CN116712055A

CN116712055A - Arrhythmia detection device and readable storage medium

Info

Publication number: CN116712055A
Application number: CN202310545779.4A
Authority: CN
Inventors: 芦雪聪; 黎冰
Original assignee: Shenzhen University
Current assignee: Shenzhen University
Priority date: 2023-05-15
Filing date: 2023-05-15
Publication date: 2023-09-08

Abstract

The application is suitable for the technical field of electrocardiosignal processing, and provides arrhythmia detection equipment and a readable storage medium, wherein a memory and a processor of the arrhythmia detection equipment are provided, computer readable instructions are stored in the memory, and the processor is used for executing the following steps when calling the computer readable instructions in the memory: classifying a heart beat set to be detected by using a preset abnormal heart beat classification model to obtain an abnormal classification result, wherein the heart beat set to be detected comprises N heart beats to be detected, and N is a positive integer greater than or equal to 2; if abnormal heart beats exist in the heart beat set to be detected, classifying the first heart beat to be detected by using a preset abnormal heart rhythm classification model to obtain an abnormal heart rhythm classification result of the first heart beat to be detected; if no abnormal heart beat exists in the heart beat set to be detected, classifying the next heart beat set to be detected by using a preset abnormal heart beat classification model. By setting the wake-up mechanism, the power consumption of the arrhythmia detection device can be reduced.

Description

Arrhythmia detection device and readable storage medium

Technical Field

The application belongs to the technical field of electrocardiosignal processing, and particularly relates to arrhythmia detection equipment and a readable storage medium.

Background

According to the statistics of world health organization and the Chinese cardiovascular health and disease report 2021, cardiovascular diseases are the first global cause of death, and meanwhile, the incidence rate and the mortality rate of the cardiovascular diseases in China are still high in the top of the ban. Arrhythmia behavior occurs mostly before cardiovascular disease patients develop, so arrhythmia detection plays a vital role in effectively preventing cardiovascular disease.

With the development of artificial intelligence technology, diagnosis of arrhythmia by using a machine learning algorithm is currently becoming a popular research. In order to obtain an accurate arrhythmia detection result, the current arrhythmia detection algorithm generally has the problems of high complexity and high power consumption, and the problems limit the application of the arrhythmia detection algorithm, for example, the current arrhythmia detection algorithm can be processed only by a cloud processor, or can be processed by a local processor but can only monitor specific patients, and the power consumption is reduced by reducing the number of adaptation groups to reduce the calculation amount of the algorithm. In order to widen the application scene of the arrhythmia algorithm under the condition of ensuring the detection accuracy, how to reduce the power consumption of the arrhythmia detection algorithm is a technical problem to be solved.

Disclosure of Invention

The embodiment of the application provides arrhythmia detection equipment and a readable storage medium, which can solve the technical problem that the arrhythmia detection algorithm in the prior art has higher power consumption, so that the application scene is limited.

In a first aspect, an embodiment of the present application provides an arrhythmia detection device, including a memory having computer readable instructions stored therein, and a processor for executing the following steps when the processor invokes the computer readable instructions in the memory:

classifying a heart beat set to be detected by using a preset abnormal heart beat classification model to obtain an abnormal classification result, wherein the abnormal classification result comprises that abnormal heart beats exist in the heart beat set to be detected or no abnormal heart beats exist in the heart beat set to be detected, the heart beat set to be detected comprises N heart beats to be detected, and N is a positive integer greater than or equal to 2;

if abnormal heart beats exist in the heart beat set to be detected, classifying the first heart beat to be detected by using a preset abnormal heart rhythm classification model to obtain an abnormal heart rhythm classification result of the first heart beat to be detected, wherein the first heart beat to be detected is any heart beat to be detected in the heart beat set to be detected, and the abnormal heart rhythm classification result represents the arrhythmia type of the first heart beat to be detected;

If no abnormal heart beat exists in the heart beat set to be detected, classifying the next heart beat set to be detected of the heart beat set to be detected by using a preset abnormal heart beat classification model.

Based on the arrhythmia detection equipment, performing abnormal heart beat judgment on the heart beat set to be detected by using a preset abnormal heart beat classification model, and triggering the preset abnormal heart beat classification model to perform abnormal heart rhythm detection on N heart beats to be detected only when abnormal heart beats exist in the heart beat set to be detected (namely at least one abnormal heart beat exists in N heart beats to be detected); and under the condition that abnormal heart beats do not exist in the heart beat set to be detected (namely, N heart beats to be detected are not abnormal heart beats), classifying the next heart beat set to be detected by using a preset abnormal heart beat classification model. Because the heart beat set to be detected comprises a plurality of heart beats to be detected, the abnormal judgment times of abnormal heart beats can be reduced, and the power consumption of arrhythmia detection equipment can be reduced; in addition, the wake-up mechanism enables the preset abnormal heart rhythm classification model to be woken up only when abnormal heart beats exist in N heart beats to be detected, so that the starting times of the abnormal heart rhythm classification model are reduced, and the power consumption of the arrhythmia detection equipment can be further reduced.

In a possible implementation manner of the first aspect, classifying the heart beat set to be measured by using a preset abnormal heart beat classification model to obtain an abnormal classification result includes: determining the heart rate variability characteristic of each of the N heart beats to be detected, wherein the heart rate variability characteristic represents the variability of the time interval between adjacent R peaks; and inputting heart rate variability characteristics of the N heart beats to be detected into a preset abnormal heart beat classification model to obtain an abnormal classification result. In the embodiment, heart beat abnormality classification is performed by replacing heart beat data per se with heart rate variability characteristics, N heart beats to be detected are reconstructed, each heart beat is expressed by a plurality of heart rate variability characteristics, the heart beat segment data after reconstruction operation can effectively reduce the difference of heart rhythm information among individuals and highlight heart rhythm abnormality in the individuals, and the heart beat abnormality classification method has positive significance for improving the classification accuracy of the neural network.

In a possible implementation manner of the first aspect, classifying the first beat to be tested by using a preset abnormal heart rhythm classification model to obtain an abnormal heart rhythm classification result of the first beat to be tested includes:

obtaining pathological characteristics of a first heart beat to be tested; and inputting the case characteristics of the first heart beat to a preset abnormal heart rhythm classification model to obtain an abnormal heart rhythm classification result of the first heart beat. In this embodiment, compared to inputting the raw ECG data to an end-to-end neural network classifier, the use of carefully screened pathology features, while introducing additional feature extraction, can effectively simplify the neural network scale and greatly reduce the computational effort.

In a possible implementation manner of the first aspect, the pathological features of the first beat to be measured include a heart rate variability feature of the first beat to be measured, a beat dimension reduction feature of the first beat to be measured, and a local QRS feature, the beat dimension reduction feature of the first beat to be measured being obtained by dimension reduction of a plurality of sampling points of the first beat to be measured, the local QRS feature of the first beat to be measured including a feature related to a QRS peak of the first beat to be measured. In this embodiment, the heart rate variability features have important pathological information and require very little computation. The dimension reduction feature is obtained by accumulation and summation of heart beats and averaging. The dimension reduction feature transforms the original high-dimension vector feature into a low-dimension vector feature, main useful information is not lost in the process, and core features in the heart beat are extracted to filter interference information such as irrelevant features, noise and the like, so that the input complexity of the neural network is reduced, and the generalization capability of the classification model is enhanced to a certain extent. Local QRS features can increase individual variability from patient to patient by increasing pathological features.

In a possible implementation manner of the first aspect, the preset abnormal heart beat classification model includes a first artificial neural network. In the embodiment, the artificial neural network has a simple structure and low power consumption.

In a possible implementation manner of the first aspect, the preset abnormal heart rhythm classification model is a hybrid neural network model, and the hybrid neural network model includes a long-short-time neural network, a second artificial neural network and a third artificial neural network; the input data of the long-short time neural network and the second artificial neural network comprise pathological features of the first heart beat to be tested, the input data of the third artificial neural network comprise output data of the long-short time neural network and output data of the third artificial neural network, and the output data of the third artificial neural network comprise abnormal heart rhythm classification results of the first heart beat to be tested.

In a possible implementation manner of the first aspect, the input data of the long-short-term neural network includes a beat dimension-reducing feature of the first beat to be measured and a local QRS feature of the first beat to be measured, and the input data of the second artificial neural network includes a heart rate variability feature of the first beat to be measured. In the embodiment, the multiplexing of the heart rate variability characteristics after the fusion of the characteristics makes the calculation amount of the network structure smaller.

In a possible implementation manner of the first aspect, the processor further performs the following steps:

acquiring electrocardiosignals of a target patient to obtain a plurality of sampling points, wherein the first sampling point comprises polarity information and time interval information, the polarity information of the first sampling point represents the amplitude change condition of the electrocardiosignals of the first sampling point and the last sampling point, and the time interval information of the first sampling point represents the sampling time change of the first sampling point and the last sampling point;

Determining whether the first sampling point is an extreme point according to the polarity information of the first sampling point and the polarity information of the last sampling point of the first sampling point;

if the first sampling point is an extreme point, acquiring a first time interval between the first sampling point and the last extreme point of the first sampling point;

if the first time interval is smaller than the duration threshold, acquiring a second time interval between the first sampling point and the last R peak of the first sampling point and a third time interval corresponding to the first sampling point, wherein the third time interval is an average RR interval between the first sampling point and W R peaks before the first sampling point, and W is a positive integer greater than 2;

if the second time interval is greater than the first threshold and smaller than the second threshold, or the second time interval is greater than the first threshold and smaller than the product of the third time interval and the first parameter, judging whether the second time interval is smaller than the product of the third time interval and the second parameter;

if the second time interval is smaller than the product of the third time interval and the second parameter, the first sampling point is determined to be an R peak.

after determining the first sampling point as an R peak, the duration threshold is updated according to the first time interval. In the embodiment, the adaptive adjustment duration threshold positioning and the adaptive adjustment of the average RR interval can effectively inhibit three kinds of noise including baseline drift, myoelectricity and motion artifact through two adaptive thresholds, so that the arrhythmia detection device can be well used for arrhythmia detection in a motion state.

if the second time interval is not greater than the first threshold or not less than the second threshold, and the second time interval is not greater than the first threshold or not less than the product of the third time interval and the second parameter, judging whether the first time interval is less than a fourth time interval, wherein the fourth time interval is the time interval between the first R peak and the last extreme point of the first R peak, and the first R peak is the last R peak of the first sampling point;

if the first time interval is smaller than the fourth time interval, the first sampling point is determined to be an R peak.

if the second time interval is not less than the product of the third time interval and the second parameter, judging whether the first time interval is less than a fifth time interval, wherein the fifth time interval is the minimum value in RR intervals among W R peaks before the first sampling point;

if the first time interval is smaller than the fifth time interval, the first sampling point is determined to be an R peak.

In a possible implementation manner of the first aspect, an absolute value of a difference between the amplitude of the first sampling point and the amplitude of a last sampling point of the first sampling point is greater than or equal to a preset amplitude threshold.

In a possible implementation manner of the first aspect, the original sample data set includes original abnormal samples and original normal samples, each original abnormal sample includes an abnormal heart beat, and each original normal sample includes a normal heart beat; the processor performs the steps of: oversampling the original abnormal samples to obtain a plurality of synthesized abnormal samples, each of the synthesized abnormal samples including an abnormal heart beat; randomly sampling a plurality of synthetic abnormal samples to obtain a target synthetic abnormal sample; and training the initial abnormal heart beat classification model and the initial abnormal heart rhythm classification model by taking a data set consisting of the target synthetic abnormal sample, the original abnormal sample and the original normal sample as a training data set to obtain a preset abnormal heart beat classification model and a preset abnormal heart rhythm classification model. The number of abnormal heart beats and normal heart beats is kept balanced by the random extraction algorithm based on sample synthesis, so that the training efficiency of the model is improved.

In a second aspect, an embodiment of the application provides an arrhythmia detection apparatus comprising means for performing the steps performed by a processor in an arrhythmia detection device as described in any of the first aspects above.

In a third aspect, embodiments of the present application provide a computer readable storage medium storing a computer program which, when executed by a processor, implements the steps performed by the processor in the arrhythmia detection device of any of the first aspects described above.

In a fourth aspect, embodiments of the present application provide a computer program product which, when run on a terminal device, causes the terminal device to perform the steps performed by a processor in an arrhythmia detection device as described in any of the first aspects above.

In a fifth aspect, an embodiment of the present application provides a chip, including: a processor for recalling and running a computer program from memory, causing an electronic device on which the chip is mounted to perform the steps performed by the processor in the arrhythmia detection device of any of the first aspects.

It will be appreciated that the advantages of the second to fifth aspects may be found in the relevant description of the first aspect, and are not described here again.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments or the description of the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a graph comparing electrocardiographic signals in a resting state and a moving state;

fig. 2 is a schematic diagram of an application scenario of an arrhythmia detection device according to an embodiment of the present application;

FIG. 3 is a system architecture diagram of an arrhythmia detection device according to an embodiment of the application;

FIG. 4 is a flowchart of an arrhythmia detection method according to an embodiment of the present application;

FIG. 5 is a schematic diagram of a heart beat reconstruction process in an arrhythmia detection method according to an embodiment of the application;

FIG. 6 is a schematic diagram showing the overall structure of a preset abnormal heart beat classification model and a preset abnormal heart rhythm classification model in an arrhythmia detection device according to an embodiment of the present application;

FIG. 7 is a graph comparing sample points using conventional Nyquist equidistant sampling and event driven sampling in an embodiment of the present application;

FIG. 8 is a flowchart of a heartbeat positioning method in an arrhythmia detection method according to an embodiment of the application;

FIG. 9 is a schematic diagram of an arrhythmia detection system according to an embodiment of the application;

FIG. 10 is a flow chart of a process for treating an imbalance-like condition in an arrhythmia detection method according to an embodiment of the application;

fig. 11 is a schematic structural diagram of an arrhythmia detection device according to an embodiment of the present application.

Detailed Description

In the following description, for purposes of explanation and not limitation, specific details are set forth such as the particular system architecture, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. It will be apparent, however, to one skilled in the art that the present application may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

It should be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.

It should also be understood that the term "and/or" as used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.

Furthermore, the terms "first," "second," "third," and the like in the description of the present specification and in the appended claims, are used for distinguishing between descriptions and not necessarily for indicating or implying a relative importance.

Reference in the specification to "one embodiment" or "some embodiments" or the like means that a particular feature, structure, or characteristic described in connection with the embodiment is included in one or more embodiments of the application. Thus, appearances of the phrases "in one embodiment," "in some embodiments," "in other embodiments," and the like in the specification are not necessarily all referring to the same embodiment, but mean "one or more but not all embodiments" unless expressly specified otherwise. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless expressly specified otherwise.

The current common arrhythmia detection schemes mainly comprise three types. The first scheme is that a patient acquires electrocardiograph data by using a dynamic electrocardiograph in a hospital to generate an electrocardiogram, and a doctor diagnoses according to the electrocardiogram on site, and the first scheme is the earliest arrhythmia detection scheme and needs the patient to acquire electrocardiograph signals from the hospital. The second scheme is that the patient wears the intelligent patch to carry out continuous electrocardiograph data acquisition for a plurality of time periods, then the intelligent patch is taken down and sent to a doctor for analysis and diagnosis, and the patient does not need to carry out electrocardiograph acquisition to a hospital in the second scheme, but the timeliness is poor because a long time difference is needed from electrocardiograph acquisition to doctor analysis and diagnosis. Thirdly, the patient wears a wearable device to collect electrocardio, and the wearable device sends the collected electrocardio data to a cloud for analysis and diagnosis; in the third scheme, the central electric data is transmitted to the cloud end in a communication mode, and the cloud end server performs analysis and diagnosis.

In general, the optimal treatment time for arrhythmia is within "golden 4 minutes" after the occurrence, so that the timeliness of arrhythmia detection is also very important. The detection of arrhythmia by using a local processor integrates the acquisition and processing of electrocardiosignals, avoids the power consumption of data transmission, realizes the real-time detection of arrhythmia and simultaneously is easy to realize the long-time passive detection.

In the current scheme for detecting arrhythmia by using a machine learning algorithm, in order to obtain an accurate arrhythmia detection result, the adopted arrhythmia detection algorithm generally has the problems of high complexity and high power consumption. In some schemes, a monitoring mode is adopted for a specific patient, and the power consumption is reduced by reducing the algorithm operation amount by reducing the adaptive population, so that arrhythmia detection based on a local processor is realized. However, the heart rhythm detection algorithm which is common to patients (applicable to general population) requires huge calculation, so that the energy consumption and the equipment area caused by the calculation are not suitable for a local processor, and the calculation is mainly realized at the cloud.

In addition, the electronics in current arrhythmia detection schemes all require the patient to be in a quiet state, while the occurrence of arrhythmias is frequent in a motor state. The electrocardiographic signals are shown in a quiet state and a motion state in fig. 1. Noise generated in a motion state: motion artifact, baseline drift and myoelectric interference, noise can seriously affect the acquisition and detection of electrocardiosignals of the heart, and a heart rhythm abnormality classification algorithm with higher complexity is needed to further increase the processing power consumption.

In order to solve the problem of high power consumption of arrhythmia detection equipment in the prior art, the embodiment of the application provides arrhythmia detection equipment and a readable storage medium, wherein the arrhythmia detection equipment carries out abnormal heart beat judgment on a heart beat set to be detected through a preset abnormal heart beat classification model, and the abnormal heart beat detection is carried out on N heart beats to be detected by triggering the preset abnormal heart beat classification model only when abnormal heart beats exist in the heart beat set to be detected (namely, at least one abnormal heart beat exists in N heart beats to be detected); and under the condition that abnormal heart beats do not exist in the heart beat set to be detected (namely, N heart beats to be detected are not abnormal heart beats), classifying the next heart beat set to be detected by using a preset abnormal heart beat classification model. The wake-up mechanism enables the preset abnormal heart rhythm classification model to be woken up only when abnormal heart beats exist in N heart beats to be detected, so that the starting times of the abnormal heart rhythm classification model are reduced, the power consumption of the arrhythmia detection equipment is reduced, the arrhythmia detection equipment can be widely applied, and for example, the arrhythmia detection equipment can adopt a local processor to process data.

The arrhythmia detection device, readable storage medium, and chip provided by the present application are described below with reference to specific examples.

Fig. 2 is a schematic diagram illustrating an application scenario suitable for an embodiment of the present application. As shown in fig. 2, the user uses the arrhythmia detection device 210 and the electronic device 220, wherein the arrhythmia detection device 210 is provided with an electrocardiograph data collector, so that electrocardiograph data collection can be performed on the user, the arrhythmia detection device 210 can perform arrhythmia detection and send arrhythmia detection results to the electronic device 220, and the electronic device 220 can display arrhythmia detection results to the user or send early warning to the user when the electronic device 220 receives arrhythmia detection results with arrhythmia.

Communication between the arrhythmia detection device 210 and the electronic device 220 may be performed through wireless communication technologies such as Bluetooth (BT) technology, wireless-fidelity (WiFi) technology, near field communication (near field communication, NFC) technology, and the like.

It should be understood that fig. 2 is merely exemplary, and should not be construed as limiting the application scenario of the embodiment of the present application, for example, in the scenario shown in fig. 2, more arrhythmia detection devices 210 and electronic devices 220 may be included.

In some embodiments, arrhythmia detection device 210 may have the functionality to display arrhythmia detection results to a user and/or to issue an alert to the user when the arrhythmia detection results are arrhythmia.

In some embodiments, the arrhythmia detection device 210 may also be a whole structure, that is, the acquisition device and the processing device are both local, and the two devices are integrally disposed to form the arrhythmia detection device 210; or the arrhythmia detection device 210 may also be in a split structure, for example, the arrhythmia detection device 210 may include a collection device installed on the surface of a human body and a processing device located at the cloud, where the collection device is communicatively connected with the processing device, and the collection device collects an electrocardiograph signal and sends the electrocardiograph signal to the processing device for processing. The application is not limited herein.

In an embodiment of the present application, arrhythmia detection device 210 may be a wearable device, which may be a smart bracelet, a wearable wrist device, or the like. Of course, the wearable device can also be a generic name for intelligently designing daily wear by applying wearable technology and developing wearable devices, such as gloves, watches, clothes, footwear and the like. The wearable device can be worn directly on the body or can be integrated into a user's clothing or accessory. The wearable device is not only a hardware device, but also can realize more powerful functions through software support, data interaction, cloud interaction and the like. The generalized wearable intelligent device comprises full functions, large size, and complete or partial functions which can be realized independent of a smart phone, such as a smart watch or a smart glasses, and is only focused on certain application functions, and needs to be matched with other devices such as the smart phone for use, such as various smart bracelets, smart jewelry and the like for physical sign monitoring.

In an embodiment of the present application, the electronic device 220 may further include: smart phones, smart televisions, tablet computers, netbooks, personal digital assistants (personal digital assistant, PDAs), computer handheld communication devices, handheld computing devices, and other portable electronic devices.

For ease of understanding, the meaning of the relevant terms in the examples of the application are briefly described below.

Arrhythmia: the abnormal activation of the sinus node or the activation occurs outside the sinus node, and the activation is slow, blocked or conducted through abnormal channels, i.e. the origin of the heart activity and/or the conduction disorder causes abnormal frequency and/or rhythm of the heart beat.

Analog front end: is the basic system building block of the sensor circuit for amplifying and/or filtering sensor signals that are typically weak and may have complex electrical configurations to support different MCUs. The analog front end is used for processing analog signals given by a signal source and digitizing the analog signals, and the main functions of the analog front end include signal amplification, frequency conversion, modulation, demodulation, adjacent frequency processing, level adjustment and control and mixing.

Electrocardiosignal: is a weak bioelectric signal generated by the heart and collected by special equipment, can reflect the change of the generation, conduction and recovery of the heart excitation, at present, a conventional electrocardiograph in clinic is a change curve of an electrocardiograph signal obtained by measuring a certain part of the surface of a human body.

Heart beat: the electrocardiosignal is composed of a series of repeatedly appeared cardiac beats, generally, each cardiac beat comprises characteristic waves such as P wave, QRS complex, T wave and the like, and the QRS complex is the most obvious characteristic wave in an electrocardiogram and represents the potential change generated in the process of ventricular septum and left ventricular depolarization.

Fig. 3 is a system architecture diagram of an arrhythmia detection device according to an embodiment of the present application, where the system architecture includes: a beat abnormality classification unit 301 and a rhythm abnormality classification unit 302. The heart beat anomaly classification unit 301 classifies a heart beat set to be detected by using a preset anomaly heart beat classification model to obtain an anomaly classification result, wherein the anomaly classification result comprises that an anomaly heart beat exists in the heart beat set to be detected or no anomaly heart beat exists in the heart beat set to be detected, the heart beat set to be detected comprises N heart beats to be detected, and N is a positive integer greater than or equal to 2. When the beat anomaly classification unit 301 determines that an abnormal beat exists in the beat set to be detected, the rhythm anomaly classification unit 302 is awakened, and after the rhythm anomaly classification unit 302 is awakened, each beat to be detected in the beat set to be detected is classified by using a preset abnormal rhythm classification model, so that an abnormal rhythm classification result of each beat to be detected is obtained, and the abnormal rhythm classification result represents the arrhythmia type of the corresponding beat to be detected; when the beat anomaly classification unit 301 determines that there is no abnormal beat in the beat set to be measured, the beat anomaly classification unit 301 classifies the next beat set to be measured by using a preset abnormal beat classification model.

It will be appreciated that the architecture illustrated in fig. 3 does not constitute a particular limitation on the system architecture of the arrhythmia detection device. The arrhythmia detection device shown in fig. 3 may be a software unit, a hardware unit, or a unit combining soft and hard, which are built in an existing electronic device, or may be integrated into the electronic device as an independent pendant, or may exist as an independent electronic device.

In other embodiments of the application, the system architecture of the arrhythmia detection device may include more or fewer units or modules than shown, or combine certain units, or split certain units, or different component units. The illustrated elements may be implemented in hardware, software, or a combination of software and hardware. The embodiments of the application are not limited in this regard.

Fig. 4 is a flowchart illustrating an example arrhythmia detection method according to the present application. The method may be performed by an arrhythmia detection device provided by an embodiment of the application. The arrhythmia detection device will be described below as an example of a wearable device. As shown in fig. 4, the arrhythmia detection method includes: s401 to S403.

S401, classifying a heart beat set to be detected by using a preset abnormal heart beat classification model to obtain an abnormal classification result, wherein the abnormal classification result comprises that abnormal heart beats exist in the heart beat set to be detected or no abnormal heart beats exist in the heart beat set to be detected, the heart beat set to be detected comprises N heart beats to be detected, and N is a positive integer greater than or equal to 2.

In the embodiment of the application, the heart beat set to be detected comprises N heart beats to be detected, and N is a positive integer greater than or equal to 2, so that the preset abnormal heart beat classification model is used for judging a plurality of heart beats to be detected at the same time. The working frequency of a preset abnormal heart beat classification model is reduced, and meanwhile, the screening operation is simpler and more efficient because the related information (bean-to-bean correlation) of front and rear heart beats is reserved among the processing data.

In some embodiments, when classifying heart beat anomalies, the method specifically comprises the following steps: firstly, determining the heart rate variability characteristic of each heart beat to be detected in N heart beats to be detected, wherein the heart rate variability characteristic represents the difference of time intervals between adjacent R peaks; and inputting heart rate variability characteristics of the N heart beats to be detected into a preset abnormal heart beat classification model to obtain an abnormal classification result. In this embodiment, heart beat reconstruction is performed on the heart beat to be measured, and one heart beat is represented by using heart rate variability characteristics. The heart beat data after reconstruction operation can effectively reduce the difference of heart rhythm information among individuals, and highlight heart rhythm abnormality in the individuals, and has positive significance for improving the classification accuracy of a preset abnormal heart beat classification model.

It will be appreciated that it may be expressed in terms of RR intervals (RR intervals refer to the time difference between adjacent two R peaks) and heart rate variability characteristics such as adjacent RR intervals.

In some embodiments, fig. 5 is a schematic diagram of a heart beat reconstruction process according to an embodiment of the present application, in which 5 heart beats to be measured in one heart beat set to be measured are obtained through R-peak positioning. The R peaks of the 5 heart beats to be measured are represented by R-peak0, R-peak1, R-peak2, R-peak3, R-peak4, respectively, as shown in FIG. 5. RR-inteval 0 denotes the RR interval between R-peak0 and the previous R peak (i.e., R peak to the left of R-peak 0), RR-inteval 1 denotes the RR interval between R-peak0 and R-peak1, RR-inteval 2 denotes the RR interval between R-peak1 and R-peak2, RR-inteval 3 denotes the RR interval between R-peak2 and R-peak3, RR-inteval 4 denotes the RR interval between R-peak3 and R-peak4, and RR-inteval 5 denotes the RR interval between R-peak4 and the next R peak of R-peak4 (i.e., R peak to the right of R-peak 4).

As can be seen from fig. 5, the heart beat to be measured corresponding to any R peak in this embodiment includes 6 heart rate variability features, namely, feature 0, feature 1, feature 2, feature 3, feature 4 and feature 5. As shown in FIG. 5, feature 0 is RR _i Representing the separation of the R peak from the previous R peak; feature 1 is RR _i+1 Representing the separation of the R peak from the latter R peak; feature 2 isRepresenting the interval between the current heart beat to be measured and the first 10 RR; feature 3 is RR _i ratio, denoted RR _i The ratio of the average value to all RR intervals before the current heart beat to be measured; feature 4 is the nerrr _i ratio, denoted RR _i The ratio of the average value to the average value of 10 RRs before the current heart beat to be measured; feature 5 is RR _index Representing adjacent RR associated information.

It will be appreciated that RR _index Equal to RR _i And R is R _i R _i-1 Dividing the difference by RR _i And R is R _i R _i-1 And, wherein R is _i R _i-1 The RR interval of the preceding R peak and the further preceding R peak is indicated.

For example, a heart beat to be measured corresponding to R-peak2 is exemplified, and RR corresponding to the heart beat to be measured _index Obtained according to formula (1):

in some embodiments, the predetermined abnormal heart beat classification model is a first artificial neural network. The preset abnormal heart beat classification model is realized based on an artificial neural network, the artificial neural network is simpler in structure and smaller in calculated amount, and light calculation can be performed by keeping a smaller network scale, so that the power consumption of the heart rhythm abnormal detection equipment is reduced.

It may be understood that, if the arrhythmia detection device in the embodiment of the present application uses a remote processor to classify abnormal beats, any available neural network in the prior art may be selected as a preset abnormal beat classification model, for example, the neural network may be: CNN (Convolutional Neural Network ), DNN (Deep Neural Networks, deep neural network), RNN (Recurrent Neural Network ), and the like, to which the present application is not limited.

S402, if abnormal heart beats exist in the heart beat set to be detected, classifying the first heart beat to be detected by using a preset abnormal heart rhythm classification model to obtain an abnormal heart rhythm classification result of the first heart beat to be detected, wherein the first heart beat to be detected is any heart beat to be detected in the heart beat set to be detected, and the abnormal heart rhythm classification result represents the arrhythmia type of the first heart beat to be detected.

In some embodiments, the process of classifying the abnormal heart rhythm of the first heart to be measured specifically includes: obtaining pathological characteristics of a first heart beat to be tested; and inputting the case characteristics of the first heart beat to a preset abnormal heart rhythm classification model to obtain an abnormal heart rhythm classification result of the first heart beat.

It will be appreciated that the selection of pathological features requires balancing the contradiction between pathological information integrity and the computational effort required for feature extraction. In some embodiments, the pathology features include three classes, respectively: heart rate variability features, beat dimension reduction features, local QRS features.

Table 1 is a list of specific pathological features involved in some embodiments of the present application, including 6 heart rate variability features, 16 beat dimension reduction features, 5 local QRS features, i.e., a total of 27 pathological features.

TABLE 1

It will be appreciated that heart rate variability features can be used to characterize HRV (Heart Rate Variability ), have important pathological information, and require very little computation. The heart beat dimension reduction feature is obtained by accumulating, summing and averaging heart beat, the heart beat dimension reduction feature transforms the original high-dimension vector feature into a low-dimension vector feature, main useful information is not lost in the process, and core features in the heart beat are extracted to filter out interference information such as uncorrelated features and noise, so that the input complexity of a preset abnormal heart rhythm classification model is reduced, and the generalization capability of the preset abnormal heart rhythm classification model is enhanced to a certain extent. Local QRS features can increase individual variability from patient to patient by increasing pathological features.

The heart beat dimension reduction feature is obtained by accumulating and averaging a plurality of sampling points at the left side and the right side of the QRS peak value corresponding to the heart beat. For example, the Q peak is P wave band with 30 sampling points to the left, and every five sampling points are accumulated into one dimension reduction feature to obtain 6 dimension reduction features; the S peak is provided with 40 sampling points to the right as T wave bands, and every five sampling points are accumulated into one dimension reduction feature to obtain 8 dimension reduction features; respectively accumulating 5 sampling points left and right of the R peak to obtain 2 dimension reduction features; a total of 16 dimension reduction features were obtained.

In some embodiments, the preset abnormal heart rhythm classification model is a hybrid neural network model including a long-short time neural network, a second artificial neural network, and a third artificial neural network; the input data of the long-short time neural network and the second artificial neural network comprise pathological features of the first heart beat to be tested, the input data of the third artificial neural network comprise output data of the long-short time neural network and output data of the third artificial neural network, and the output data of the third artificial neural network comprise abnormal heart rhythm classification results of the first heart beat to be tested. The long and short time neural network, the second artificial neural network and the third artificial neural network are all network structures with simpler structures, and the model complexity is lower.

In some embodiments, the input data of the long-short-term neural network includes heart beat dimension reduction features of the first heart beat to be measured and QRS features of a part of the first heart beat to be measured, and the input data of the second artificial neural network includes heart rate variability features of the first heart beat to be measured. The heart rate variability features are used as input data of the second artificial neural network, and are reused after being applied to abnormal heart beat classification, so that the data utilization rate is improved.

Fig. 6 is a schematic diagram of the overall structure of a preset abnormal heart beat classification model and a preset abnormal heart rhythm classification model according to an embodiment of the application.

As shown in fig. 6, the AHC represents a first artificial neural network used by a preset abnormal heart beat classification model, and the AHC includes three full-connection layers, and the scale of the first artificial neural network is optimized to be 15×8×2 while maintaining accurate classification accuracy through repeated training evaluation. The neuron outputs of the input layer and the hidden layer are both subjected to a ReLu activation function of fixed-point quantization. The output layer neuron adopts a softMax activation function to calculate the corresponding probability of each classification and judge the classification result. The operations that each neuron contains can be expressed as formula (2):

y＝Act(∑w _i,j x _i,j +θ _i ) (2)

wherein w is _i,j The j-th input weight, θ, representing the i-th neuron _i Representing the bias of the ith neuron. X is x _i,j The j-th input data representing the i-th neuron.

As shown in fig. 6, where CAC represents a hybrid neural network (LSTM-ANN) used by a preset abnormal heart rhythm classification model. As shown in fig. 6, the hybrid neural network specifically includes: a long short time neural network (LSTM), a second artificial neural network (ANN 1), and a third artificial neural network (ANN 2). As shown in fig. 6, first, further feature extraction is performed on the heart beat dimension reduction feature and the heart beat local feature (namely, local QRS feature) of the heart beat to be measured by using LSTM to obtain a feature group a; meanwhile, carrying out further feature extraction on 6 heart rate variability features (namely reconstruction data) of the heart beat to be detected through the ANN1 to obtain a feature group B; and then combining the feature group A and the feature group B to form a new feature group, and inputting the new feature group into the ANN2 for final arrhythmia classification. In the classification process shown in fig. 6, the redundant features of the ECG are removed after multiple feature dimension reduction, so that a better abnormal heart rhythm classification result can be achieved by using only LSTM with the hidden layer number of 20 (n_h=20) and two artificial neural networks (ANN 1 and ANN 2). Wherein the ANN1 network scale is optimized to be 6×4, the ANN2 network scale is optimized to be 15×8×4, and the LSTM network scale is: input_size=21, hidden_layer=20, output_size=11, i.e., the input layer of LSTMCell has 21 neurons, the layer number of hidden layers is 20, and the output layer has 11 neurons.

S403: if no abnormal heart beat exists in the heart beat set to be detected, classifying the next heart beat set to be detected by using a preset abnormal heart beat classification model.

In the embodiment of the application, under the condition that abnormal heart beats do not exist in the heart beat set to be detected, the abnormal heart beat set to be detected is subjected to abnormal classification by utilizing a preset abnormal heart beat classification model without waking up the heart rhythm abnormal classification model.

It will be appreciated that each set of beats to be measured may be arranged according to time, and the next set of beats to be measured refers to the next set of beats to be measured in time, especially in the case of real-time arrhythmia detection of a user, the sets of beats to be measured are generally arranged according to the chronological order of time. Of course, the detection sequence of the heart beat set to be detected can be modified as required, and the application is not limited to the detection sequence.

In some embodiments, the heart beats to be measured included in different heart beat sets to be measured are different, and specifically, sampling times of sampling points corresponding to the heart beats to be measured may be different. The sampling times of the sampling points of the heart beats to be measured in the same heart beat set to be measured can be continuous.

In the arrhythmia detection method, a wake-up mechanism is set, and an abnormal heart rhythm classification model preset only when abnormal heart beats exist in N heart beats to be detected in the heart beat set to be detected is started, so that the starting times of the abnormal heart rhythm classification model are reduced, the power consumption of arrhythmia detection equipment is reduced, the application of the arrhythmia detection equipment is wider, and for example, the arrhythmia detection equipment can adopt a local processor to process data.

In some embodiments, the arrhythmia detection method further includes a beat positioning process (i.e. a determination process of a beat to be detected) before performing beat anomaly classification, and the beat positioning process is exemplarily described below with reference to the accompanying drawings.

It will be appreciated that the acquisition of the electrocardiographic signal is required prior to the heart beat positioning, and that one sampling point is obtained for each acquisition of the electrocardiographic signal.

In some embodiments of the application, event-driven sampling is employed to sample the electrocardiographic signals. Specifically, each sampling point includes polarity information and time interval information with the sampling point, wherein the polarity information represents amplitude change conditions of electrocardiosignals of the current sampling point and a last sampling point, the time interval information represents sampling time change of the current sampling point and the last sampling point, the last sampling point is adjacent to the current sampling point in sampling time, and the sampling time of the last sampling is earlier.

In some embodiments, the event driven sampling method is specifically: and when the change of the amplitude of the electrocardiosignal is greater than or equal to a preset amplitude threshold value, sampling is carried out once to obtain a sampling point. Therefore, the difference between the amplitudes of two adjacent sampling points is larger than or equal to a preset amplitude threshold value.

For example, the difference in amplitude between adjacent sample points in event driven sampling may be a fixed value.

Fig. 7 is a graph comparing sample points using conventional nyquist equidistant sampling and event driven sampling in an embodiment of the present application. Traditional nyquist equidistant sampling refers to sampling once every fixed time interval; the event-driven sampling adopted in the embodiment of the application is to perform one sampling under the condition that the amplitude variation reaches a certain value (namely the minimum variation amplitude), and the time interval between two sampling points is deltat. As shown in FIG. 7, compared with the traditional Nyquist equidistant sampling, the number of sampling points is greatly reduced after the event-driven sampling, so that the sampling efficiency is greatly improved and the wake-up power consumption is reduced.

After the sampling process is introduced, a heart beat positioning process (i.e., a process of determining a heart beat to be measured) will be described with reference to the drawings.

Fig. 8 is a flowchart of a heartbeat positioning method according to an embodiment of the application. As shown in fig. 8, the heartbeat positioning method includes:

s801, judging whether the polarity of the first sampling point is changed relative to the polarity of the last sampling point.

For example, whether the first sampling point is an extreme point may be determined according to the polarity information of the first sampling point and the polarity information of the last sampling point of the first sampling point.

For example, as shown in fig. 8, when the polarity of the first sampling point is 1 or 0, when the polarity of the first sampling point is 1, it means that the amplitude of the first sampling point is greater than the amplitude of the previous sampling point of the first sampling point, and if the polarity of the first sampling point is 0, the amplitude of the first sampling point is greater than the amplitude of the previous sampling point of the first sampling point.

In some embodiments, according to the polarity information of the first sampling point and the polarity information of the last sampling point of the first sampling point, it is determined whether the first sampling point is an extremum point, specifically, the polarity of the first sampling point is multiplied by the polarity value of the last sampling point, if the product is 0, it indicates that the first sampling point is an extremum point, and if the product is 1, it indicates that the first sampling point is not an extremum point.

S802, if S801 judges yes, duration is acquired; duration is: a first time interval between the first sampling point and a last extreme point of the first sampling point.

As shown in fig. 8, if no is determined in S801, the process returns to step S801 to determine the next sampling point of the first sampling point.

S803, judging whether the first condition Duration < Dur-Threshold is met. Duration < Dur-Threshold indicates whether the first time interval is less than a Duration Threshold;

S804, if S803 judges yes, R-R Dist and Avg RR are obtained. R-R Dist represents a second time interval between the first sampling point and the last R peak of the first sampling point, avg RR represents a third time interval corresponding to the first sampling point, the third time interval is an average RR interval between the first sampling point and W R peaks before the first sampling point, and W is a positive integer greater than 2. The value of W may be selected empirically, and is not particularly limited herein.

As shown in fig. 8, if no is determined in S803, the process returns to step S801 to determine the next sampling point of the first sampling point.

S805, determining whether the following second condition is satisfied, where the second condition is as follows: t1 < R-R Dist < T2 or T1 < R-R Dist < K1×AvgRR. T1 < R-R Dist < T2 or T1 < R-R Dist < K1×Avg RR means: the second time interval is greater than the first threshold and less than the second threshold, or the second time interval is greater than the first threshold and less than the product of the third time interval and the first parameter.

If yes in S806, S805 determines whether or not a third condition is satisfied, the third condition is as follows: R-R Dist < K2×Avg RR. R-R Dist < K2×Avg RR means: the second time interval is smaller than the product of the third time interval and the second parameter.

In the embodiment of the application, K1 is less than K2, and K1 and K2 are positive numbers. In an alternative embodiment, k1=0.5, k2=1.66.

S807, if the determination in S806 is yes, determining that the first sampling point is an R peak.

S808, if the determination in S806 is yes, the Duration Threshold (Dur-Threshold) is updated according to the first time interval (Duration).

As shown in fig. 8, the heartbeat positioning method of the sampling point further includes:

s809, after determining that the first sampling point is an R peak, determining the last extreme point of the first sampling point as a Q peak, and determining the last extreme point of the first sampling point as an S peak, wherein the R peak, the Q peak and the S peak belong to the same heart beat to be detected.

S810, if no in S805, determining whether a fourth condition is satisfied, the fourth condition is as follows: duration < Last R-peak Duration. Duration < Last R-peak Duration indicates that the first time interval (Duration) is smaller than the fourth time interval (Last R-peak Duration), the fourth time interval being the time interval between the first R-peak and the Last extreme point of the first R-peak, the first R-peak being the Last R-peak of the first sampling point.

If S810 is determined to be yes, the first sampling point is determined to be an R peak.

If the determination at step S810 is negative, the process returns to step S801 to determine the next sampling point of the first sampling point.

If the determination in S811 is negative, it is determined whether the fifth condition is satisfied, where the fifth condition is: duration < Min Duration. Duration < Min Duration means that the first time interval is smaller than the fifth time interval (Min Duration), which is the minimum value among RR intervals between W R peaks before the first sampling point.

If S811 determines that it is true, the first sampling point is determined to be the R peak.

If the determination in S811 is negative, the routine returns to step S801 to determine the next sampling point of the first sampling point.

It will be appreciated that the above-described process of the beat locating method mainly involves locating the peak of the QRS wave, which occupies a major part of a beat. The P-wave and T-wave may be determined empirically, for example, a number of extreme points after the S-peak may be determined as the T-peak, and a number of extreme points before the Q-peak may be determined as the P-peak according to statistical data, which is not described in detail in the present application.

The terms in fig. 8 are exemplary described below.

Sample point polarity: as shown in fig. 8, if the amplitude of the target sampling point is greater than the amplitude of the last sampling point of the target sampling point, the polarity of the target sampling point is 1, and if the amplitude of the target sampling point is smaller than the amplitude of the last sampling point of the target sampling point, the polarity of the target sampling point is 0.

Extreme points: if the polarity of the target sampling point is different from the polarity of the last sampling point of the target sampling point, the target sampling point is judged to be an extreme point (for example, the product of the polarities is 0), and if the polarity of the target sampling point is the same as the polarity of the last sampling point of the target sampling point, the target sampling point is judged not to be the extreme point.

As shown in fig. 8, the Duration is a first time interval, which represents a sampling time interval between the last extreme points of the first sampling point; R-R Dist is a second time interval representing the sampling time interval between the first sampling point and the last R peak of the first sampling point.

In some embodiments, if the R peak is determined to be R peak in step S806 shown in fig. 8 when the R peak determination is performed on the sampling point S, the duration threshold needs to be updated at this time. The duration Threshold (Dur-Threshold) may be updated according to equations (3), (4) and (5):

SP _s+1 ＝SP _s -[0.25×(SP _s -Duration _s )] (3)

NP _s+1 ＝NP _s -[0.25×(NP _s -Duration _s )] (4)

Dur_Threshold _s+1 ＝SP _s+1 +[0.25×(NP _s+1 -SP _s+1 )] (5)

in the formulae (3) - (5), SP _s Represents the SP constant, SP corresponding to the sampling point S _s+1 Represents the updated SP constant, duration _s A first time interval representing a sampling point S; NP (NP) _s NP constant corresponding to sample point S, NP _s+1 Represents the updated SP constant, dur_Threshold _s+1 Representing the updated duration threshold.

It is understood that the initial value of SP, the initial value of NP, and the initial value of the duration Threshold Dur_Threshold may all be set according to experimental data.

In some embodiments, the initial value of SP and the initial value of NP are both 255 and the initial value of duration Threshold dur_threshold is 2300.

It should be noted that, because the content of information interaction and execution process between the devices/apparatuses/units in the embodiments of the present application is based on the same concept as the embodiments of the method of the present application, specific functions and technical effects thereof may be referred to in the method embodiment section, and will not be described herein again.

It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions. The functional units and modules in the embodiment may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit, where the integrated units may be implemented in a form of hardware or a form of a software functional unit. In addition, the specific names of the functional units and modules are only for distinguishing from each other, and are not used for limiting the protection scope of the present application. The specific working process of the units and modules in the above system may refer to the corresponding process in the foregoing method embodiment, which is not described herein again.

For ease of understanding, the architecture of an arrhythmia detection system in an embodiment of the application is described below with reference to the accompanying drawings. As shown in fig. 9, the system includes an analog front end, a QRS beat locator, an imbalance-like processor, an abnormal beat detector, and an abnormal heart rhythm detector; the abnormal heart beat detector comprises a heart beat reconstructor and an abnormal heart beat classifier, and the abnormal heart beat detector comprises a feature fusion device and an abnormal heart rhythm classifier.

In the arrhythmia detection system shown in fig. 9, the electrode acquires potential data from the surface of a human body and transmits the potential data to an analog front end, the analog front end performs false driving sampling on the potential data to obtain sampling point signals (such as 'polarity' and 'Δt' shown in fig. 9, the polarity represents polarity information of the sampling points, and the Δt represents time interval between the two sampling points), the signals are used as input of the arrhythmia detection system, and QRS heartbeat positioning is performed on an electrocardiosignal by using a high-robustness heartbeat positioning method of which threshold adaptation is performed in a QRS heartbeat positioner based on R peak slope change, wherein the adaptive threshold comprises an adaptive adjustment duration threshold and an adaptive adjustment average RR interval, and three kinds of noises including baseline drift, myoelectricity and motion artifact can be effectively restrained through two adaptive thresholds. After the heart beat positioning is completed, the electrocardiosignals are divided into independent heart beats. And then, performing class unbalance processing on a plurality of obtained independent heart beats (namely, each independent heart beat is a sample) by using a class unbalance processor, and keeping the sample balance between the abnormal samples and the normal samples in the final training data set through the proposed abnormal sample synthesis and random sampling algorithm. And then reconstructing the heart beat after sample balancing, replacing the heart beat by heart rate variability characteristics, and inputting segments formed by five continuous reconstructed heart beats into an abnormal heart beat classifier (a first artificial neural network ANN) for performing classification processing on the heart beat segments. When abnormal heart beat segments exist, an abnormal heart beat detector is awakened, a feature fusion device is started, heart beats are subjected to feature dimension reduction, meanwhile, local features of the heart beats are extracted, heart rate variability features are multiplexed and input into an abnormal heart rhythm classifier (LSTM_ANN hybrid neural network) together to conduct arrhythmia four classification, N in the arrhythmia four classification represents normal or bundle branch conduction block beats, SVEB represents supraventricular abnormal beats, VEB represents ventricular abnormal beats, and F represents fusion beats.

The arrhythmia detection system shown in fig. 9 is divided into three stages of wake-up: 1. QRS positioning awakening is started after the heart beat segment is received; 2. the abnormal heart beat detector wakes up, and the abnormal heart beat detector wakes up and starts up only after the sample is balanced; 3. the abnormal heart rhythm detector wakes up, and the abnormal heart rhythm detector wakes up to be started only when the abnormal heart beat segment exists.

It can be understood that the quasi-unbalance processor in the arrhythmia detection system is started only in the model training process, so that the quantity of normal heart beats and abnormal heart beats in the training sample data set is balanced, and the model training effect is ensured; in the application process of the arrhythmia detection system, the quasi-unbalance processing is not needed, so that the quasi-unbalance processing is closed, and the heart beat output by the QRS heart beat positioner directly enters the abnormal heart beat detector.

For ease of understanding, an exemplary description of the class imbalance process is provided below, and FIG. 10 is a flow chart of the class imbalance process. As shown in fig. 10, wherein the original sample data set includes original abnormal samples and original normal samples, each original abnormal sample includes an abnormal heart beat, and each original normal sample includes a normal heart beat; the class imbalance process is generally as follows:

(1) And (3) oversampling the original abnormal samples to obtain a plurality of synthesized abnormal samples, wherein each synthesized abnormal sample comprises an abnormal heart beat. It can be understood that the oversampling is to use SMOTE algorithm for processing, and the specific processing procedure of the SMOTE algorithm is a conventional technology, which is not described herein.

(2) Randomly sampling the plurality of synthesized abnormal samples to obtain the target synthesized abnormal samples.

(3) And taking the target synthesized abnormal sample and the original abnormal sample as the final class balanced abnormal sample, and taking the original normal sample as the class balanced normal sample.

It should be understood that, in general, the number of original normal samples and the number of original abnormal samples in the original sample data set are multiple, and the number of original normal samples is far greater than the number of original abnormal samples, in order to ensure balance between the abnormal samples and the normal samples in the training data set, the embodiment of the present application performs oversampling on the abnormal samples to increase the number of abnormal samples.

It will be appreciated that the proportion of random samples may be set as desired. In some embodiments, the random sampling ratio is set such that the ratio of the number of the most-obtained class-balanced abnormal samples to the number of class-balanced normal samples is approximately 1:1.

Illustratively, in one embodiment, the number of original abnormal samples in the original sample dataset is 7519, the number of original normal samples is 2529, the number of synthesized abnormal samples is 16634, the random sampling ratio is 30%, and the number of obtained target synthesized abnormal samples is 4990; the number of abnormal samples in the final class balance is equal to the number of normal samples in the class balance.

It will be appreciated that in the arrhythmia detection system shown in fig. 9, the abnormal heart beat detector and the abnormal heart rhythm detector may be hardware-shared, so that the hardware cost may be reduced.

Performance verification is also performed for the arrhythmia detection system shown in fig. 9 in an embodiment of the present application. The verification result will be described below.

1. Verification of heart beat positioning

The MIT-BIH data set was used to evaluate the performance of the beat positioning algorithm to be tried initially in a static scenario and the data set contained 44 30 minute electrocardiographic recordings each tested in detail. The results are shown in Table 2, and the specific performance evaluation indexes include: TP (True positive) the algorithm accurately locates the number of correct beats, FN (False negative) the algorithm misplaces beats, FP (False positive) the algorithm missed beats, se sensitivity = TP/(tp+fn), positive prediction rate = TP/(tp+fp) and DER error rate = (fp+fn)/(tp+fn). The evaluation result shows that the heart beat positioning algorithm tried at present has better performance on all patient data in a data set, the sensitivity is generally maintained at 98% -100%, and the average value is 98.62%; the positive prediction rate is 98.88% at 99% -100% average value; the overall error rate is only 2.488%.

The performance of the beat positioning algorithm in dynamic scenarios was evaluated using the MIT-BIH-NST database, and 8 records at different signal-to-noise ratios were tested, as shown in Table 3. The test results showed that even in the absence of pretreatment, the average sensitivity (Sen) in noisy environments was 97.38% and the average positive prediction (p+) was 97.08%. In particular, when the subject runs at 7 km/h, the signal-to-noise ratio is only 6dB, the algorithm sensitivity (Sen) can reach 94.48%, and the positive predictive rate (p+) can reach 93.54%. Meanwhile, the time complexity T (n) of the algorithm and the data length n only have a linear relation, so that the implementation cost of hardware is greatly reduced.

TABLE 2

TABLE 3 Table 3

2. Accurate arrhythmia classification

Under the condition of adopting AAMI standard patient general classification, firstly carrying out abnormal heart beat classification (AHC) on the reconstructed heart beat segments, waking up an abnormal heart rhythm detector under the condition of abnormal heart beat, carrying out dimension reduction and feature extraction on the heart beat, carrying out feature fusion on multiplexing heart rate variability features, and inputting the feature fusion into a hybrid neural network for four classification. By the two-stage classification mode, redundant calculation amount caused by multi-continuous shooting segments of all normal heart beats can be reduced as much as possible on the important premise of avoiding abnormal heart beat omission. Meanwhile, as the class unbalance processor is introduced to balance the sample, the balance of the classification result is greatly improved.

Table 4 shows the confusion matrix and classification results for its abnormal beat classification. Ab Sen indicates that abnormal beat classification identifies 96.21% of abnormal beat segments and wakes up the abnormal heart rhythm detector to further classify abnormal beats. N Sen represents a pure frequently-repeated shooting segment of 84.32% that can be correctly recognized. Namely, the processor omits 3.79 percent of multi-continuous shooting fragments with abnormal heart beats under the low-power consumption working mode (closing the abnormal heart rhythm detector); although 15.68% of the redundant computation of a normal heart beat abnormal heart rhythm detector is introduced, the operating frequency and power consumption of the abnormal heart rhythm detector are greatly reduced by a wake-up mechanism. The classification accuracy ACC of the overall abnormal beat classification (CAC) was 87.72%. Meanwhile, in order to evaluate the performance of the two classification systems, the two parameters of Pre and Sen are considered, the performance index F1-socre is used for evaluation, the F1-socre of the overall abnormal heart beat classification is 90.75%, and the classification performance is good.

TABLE 4 Table 4

Table 5 shows the confusion matrix and classification results plotted by AHC+CAC. Because the number of F beats is too small, the performance index is mainly focused on N, SVEB and VEB classification. According to the confusion matrix, the overall classification performance of N classes of heart beats is that the sensitivity Sen is 94.56%, the positive prediction rate P+ is 99.22%, and the specificity Spr is 93.60%; the VEB classification sensitivity Sen was 95.84%, the positive predictive rate p+ was 86%, and the specificity Spr was 96..9%; SVEB sensitivity Sen was 77.05%, positive predictive rate P+ was 47.75%, specificity Spr96.9%. The overall accuracy is 93.47%, the abnormal heart rhythm sensitivity is 89.1%, and the classification performance is good, so that the universal standard of patients is met.

TABLE 5

3. Robustness testing of arrhythmia classification

In order to verify the robustness of arrhythmia classification, an MIT-BIH arrhythmia database and an MIT-BIH noise pressure database are used in a superposition combination mode for verification, and because three noise records of the MIT-BIH noise pressure database exist in a large number of low-frequency components, the noise amplitude measured at the moment can be virtually high, and after local direct current components are removed, the average value of noise signals in every 10 seconds is calculated as the noise amplitude based on the calculated signal amplitude and the noise amplitude, and noise electrocardiosignals with any signal to noise ratio can be obtained. Finally, three noise signals are respectively added to 44 records of the MIT-BIH arrhythmia database based on the above description, and corresponding arrhythmia classification is carried out. Table five shows the test results for arrhythmia at original signal, 24db noise, 18db noise, 12db noise, and 6db noise.

It can be seen from table 6 that it can have an overall accuracy of 90.1% even at a noise of 6db, and that it can maintain accuracy of 94.2% and 93.2% for V classification (VEB) and S classification (SVEB) (main reference index for abnormal heart rhythm classification) as well, with good performance.

TABLE 6

Fig. 11 is a schematic structural diagram of an arrhythmia detection device according to an embodiment of the present application. As shown in fig. 11, the arrhythmia detection device 110 of this embodiment includes: at least one processor 1100 (only one processor is shown in fig. 11), a memory 1101, and a computer program 1102 stored in the memory 1101 and executable on the at least one processor 1100, the processor 1100 implementing the steps in any of the arrhythmia detection method embodiments described above when executing the computer program 1102.

It will be appreciated by those skilled in the art that fig. 11 is merely an example of arrhythmia detection device 110 and is not intended to limit arrhythmia detection device 110, and may include more or fewer components than shown, or may combine certain components, or may include different components, such as input-output devices, network access devices, etc.

The processor 1100 may be a central processing unit (Central Processing Unit, CPU), the processor 1100 may also be other general purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate arrays (fiiel-Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The memory 1101 may in some embodiments be an internal storage unit of the arrhythmia detection device 110, such as a hard disk or memory of the arrhythmia detection device 110. The memory 1101 may also be an external storage device of the arrhythmia detection device 110 in other embodiments, such as a plug-in hard disk, smart Media Card (SMC), secure Digital (SD) Card, flash Card (Flash Card) or the like, which are provided on the arrhythmia detection device 110. Further, the memory 1101 may also include both internal and external storage units of the arrhythmia detection device 110. The memory 1101 is used to store an operating system, application programs, boot loader (BootLoader), data, and other programs, such as program codes of the computer programs. The memory 1101 may also be used to temporarily store data that has been output or is to be output.

Embodiments of the present application also provide a computer readable storage medium storing a computer program which, when executed by a processor, implements steps for implementing the embodiments of the method for detecting arrhythmia described above.

Embodiments of the present application provide a computer program product which, when run on a mobile terminal, causes an arrhythmia detection device to perform steps that enable the various arrhythmia detection method embodiments described above to be carried out.

The embodiment of the application also provides a chip in the electronic equipment, which comprises: a processing unit, which may be, for example, a processor, and a communication unit, which may be, for example, an input/output interface, pins or circuitry, etc. The processing unit may execute computer instructions to cause the electronic device to perform any of the arrhythmia detection methods provided by the embodiments of the application described above.

Optionally, the computer instructions are stored in a storage unit.

Alternatively, the storage unit is a storage unit in the chip, such as a register, a cache, etc., and the storage unit may also be a storage unit in the terminal located outside the chip, such as a ROM or other type of static storage device that can store static information and instructions, a random RAM, etc. The processor mentioned in any of the above may be a CPU, a microprocessor, an ASIC, or one or more integrated circuits for controlling the execution of the program of the above-mentioned feedback information transmission method. The processing unit and the storage unit may be decoupled and respectively disposed on different physical devices, and the respective functions of the processing unit and the storage unit are implemented by wired or wireless connection, so as to support the system chip to implement the various functions in the foregoing embodiments. Alternatively, the processing unit and the memory may be coupled to the same device.

The arrhythmia detection device, the computer readable storage medium, the computer program product or the chip provided in this embodiment are used to execute the corresponding method provided above, so that the beneficial effects thereof can be referred to the beneficial effects in the corresponding method provided above, and will not be described herein.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the present application may implement all or part of the flow of the method of the above embodiments, and may be implemented by a computer program to instruct related hardware, where the computer program may be stored in a computer readable storage medium, and when the computer program is executed by a processor, the computer program may implement the steps of each of the method embodiments described above. Wherein the computer program comprises computer program code which may be in source code form, object code form, executable file or some intermediate form etc. The computer readable medium may include at least: the computer program code can be carried to any entity or device, a recording medium, a computer Memory, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), an electrical carrier signal, a telecommunication signal, and a software distribution medium of the walker-based fall prevention walker device/terminal equipment. Such as a U-disk, removable hard disk, magnetic or optical disk, etc. In some jurisdictions, computer readable media may not be electrical carrier signals and telecommunications signals in accordance with legislation and patent practice.

In the foregoing embodiments, the descriptions of the embodiments are emphasized, and in part, not described or illustrated in any particular embodiment, reference is made to the related descriptions of other embodiments.

Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

In the embodiments provided in the present application, it should be understood that the disclosed apparatus/network device and method may be implemented in other manners. For example, the apparatus/network device embodiments described above are merely illustrative, e.g., the division of the modules or units is merely a logical functional division, and there may be additional divisions in actual implementation, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection via interfaces, devices or units, which may be in electrical, mechanical or other forms.

The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

The above embodiments are only for illustrating the technical solution of the present application, and not for limiting the same; although the application has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present application, and are intended to be included in the scope of the present application.

Claims

1. An arrhythmia detection device comprising a memory having stored therein computer readable instructions and a processor for executing the following steps when the processor invokes the computer readable instructions in the memory:

Classifying a heart beat set to be detected by using a preset abnormal heart beat classification model to obtain an abnormal classification result, wherein the abnormal classification result comprises abnormal heart beats in the heart beat set to be detected or no abnormal heart beats in the heart beat set to be detected, and the heart beat set to be detected comprises N heart beats to be detected, wherein N is a positive integer greater than or equal to 2;

if abnormal heart beats exist in the heart beat set to be detected, classifying a first heart beat to be detected by using a preset abnormal heart rhythm classification model to obtain an abnormal heart rhythm classification result of the first heart beat to be detected, wherein the first heart beat to be detected is any heart beat to be detected in the heart beat set to be detected, and the abnormal heart rhythm classification result represents the arrhythmia type of the first heart beat to be detected;

2. The arrhythmia detection device as claimed in claim 1, wherein classifying the set of beats to be detected using a preset abnormal beat classification model to obtain an abnormal classification result comprises:

Determining heart rate variability characteristics of each of the N heart beats to be detected, the heart rate variability characteristics representing variability in time intervals between adjacent R peaks;

inputting heart rate variability characteristics of the N heart beats to be detected into a preset abnormal heart beat classification model to obtain an abnormal classification result;

the classifying the first heart beat to be tested by using a preset abnormal heart rhythm classification model to obtain an abnormal heart rhythm classification result of the first heart beat to be tested comprises the following steps:

obtaining pathological characteristics of the first heart beat to be tested;

and inputting the case characteristics of the first heart beat to a preset abnormal heart rhythm classification model to obtain an abnormal heart rhythm classification result of the first heart beat.

3. An arrhythmia detection device as claimed in claim 2 wherein the pathological features of the first beat to be detected include heart rate variability features, beat dimension reduction features and local QRS features of the first beat to be detected, the beat dimension reduction features of the first beat to be detected being obtained by dimension reduction at a plurality of sampling points of the first beat to be detected, the local QRS features of the first beat to be detected including features relating to QRS peaks of the first beat to be detected.

4. The arrhythmia detection device as defined in claim 3 wherein the preset abnormal heart beat classification model comprises a first artificial neural network;

the preset abnormal heart rhythm classification model is a hybrid neural network model, and the hybrid neural network model comprises a long-short-time neural network, a second artificial neural network and a third artificial neural network;

the input data of the long-short time neural network and the second artificial neural network comprise pathological features of the first heart beat to be detected, the input data of the third artificial neural network comprise output data of the long-short time neural network and output data of the third artificial neural network, and the output data of the third artificial neural network comprise abnormal heart rhythm classification results of the first heart beat to be detected;

the input data of the long-short time neural network comprises heart beat dimension reduction characteristics of the first heart beat to be detected and partial QRS characteristics of the first heart beat to be detected, and the input data of the second artificial neural network comprises heart rate variability characteristics of the first heart beat to be detected.

5. The arrhythmia detection device of claim 1 wherein the processor further performs the steps of:

Acquiring electrocardiosignals of a target patient to obtain a plurality of sampling points, wherein a first sampling point comprises polarity information and time interval information, the polarity information represents the amplitude change condition of the electrocardiosignals of the first sampling point and a last sampling point, the time interval information represents the sampling time change of the first sampling point and the last sampling point, and the first sampling point is any one of the plurality of sampling points;

determining whether the first sampling point is an extreme point or not according to the polarity information of the first sampling point and the polarity information of the last sampling point of the first sampling point;

if the first sampling point is an extreme point, a first time interval between the first sampling point and the last extreme point of the first sampling point is acquired;

if the first time interval is smaller than a duration threshold, acquiring a second time interval between the first sampling point and the last R peak of the first sampling point and a third time interval corresponding to the first sampling point, wherein the third time interval is an average RR interval between the first sampling point and W R peaks before the first sampling point, and W is a positive integer greater than 2;

and if the second time interval is smaller than the product of the third time interval and the second parameter, determining the first sampling point as an R peak.

6. The arrhythmia detection device of claim 5 wherein the processor further performs the steps of:

and after the first sampling point is determined to be R peak, updating the duration threshold according to the first time interval.

7. The arrhythmia detection device of claim 5 wherein the processor further performs the steps of:

if the second time interval is not greater than the first threshold or not less than the second threshold, and the second time interval is not greater than the first threshold or not less than the product of a third time interval and a second parameter, judging whether the first time interval is less than a fourth time interval, wherein the fourth time interval is a time interval between a first R peak and a last extreme point of the first R peak, and the first R peak is a last R peak of the first sampling point;

If the first time interval is smaller than the fourth time interval, determining the first sampling point as an R peak;

if the second time interval is not smaller than the product of the third time interval and the second parameter, judging whether the first time interval is smaller than a fifth time interval, wherein the fifth time interval is the minimum value in RR intervals among W R peaks before the first sampling point;

and if the first time interval is smaller than the fifth time interval, determining the first sampling point as an R peak.

8. The arrhythmia detection device of claim 5 wherein an absolute value of a difference between the magnitude of the first sampling point and the magnitude of a previous sampling point from the first sampling point is greater than or equal to a preset magnitude threshold.

9. The arrhythmia detection device as claimed in any one of claims 1 to 8 wherein the raw sample data set comprises raw abnormal samples and raw normal samples, each raw abnormal sample comprising an abnormal heart beat and each raw normal sample comprising a normal heart beat; the processor performs the steps of:

oversampling the original abnormal samples to obtain a plurality of synthesized abnormal samples, each of the synthesized abnormal samples including an abnormal heart beat;

Randomly sampling the plurality of synthetic abnormal samples to obtain a target synthetic abnormal sample;

and training an initial abnormal heart beat classification model and an initial abnormal heart rhythm classification model by taking a data set formed by the target synthesized abnormal sample, the original abnormal sample and the original normal sample as a training data set to obtain the preset abnormal heart beat classification model and the preset abnormal heart rhythm classification model.

10. A computer readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, performs the steps performed by the processor in the arrhythmia detection device of any one of claims 1 to 9.