用深度卷积神经网络对预设识别模型进行训练,得到R波识别模型；根据R波位置找到R波高峰,根据R波高峰位置得到位置序列向量V,进而得到R-R间期的差值向量V1；根据差值向量V1进行标记,若判定为房颤则记为1,否则记为0,进而得到判定结果集合；利用分类器对结果集合进行分类训练,得到房颤识别模型；利用R波识别模型和房颤识别模型对新的心电图12导联ECG原始信号进行识别,进而监测出该段心电图是否有房颤,本发明具有能快速判断是否有房颤的优点。(The invention discloses an atrial fibrillation monitoring method based on ECG signals, which relates to the technical field of electrocardiogram analysis, and comprises the steps of drawing 12-lead ECG original signals on the same canvas and storing the signals as ECG signal pictures; marking waveforms and R wave peaks in an ECG signal picture, and training a preset identification model by using a deep convolutional neural network to obtain an R wave identification model; finding an R wave peak according to the R wave position, obtaining a position sequence vector V according to the R wave peak position, and further obtaining a difference vector V1 of an R-R interval; marking according to the difference vector V1, if the atrial fibrillation is judged, marking as 1, otherwise, marking as 0, and further obtaining a judgment result set; carrying out classification training on the result set by using a classifier to obtain an atrial fibrillation recognition model; the R wave identification model and the atrial fibrillation identification model are utilized to identify the new electrocardiogram 12-lead ECG original signal so as to monitor whether atrial fibrillation exists in the section of electrocardiogram.)
1. An ECG signal-based atrial fibrillation monitoring method, comprising:
s1, training to obtain an R wave recognition model:
s1.1: acquiring an electrocardiogram 12-lead ECG original signal, preprocessing the ECG original signal, then, drawing the ECG original signal on the same canvas, and then, storing the canvas as an ECG signal picture;
s1.2: marking each waveform and an R wave peak in an ECG signal picture, inputting the ECG signal picture into a preset identification model, and training the preset identification model by using a deep convolution neural network to obtain an R wave identification model which is output as an R wave position;
s2, training to obtain an atrial fibrillation recognition model:
s2.1: finding an R wave peak according to the R wave position, obtaining a position sequence vector V according to the R wave peak position, and calculating the position sequence vector V to obtain a difference vector V1 of an R-R interval;
s2.2: marking according to the difference vector V1, if the atrial fibrillation is judged, marking as 1, otherwise, marking as 0, and further obtaining a judgment result set;
s2.3: carrying out classification training on the result set by using a classifier to obtain an atrial fibrillation identification model for judging atrial fibrillation;
s3, judging and identifying:
and identifying the new electrocardiogram 12-lead ECG original signal by utilizing an R wave identification model and an atrial fibrillation identification model, and further monitoring whether the section of electrocardiogram has atrial fibrillation.
2. The method for monitoring atrial fibrillation based on ECG signals according to claim 1, wherein in S1.1, the preprocessing of the ECG raw signals of 12 leads of electrocardiogram is specifically: clutter is filtered out firstly, then a base line is removed to obtain a better waveform, and normalization processing is carried out on the heart rate.
3. The method for monitoring atrial fibrillation based on ECG signals according to claim 1, wherein in S1.1, when the preprocessed electrocardiogram 12-lead ECG raw signals are plotted on a canvas, the x-axis is set as a time axis and the y-axis is set as an amplitude value.
4. An atrial fibrillation monitoring method according to claim 3, wherein in S1.2, the R wave peaks are marked as the values of the time points, and the rest of the points are marked as 0.
5. The method for monitoring atrial fibrillation according to claim 1, wherein in S1.2, deep convolutional neural networks are used for training for classification, and the deep convolutional neural networks used for training comprise but are not limited to VGG network models or ZF network models.
6. The method according to claim 1, wherein in S2.1, the length of the predetermined position sequence vector V is k, and the position sequence vector V is R1,R2,…,RnWhen n is<k, zero padding is performed on the position sequence vector V.
7. An ECG signal based atrial fibrillation monitoring method according to claim 6, wherein the difference vector V1 is (R)2-R1)/(R2+R1),(R3-R2)/(R3+R2),…,(Rn-Rn-1)/(Rn+Rn-1)。
8. An ECG signal based atrial fibrillation monitoring method according to claim 1, wherein in S2.3, the classifier is a Bi-LSTM model.
9. The atrial fibrillation monitoring method according to claim 1, wherein the step S3 is specifically as follows: the method comprises the steps of processing a new electrocardiogram 12-lead ECG original signal through S1.1 to obtain an ECG signal picture to be identified, finding out the R wave position of the ECG signal picture to be identified by using an R wave identification model, obtaining a new difference vector V1 through S2.1, identifying the new difference vector V1 by using an atrial fibrillation identification model, and further monitoring whether atrial fibrillation exists in the section of electrocardiogram.
10. The method for monitoring atrial fibrillation according to any one of claims 1 to 9, wherein after the R-wave identification model finds the R-wave position, small pictures of the R-wave are taken out through sliding windows, so as to find the peak of the R-wave.
Technical Field
The invention relates to the technical field of electrocardiogram analysis, in particular to an atrial fibrillation monitoring method based on an ECG signal.
Background
The ECG signal is an electrocardiographic signal collected by an electrocardiograph, which is generally called as an electrocardiogram, each cardiac cycle in the electrocardiogram is composed of a series of regular waveforms, which are P-wave, QRS complex and T-wave, and the start point, end point, peak, trough and period of these waveforms record detailed information of the heart activity state, respectively, so as to provide an important analysis basis for diagnosing heart diseases. A normal person normally has a cardiac cycle of about 0.80s, i.e. a cycle of about 0.80s of the ECG signal.
The P wave is generated by the activation of the atrium, the latter half is mainly generated by the left atrium, the normal P wave lasts for 0.08s to 0.11s, and the waveform is small and round; the QRS complex reflects the potential change of the depolarization process of the left ventricle and the right ventricle, the QRS complex is the wave band with the most drastic change in the electrocardiogram and consists of three closely connected waves, the first wave is a Q wave with a downward waveform, the next wave is an R wave with a high and sharp upward waveform, the last wave is an S wave with a downward waveform, the QRS complex generally lasts for 0.06S to 0.10S, and the amplitude change of the waveform is large; the T wave represents the potential change in the ventricular repolarization process, is a wave with lower amplitude after S, has a flat shape, and cannot be too low on an electrocardiogram taking R waves as a main part; the U wave is positioned behind the T wave and represents the subsequent potential of the ventricle, the direction of the U wave is consistent with that of the T wave, the amplitude of the U wave is lower than that of the T wave, and sometimes the waveform is not obvious.
Currently, it is common to pass the ECG one-dimensional signal: time and amplitude are used as input of a recognition system, a model is used for recognizing sequence signals, however, the form variation of PQRS waves is relatively small at any time, the existing model is difficult to recognize the sequence signals, and most of the existing models adopt single leads or a few leads for recognition, so that the utilization rate of information is low.
Disclosure of Invention
The invention aims to: in order to solve the problem that the identification of ECG one-dimensional signals by a model of the existing identification system is difficult due to the relatively small morphological variation of PQRS waves, the invention provides an atrial fibrillation monitoring method based on ECG signals.
The invention specifically adopts the following technical scheme for realizing the purpose:
an ECG signal based atrial fibrillation monitoring method, comprising:
s1, training to obtain an R wave recognition model:
s1.1: acquiring an electrocardiogram 12-lead ECG original signal, preprocessing the ECG original signal, then, drawing the ECG original signal on the same canvas, and then, storing the canvas as an ECG signal picture;
s1.2: marking each waveform and an R wave peak in an ECG signal picture, inputting the ECG signal picture into a preset identification model, and training the preset identification model by using a deep convolution neural network to obtain an R wave identification model which is output as an R wave position;
s2, training to obtain an atrial fibrillation recognition model:
s2.1: finding an R wave peak according to the R wave position, obtaining a position sequence vector V according to the R wave peak position, and calculating the position sequence vector V to obtain a difference vector V1 of an R-R interval;
s2.2: marking according to the difference vector V1, if the atrial fibrillation is judged, marking as 1, otherwise, marking as 0, and further obtaining a judgment result set;
s2.3: carrying out classification training on the result set by using a classifier to obtain an atrial fibrillation identification model for judging atrial fibrillation;
s3, judging and identifying:
and identifying the new electrocardiogram 12-lead ECG original signal by utilizing an R wave identification model and an atrial fibrillation identification model, and further monitoring whether the section of electrocardiogram has atrial fibrillation.
Further, in S1.1, the preprocessing of the ECG raw signals of 12 leads of the electrocardiogram is specifically: clutter is filtered out firstly, then a base line is removed to obtain a better waveform, and normalization processing is carried out on the heart rate.
Further, in S1.1, when the preprocessed electrocardiogram 12-lead ECG raw signal is plotted on the canvas, the x-axis is set as the time axis, and the y-axis is set as the amplitude value.
Further, in S1.2, the peak of the R wave is marked as the value of the time point, and the remaining points are marked as 0.
Further, in S1.2, deep convolutional neural network is used for training for classification, and the deep convolutional neural network used for training includes, but is not limited to, a VGG network model or a ZF network model.
Further, in S2.1, the length of the preset position sequence vector V is k, and the position sequence vector V is R1,R2,…,RnWhen n is<k, zero padding is performed on the position sequence vector V.
Further, the difference vector V1 is (R)2-R1)/(R2+R1),(R3-R2)/(R3+R2),…,(Rn-Rn-1)/(Rn+Rn-1)。
Further, in S2.3, the classifier is a Bi-LSTM model.
Further, the S3 specifically includes: the method comprises the steps of processing a new electrocardiogram 12-lead ECG original signal through S1.1 to obtain an ECG signal picture to be identified, finding out the R wave position of the ECG signal picture to be identified by using an R wave identification model, obtaining a new difference vector V1 through S2.1, identifying the new difference vector V1 by using an atrial fibrillation identification model, and further monitoring whether atrial fibrillation exists in the section of electrocardiogram.
Furthermore, after the R wave position is found out by the R wave identification model, the R wave small picture is cut out through the sliding window, and then the R wave peak is found.
The invention has the following beneficial effects:
1. according to the invention, all ECG original signals are painted on the same canvas, then the canvas is stored as an ECG signal picture, R wave extraction is carried out by marking each waveform in the ECG signal picture, R wave can be found out quickly and accurately, and then whether atrial fibrillation exists in the corresponding electrocardiogram is judged quickly.
2. The invention adopts 12 leads for identification, can maximally utilize the existing information and can quickly judge whether atrial fibrillation exists.
Drawings
FIG. 1 is a pictorial illustration of an ECG signal according to an embodiment of the present invention.
Detailed Description
For a better understanding of the present invention by those skilled in the art, the present invention will be described in further detail below with reference to the accompanying drawings and the following examples.