CN115633965A

CN115633965A - Electrocardiosignal detection device and equipment

Info

Publication number: CN115633965A
Application number: CN202110815870.4A
Authority: CN
Inventors: 胡静; 赵巍
Original assignee: Guangzhou Xicoo Medical Technology Co ltd; Guangzhou Shiyuan Electronics Thecnology Co Ltd; Guangzhou Shiyuan Artificial Intelligence Innovation Research Institute Co Ltd
Current assignee: Guangzhou Xicoo Medical Technology Co ltd; Guangzhou Shiyuan Electronics Thecnology Co Ltd; Guangzhou Shiyuan Artificial Intelligence Innovation Research Institute Co Ltd
Priority date: 2021-07-19
Filing date: 2021-07-19
Publication date: 2023-01-24

Abstract

The embodiment of the application discloses electrocardiosignal detection device and equipment, it includes: the signal acquisition module is used for acquiring a first electrocardiosignal to be detected; the characteristic extraction module is used for inputting the first electrocardiosignal to the first neural network and extracting the depth characteristic of the first electrocardiosignal by the first neural network; the prior acquisition module is used for acquiring preset electrocardio prior characteristics; the characteristic fusion module is used for fusing the depth characteristic and the electrocardio priori characteristic to obtain a fusion characteristic; and the signal detection module is used for inputting the fusion characteristics to the second neural network, and outputting the detection result of the first electrocardiosignal by the second neural network according to the fusion characteristics. The device can solve the technical problem that the portable electrocardio detector cannot analyze complex electrocardio signals with high performance in the related technology.

Description

Electrocardiosignal detection device and equipment

Technical Field

The embodiment of the application relates to the technical field of electrocardiosignal analysis, in particular to an electrocardiosignal detection device and equipment.

Background

Electrocardiosignals can reflect the electrophysiological process of heart activity and are often used to assist in the diagnosis of heart diseases. A portable electrocardiograph is widely used as an electrocardiographic signal detection device. The portable electrocardio detector can regularly record electrocardiosignal data at any time and any place in daily life of people, and carry out calculation and analysis to determine whether the electrocardiosignal is abnormal or not, thereby realizing the monitoring of heart activity.

In order to improve the performance of a portable electrocardiograph, it is necessary to enable the portable electrocardiograph to analyze complex electrocardiographic signals. In some related technologies, a machine learning technique is applied to a portable electrocardiograph detector, so that the portable electrocardiograph detector has the capability of analyzing complex electrocardiograph signals. For example, an artificial neural network and a support vector machine are applied to a portable electrocardiograph to identify the abnormal type of a complex electrocardiograph signal, however, in such a manner, professional domain knowledge needs to be used for constructing artificial features, so that a detection result is easily interfered by human factors, for example, the problems of incomplete feature dimension and weak feature expression capability are easily caused when the artificial features are constructed, and the performance of the portable electrocardiograph is further influenced. For another example, a deep neural network is applied to a portable electrocardiograph to detect an abnormality of an electrocardiographic signal, but the deep neural network requires a large number of training samples, and when the amount of data of the abnormal electrocardiographic signal as the training sample is small, the trained deep neural network has a large performance defect, which further affects the performance of the portable electrocardiograph.

In conclusion, how to make a portable electrocardiogram detector analyze complex electrocardiogram signals with high performance becomes a technical problem which needs to be solved urgently.

Disclosure of Invention

The embodiment of the application provides an electrocardiosignal detection device and equipment, which are used for solving the technical problem that the portable electrocardiosignal detector cannot analyze complex electrocardiosignals with high performance in the related technology.

In a first aspect, an embodiment of the present application provides an electrocardiograph signal detection apparatus, including:

the signal acquisition module is used for acquiring a first electrocardiosignal to be detected;

the feature extraction module is used for inputting the first electrocardiosignal to a first neural network and extracting the depth feature of the first electrocardiosignal by the first neural network;

the prior acquisition module is used for acquiring preset electrocardio prior characteristics;

the feature fusion module is used for fusing the depth feature and the electrocardio prior feature to obtain a fusion feature;

and the signal detection module is used for inputting the fusion characteristics to a second neural network, and the second neural network outputs the detection result of the first electrocardiosignal according to the fusion characteristics.

In a second aspect, an embodiment of the present application further provides an electrocardiograph signal detection apparatus, including:

one or more processors;

a memory for storing one or more programs;

when executed by the one or more processors, cause the one or more processors to perform the calculations of the cardiac signal detection apparatus according to the first aspect.

According to the electrocardiosignal detection device and equipment, the signal acquisition module is used for acquiring a first electrocardiosignal to be detected, the feature extraction module is used for extracting the depth feature of the first electrocardiosignal by utilizing the first neural network, the prior acquisition module is used for acquiring the preset electrocardio prior feature, the feature fusion module is used for fusing the depth feature and the electrocardio prior feature, and the signal detection module is used for inputting the fused feature obtained after fusion into the second neural network to obtain the detection result of the first electrocardiosignal. The electrocardio priori characteristics are set, the electrocardio priori characteristics and the depth characteristics are fused to obtain fusion characteristics with higher distinguishing capability, the situation that the first neural network and the second neural network have performance defects when the data volume of abnormal electrocardio signals serving as training samples is less is avoided, the electrocardio signal detection device has higher practicability, and the problems of incomplete characteristic dimensionality and weaker characteristic expression capability when the electrocardio priori characteristics are independently used are solved.

Drawings

Fig. 1 is a schematic structural diagram of an electrocardiograph signal detection device according to an embodiment of the present application;

fig. 2 is a schematic structural diagram of a residual convolution network according to an embodiment of the present application;

FIG. 3 is a schematic diagram of a deep neural network provided in an embodiment of the present application;

FIG. 4 is a flow chart illustrating processing of an ECG signal according to an embodiment of the present application;

fig. 5 is a schematic structural diagram of an electrocardiograph signal detection device according to an embodiment of the present application.

Detailed Description

The present application will be described in further detail with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are for purposes of illustration and not limitation. It should be further noted that, for the convenience of description, only some of the structures associated with the present application are shown in the drawings, not all of them.

An embodiment of the present application provides an electrocardiographic signal detection device, which can be used for detecting electrocardiographic signals to analyze whether the electrocardiographic signals are abnormal. The electrocardiosignal detection device can be integrated in electrocardiosignal detection equipment. The electrocardiosignal detection device can be formed by two or more physical entities or can be formed by one physical entity. The electrocardiosignal detection equipment can be an electrocardiogram machine, an electrocardio detector, a portable electrocardio detector and the like. In one embodiment, a portable electrocardiograph is taken as an example for description, and the portable electrocardiograph can acquire electrocardiographic signals of a human body and analyze the electrocardiographic signals to obtain an abnormal detection result of the electrocardiographic signals.

For example, in order to facilitate understanding of the operation of the cardiac signal detection device, in one embodiment, the cardiac signal detection device may detect Atrial Fibrillation (AF) through the cardiac signal.

Atrial fibrillation is characterized by disordered atrial activity and subsequent complications such as stroke, myocardial infarction and the like, has high disability rate and death rate and seriously harms human health and life, and has the frequency of atrial excitation reaching 300-600 times/minute so that the heartbeat frequency is usually fast and irregular and sometimes reaches 100-160 times/minute, is much faster than that of normal people and is absolutely irregular, and the atrium loses the effective contraction function. At the onset of atrial fibrillation, two important clinical manifestations are: 1. the P wave in the electrocardiosignal disappears and is replaced by atrial fibrillation wave (f wave) with absolutely irregular shape, space and amplitude; 2. the QRS wave intervals in the electrocardiosignals are absolutely irregular, and the forms and amplitudes of the QRS wave intervals can be frequently different, namely RR intervals in the electrocardiosignals are absolutely irregular. It can be understood that the electrocardiographic signal is composed of a plurality of heart beats, wherein a heart beat refers to a segment of signal containing a P wave, a QRS wave and a T wave. The P-wave represents the electrical activity of the atrial polarization. The QRS wave and the T wave represent ventricular polarized and repolarized electrical activity, respectively. The first downward wave in a QRS wave is the Q wave, the upward wave is the R wave, and the next downward wave is the S wave. Sometimes, the heartbeat includes a U wave, which is a wide and low wave appearing from 0.02s to 0.04s after the T wave. The RR interval refers to a time limit (time length) between two R waves in the electrocardiographic signal. In one embodiment, when the cardiac signal detection device detects atrial fibrillation, the two clinical manifestations are used as references to construct the electrocardio prior characteristics, so that the cardiac signal detection device may process complex cardiac signals.

Fig. 1 is a schematic structural diagram of an electrocardiograph signal detection apparatus according to an embodiment of the present application. Referring to fig. 1, the electrocardiographic signal detecting device includes: a signal acquisition module 101, a feature extraction module 102, a priori acquisition module 103, a feature fusion module 104, and a signal detection module 105.

The signal acquisition module 101 is configured to acquire a first electrocardiosignal to be detected; the feature extraction module 102 is configured to input the first cardiac signal to a first neural network, and extract a depth feature of the first cardiac signal by the first neural network; the prior acquisition module 103 is used for acquiring preset electrocardio prior characteristics; the feature fusion module 104 is configured to fuse the depth feature and the electrocardiographic prior feature to obtain a fusion feature; and the signal detection module 105 is used for inputting the fusion characteristics into the second neural network, and the second neural network outputs the abnormal detection result of the first electrocardiosignal according to the fusion characteristics.

For example, the signal obtaining module 101 may obtain an electrocardiogram signal to be currently detected, and in an embodiment, the electrocardiogram signal to be currently detected is recorded as a first electrocardiogram signal. Optionally, the length of the first cardiac signal is a preset length, and if the preset length is 10s, the first cardiac signal is a cardiac signal having a length of 10s. The first cardiac signal may be one or more, and generally, one first cardiac signal includes a plurality of heartbeats.

In one embodiment, the electrocardiosignal detection device is provided with a device for collecting electrocardiosignals, so that the electrocardiosignals of a human body can be directly collected through the device. Optionally, the device for acquiring an electrocardiographic signal may be a lead, a sensor, or the like, which is not limited in the embodiment. In the technical terms of electrocardiography, the placement position of the electrodes on the body surface of the human body and the connection mode of the electrodes and the amplifier when recording the electrocardiogram are called as leads of the electrocardiogram, and according to the number of lead channels, the electrocardiographic signals obtained through leads can be divided into single-lead electrocardiographic signals (i.e. the electrocardiographic signals collected through one lead) and multi-lead electrocardiographic signals (i.e. the electrocardiographic signals collected through a plurality of leads), wherein the multi-lead electrocardiographic signals can be considered to be composed of a plurality of single-lead electrocardiographic signals, and the number of leads of the common multi-lead electrocardiographic signals is three leads, six leads, twelve leads, eighteen leads and the like. In one embodiment, the signal obtaining module 101 may obtain a currently acquired cardiac signal as a first cardiac signal to be detected. Generally, the electrocardiographic signal directly acquired from a human body is an analog signal, and the electrocardiographic signal required in the subsequent detection is a digital signal, that is, the first electrocardiographic signal needs to be a digital signal, so in an embodiment, the signal acquisition module 101 further needs to preprocess the acquired electrocardiographic signal to obtain the first electrocardiographic signal meeting the requirement. Optionally, during the preprocessing, after the signal obtaining module 101 obtains the electrocardiographic signal (analog signal) directly collected from the human body, the electrocardiographic signal is subjected to impedance matching, filtering, amplifying and the like, so as to improve the quality of the electrocardiographic signal, then the signal obtaining module 101 converts the analog electrocardiographic signal into a digital electrocardiographic signal by using an analog-to-digital converter, then the signal obtaining module 101 filters noise interference in the electrocardiographic signal, and resamples the electrocardiographic signal, so as to resample the electrocardiographic signal to a target frequency, wherein a value of the target frequency can be set according to an actual situation. Then, the signal obtaining module 101 cuts the resampled electrocardiograph signal into a preset length to obtain at least one electrocardiograph signal segment, and then performs normalization processing on the electrocardiograph signal segment to obtain a first electrocardiograph signal to be processed.

After obtaining the first electrocardiosignal, the signal obtaining module 101 sends the first electrocardiosignal to the feature extraction module 102. The feature extraction module 102 extracts depth features of the first cardiac signal. Wherein the depth feature is a feature related to heartbeat. In one embodiment, the feature extraction module 102 may extract the depth features of the first cardiac signal using a neural network, and denote the neural network used by the feature extraction module 102 as the first neural network. The first neural Network may adopt a residual convolutional Network (resource), a VGG Network (Visual Geometry Group Network), an AlexNet, or the like. In one embodiment, the first neural network is described by taking a residual convolutional network as an example, and in this case, the first neural network includes at least one residual convolutional network. For example, fig. 2 is a schematic structural diagram of a residual convolutional network provided in an embodiment of the present application, and referring to fig. 2, a residual convolutional network includes: IDConvolition (ID convolution layer), batch Normalization (Batch Normalization layer), reLU function, IDConvolition (ID convolution layer), batch Normalization (Batch Normalization layer), and Downsample (downsampling layer). In one embodiment, the number and parameters of the residual convolutional networks in the first neural network may be set in combination with the actual requirements. Optionally, the first neural network may further include a module consisting of idcontribution, batch Normalization, reLU function, and MaxPool (maximum pooling layer) in addition to the at least one residual convolutional network. After the first electrocardiosignal is input to the first neural network, the first electrocardiosignal is convoluted through the modules, and then the first electrocardiosignal is input to the residual convolution network to obtain a better depth characteristic. Wherein, maxPool's size can set up according to actual demand. It should be noted that the first neural network used by the feature extraction module 102 is a neural network that is trained and can be used for deploying applications.

Illustratively, in order to improve the detection capability of the electrocardiosignal detection device, an electrocardio prior characteristic is used, wherein the characteristic category of the electrocardio prior characteristic is set manually, namely the characteristic category related to the electrocardio signal abnormal type detectable by the electrocardiosignal detection device is set as the electrocardio prior characteristic in combination with the electrocardiosignal abnormal type. For example, when the electrocardiographic signal detection device detects atrial fibrillation, the feature class of the electrocardiographic prior feature is related to the clinical manifestation of the electrocardiographic signal at the onset of atrial fibrillation. For example, for atrial fibrillation, RR interval irregularity of an ecg signal is a more significant clinical manifestation, and therefore, the standard deviation of the RR interval may be used as a feature category of the ecg priori features to indicate whether the RR interval is regular or not according to the standard deviation of the RR interval. In one embodiment, the electrocardiographic signals in the public data set are used to construct an electrocardiographic prior signature. For example, an electrocardiographic signal in the public data set is obtained, each RR interval in the electrocardiographic signal is determined, and then a standard deviation of the RR intervals is calculated to obtain an electrocardiographic prior characteristic. It can be understood that besides the standard deviation of the RR intervals, other electrocardio-prior characteristics can be constructed by combining the clinical manifestations of atrial fibrillation so as to enrich the characteristic classes of the electrocardio-prior characteristics. Optionally, after the construction of the prior electrocardiographic feature is completed, the prior electrocardiographic feature is normalized for subsequent use, where an embodiment of an implementation manner of normalization is not limited, for example, min-max normalization is adopted, so that a value of the prior electrocardiographic feature is mapped to a value between [0 and 1 ]. Optionally, the electrocardio prior characteristic may be constructed by an electrocardiosignal detection device, and at this time, the electrocardiosignal detection device further includes a relevant module for constructing the electrocardio prior characteristic, or the electrocardio prior characteristic may be constructed by other devices, and is sent to the electrocardiosignal detection device after the construction is completed, so as to be used by the electrocardiosignal detection device in the electrocardiosignal detection device. In one embodiment, the apriori acquisition module 103 directly acquires the constructed electrocardiographic apriori characteristics. It should be noted that, the occurrence time sequence between the acquisition of the apriori electrocardiographic feature by the apriori acquisition module 103 and the acquisition of the first electrocardiographic signal by the signal acquisition module 101 is not limited in this embodiment.

Illustratively, the feature fusion module 104 obtains the depth feature extracted by the feature extraction module 102 and the prior electrocardiographic feature obtained by the prior acquisition module 103, and fuses the depth feature and the prior electrocardiographic feature. Optionally, a splicing fusion mode is adopted to fuse the depth characteristic and the electrocardio prior characteristic. For example, the depth feature is represented as x _DF The prior characteristic of electrocardio is expressed as x _HF Then the fusion feature obtained after splicing fusion is represented as x _Fusion And x is _Fusion ＝[x _DF ，x _HF ]。

Illustratively, the signal detection module 105 obtains the fusion feature obtained by the feature fusion module 104, and obtains the detection result of the first cardiac signal through the fusion feature. In one embodiment, when the cardiac signal detection device can detect atrial fibrillation, the detection result may be that the first cardiac signal is of an atrial fibrillation type or that the first cardiac signal is of a non-atrial fibrillation type. In one embodiment, the signal detection module 105 processes the fusion feature using a neural network and obtains a detection result of the first cardiac signal, and at this time, the currently used neural network is referred to as a second neural network. Optionally, the structure of the second neural network may be set according to actual conditions, and in one embodiment, the second neural network includes a Bi-directional Long Short-Term Memory network layer (bilst), a scatter layer, and a sense layer. The BiLSTM is composed of a forward LSTM and a backward LSTM, the BiLSTM is used for enhancing the electrocardio prior characteristics, the Flatten layer is used for realizing multidimensional input in a one-dimensional mode, and the Dense layer is a full connection layer. Optionally, the BiLSTM layer may be replaced by a Long Short-Term Memory (LSTM) layer, a Recurrent Neural Network (RNN) layer, a Time Convolutional Network (TCN) layer, or the like. In one embodiment, the second neural network processes the fused features, which may be understood as the fused features, and uses the classification result as the detection result, for example, when the ecg device detects atrial fibrillation, the classification of the second neural network includes atrial fibrillation type and non-atrial fibrillation type. At this time, a SoftMax function can be further used behind the Dense layer of the second neural network, so that a classification result is obtained through the SoftMax function, and a detection result is obtained. The parameters of each layer in the second neural network can be set according to actual requirements. It should be noted that the second neural network used by the signal detection module 105 is a neural network that is trained and can be used to deploy an application.

It is understood that the first neural network and the second neural network may constitute a deep neural network and be deployed in the cardiac electrical signal detection device for use by the feature extraction module 102 and the signal detection module 105. For example, fig. 3 is a schematic diagram of a deep neural network provided in an embodiment of the present application. Referring to fig. 3, the deep neural network adopts an end-to-end LSTM + CNN (convolutional neural network) network structure, where the CNN is a first neural network, the CNN includes modules composed of idcontribution, batch Normalization, reLU function, and MaxPool, and also includes 4 residual convolutional networks, and a second neural network is connected behind the 4 residual convolutional networks, where the second neural network includes a BiLSTM layer, a scatter layer, a Dense layer, and a SoftMax function, which are sequentially placed. The CNN may extract a depth feature from the first cardiac signal. And then, the depth features extracted by the CNN and the electrocardio prior features are fused and input into a BilSTM layer to realize the enhancement of the electrocardio prior features, and then, the subsequent Flatten layer, the Dense layer and the SoftMax function are facilitated to output the detection result. Optionally, the number of neurons in the BiLSTM layer is 32, the number of neurons in the detect layer is 64, and the size and channel of the convolution kernel of IDConvolution in the residual convolution network are set to 11 and 64, respectively. In the module composed of IDConvolation, batch Normalization, reLU function and MaxPool, the convolution kernel size and channel of IDConvolation are set to 11 and 128 respectively, and the size of MaxPool is 2.

It is understood that, in the above embodiments, the description is given by taking the example of detecting atrial fibrillation, in practical applications, the electrocardiographic signal detecting device may also detect other abnormal types, for example, the electrocardiographic signal detecting device may also detect atrial flutter, in this case, the electrocardiographic prior characteristic is also related to clinical performance of the atrial flutter, and the detection result may also include whether the first electrocardiographic signal is atrial flutter.

The technical means that the signal acquisition module is used for acquiring a first electrocardiosignal to be detected, the feature extraction module is used for extracting the depth feature of the first electrocardiosignal by using the first neural network, the priori acquisition module is used for acquiring the preset electrocardio priori feature, the feature fusion module is used for fusing the depth feature and the electrocardio priori feature, and the signal detection module is used for inputting the fused feature obtained after fusion into the second neural network to obtain the detection result of the first electrocardiosignal solves the technical problem that the portable electrocardiosignal detector cannot analyze complex electrocardiosignals with high performance. The electrocardio priori characteristics are set, the electrocardio priori characteristics and the depth characteristics are fused to obtain fusion characteristics with stronger distinguishing capability, the situation that the first neural network and the second neural network have performance defects when the data volume of abnormal electrocardiosignals serving as training samples is less is avoided, the electrocardiosignal detection device has stronger practicability, and the problems that the characteristic dimensionality is incomplete and the characteristic expression capability is weaker when the electrocardio priori characteristics are used independently can be avoided. The CNN-LSTM deep neural network consisting of the first neural network and the second neural network can improve the detection accuracy, avoid the limitation caused by local perception of the first neural network when the first neural network (CNN) is used independently, and reduce the interference of noise in the electrocardio prior characteristics on the detection result and the interference of invalid characteristics on the detection result.

In one embodiment, the electrocardiographic signal detection device constructs an electrocardiographic prior characteristic, and at this time, the electrocardiographic signal detection device further includes: the heart beat label acquisition module is used for acquiring a second cardiac electric signal with a heart beat label, and the heart beat label is used for labeling the heart beat position of each heart beat in the second cardiac electric signal; and the prior characteristic determining module is used for obtaining the electrocardio prior characteristic based on the second electrocardiosignal according to the heart beat position in the heart beat mark.

Illustratively, the electrocardiographic signal used for constructing the electrocardiographic prior characteristic is recorded as a second electrocardiographic signal, and optionally, the second electrocardiographic signal is an electrocardiographic signal in the public data set. In one embodiment, the second cardiac signal has a heartbeat label, wherein the heartbeat label is used to show a heartbeat position in the second cardiac signal, and the heartbeat position may include: p wave position, QRS wave position, and T wave position, or include: QRS wave location. Optionally, the used public data set may or may not include a heartbeat label, for example, when the public data set MIT-AR is used, all the electrocardiographic signals in the MIT-AR have the heartbeat label, and for example, when the public data set CPSC2018 is used, the electrocardiographic signals in the CPSC2018 do not have the heartbeat label. When the public data set contains the heartbeat label, the heartbeat label acquisition module can directly acquire the electrocardiosignal in the public data set as a second electrocardiosignal and acquire the corresponding heartbeat label. When the public data set does not contain the heartbeat label, the heartbeat label acquisition module detects a second electrocardiosignal after acquiring the second electrocardiosignal in the public data set so as to obtain a heartbeat position and construct a corresponding heartbeat label. At this moment, heart claps mark collection module and includes: the first signal acquisition unit is used for acquiring a second electrocardiosignal; and the heartbeat labeling unit is used for identifying the QRS wave in the second electrocardiosignal and determining the heartbeat label of the second electrocardiosignal according to the position of the QRS wave in the second electrocardiosignal. Illustratively, a second cardiac signal is acquired in the disclosed data set by a second signal acquisition unit. And then, the heart beat marking unit detects the second electrocardiosignal to obtain a QRS wave position, and obtains a heart beat position according to the QRS wave position, so as to obtain a heart beat mark. It can be understood that a heart beat comprises a group of continuous P waves, QRS waves and T waves, the position of the QRS waves is located, and the positions of the P waves and the T waves in the heart beat can be estimated according to the position of the QRS waves. The embodiment of the technical means used by the heartbeat tagging unit to detect the second electrocardiosignal to obtain the QRS wave position is not limited, for example, the heartbeat tagging unit identifies the second electrocardiosignal by using a pre-trained neural network to obtain the QRS wave position.

Illustratively, when the prior characteristic determining module constructs the electrocardiographic prior characteristic, the length of the second electrocardiographic signal is set to be a preset length, so as to ensure that the length of the second electrocardiographic signal is the same as the length of the first electrocardiographic signal. Optionally, after the prior characteristic determining module obtains the second cardiac electric signal sent by the heartbeat label collecting module, a section of cardiac electric signal with a preset length is intercepted. And then, the prior characteristic determining module constructs the electrocardio prior characteristic by using heartbeat labeling. In one embodiment, the corresponding electrocardio prior characteristics are constructed by combining with the heart beat position in the heart beat label, for example, when the electrocardio prior characteristics are RR interval standard deviations, the prior characteristic determination module determines the positions of the R waves in the second electrocardiosignal according to the heart beat position, obtains RR intervals according to the positions of the R waves, and further calculates the standard deviations of the RR intervals. Illustratively, after the prior characteristic determining module constructs the electrocardio prior characteristic, min-max normalization is adopted to normalize the electrocardio prior characteristic, and the normalized electrocardio prior characteristic is stored for the prior acquiring module to acquire.

Optionally, during each time of electrocardiographic signal detection, the electrocardiographic prior feature can be constructed by the cardiac beat marking acquisition module and the prior feature determination module. Or the heart beat label acquisition module and the prior characteristic determination module construct the electrocardio prior characteristic and store the electrocardio prior characteristic so as to directly use the constructed electrocardio prior characteristic during each electrocardiosignal detection.

It can be understood that the construction process of the electrocardio prior characteristic is described by taking the example that the electrocardio signal detection device constructs the electrocardio prior characteristic by itself. In practical application, the electrocardiogram prior characteristics can be constructed by other devices, and at this time, the electrocardiogram detection device can directly store the constructed electrocardiogram prior characteristics so that the prior acquisition module 103 can directly acquire the electrocardiogram prior characteristics.

In one embodiment, the electrocardiographic signal detection device is described as detecting atrial fibrillation, where the electrocardiographic prior characteristics include: at least one of standard deviation of RR interval, coefficient of variation of RR interval, standard deviation of PR interval variability, and standard deviation of P-wave variability. Accordingly, the first cardiac signal is either of atrial fibrillation type or non-atrial fibrillation type.

From the physiological aspect, the RR interval of the electrocardiosignals is absolutely irregular when the atrial fibrillation occurs, so that the RR interval standard deviation can be used as the electrocardio priori characteristic to evaluate whether the RR interval is regular or not according to the RR interval standard deviation, and further judgment of the atrial fibrillation is carried out. The RR interval refers to a time limit (time length) between two R waves in the electrocardiographic signal.

In one embodiment, the electrocardiographic prior features comprise: RR interval standard deviation, the priori characteristic determination module includes: the first extraction unit is used for extracting each RR interval in the second electrocardiosignal according to the heart beat position in the heart beat mark, wherein the RR interval refers to the time limit between two adjacent R waves; and the first standard deviation calculation unit is used for calculating a first standard deviation of the RR interval and taking the first standard deviation as the standard deviation of the RR interval.

Illustratively, the first extraction unit finds the position of each R-wave in the second cardiac signal according to the cardiac beat position in the cardiac beat label, and calculates the time limit (time length) between two adjacent R-waves according to the position of each R-wave to obtain the RR interval. At this time, there is one RR interval per two R-waves. It should be noted that the second cardiac signal includes a plurality of RR intervals. And then, the first standard deviation calculating unit acquires each RR interval obtained by the first extracting unit and calculates the standard deviation of each RR interval, and the standard deviation can reflect the dispersion degree of each RR interval. In one embodiment, the currently calculated standard deviation is taken as the first standard deviation, and the first standard deviation is taken as the RR interval standard deviation.

From the physiological aspect, the RR interval variation coefficient of the electrocardiosignals can reflect RR interval irregularity to a certain extent when atrial fibrillation occurs, so that the RR interval variation coefficient can be used as the electrocardio priori characteristic to judge atrial fibrillation. Wherein, the RR interval variation coefficient refers to the ratio of the standard deviation of the RR interval to the average value of the RR interval.

In one embodiment, the electrocardiographic prior features comprise: RR interval coefficient of variation, the priori characteristic confirms the module includes: the second extraction unit is used for extracting each RR interval from the second electrocardiosignal according to the heart beat position in the heart beat mark; a first parameter determination unit, configured to calculate a second standard deviation and a first average value of the RR intervals; and a coefficient of variation determining unit, configured to obtain the coefficient of variation during the RR period according to the second standard deviation and the first average value.

For example, the second extraction unit has the same function as the first extraction unit, and is not described herein again. After the second extracting unit extracts the RR intervals, the first parameter determining unit calculates the standard deviation of each RR interval, and records the currently calculated standard deviation as the second standard deviation. The first parameter determination unit further calculates an average value of each RR interval, and records the currently calculated average value as a first average value. Then, the coefficient of variation determining unit obtains the second standard deviation and the first average value calculated by the first parameter determining unit, and then calculates a ratio of the second standard deviation to the first average value, and takes the ratio as the coefficient of variation of RR interval.

From the physiological point of view, when the atrial fibrillation occurs, the clinical manifestation of the electrocardiosignal is that the P wave disappears, at this time, the position of the P wave in the electrocardiosignal cannot be detected, and further the PR interval of the electrocardiosignal is irregular, wherein the PR interval refers to the time limit (time length) between the P wave in the electrocardiosignal and the starting point of the adjacent QRS wave. Since the standard deviation of variability of the PR interval can assess whether the PR interval is regular, in one embodiment, the standard deviation of variability of the PR interval is used as an electrocardiographic prior feature to make the determination of atrial fibrillation. The ratio of the PR interval of one heart beat to the average PR interval of the cardiac signal can be used as the PR interval variability.

In one embodiment, the electrocardiographic prior features include: a standard deviation of variability of PR intervals, the a priori characteristic determination module comprising: a third extraction unit, configured to extract PR intervals in the second cardiac electrical signal according to the heart beat position in the heart beat label, where the PR intervals refer to time periods from a P-wave start point to an adjacent QRS-wave start point; a second parameter determination unit for calculating a second average of the PR intervals; a first variability calculation unit for calculating a first ratio of each PR interval to the second average value, and taking the first ratio as the PR interval variability; a second standard deviation calculation unit for calculating a third standard deviation of the variability of the PR interval and taking the third standard deviation as the standard deviation of the variability of the PR interval.

Illustratively, the third extracting unit finds the positions of the P waves and the positions of the QRS waves in the second electrocardiosignal according to the heart beat position in the heart beat label, then obtains the position of the start point of the P wave according to the P wave position, obtains the position of the start point of the QRS wave according to the QRS wave position, and then takes the time limit (time length) from the position of the start point of the P wave to the position of the start point of the QRS wave (i.e., the QRS wave adjacent to the P wave) in the same heart beat as the PR interval. In one embodiment, a reference point detection algorithm is used to obtain the position of the P wave starting point and the position of the QRS wave starting point, that is, the P wave starting point and the QRS wave starting point are used as reference points, and the positions of the P wave starting point and the QRS wave starting point are obtained by combining the position characteristics of the reference points in the electrocardiosignal (such as the position range of the P wave starting point generally appearing in the heartbeat), so as to obtain the PR interval. It can be understood that when the P wave disappears, the evaluation difficulty of PR interval irregularity is larger, therefore, the time limit (time length) from the position of the starting point of the P wave to the position of the starting point of the QRS wave is taken as the PR interval, so that the reliability of the PR interval can be ensured, and further, the reliability of evaluating the PR interval irregularity can be ensured. And then, the second parameter determining unit acquires each PR interval obtained by the third extracting unit, calculates the average value of each PR interval and marks the currently calculated average value as a second average value. And then, the first variability calculating unit is used for calculating a ratio between each PR interval and the second average value, and recording the ratio as a first ratio, wherein each PR interval corresponds to one first ratio, and the first ratio corresponding to each PR interval can be used as the PR interval variability of the corresponding PR interval. Then, the second standard deviation calculating unit obtains PR interval variability of each PR interval, and calculates a standard deviation of each PR interval variability, and at this time, the currently calculated standard deviation is recorded as a third standard deviation, where a calculation manner of the third standard deviation is the same as a calculation manner of the first standard deviation and the second standard deviation, which is not described herein again. The third standard deviation may be considered as the standard deviation of the variability of the PR interval.

From a physiological point of view, the clinical manifestations of the electrocardiosignals, in addition to the disappearance of the P-wave, also appear as f-waves at the onset of atrial fibrillation. However, the difficulty of detecting the disappearance of the P wave and the appearance of the f wave in the electrocardiograph signal is high, and the main difficulty is that the f wave is irregular, so that the difficulty of extracting the f wave is high, and thus, the method is not beneficial to judging whether atrial fibrillation occurs in a mode of extracting the f wave. However, since the difference in P-wave variability between the atrial-fibrillation electrocardiographic signal and the non-atrial-fibrillation electrocardiographic signal is large due to the disappearance of the P-wave, atrial fibrillation can be detected based on the P-wave variability, which is the ratio of the width of the P-wave in the current heart beat to the average value of the widths of the P-waves in the electrocardiographic signal. The P wave width refers to the time length of P waves in the electrocardiosignals, and each P wave corresponds to one P wave width. In one embodiment, the standard deviation of variability of the P-wave in the second cardiac electrical signal is used as an electrocardiographic prior feature to make the determination of atrial fibrillation.

In one embodiment, the electrocardiographic prior features include: the standard deviation of P wave variability, the prior feature determination module comprises: a fourth extraction unit, configured to extract the P-wave width of each P-wave in the second cardiac electric signal according to the cardiac beat position in the cardiac beat label; the parameter determining unit is used for calculating a third average value of the P wave width; the second variability calculating unit is used for calculating a second ratio of each P wave width to the third average value, and taking the second ratio as P wave variability; and the third standard deviation calculation unit is used for counting a fourth standard deviation of the P wave variability and taking the fourth standard deviation as the standard deviation of the P wave variability.

Illustratively, the fourth extraction unit finds the position of each P wave in the second cardiac electric signal according to the cardiac beat position in the cardiac beat label, and obtains the P wave width of each P wave according to the P wave position. Optionally, the P-wave width is obtained according to a starting point of the P-wave, and in an embodiment, a reference point detection algorithm is used to obtain positions of the starting point and the ending point of the P-wave, so as to obtain the P-wave width according to the position of the starting point and the position of the ending point of the P-wave, so as to ensure reliability of the P-wave width. The reference point detection algorithm is the same as the reference point detection algorithm used for calculating the standard deviation of the interval variability of the PR, and is not described herein again. And then, the third parameter determining unit acquires each P wave width obtained by the fourth extracting unit, calculates an average value of each P wave width, and records the currently calculated average value as a third average value. And then, the second variability calculating unit is used for calculating the ratio between the width of each P wave and the third average value, and recording the ratio as a second ratio, wherein each P wave corresponds to one second ratio, and the second ratio corresponding to the P wave can be used as the P-wave variability of the P wave. Then, the third standard deviation calculating unit obtains the P-wave variability of each P-wave, and calculates the standard deviation of each P-wave variability, and at this time, the currently calculated standard deviation is recorded as a fourth standard deviation, where the calculation method of the fourth standard deviation is the same as the calculation methods of the first standard deviation, the second standard deviation, and the third standard deviation, and details are not described here. The fourth standard deviation can be considered as the standard deviation of the variability of the P-wave.

In practical application, one or more electrocardiographic prior features can be selected from the four electrocardiographic prior features according to practical requirements. When a plurality of electrocardio prior characteristics are selected, the plurality of electrocardio prior characteristics are determined by the same second electrocardio signal, and at the moment, the unit for calculating the electrocardio prior characteristics can be multiplexed. For example, when the electrocardiographic prior feature includes an RR-period standard deviation and an RR-period variation coefficient, the first extraction unit and the second extraction unit may be the same unit, the first standard deviation and the second standard deviation are the same value, and the first parameter determination unit and the first standard deviation calculation unit may be the same unit, which may calculate the standard deviation during the RR period to obtain the RR-period standard deviation, and may also obtain an average value during the RR period, so that the variation coefficient determination unit may calculate the RR-period variation coefficient. It should be noted that the more the types of the electrocardiographic prior features are, the more accurate the detection result of the first electrocardiographic signal is, and for example, when the four electrocardiographic prior features are used simultaneously, the most accurate the detection result of the first electrocardiographic signal is. After the electrocardio priori characteristics are built, the electrocardio priori characteristics can be used by a priori acquisition module, then the characteristic fusion module realizes characteristic fusion, and a detection result that the first electrocardiosignal is in an atrial fibrillation type or the first electrocardiosignal is in a non-atrial fibrillation type is obtained by the signal detection module.

It can be understood that, for the deep neural network (as shown in fig. 3) composed of the first neural network and the second neural network, the first electrocardiosignal can also be detected based on the depth features directly without adding the electrocardio-prior features. Taking atrial fibrillation detection (AF _ detection) as an example, the deep neural network is tested without adding the electrocardiographic prior feature, and the performance of the deep neural network is shown in table 1.

TABLE 1

The performance of the deep neural network is denoted as Baseline1, and table 1 shows the Sensitivity (SE) and Specificity (SP) of each item of test data when the deep neural network detects atrial fibrillation.

Considering that when the electrocardio prior characteristics are added, the depth characteristics and the electrocardio prior characteristics need to be fused, therefore, the fusion needs are combined, global pooling is carried out on the depth neural network, at this time, by taking atrial fibrillation detection (AF _ detection) as an example, when the electrocardio prior characteristics are not added, the depth neural network for global pooling is tested, and the performance of the depth neural network is shown in table 2.

TABLE 2

The performance of the deep neural network for global-posing is denoted as Baseline2, and table 2 shows the Sensitivity (SE) and Specificity (SP) of each test data when the deep neural network for global-posing detects atrial fibrillation. Meanwhile, baseline1 (performance of deep neural network without global-poling) is also shown in Table 2. Comparing various data of Baseline1 and Baseline2 in Table 2 shows that when no electrocardio prior characteristics are added, the performance of the deep neural network for global-posing is reduced to a certain extent, the reduction range is about 1-2%, and the main reason for the reduction is that when global-posing is carried out, global-posing compresses and reduces the dimension of the deep neural network, so that the performance of the deep neural network is influenced to a certain extent.

In one embodiment, taking deep neural network detection (AF _ detection) as an example, the RR interval standard deviation is used as an electrocardiographic prior feature, and the RR interval standard deviation is added to the deep neural network for global-posing, that is, the depth feature obtained by CNN and the RR interval standard deviation are fused and input to the BiLSTM layer. At this time, after testing the deep neural network, the performance of the deep neural network is shown in table 3.

TABLE 3

After adding the RR interval standard deviation, the performance of the deep neural network is recorded as B2+ RR _ sdnn, where RR _ sdnn represents the RR interval standard deviation, and table 3 shows the Sensitivity (SE) and Specificity (SP) of each test data when the deep neural network added with the RR interval standard deviation is used for detecting atrial fibrillation. Meanwhile, table 3 also shows Baseline1 and Baseline2 in table 2. As can be seen from table 3, most of the data in B2+ rr _ sdnn are improved to some extent compared to Baseline1, that is, the performance of the deep neural network is improved to some extent, reference may be made to the column in table 3 where P (our + B1) is located, which shows a significant correlation between Baseline1 and data in B2+ rr _ sdnn.

In one embodiment, taking atrial fibrillation detection (AF _ detection) as an example, the RR interval coefficient of variation is used as an electrocardiographic prior feature, and the RR interval coefficient of variation is added to the deep neural network for global-posing, that is, the depth feature obtained by CNN and the RR interval coefficient of variation are fused and then input to the BiLSTM layer for atrial fibrillation detection. At this time, after testing the deep neural network, the performance of the deep neural network is shown in table 4.

TABLE 4

After the RR interval coefficient of variation is added, the performance of the deep neural network is marked as B2+ RR _ cv, RR _ cv represents the RR interval coefficient of variation, and table 4 shows the Sensitivity (SE) and Specificity (SP) of each item of test data when the deep neural network added with the RR interval coefficient of variation performs atrial fibrillation detection. Meanwhile, table 4 also shows Baseline1 and Baseline2 in table 2. As can be seen from table 4, most of the data in B2+ rr _ cv are improved to some extent compared to Baseline1, and reference may be made to the column of table 4P (ours + B1), which shows the significant correlation between Baseline1 and each item of data in B2+ rr _ cv.

In one embodiment, taking atrial fibrillation detection (AF _ detection) as an example, the RR period standard deviation and the RR interval variation coefficient are used as the electrocardiographic prior characteristics, and the RR period standard deviation and the RR interval variation coefficient are added to the deep neural network for global-posing, that is, the depth characteristics obtained by CNN, the RR period standard deviation and the RR interval variation coefficient are fused and then input to the BiLSTM layer for atrial fibrillation detection. The performance of the deep neural network when tested is shown in table 5.

TABLE 5

After the RR period standard deviation and the RR interval variation coefficient are added, the performance of the deep neural network is recorded as B2+ RR _ cv + RR _ sdnn, and table 5 shows the Sensitivity (SE) and Specificity (SP) of each item of test data when the deep neural network added with the RR period standard deviation and the RR interval variation coefficient performs atrial fibrillation detection. Meanwhile, the performance of the deep neural network when only RR interval standard deviation is added (data items under B2+ RR _ sdnn in table 5) and the performance of the deep neural network when only RR interval variation coefficient is added (data items under B2+ RR _ cv in table 5) are also shown in table 5. As can be seen from table 5, after the standard deviation during RR and the RR interval coefficient of variation are added simultaneously, the performance of the deep neural network is better than the performance of the deep neural network only with the standard deviation during RR or the RR interval coefficient of variation, so that the standard deviation during RR and the RR interval coefficient of variation can be used jointly as the electrocardiographic prior characteristics.

In one embodiment, taking atrial fibrillation detection (AF _ detection) as an example, the standard deviation of PR interval variability is used as an electrocardiographic prior feature, and the standard deviation of PR interval variability is added to the deep neural network for global-firing, that is, the depth feature obtained by CNN and the standard deviation of PR interval variability are fused and then input to the BiLSTM layer for atrial fibrillation detection. The performance of the deep neural network after testing is shown in table 6.

TABLE 6

After the standard deviation of the PR interval variability is added, the performance of the deep neural network is recorded as B2+ PR _ csdnn, the PR _ csdnn represents the standard deviation of the PR interval variability, and the Sensitivity (SE) and the Specificity (SP) of each data when the deep neural network added with the standard deviation of the PR interval variability is used for detecting atrial fibrillation are shown in table 6. Meanwhile, table 6 also shows Baseline1 and Baseline2 in table 2. As can be seen from table 6, compared to Baseline1, most of the data in B2+ pr _ csdnn are improved to some extent, that is, the performance of the deep neural network is improved to some extent, reference may be made to the column in table 6 where P (our + B1) is located, which shows a significant correlation between Baseline1 and data in B2+ pr _ csdnn.

In one embodiment, taking atrial fibrillation detection (AF _ detection) as an example, the standard deviation of RR period, the RR interval variation coefficient and the standard deviation of PR interval variability are used as the electrocardiographic prior features, and the standard deviation of RR period, the RR interval variation coefficient and the standard deviation of PR interval variability are added to the deep neural network for global-posing, that is, the depth features obtained by CNN and the standard deviations of RR period, RR interval variation coefficient and PR interval variability are fused and input to the BiLSTM layer for atrial fibrillation detection. After testing the deep neural network, the performance of the deep neural network is shown in table 7.

TABLE 7

After adding the standard deviation of the RR period, the RR interval variation coefficient and the standard deviation of the PR interval variability, the performance of the deep neural network is recorded as B2+ RR _ cv + RR _ sdnn + PR _ csdnn, and the Sensitivity (SE) and the Specificity (SP) of each item of test data when atrial fibrillation is detected by adding the deep neural network with the standard deviation of the RR period, the RR interval variation coefficient and the standard deviation of the PR interval variability are shown in table 7. Also shown in table 7 are the performance of the deep neural network when only the RR interval standard deviation is added (data under B2+ RR _ sdnn in table 7) and the performance of the deep neural network when only the RR interval standard deviation and the RR interval coefficient of variation are added (data under B2+ RR _ cv + RR _ sdnn in table 7). As can be seen from table 7, after adding the standard deviation of RR period, the RR interval variation coefficient and the standard deviation of PR interval variability at the same time, the performance of the deep neural network is better than the performance of the deep neural network only adding the standard deviation of RR period or only adding the standard deviation of RR interval and the RR interval variation coefficient, so that the standard deviation of RR period, the RR interval variation coefficient and the standard deviation of PR interval variability can be used jointly as the electrocardiographic prior characteristic.

In one embodiment, taking atrial fibrillation detection (AF _ detection) as an example, the standard deviation of P-wave variability is used as an electrocardiographic prior feature, and the standard deviation of P-wave variability is added to the deep neural network for global-posing, that is, the depth feature obtained by CNN and the standard deviation of P-wave variability are fused and input to the BiLSTM layer for atrial fibrillation detection. The performance of the deep neural network after testing is shown in table 8.

TABLE 8

The performance of the deep neural network is recorded as B2+ P _ csdnn after the standard deviation of P-wave variability is added, P _ csdnn represents the standard deviation of P-wave variability, and the Sensitivity (SE) and Specificity (SP) of each test data when the deep neural network added with the standard deviation of P-wave variability is used for detecting atrial fibrillation are shown in table 8. Meanwhile, table 8 also shows Baseline1 and Baseline2 in table 2. As can be seen from table 8, compared to Baseline1, most of the data in B2+ P _ csdnn are improved to some extent, that is, the performance of the deep neural network is improved to some extent, reference may be made to the column in table 8 where P (our + B1) is located, which shows the significant correlation between Baseline1 and data in B2+ P _ csdnn.

In one embodiment, taking atrial fibrillation detection (AF _ detection) as an example, the RR period standard deviation, the RR interval variation coefficient, the standard deviation of PR interval variation and the standard deviation of P-wave variation are used as the ecg features, and the RR period standard deviation, the RR interval variation coefficient, the standard deviation of PR interval variation and the standard deviation of P-wave variation are added to the above global-posing deep neural network, that is, the CNN depth features are fused with the RR period standard deviation, the RR interval variation coefficient, the standard deviation of PR interval variation and the standard deviation of P-wave variation and then input to the BiLSTM layer to test the deep neural network when atrial fibrillation is detected, and at this time, the performance of the deep neural network is shown in table 9.

TABLE 9

After adding the standard deviation of the RR period, the RR interval variation coefficient, the standard deviation of the PR interval variability and the standard deviation of the P-wave variability, the performance of the deep neural network is recorded as B2+ RR _ cv + RR _ sdnn + PR _ csdnn + P _ csdnn, and the Sensitivity (SE) and the Specificity (SP) of each data are shown in table 9 when atrial fibrillation is detected by the deep neural network to which the standard deviation of the RR period, the RR interval variation coefficient, the standard deviation of the PR interval variability and the standard deviation of the P-wave variability are added. Meanwhile, the performance of the deep neural network when only RR interval standard deviation is added (each item of data under B2+ RR _ sdnn in table 9), the performance of the deep neural network when only RR interval standard deviation and RR interval variation coefficient are added (each item of data under B2+ RR _ cv + RR _ sdnn in table 9), and the performance of the deep neural network when only RR interval standard deviation, RR interval variation coefficient and PR interval variation are added (each item of data under B2+ RR _ cv + RR _ sdnn + PR _ csdnn in table 9). As can be seen from table 9, after the standard deviation of RR period, the RR interval variation coefficient, the standard deviation of PR interval variability, and the standard deviation of P-wave variability are added simultaneously, the performance of the deep neural network is better than the performance of the deep neural network only adding the standard deviation of RR period, only adding the standard deviation of RR interval and the RR interval variation coefficient, and only adding the standard deviation of RR period, the RR interval variation coefficient, and the standard deviation of P-wave variability, so that the standard deviation of RR period, the RR interval variation coefficient, the standard deviation of PR interval variability, and the standard deviation of P-wave variability can be used jointly as the electrocardiographic prior characteristics.

The construction of the electrocardio prior characteristics combines two clinical expressions during atrial fibrillation, avoids the condition that atrial fibrillation is difficult to detect due to disappearance of P waves and weak f wave signals, also avoids the detection interference of irregular RR intervals caused by other abnormal types on atrial fibrillation, avoids the condition of misjudgment of the abnormal types, and improves the robustness of atrial fibrillation detection.

In one embodiment, the signal acquisition module 101 comprises: the second signal acquisition unit is used for acquiring an original third electrocardiosignal, and the third electrocardiosignal is an analog electrocardiosignal; the analog-to-digital conversion unit is used for converting the third electrocardiosignal into a fourth electrocardiosignal, and the fourth electrocardiosignal is a digital electrocardiosignal; the interference removing unit is used for inputting the fourth electrocardiosignal into a band-pass filter with a set cut-off frequency so as to remove noise interference in the fourth electrocardiosignal; the resampling unit is used for resampling the fourth electrocardiosignal without noise interference and cutting the fourth electrocardiosignal into electrocardiosignal segments with set length; and the normalization unit is used for performing normalization processing on the electrocardiosignal segments to obtain a first electrocardiosignal to be detected.

The signal acquisition module 101 includes a second signal acquisition unit, an analog-to-digital conversion unit, an interference removal unit, a resampling unit, and a normalization unit when acquiring the first cardiac signal. The second signal acquisition unit is used for acquiring electrocardiosignals of a human body, and the electrocardiosignals can be regarded as the acquired original electrocardiosignals. In one embodiment, the cardiac signal is identified as a third cardiac signal. The third electrocardiosignal is the collected simulated electrocardiosignal. It is understood that the third cardiac signal may be acquired by the second signal acquisition unit using a lead, a sensor, or the like. The third cardiac signal includes various noises and has a rough and unsmooth waveform. In an embodiment, the second signal acquisition unit may perform impedance matching, filtering, amplifying, and the like on the third cardiac signal by using an analog circuit to improve quality of the third cardiac signal, where a specific structural embodiment of the analog circuit is not limited. And then, the analog-to-digital conversion unit acquires a third electrocardiosignal in the second signal acquisition unit and converts the third electrocardiosignal into a digital electrocardiosignal by using the analog-to-digital converter. The specific structural embodiment of the analog-to-digital converter is not limited. In one embodiment, the digital cardiac signal obtained by the analog-to-digital conversion unit is denoted as a fourth cardiac signal. Optionally, the analog-to-digital conversion unit stores the fourth cardiac signal into a memory of the cardiac signal detection device, so as to be used by the cardiac signal detection device.

The interference removing unit is used for obtaining the fourth electrocardiosignal and removing noise interference in the fourth electrocardiosignal. In one embodiment, the interference removing unit removes noise interference in the fourth cardiac signal using a FIR (Finite Impulse Response) band-pass filter, and a cut-off frequency of the FIR band-pass filter can be determined according to the noise. Optionally, when the interference removing unit removes the low-frequency noise and the power frequency noise of the fourth electrocardiographic signal, two 50-order FIR band-pass filters are used, that is, the fourth electrocardiographic signal is sequentially input to the two 50-order FIR band-pass filters, so as to remove the low-frequency noise and the power frequency noise in the fourth electrocardiographic signal respectively. The low-frequency noise is long-wave noise, that is, noise with a long wavelength. The power frequency noise is the noise generated when the electrocardiosignal detection equipment works. In one embodiment, the cutoff frequency of the FIR band-pass filter for removing low-frequency noise is set to 0.8Hz, and the cutoff frequency of the FIR band-pass filter for removing power-frequency noise is set to 48Hz. Alternatively, two FIR band pass filters may be configured in the cardiac signal detection device for use by the interference removal unit.

The resampling unit is configured to resample the fourth electrocardiosignal from which the noise is removed, so that the resampled fourth electrocardiosignal meets a requirement of a set frequency, where the set frequency may be set according to an actual requirement, and optionally, the set frequency is equal to a frequency of a signal that can be processed by a deep neural network formed by the first neural network and the second neural network. In one embodiment, the frequency is set to 250Hz, i.e. the fourth cardiac signal is resampled to a 250Hz fourth cardiac signal. It is understood that the resampling may be implemented by interpolation, and the embodiment does not limit this. Then, the resampling unit cuts the resampled fourth ecg signal to cut the resampled fourth ecg signal into ecg signal segments with a set length, where the set length may be set according to an actual situation, and in an embodiment, the set length is 10s, that is, the fourth ecg signal is cut into at least one ecg signal segment, and the ecg signal segment has a length of 10s. Optionally, during cutting, if the length of the electrocardiographic signal segment is less than 10s, the length of the electrocardiographic signal segment can be compensated to 10s in a zero compensation manner. If the length of the electrocardiosignal segment exceeds 10s, the electrocardiosignal segment is cut off, namely the electrocardiosignal part exceeding 10s is cut off.

The normalization unit is used for performing normalization processing on the electrocardiosignal fragments obtained by the resampling unit to obtain a first electrocardiosignal to be detected. After normalization, each value in the first electrocardiosignal is between [0,1 ]. The normalization processing mode can be set according to actual conditions, in one embodiment, the electrocardiographic signal segments are normalized by z-score, and in this case, the calculation formula during normalization processing can be:

wherein mu represents the mean value of the electrocardiosignal segment, sigma represents the standard deviation of the electrocardiosignal segment, x represents the numerical value in the electrocardiosignal segment, and x represents _norm The values normalized to x are shown.

It is understood that, in practical applications, other normalization processing manners may also be adopted, and the embodiment does not limit this.

The second signal acquisition unit acquires the third electrocardiosignal, and the analog-to-digital conversion unit, the interference removal unit, the resampling unit and the normalization unit respectively perform pretreatment of analog-to-digital conversion, noise removal, resampling, cutting and normalization on the third electrocardiosignal, so that the first electrocardiosignal to be detected for the first neural network can be obtained.

The detection process of the electrocardiosignal detection device is exemplarily described below. The detection process is described by way of example for detecting atrial fibrillation. Fig. 4 is a processing flow chart of the electrocardiographic signals according to an embodiment of the present application, and referring to fig. 4, after the second signal acquisition unit acquires the third electrocardiographic signal, the third electrocardiographic signal is subjected to preprocessing (analog-to-digital conversion, noise interference removal, and resampling) and cutting by using the analog-to-digital conversion unit, the interference removal unit, and the resampling unit in sequence to obtain three electrocardiographic signal segments, and the three electrocardiographic signal segments are normalized by using the normalization unit to obtain three first electrocardiographic signals with a length of 10s. Then, the feature extraction module extracts the depth features of the first electrocardiosignals by using the first neural network, then the priori acquisition module acquires electrocardio priori features, the feature fusion module fuses the depth features and the electrocardio priori features to obtain fusion features, and then the signal detection module inputs the fusion features into the second neural network to determine the detection results of the three first electrocardiosignals, wherein the detection results of the three first electrocardiosignals are all the first electrocardiosignals which are not in atrial fibrillation type (Non-AF).

It should be noted that, in the embodiment of the electrocardiograph signal detection apparatus, each unit and each module included in the embodiment are only divided according to functional logic, but are not limited to the above division, as long as the corresponding function can be realized; in addition, specific names of the functional units are only used for distinguishing one functional unit from another, and are not used for limiting the protection scope of the application.

An embodiment of the present application further provides an electrocardiograph signal detection device, and fig. 5 is a schematic structural diagram of the electrocardiograph signal detection device provided in an embodiment of the present application. As shown in fig. 5, the cardiac signal detecting apparatus includes a processor 20, a memory 21, an input device 22, and an output device 23; the number of the processors 20 in the cardiac signal detection device may be one or more, and one processor 20 is taken as an example in fig. 5. The processor 20, the memory 21, the input means 22 and the output means 23 of the non-invasive blood glucose measuring apparatus may be connected by a bus or other means, and fig. 5 illustrates the connection by a bus as an example.

The memory 21 is used as a computer-readable storage medium for storing software programs, computer-executable programs, and modules, such as program instructions/modules used when the cardiac signal detection apparatus in the embodiments of the present application operates (for example, the signal acquisition module 101, the feature extraction module 102, the prior acquisition module 103, the feature fusion module 104, and the signal detection module 105 in the cardiac signal detection apparatus). The processor 20 executes various functional applications and data processing of the electrocardiograph signal detection device by operating the software programs, instructions and modules stored in the memory 21, that is, implements the operation process of the electrocardiograph signal detection apparatus.

The memory 21 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created from use of the electrocardiographic signal detecting device, and the like. Further, the memory 21 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid state storage device. In some examples, the memory 41 may further include memory located remotely from the processor 20, which may be connected to the cardiac signal detection device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The input device 22 may be used for receiving input digital or character information and generating key signal inputs related to user settings and function control of the cardiac signal detection apparatus, and may further include leads, sensors, etc. for collecting cardiac signals. The output device 23 may include a display device such as a display screen.

The electrocardiosignal detection equipment can comprise the electrocardiosignal detection device to realize the detection of electrocardiosignals. Optionally, the electrocardiograph signal detection device may be an electrocardiograph, a portable electrocardiograph detector, or the like. It can be understood that the electrocardiosignal detection device and the electrocardiosignal detection device have the same functions and beneficial effects, and the electrocardiosignal detection device can be referred to for technical details which are not mentioned in the electrocardiosignal detection device.

It is to be noted that the foregoing is only illustrative of the preferred embodiments of the present application and the technical principles employed. It will be understood by those skilled in the art that the present application is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the application. Therefore, although the present application has been described in more detail with reference to the above embodiments, the present application is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims

1. An electrocardiographic signal detection device characterized by comprising:

2. The cardiac signal detection device according to claim 1, further comprising:

the heart beat mark acquisition module is used for acquiring a second cardiac electrical signal with a heart beat mark, and the heart beat mark is used for marking the heart beat position of each heart beat in the second cardiac electrical signal;

and the prior characteristic determining module is used for obtaining the electrocardio prior characteristic based on the second electrocardiosignal according to the heart beat position in the heart beat mark.

3. The cardiac signal detection apparatus as set forth in claim 2, wherein the apriori cardiac features comprise: at least one of standard deviation of RR interval, coefficient of variation of RR interval, standard deviation of PR interval variability, and standard deviation of P-wave variability.

4. The cardiac signal detection apparatus as claimed in claim 3, wherein the cardioelectric prior features comprise: the standard deviation of the RR interval is shown,

the prior characteristic determination module comprises:

a first extraction unit, configured to extract, according to a heartbeat position in the heartbeat label, each RR interval in the second cardiac electrical signal, where the RR interval is a time limit between two adjacent R waves;

the first standard deviation calculation unit is used for calculating a first standard deviation of the RR interval and taking the first standard deviation as an RR interval standard deviation;

the electrocardio prior characteristics comprise: the coefficient of variation of the RR interval is,

the prior characteristic determination module comprises:

the second extraction unit is used for extracting each RR interval from the second electrocardiosignal according to the heart beat position in the heart beat mark;

a first parameter determination unit, configured to calculate a second standard deviation and a first average of the RR intervals;

a coefficient of variation determining unit, configured to obtain an RR-period coefficient of variation according to the second standard deviation and the first average;

the electrocardio prior characteristics comprise: the standard deviation of the variability of the PR interval,

the prior characteristic determination module comprises:

a third extraction unit, configured to extract PR intervals from the second cardiac electrical signal according to a heart beat position in the heart beat label, where the PR intervals are time periods from a P-wave start point to an adjacent QRS-wave start point;

a second parameter determination unit for calculating a second average of the PR intervals;

a first variability calculation unit for calculating a first ratio of each of the PR intervals to the second average value, and taking the first ratio as a PR interval variability;

a second standard deviation calculation unit for calculating a third standard deviation of the variability of the PR intervals and taking the third standard deviation as the standard deviation of the variability of the PR intervals;

the electrocardio prior characteristics comprise: the standard deviation of the variability of the P-wave,

the prior characteristic determination module comprises:

a fourth extraction unit, configured to extract a P-wave width of each P-wave from the second cardiac electric signal according to the cardiac beat position in the cardiac beat label;

a third parameter determination unit for calculating a third average value of the P-wave width;

a second variability calculating unit, configured to calculate a second ratio of each P-wave width to the third average value, and use the second ratio as P-wave variability;

and the third standard deviation calculation unit is used for calculating a fourth standard deviation of the P wave variability and taking the fourth standard deviation as the standard deviation of the P wave variability.

5. The apparatus according to claim 2, wherein the cardiac beat label collecting module comprises:

the first signal acquisition unit is used for acquiring a second electrocardiosignal;

and the heartbeat labeling unit is used for identifying the QRS wave in the second electrocardiosignal and determining the heartbeat label of the second electrocardiosignal according to the position of the QRS wave in the second electrocardiosignal.

6. The cardiac signal detection apparatus as claimed in claim 1, wherein the first neural network comprises at least one residual convolutional network.

7. The cardiac signal detection apparatus according to claim 1, wherein the second neural network comprises a bidirectional long-short term memory network layer, a Flatten layer, and a Dense layer.

8. The cardiac signal detection device according to claim 1, wherein the signal acquisition module comprises:

the second signal acquisition unit is used for acquiring an original third electrocardiosignal, and the third electrocardiosignal is an analog electrocardiosignal;

the analog-to-digital conversion unit is used for converting the third electrocardiosignal into a fourth electrocardiosignal, and the fourth electrocardiosignal is a digital electrocardiosignal;

the interference removing unit is used for inputting the fourth electrocardiosignal into a band-pass filter with a set cut-off frequency so as to remove noise interference in the fourth electrocardiosignal;

the resampling unit is used for resampling the fourth electrocardiosignal without noise interference and cutting the fourth electrocardiosignal into electrocardiosignal segments with set length;

and the normalization unit is used for performing normalization processing on the electrocardiosignal fragments to obtain a first electrocardiosignal to be detected.

9. The apparatus according to claim 3, wherein the first cardiac signal is an atrial fibrillation type or the first cardiac signal is a non-atrial fibrillation type.

10. An electrocardiographic signal detection apparatus characterized by comprising:

one or more processors;

a memory for storing one or more programs;

when executed by the one or more processors, cause the one or more processors to perform the calculations of the cardiac signal detection apparatus as recited in any one of claims 1-9.