CN115956924A - A method, device, electronic device and medium for processing electrocardiographic signals - Google Patents

Abstract

The invention discloses an electrocardiosignal processing method, an electrocardiosignal processing device, an electronic device and a medium, wherein the method comprises the following steps: acquiring an electrocardiosignal to be detected; dividing the electrocardiosignal to obtain a plurality of electrocardiosignal segments with the same length, wherein each electrocardiosignal segment at least comprises a heart beat pulse; and respectively inputting each electrocardiosignal segment into a pre-trained detection model, and determining the relative position information of each heart beat pulse in each electrocardiosignal segment and the category information of each heart beat pulse, wherein the category information reflects the abnormal condition of the heart rhythm. The method can simultaneously position and classify heart beats, further improves the classification performance by utilizing the potential dependency relationship between heart beats, and is beneficial to improving the accuracy of electrocardiosignal processing.

Description

Electrocardiogram signal processing method, device, electronic equipment and medium

Technical Field

The present invention relates to the field of data processing technology, and in particular to an electrocardiogram signal processing method, device, electronic equipment and medium.

Background Art

Electrocardiogram, also known as electrocardiogram, is a non-invasive monitoring technology for cardiac electrophysiological activity. With the development of machine learning technology, machine learning technology is often used to solve the classification problem of ECG signals. Existing automatic analysis technologies of ECG signals based on machine learning are mainly divided into two categories: the first category is to manually extract heartbeat features based on expert knowledge, and then classify the heartbeat based on the extracted features. Specifically, commonly used features include waveforms of each characteristic band, statistical values, wavelet transform features, etc. The second method is to automatically extract features from a given waveform based on a deep learning method, and classify it based on the automatically extracted features. However, the heartbeat classification task actually includes two subtasks: heartbeat location and classification. The heartbeat location needs to be extracted first and then classified. Existing technologies often require additional heartbeat location algorithms for support.

The existing automatic ECG signal analysis technology based on machine learning has the following disadvantages: the first type of method mentioned above relies on a lot of expert knowledge in the medical field, which not only requires a lot of human resources, but also the analysis effect is limited to the cognition of human experts; feature engineering requires a huge workload, and it is difficult to find the best feature subset; feature extraction is not robust enough and has poor scalability, and feature subsets need to be rebuilt for different tasks. The second type of method benefits from the ability of the stochastic gradient descent method to automatically extract appropriate features for different tasks and has good robustness to noise, but the current technology requires the use of additional heartbeat location algorithms to extract the heartbeat position, which consumes additional computing power, increases the difficulty of application deployment, cannot avoid the errors caused by low positioning accuracy, and cannot use the relevant information between heartbeats.

Therefore, how to predict the location and category of all heartbeats in a given ECG signal more quickly and efficiently and improve the ECG signal processing performance has become a technical challenge.

Summary of the invention

The embodiments of the present application provide an electrocardiogram signal processing method, device, electronic device and medium. The method can simultaneously locate and classify heartbeats, and further improve the classification performance by utilizing the potential dependencies between heartbeats, which is beneficial to improving the accuracy of electrocardiogram signal processing.

In a first aspect, the present invention provides the following technical solution through an embodiment of the present invention:

A method for processing an electrocardiogram signal comprises: obtaining an electrocardiogram signal to be detected; dividing the electrocardiogram signal to obtain a plurality of electrocardiogram signal segments of the same length, wherein each electrocardiogram signal segment contains at least one heartbeat pulse; inputting each electrocardiogram signal segment into a pre-trained detection model to determine the relative position information of each heartbeat pulse in each electrocardiogram signal segment and the category information of each heartbeat pulse, wherein the category information reflects the abnormal heart rhythm condition.

Preferably, the detection model is trained according to the following steps: obtaining an initial ECG signal for training; dividing the initial ECG signal to obtain a plurality of ECG signal segment samples, wherein the ECG signal segment samples contain at least one heart pulse, and the length of the ECG signal segment samples is equal to the length of the ECG signal segment; labeling each ECG signal segment sample to obtain a label set for each ECG signal segment sample, wherein the label set includes a position label and a category label for each heart beat in each ECG signal segment sample; training a pre-constructed machine learning model based on the ECG signal segment samples and the label set to obtain the detection model.

Preferably, the detection model includes: a one-dimensional convolution module, a self-attention encoding module, a self-attention decoding module and a fully connected output module. The ECG signal segment is input into a pre-trained detection model to determine the relative position information of each heart pulse in the ECG signal segment and the category information of each heart pulse, including: extracting features of the input ECG signal segment through the one-dimensional convolution module; extracting and fusing features of the output result of the one-dimensional convolution module through the self-attention encoding module and the self-attention decoding module; processing the features output by the self-attention decoding module through the fully connected output module, and outputting the relative position information of each heart pulse in the ECG signal segment and the category information of each heart pulse.

Preferably, each ECG signal segment sample is labeled, including: obtaining the position of the heartbeat characteristic point of each heartbeat pulse in each ECG signal segment; determining the position label of each heartbeat based on the distance between the position of the heartbeat characteristic point and the starting position of the ECG signal segment where the heartbeat is located; and for each heartbeat pulse in each ECG signal segment sample, determining a category from a plurality of preset heart rhythm abnormality categories as the category label of the heartbeat pulse.

Preferably, the pre-constructed machine learning model is trained based on the ECG signal segment samples and the label set, including: inputting the ECG signal segment samples into the machine learning model to obtain a prediction set, the prediction set including predicted position information and predicted category information of each heart beat in each input ECG signal segment sample; based on the prediction set output by the machine learning model and the label set, obtaining the best match between the prediction set and the label set; according to the best matching result, obtaining the total loss of the prediction set and the label set, the total loss including category loss and position loss; training the machine learning model based on the total loss to obtain the detection model.

Preferably, the prediction set based on the output of the machine learning model and the label set are used to obtain the best match between the prediction set and the label set, including: matching each element in the label set with each element in the prediction set in turn to obtain a match with minimal error, wherein the matching process includes: matching the relative position information of the label set with the relative position information of the prediction set, and matching the category information of the label set with the category information of the prediction set; determining the match with the smallest error as the best match.

Preferably, dividing the ECG signal to obtain a plurality of ECG signal segments of the same length comprises: dividing the ECG signal based on a sliding window of a preset length to obtain a plurality of ECG signal segments of the same length.

In a second aspect, the present invention provides the following technical solution through an embodiment of the present invention:

An electrocardiogram signal processing device, comprising:

An acquisition module, used for acquiring the electrocardiogram signal to be detected;

A division module, used for dividing the electrocardiogram signal to obtain a plurality of electrocardiogram signal segments of the same length, wherein each electrocardiogram signal segment contains at least one heart beat pulse;

The determination module is used to input each ECG signal segment into a pre-trained detection model to determine the relative position information of each heart beat pulse in each ECG signal segment and the category information of each heart beat pulse, wherein the category information reflects the abnormal heart rhythm condition.

In a third aspect, the present invention provides the following technical solution through an embodiment of the present invention:

An electronic device comprises: a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method described in the first aspect when executing the program.

In a fourth aspect, the present invention provides the following technical solution through an embodiment of the present invention:

A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method described in any one of the first aspects above.

One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

An ECG signal processing method, device, electronic device and medium provided by an embodiment of the present invention first obtain an ECG signal to be detected; divide the ECG signal to obtain multiple ECG signal segments of the same length, wherein each ECG signal segment contains at least one heartbeat pulse; input each ECG signal segment into a pre-trained detection model to determine the relative position information of each heartbeat pulse in each ECG signal segment, as well as the category information of each heartbeat pulse, wherein the category information reflects the abnormal heart rhythm condition. Thus, by inputting the divided ECG signal into the pre-trained detection model, the relative position information and category information of each heartbeat pulse in each ECG signal segment are obtained. This method can simultaneously predict the position and category of all heartbeats in the ECG signal to be detected end-to-end, without the support of an additional heartbeat positioning algorithm, and can reduce the amount of calculation and the complexity of application deployment. And because it can process ECG signals containing multiple heartbeats and simultaneously calculate the positions and categories of all heartbeats in the entire ECG signal, it is possible to obtain the potential dependencies between the heartbeats and use the potential dependencies to further improve the accuracy of heartbeat classification, which is beneficial to improving the accuracy of ECG signal processing.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

FIG1 is a flow chart of an electrocardiogram signal processing method provided by an embodiment of the present invention;

FIG2 is a schematic diagram of the division of ECG signal segments provided by an embodiment of the present invention;

FIG3 is a schematic diagram of the structure of a machine learning model provided by an embodiment of the present invention;

FIG4 is a schematic diagram of the structure of a machine learning model control process provided by an embodiment of the present invention;

FIG5 is a schematic diagram of a detection result based on an electrocardiogram signal processing method provided by an embodiment of the present invention;

FIG6 is a schematic diagram of the structure of an electrocardiogram signal processing device provided by an embodiment of the present invention;

FIG7 is a schematic diagram of the structure of an electronic device without a dedicated acceleration circuit provided by an embodiment of the present invention;

FIG8 is a schematic diagram of the structure of an electronic device with a dedicated acceleration circuit provided in an embodiment of the present invention.

DETAILED DESCRIPTION

The embodiments of the present application provide an electrocardiogram signal processing method, device, electronic device and medium. The method can simultaneously locate and classify heartbeats, and further improve the classification performance by utilizing the potential dependencies between heartbeats, which is beneficial to improving the accuracy of electrocardiogram signal processing.

The technical solution of the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

A method for processing an electrocardiogram signal comprises: obtaining an electrocardiogram signal to be detected, dividing the electrocardiogram signal to obtain a plurality of electrocardiogram signal segments of the same length, wherein each electrocardiogram signal segment contains at least one heartbeat pulse; inputting each electrocardiogram signal segment into a pre-trained detection model, respectively, determining a position label of each heartbeat pulse in each electrocardiogram signal segment, and a category label of each heartbeat pulse, wherein the category label reflects the abnormal heart rhythm condition.

In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

It should be noted that the ECG signal processing method provided in the embodiment of the present application is applicable to wearable devices such as smart watches and smart bracelets for real-time monitoring of a person's heart health status. In addition, the heartbeat pulse in the present application represents the fluctuation of the heart rhythm, and one heartbeat represents one fluctuation of the heart rhythm.

In a first aspect, an embodiment of the present invention provides an electrocardiogram signal processing method. Specifically, as shown in FIG. 1 , the method includes the following steps S101 to S103 .

Step S101, obtaining an electrocardiogram signal to be detected.

In a specific implementation process, the ECG signal includes at least ECG data of one lead channel. Specifically, the ECG signal of each lead channel can be collected separately to obtain original ECG signal data.

As an optional embodiment, after acquiring the original ECG signal data, it also includes: preprocessing the original ECG signal to obtain a preprocessed ECG signal. Specifically, the preprocessing includes noise reduction processing and/or data regularization processing. The embodiment of the present application takes the processing of ECG data of a single lead channel as an example. The preprocessing process includes: using Gaussian filtering to reduce the noise of the original ECG signal, and performing data regularization processing to reduce the complexity of the model by dynamically adjusting the values of the model parameters to obtain ECG information. Of course, the preprocessing process can also include differentiation, moving window integration, etc.

Step S102 , dividing the ECG signal to obtain a plurality of ECG signal segments of the same length, wherein each ECG signal segment contains at least one heartbeat pulse.

In a specific embodiment, the electrocardiogram signal to be detected is divided into a plurality of continuous electrocardiogram signal segments of a preset length, including: dividing the segments by using a sliding window method.

As an optional embodiment, the ECG signal is divided based on a sliding window of a preset length to obtain multiple ECG signal segments of the same length. For example, the preset length may be 2-10 seconds, and preferably, the preset length provided in the embodiment of the present application is 3 seconds.

Of course, as another optional embodiment, the ECG signal may be cut into multiple segments of the same length by machine cutting.

Step S103, input each ECG signal segment into a pre-trained detection model, determine the position label of each heart beat pulse in each ECG signal segment, and the category label of each heart beat pulse, wherein the category label reflects the abnormal heart rhythm condition. For example, the category label here may include: atrial fibrillation, tachycardia, ectopic beats, premature beats, etc.

In a specific embodiment, the detection model is trained according to the following steps: obtaining an initial ECG signal for training; dividing the initial ECG signal to obtain a plurality of ECG signal segment samples, wherein the ECG signal segment sample contains at least one heart beat pulse, and the length of the ECG signal segment sample is equal to the length of the ECG signal segment; labeling each ECG signal segment sample to obtain a label set for each ECG signal segment sample, wherein the label set includes a position label and a category label for each heart beat in each ECG signal segment sample; training a pre-constructed machine learning model based on the ECG signal segment samples and the label set to obtain a detection model.

Specifically, after obtaining the initial ECG signal for training, the method further includes: preprocessing the initial ECG information to obtain the preprocessed initial ECG signal. Then, based on the same division method as the ECG signal to be detected, a plurality of ECG signal segment samples are obtained. Furthermore, each ECG signal segment sample is labeled to obtain a label set of each ECG signal segment sample.

In a specific embodiment, the labeling process for each ECG signal segment sample includes: obtaining the position of the heartbeat characteristic point of each heartbeat pulse in each ECG signal segment; and determining the position label of each heartbeat based on the distance between the position of the heartbeat characteristic point and the starting position of the ECG signal segment where the heartbeat is located.

Specifically, the position of each heartbeat in the ECG signal can be identified through the QRS complex recognition algorithm, and the distance between at least one characteristic point describing the heartbeat position and the starting position of the ECG segment where the heartbeat is located is normalized to between 0 and 1, and the normalized distance data is used as the position label of the heartbeat.

For example, as shown in FIG2 , the embodiment of the present application selects each heartbeat pulse and the dividing point of the two adjacent heartbeat pulses as two feature points, that is, L10 in the figure is an adjacent dividing point of the heartbeat pulse, and L11 is another adjacent dividing point of the heartbeat pulse. Specifically, the dividing point S _i between adjacent heartbeats is defined as: S _i = α×P _i-1 +(1-α)×P _i , wherein P _i-1 and P _i are the peak positions of the R waves in the two adjacent heartbeat QRS complexes, respectively, which can be obtained by the QRS complex recognition algorithm or manual annotation; α is a preset hyperparameter, ranging from 0 to 1, and is set to 0.4 in this embodiment. The distance between the feature point and the starting position of the ECG segment where the heartbeat is located is used as the position label of the heartbeat.

In a specific embodiment, the labeling process for each ECG signal segment sample further includes: for each heartbeat pulse in each ECG signal segment sample, determining a category from a plurality of preset heart rhythm abnormality categories as a category label for the heartbeat pulse.

Specifically, the category feature points of each heartbeat pulse in each ECG signal segment sample are obtained, and each heartbeat pulse is labeled based on the category feature points to determine the category of the heartbeat pulse. Of course, at least one label can also be manually selected from a plurality of preset heart rhythm abnormality categories as the category label of the heartbeat.

Furthermore, the training of a pre-constructed machine learning model based on the ECG signal segment samples and the label set may include: inputting the ECG signal segment samples into the machine learning model to obtain a prediction set, the prediction set including the predicted position information and predicted category information of each heart beat in each input ECG signal segment sample; based on the prediction set and the label set output by the machine learning model, obtaining the best match between the prediction set and the label set; according to the best matching result, obtaining the total loss of the prediction set and the label set, the total loss including the category loss and the position loss; and training the machine learning model based on the total loss to obtain a detection model.

Specifically, the prediction set and label set outputted by the machine learning model are used to obtain the best match between the prediction set and the label set, including:

Match each element in the label set with each element in the prediction set in turn to obtain a match with the smallest error, wherein the matching process includes: matching the relative position information of the label set with the relative position information of the prediction set, and matching the category information of the label set with the category information of the prediction set; determining the match with the smallest error as the best match.

Specifically, the comprehensive error caused by matching every two elements in the prediction set and the label set is calculated. The comprehensive error is obtained by linearly combining the position error and the category error:

代表i元素和j元素匹配带来的位置误差，本实施例中选择L1损失作为位置误差，

代表i元素和j元素匹配带来的类别误差，其中i元素来自预测集合，j元素来自标签集合。in,

represents the position error caused by the matching of the i element and the j element. In this embodiment, the L1 loss is selected as the position error.

Represents the category error caused by the matching of element i and element j, where element i comes from the prediction set and element j comes from the label set.

Then, the best match that minimizes the matching error is calculated based on the comprehensive error. For example, this embodiment uses the Hungarian algorithm to calculate the best match.

It should be noted that when the machine learning model outputs a preset number of heartbeat position information and category prediction information sets, this preset number is often not less than the maximum number of heartbeats that may appear in the ECG signal segment sample, so that each element in the ECG signal segment sample label set corresponds to an element in the prediction set. For example, in this embodiment, the preset number is 8, and the maximum number of heartbeats contained in an ECG signal segment in the label set is 8.

In addition, during the matching process, the output categories include an additional no-target category in addition to the preset multiple abnormal heart rhythm categories, that is, no heartbeat waveform appears at the detected location. Then, the best match between the prediction set and the label set is calculated.

Furthermore, based on the best matching result, the total loss of the prediction set and the label set is obtained, including: the total loss is formed by linear combination:

L _total ＝λ _l1 L _l1 +λ _giou L _giou +λ _class L _class

In this embodiment of the present application, KL divergence is used as the class loss L _class , L _l1 is the L1 regularization loss, and L _giou is a customized one-dimensional extended intersection-over-union loss.

Preferably, in this embodiment, L _l1 and L _giou are used together as position loss, and are defined as:

L _l1 (D _p ,D _gt )=||b _p -b _gt || ₁

Wherein, D _gt is an element in the label set, D _p is an element in the prediction set that matches the element in the label set, b _gt and b _p respectively represent the intervals covered by the two feature points defined in this embodiment, A function is the minimum interval length required to cover the two intervals, and δ is a very small constant.

After obtaining the total loss of the prediction set and the label set, the total loss is used to train the machine learning model based on the back propagation algorithm and the stochastic gradient descent algorithm to obtain the detection model.

Therefore, the present application inputs the ECG signal segments obtained based on step S102 into the above detection model respectively, and determines the position label of each heart pulse in each ECG signal segment, and the category label of each heart pulse.

Specifically, the category label of each heart pulse in each ECG signal segment is obtained, including: when a heart pulse contains multiple category labels, the probabilities of multiple category labels in the heart pulse are compared, and the category label that is greater than a preset probability threshold is used as the category label of the heart pulse, thereby determining the category of the heart pulse.

It should be noted that, as shown in Figure 3, the machine learning model 100 described in the embodiment of the present application may include: a one-dimensional convolution module 102, a self-attention encoding module 104, a self-attention decoding module 105, and a fully connected output module 106.

Among them, the ECG signal segment is input into a pre-trained detection model to determine the relative position information of each heart pulse in the ECG signal segment and the category information of each heart pulse, including: extracting features of the input ECG signal segment through a one-dimensional convolution module 102; extracting and fusing features of the first ECG signal segment through a self-attention encoding module 104 and a self-attention decoding module 105; and the fully connected output module 106 processes the features output by the self-attention decoding module 105, and outputs the relative position information of each heart pulse in the ECG signal segment and the category information of each heart pulse.

Specifically, the one-dimensional convolution module 102 is a network structure formed by stacking block structures. Preferably, each block includes a point convolution layer for channel expansion and an activation function for correction, a channel convolution layer and an activation function for correction, a point convolution layer for channel reduction, and a channel attention module, and each block is connected using a residual operation.

Among them, the self-attention encoding module 104 and the self-attention decoding module 105 are used to further extract and fuse the output of the one-dimensional convolution layer 102, and use the self-attention mechanism to extract the interdependent information of different regions; the encoding and decoding layers are network structures stacked by block structures. In this embodiment, each block contains a number of self-attention modules based on the attention mechanism, which are used for the modified layer regularization and multi-layer perceptron modules, and are used between regularization layers.

Specifically, as shown in FIG4 , the self-attention calculation module selected in this embodiment specifically includes: a fully connected layer for mapping the input feature map into three sets of inputs V, K and Q, wherein K and Q are sequentially subjected to matrix multiplication, scaling and softmax layers, and the obtained output is then matrix multiplied with V to obtain the output, which can be expressed as follows using a mathematical formula:

Where d _k represents the number of columns of the K matrix.

Furthermore, the machine learning model described in the present application also includes a position encoding module 103. Since the encoding module 104 and the decoding module 105 based on the self-attention mechanism cannot know the influence of the sequence order in the input, the order of the sequence can help the model to detect ECG signals with higher accuracy and achieve higher accuracy ECG information processing. Because according to statistics, there is a higher probability of abnormal heartbeats occurring within a period of time after abnormal heartbeats occur. Therefore, in this embodiment, a commonly used sine-cosine position encoding module 103 is provided to provide position encoding information. The position encoding is the same size as the feature map output by the one-dimensional convolution module, and is provided to the encoding module as input after addition. The position encoding is defined as:

Among them, d _model is a hyperparameter of the machine learning model, which is used to control the shape and size of the intermediate calculation results.

Among them, the fully connected output module 106 is used to process the features output by the decoding layer 105 based on the self-attention mechanism, and output the relative position information and category probability prediction of a fixed number of heartbeats, and the weights of different positions are shared. In addition to the preset N classifications, the prediction of the no-target class is also included, indicating that no heartbeat waveform is detected in the target detection area.

Specifically, in this embodiment, two common fully connected layers are used to process all outputs of the decoding module based on the self-attention mechanism, and the probability of the predicted category and the regression value of the position feature point of the heartbeat are output respectively.

As shown in FIG5 , it is the detection result based on the electrocardiogram signal processing method. The horizontal axis represents time, the vertical axis represents heart rhythm intensity, and N, S, V and F marked in the figure are the categories of heartbeats. This embodiment classifies heartbeats with reference to the standards proposed by the American Association for the Advancement of Medical Instrumentation (AAMI). It can be seen that the position and category of each heartbeat are detected at the same time.

The ECG signal processing method provided in the present application can simultaneously predict the positions and categories of all heartbeats in a given ECG signal in an end-to-end manner without the support of an additional heartbeat localization algorithm, which can reduce the amount of calculation and the complexity of application deployment. In addition, the model is trained through a stochastic gradient descent algorithm, and the positioning loss and classification loss are used to optimize the machine learning model, which can reduce the classification performance loss caused by inaccurate positioning, and can automatically extract features suitable for the task according to different task requirements.

In addition, the ECG signal processing method provided in the present application can process ECG signals containing multiple heartbeats. A forward reasoning operation can simultaneously calculate the positions and categories of all heartbeats in the entire ECG signal, and further improve the accuracy of heartbeat classification by utilizing the potential dependencies between these heartbeats through the self-attention mechanism.

In summary, the ECG signal processing method provided by the embodiment of the present invention can simultaneously locate and classify heartbeats, and further improve the classification performance by utilizing the potential dependency between heartbeats, which is beneficial to improving the accuracy of ECG signal processing.

In the second aspect, based on the same inventive concept, this embodiment provides an electrocardiogram signal processing device, as shown in FIG6 , comprising:

An acquisition module 401 is used to acquire an electrocardiogram signal to be detected;

A division module 402 is used to divide the ECG signal to obtain a plurality of ECG signal segments of the same length, wherein each ECG signal segment contains at least one heart beat pulse;

The determination module 403 is used to input each ECG signal segment into a pre-trained detection model to determine the relative position information of each heart beat pulse in each ECG signal segment and the category information of each heart beat pulse, wherein the category information reflects the abnormal heart rhythm condition.

As an optional embodiment, the determining module 403 specifically includes:

An acquisition submodule, used for acquiring an initial ECG signal for training;

A division submodule, used for dividing the initial ECG signal to obtain a plurality of ECG signal segment samples, wherein the ECG signal segment samples contain at least one heart beat pulse, and the length of the ECG signal segment samples is equal to the length of the ECG signal segment;

A labeling processing submodule is used to label each ECG signal segment sample to obtain a label set for each ECG signal segment sample, wherein the label set includes a position label and a category label for each heart beat in each ECG signal segment sample;

The training submodule is used to train a pre-built machine learning model based on the ECG signal segment samples and the label set to obtain a detection model.

As an optional embodiment, the detection model in the determination module 403 includes: a one-dimensional convolution module, a self-attention encoding module, a self-attention decoding module and a fully connected output module. The determination module 403 is used to: extract features of the input ECG signal segment through the one-dimensional convolution module; extract and fuse features of the output result of the one-dimensional convolution module through the self-attention encoding module and the self-attention decoding module; process the features output by the self-attention decoding module through the fully connected output module, and output the relative position information of each heart pulse in the ECG signal segment and the category information of each heart pulse.

As an optional embodiment, the labeling processing submodule is used to: obtain the position of the heartbeat characteristic point of each heartbeat pulse in each ECG signal segment; determine the position label of each heartbeat based on the distance between the position of the heartbeat characteristic point and the starting position of the ECG signal segment where the heartbeat is located; and for each heartbeat pulse in each ECG signal segment sample, determine a category from a plurality of preset heart rhythm abnormality categories as the category label of the heartbeat pulse.

As an optional embodiment, the training submodule is used to: input ECG signal segment samples into a machine learning model to obtain a prediction set, the prediction set including predicted position information and predicted category information of each heart beat in each input ECG signal segment sample; based on the prediction set and label set output by the machine learning model, obtain the best match between the prediction set and the label set; based on the best matching result, obtain the total loss of the prediction set and the label set, the total loss includes category loss and position loss; train the machine learning model based on the total loss to obtain a detection model.

As an optional embodiment, the training submodule is used to: match each element in the label set with each element in the prediction set in turn to obtain a match with the smallest error, wherein the matching process includes: matching the relative position information of the label set with the relative position information of the prediction set, and matching the category information of the label set with the category information of the prediction set; and determining the match with the smallest error as the best match.

As an optional embodiment, the division module 402 is used to: control the sliding window to divide the ECG signal so that the sliding window obtains an ECG signal segment after each preset time, and obtains multiple ECG signal segments of the same length.

The above modules can be implemented by software codes, in which case, the above modules can be stored in the memory of the control device. The above modules can also be implemented by hardware such as integrated circuit chips.

An electrocardiogram signal processing device provided in an embodiment of the present invention has the same implementation principle and technical effects as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference may be made to the corresponding contents in the aforementioned method embodiment.

In the third aspect, based on the same inventive concept, this embodiment provides an electronic device 500, as shown in Figure 7, comprising: a memory 501, a processor 502, and a computer program 503 stored in the memory and executable on the processor, wherein when the processor 502 executes the program, the steps of the electrocardiogram signal processing method described in the first aspect are implemented.

Furthermore, as shown in FIG8 , the electronic device also includes a dedicated acceleration circuit 504, which is connected to the processor 502 and is used to process computationally intensive tasks in ECG signal processing, including but not limited to convolution operations, matrix multiplication operations, vector operations, exponential operations, quantization operations, etc., and to perform hardware acceleration on these computational tasks.

The dedicated acceleration circuit can reduce system power consumption and cost, so that the method can be deployed on wearable devices such as smart watches and smart bracelets to monitor a person's heart health status in real time.

In the fourth aspect, based on the same inventive concept, this embodiment provides a non-temporary computer-readable storage medium. When the instructions in the storage medium are executed by the processor of the electronic device 500, the electronic device 500 can perform an electrocardiogram signal processing method, including the steps of any one of the methods described in the first aspect.

Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed in this disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the following claims.

Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

The present invention is described with reference to the flowchart and/or block diagram of the method, device (system) and computer program product according to the embodiment of the present invention. It should be understood that each process and/or box in the flowchart and/or block diagram, as well as the combination of the process and/or box in the flowchart and/or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate a module for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction module that implements the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims (10)

1. An electrocardiographic signal processing method, comprising:

acquiring an electrocardiosignal to be detected;

dividing the electrocardiosignals to obtain a plurality of electrocardiosignal segments with the same length, wherein each electrocardiosignal segment at least comprises a heart beat pulse;

and respectively inputting each electrocardiosignal segment into a pre-trained detection model, and determining the relative position information of each heart beat pulse in each electrocardiosignal segment and the category information of each heart beat pulse, wherein the category information reflects the abnormal condition of the heart rhythm.

2. The method of claim 1, wherein the detection model is trained according to the following steps:

acquiring an initial electrocardiosignal for training;

dividing the initial electrocardiosignal to obtain a plurality of electrocardiosignal fragment samples, wherein the electrocardiosignal fragment samples at least comprise a heartbeat pulse, and the length of the electrocardiosignal fragment samples is equal to that of the electrocardiosignal fragments;

marking each electrocardiosignal fragment sample to obtain a label set of each electrocardiosignal fragment sample, wherein the label set comprises a position label and a category label of each heart beat in each electrocardiosignal fragment sample;

and training a pre-constructed machine learning model based on the electrocardiosignal fragment sample and the label set to obtain the detection model.

3. The method of claim 1, wherein the detection model comprises: the method comprises the following steps of inputting the electrocardiosignal segment into a pre-trained detection model, and determining the relative position information of each heart beat pulse in the electrocardiosignal segment and the category information of each heart beat pulse, wherein the method comprises the following steps:

performing feature extraction on the input electrocardiosignal segment through the one-dimensional convolution module;

performing feature extraction and fusion on the output result of the one-dimensional convolution module through the self-attention coding module and the self-attention decoding module;

and processing the characteristics output by the self-attention decoding module through the full-connection output module, and outputting the relative position information of each heart pulse in the electrocardiosignal segment and the category information of each heart pulse.

4. The method according to claim 2, wherein said marking of each sample of segments of the cardiac signal comprises:

acquiring the position of a heart beat characteristic point of each heart beat pulse in each electrocardiosignal segment;

determining a position label of each heartbeat based on the distance between the position of the feature point of the heartbeat and the initial position of the electrocardiosignal segment where the heartbeat is located;

for each heart beat pulse in each electrocardiosignal segment sample, determining a class from a plurality of preset abnormal heart rhythm classes as a class label of the heart beat pulse.

5. The method of claim 2, wherein training a pre-constructed machine learning model based on the samples of the cardiac electrical signal segments and the set of labels comprises:

inputting the electrocardiosignal segment samples into the machine learning model to obtain a prediction set, wherein the prediction set comprises the prediction position information and the prediction category information of each heart beat in the input electrocardiosignal segment samples;

based on a prediction set output by the machine learning model and the label set, obtaining the best matching of the prediction set and the label set;

obtaining the total loss of the prediction set and the label set according to the best matching result, wherein the total loss comprises category loss and position loss;

and training the machine learning model based on the total loss to obtain the detection model.

6. The method of claim 5, wherein deriving the best match of the prediction set to the set of labels based on the prediction set output by the machine learning model and the set of labels comprises:

and sequentially matching each element in the label set with each element in the prediction set to obtain a match with the minimum error, wherein the matching process comprises the following steps: matching the relative position information of the labelsets with the relative position information of the prediction sets, and matching the category information of the labelsets with the category information of the prediction sets;

and determining the match with the minimum error as the best match.

7. The method of claim 1, wherein said dividing said cardiac signal into a plurality of cardiac signal segments of equal length comprises:

and dividing the electrocardiosignals based on a sliding window with a preset length to obtain a plurality of electrocardiosignal segments with the same length.

8. An electrocardiographic signal processing apparatus, comprising:

the acquisition module is used for acquiring the electrocardiosignals to be detected;

the dividing module is used for dividing the electrocardiosignals to obtain a plurality of electrocardiosignal segments with the same length, wherein each electrocardiosignal segment at least comprises a heartbeat pulse;

and the determining module is used for respectively inputting each electrocardiosignal segment into a pre-trained detection model, and determining the relative position information of each heart pulse in each electrocardiosignal segment and the category information of each heart pulse, wherein the category information reflects the abnormal condition of the heart rhythm.

9. An electronic device, comprising: memory, processor and computer program stored on the memory and executable on the processor, the processor implementing the steps of the method of any one of claims 1 to 7 when executing the program.

10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.

Images