阅读说明：本技术 心电数据的处理方法、装置、存储介质和电子设备 (Electrocardio data processing method and device, storage medium and electronic equipment ) 是由 朱宝峰 何光宇 程万军 于 2020-12-22 设计创作，主要内容包括：本公开涉及一种心电数据的处理方法、装置、存储介质和电子设备,应用于电子信息处理技术领域,该方法包括：获取待处理的目标心电数据,根据目标心电数据,通过预先训练的自编码器确定能够表征目标心电数据的目标心电特征向量,根据目标心电特征向量和预先训练的自组织映射网络,通过预设的分类算法确定目标心电数据对应的目标体征类型,自组织映射网络为根据预设的训练数据集训练得到的。本公开可以通过自编码器、自组织映射网络,结合分类算法,来智能地确定目标心电数据对应的目标体征类型,不需要人工参与,能够提高心电数据的处理效率和准确度。(The disclosure relates to a method and a device for processing electrocardiographic data, a storage medium and an electronic device, which are applied to the technical field of electronic information processing, wherein the method comprises the following steps: the method comprises the steps of obtaining target electrocardiogram data to be processed, determining a target electrocardiogram characteristic vector capable of representing the target electrocardiogram data through a pre-trained self-encoder according to the target electrocardiogram data, determining a target sign type corresponding to the target electrocardiogram data through a preset classification algorithm according to the target electrocardiogram characteristic vector and a pre-trained self-organized mapping network, wherein the self-organized mapping network is obtained through training according to a preset training data set. The method and the device can intelligently determine the target sign type corresponding to the target electrocardiogram data through the self-encoder and the self-organizing mapping network in combination with a classification algorithm, do not need manual participation, and can improve the processing efficiency and accuracy of the electrocardiogram data.)

1. A method for processing electrocardiographic data, the method comprising:

acquiring target electrocardiogram data to be processed;

according to the target electrocardiogram data, determining a target electrocardiogram feature vector capable of representing the target electrocardiogram data through a pre-trained self-encoder;

and determining a target sign type corresponding to the target electrocardiogram data through a preset classification algorithm according to the target electrocardiogram feature vector and a pre-trained self-organized mapping network, wherein the self-organized mapping network is obtained by training according to a preset training data set.

2. The method according to claim 1, wherein said determining, from said target electrocardiographic data, a target electrocardiographic feature vector capable of characterizing said target electrocardiographic data by a pre-trained self-encoder comprises:

filtering the target electrocardio data, and performing peak detection on the filtered target electrocardio data to extract a second number of heartbeat data;

performing dimensionality reduction processing on a third quantity of heartbeat data in the second quantity of heartbeat data to obtain a heartbeat data sequence, wherein the third quantity is smaller than or equal to the second quantity;

and inputting the heartbeat data sequence into the self-encoder to obtain the target electrocardio characteristic vector.

3. The method according to claim 2, wherein said inputting the heartbeat data sequence into the self-encoder to obtain the target cardiac electrical feature vector comprises:

dividing the heartbeat data sequence into a plurality of heartbeat data segments according to a specified time length;

inputting each heartbeat data segment into the self-encoder to obtain a sub-feature vector capable of representing the heartbeat data segment;

and splicing the sub-feature vectors corresponding to each heartbeat data segment to obtain the target electrocardio feature vector.

4. The method according to claim 1, wherein the determining, according to the target electrocardiographic feature vector and a pre-trained self-organizing mapping network, a target sign type corresponding to the target electrocardiographic data by a preset classification algorithm includes:

determining the distance from the target electrocardio characteristic vector to each of a first number of regions corresponding to the self-organizing mapping network, wherein the first number is determined when the self-organizing mapping network is trained according to the training data set;

and taking the sign type corresponding to the target region as the target sign type, wherein the target region is the region with the minimum distance with the target electrocardio characteristic vector.

5. The method of claim 4, wherein the self-organizing map network is trained by:

acquiring the training data set, wherein the training data set comprises a plurality of training electrocardiogram data;

according to any training electrocardiogram data, determining the optimal matching node matched with the training electrocardiogram data in each node included in the network to be trained;

determining neighborhood nodes in the topological neighborhood of the best matching node;

updating the weights of the neighborhood nodes and the optimal matching nodes according to the distances between the neighborhood nodes and the optimal matching nodes;

and repeatedly executing the step of determining the best matching node matched with the training electrocardiogram data in each node included in the network to be trained according to any training electrocardiogram data, and updating the weights of the neighborhood nodes and the best matching node until the network to be trained meets a preset training end condition to obtain the self-organizing mapping network.

6. An apparatus for processing electrocardiographic data, the apparatus comprising:

the acquisition module is used for acquiring target electrocardiogram data to be processed;

the processing module is used for determining a target electrocardio characteristic vector capable of representing the target electrocardio data through a pre-trained self-encoder according to the target electrocardio data;

and the determining module is used for determining the target sign type corresponding to the target electrocardiogram data through a preset classification algorithm according to the target electrocardiogram feature vector and a pre-trained self-organized mapping network, wherein the self-organized mapping network is obtained by training according to a preset training data set.

7. The apparatus of claim 6, wherein the processing module comprises:

the extraction submodule is used for filtering the target electrocardio data and carrying out peak detection on the filtered target electrocardio data so as to extract a second number of heartbeat data;

the dimension reduction sub-module is used for performing dimension reduction processing on a third quantity of heartbeat data in the second quantity of heartbeat data to obtain a heartbeat data sequence, wherein the third quantity is smaller than or equal to the second quantity;

and the coding sub-module is used for inputting the heartbeat data sequence into the self-coder to obtain the target electrocardio characteristic vector.

8. The apparatus of claim 7, wherein the encoding submodule is configured to:

dividing the heartbeat data sequence into a plurality of heartbeat data segments according to a specified time length;

inputting each heartbeat data segment into the self-encoder to obtain a sub-feature vector capable of representing the heartbeat data segment;

and splicing the sub-feature vectors corresponding to each heartbeat data segment to obtain the target electrocardio feature vector.

9. 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 5.

10. An electronic device, comprising:

a memory having a computer program stored thereon;

a processor for executing the computer program in the memory to carry out the steps of the method of any one of claims 1 to 5.

Technical Field

The present disclosure relates to the field of electronic information processing technologies, and in particular, to a method and an apparatus for processing electrocardiographic data, a storage medium, and an electronic device.

Background

Heart-related diseases are one of the major life-threatening diseases for human beings, and the research on heart-related diseases has long been an important topic in the medical field. An electrocardiogram (abbreviated as ECG) is a simple continuous record of electrical activity generated by the heart, and can reflect the physical sign states of various parts of the heart to a certain extent, thereby effectively assisting doctors in judging the heart states of users. However, to judge the heart state of the user through the electrocardiographic data, the requirements on the capability and experience of the doctor are high, and erroneous judgment or missed judgment is easily caused.

Disclosure of Invention

In order to solve the problems in the related art, the present disclosure provides a method and an apparatus for processing electrocardiographic data, a storage medium, and an electronic device.

In order to achieve the above object, according to a first aspect of the embodiments of the present disclosure, there is provided a method for processing electrocardiographic data, the method including:

acquiring target electrocardiogram data to be processed;

according to the target electrocardiogram data, determining a target electrocardiogram feature vector capable of representing the target electrocardiogram data through a pre-trained self-encoder;

and determining a target sign type corresponding to the target electrocardiogram data through a preset classification algorithm according to the target electrocardiogram feature vector and a pre-trained self-organized mapping network, wherein the self-organized mapping network is obtained by training according to a preset training data set.

Optionally, the determining, according to the target electrocardiographic data, a target electrocardiographic feature vector capable of characterizing the target electrocardiographic data by a pre-trained self-encoder includes:

filtering the target electrocardio data, and performing peak detection on the filtered target electrocardio data to extract a second number of heartbeat data;

performing dimensionality reduction processing on a third quantity of heartbeat data in the second quantity of heartbeat data to obtain a heartbeat data sequence, wherein the third quantity is smaller than or equal to the second quantity;

and inputting the heartbeat data sequence into the self-encoder to obtain the target electrocardio characteristic vector.

Optionally, the inputting the heartbeat data sequence into the self-encoder to obtain the target electrocardiogram feature vector includes:

dividing the heartbeat data sequence into a plurality of heartbeat data segments according to a specified time length;

inputting each heartbeat data segment into the self-encoder to obtain a sub-feature vector capable of representing the heartbeat data segment;

and splicing the sub-feature vectors corresponding to each heartbeat data segment to obtain the target electrocardio feature vector.

Optionally, the determining, according to the target electrocardiogram feature vector and a pre-trained self-organizing map network, a target sign type corresponding to the target electrocardiogram data through a preset classification algorithm includes:

determining the distance from the target electrocardio characteristic vector to each of a first number of regions corresponding to the self-organizing mapping network, wherein the first number is determined when the self-organizing mapping network is trained according to the training data set;

and taking the sign type corresponding to the target region as the target sign type, wherein the target region is the region with the minimum distance with the target electrocardio characteristic vector.

Optionally, the self-organizing map network is trained by:

acquiring the training data set, wherein the training data set comprises a plurality of training electrocardiogram data;

according to any training electrocardiogram data, determining the optimal matching node matched with the training electrocardiogram data in each node included in the network to be trained;

determining neighborhood nodes in the topological neighborhood of the best matching node;

updating the weights of the neighborhood nodes and the optimal matching nodes according to the distances between the neighborhood nodes and the optimal matching nodes;

and repeatedly executing the step of determining the best matching node matched with the training electrocardiogram data in each node included in the network to be trained according to any training electrocardiogram data, and updating the weights of the neighborhood nodes and the best matching node until the network to be trained meets a preset training end condition to obtain the self-organizing mapping network.

According to a second aspect of the embodiments of the present disclosure, there is provided an apparatus for processing electrocardiographic data, the apparatus including:

the acquisition module is used for acquiring target electrocardiogram data to be processed;

the processing module is used for determining a target electrocardio characteristic vector capable of representing the target electrocardio data through a pre-trained self-encoder according to the target electrocardio data;

and the determining module is used for determining the target sign type corresponding to the target electrocardiogram data through a preset classification algorithm according to the target electrocardiogram feature vector and a pre-trained self-organized mapping network, wherein the self-organized mapping network is obtained by training according to a preset training data set.

Optionally, the processing module includes:

the extraction submodule is used for filtering the target electrocardio data and carrying out peak detection on the filtered target electrocardio data so as to extract a second number of heartbeat data;

the dimension reduction sub-module is used for performing dimension reduction processing on a third quantity of heartbeat data in the second quantity of heartbeat data to obtain a heartbeat data sequence, wherein the third quantity is smaller than or equal to the second quantity;

and the coding sub-module is used for inputting the heartbeat data sequence into the self-coder to obtain the target electrocardio characteristic vector.

Optionally, the encoding submodule is configured to:

dividing the heartbeat data sequence into a plurality of heartbeat data segments according to a specified time length;

inputting each heartbeat data segment into the self-encoder to obtain a sub-feature vector capable of representing the heartbeat data segment;

and splicing the sub-feature vectors corresponding to each heartbeat data segment to obtain the target electrocardio feature vector.

Optionally, the determining module includes:

a first determining submodule, configured to determine a distance from the target electrocardiogram feature vector to each of a first number of regions corresponding to the self-organizing map network, where the first number is determined when the self-organizing map network is trained according to the training data set;

and the second determining submodule is used for taking the sign type corresponding to the target area as the target sign type, wherein the target area is the area with the minimum distance with the target electrocardio characteristic vector.

Optionally, the self-organizing map network is trained by:

acquiring the training data set, wherein the training data set comprises a plurality of training electrocardiogram data;

according to any training electrocardiogram data, determining the optimal matching node matched with the training electrocardiogram data in each node included in the network to be trained;

determining neighborhood nodes in the topological neighborhood of the best matching node;

updating the weights of the neighborhood nodes and the optimal matching nodes according to the distances between the neighborhood nodes and the optimal matching nodes;

and repeatedly executing the step of determining the best matching node matched with the training electrocardiogram data in each node included in the network to be trained according to any training electrocardiogram data, and updating the weights of the neighborhood nodes and the best matching node until the network to be trained meets a preset training end condition to obtain the self-organizing mapping network.

According to the technical scheme, the target electrocardiogram data to be processed is obtained, the pre-trained self-organizing mapping network is used for determining the target electrocardiogram characteristic vector capable of representing the target electrocardiogram data according to the target electrocardiogram data and the pre-trained self-organizing encoder, and the target sign type corresponding to the target electrocardiogram data is determined through the preset classification algorithm according to the target electrocardiogram characteristic vector and the pre-trained self-organizing mapping network, wherein the self-organizing mapping network is obtained through training according to the preset training data set. The method and the device can intelligently determine the target sign type corresponding to the target electrocardiogram data through the self-encoder and the self-organizing mapping network in combination with a classification algorithm, do not need manual participation, and can improve the processing efficiency and accuracy of the electrocardiogram data.

Additional features and advantages of the disclosure will be set forth in the detailed description which follows.

Drawings

The accompanying drawings, which are included to provide a further understanding of the disclosure and are incorporated in and constitute a part of this specification, illustrate embodiments of the disclosure and together with the description serve to explain the disclosure without limiting the disclosure. In the drawings:

FIG. 1 is a flow chart illustrating a method of processing electrocardiographic data according to an exemplary embodiment;

FIG. 2 is a flow chart of one step 102 shown in the embodiment of FIG. 1;

FIG. 3 is a flow diagram illustrating a manner of training an ad hoc mapping network in accordance with an exemplary embodiment;

FIG. 4 is a block diagram illustrating an apparatus for processing electrocardiographic data according to an exemplary embodiment;

FIG. 5 is a block diagram of one type of processing module shown in the embodiment of FIG. 4;

FIG. 6 is a block diagram of a determination module shown in the embodiment of FIG. 4;

FIG. 7 is a block diagram illustrating an electronic device in accordance with an example embodiment.

Detailed Description

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the exemplary embodiments below are not intended to represent all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

Fig. 1 is a flow chart illustrating a method of processing electrocardiographic data according to an exemplary embodiment. As shown in fig. 1, the method may include the steps of:

step 101, obtaining target electrocardiogram data to be processed.

For example, firstly, the ECG device may acquire data from a designated position on the body surface of the target user by using a designated number of leads according to a designated sampling frequency and a designated sampling duration, so as to acquire target electrocardiographic data to be processed. The target electrocardiographic data may be understood as data of a change with time of an electric potential included in an electrocardiogram generated by an ECG device. For example, twelve leads (including I, II, III, aVR, aVL, aVF, V1, V2, V3, V4, V5 and V6) can be adopted by an ECG device according to a sampling frequency of 500Hz and a sampling time length of 1min to acquire the target electrocardio data. For another example, the ECG device may acquire the target electrocardiographic data by acquiring data of the designated location on the body surface of the target user through a single lead according to a sampling frequency of 300Hz and a sampling duration of 1 min. After the target electrocardiographic data is obtained, the target electrocardiographic data may be stored, for example, the target electrocardiographic data may be stored in a Matlab (english: Matrix Laboratory, chinese: Matrix Laboratory) file.

And step 102, determining a target electrocardiogram feature vector capable of representing the target electrocardiogram data through a pre-trained self-encoder according to the target electrocardiogram data.

For example, in order to reduce the requirements on the ability and experience of a doctor when the heart state of the target user is determined through the electrocardiographic data, the physical sign type corresponding to the electrocardiographic data can be automatically determined by classifying the electrocardiographic data. Specifically, the target electrocardiographic data can be processed to extract main features contained in the target electrocardiographic data, and a target electrocardiographic feature vector capable of representing the target electrocardiographic data is generated. For example, the target electrocardiographic data may be filtered to remove the interference in the target electrocardiographic data. And then extracting heartbeat data corresponding to each heartbeat from the filtered target electrocardio data, and coding the extracted heartbeat data through a pre-trained self-coder (English: automatic coder, abbreviation: AE) to obtain a target electrocardio characteristic vector.

And 103, determining a target sign type corresponding to the target electrocardiogram data through a preset classification algorithm according to the target electrocardiogram feature vector and a pre-trained self-organizing mapping network.

The self-organizing mapping network is obtained by training according to a preset training data set.

Furthermore, because a large amount of labeling of the electrocardiographic data requires high labor cost, a large amount of available data may be lacking when the electrocardiographic data is classified by adopting a supervised learning manner. Therefore, the electrocardiogram data can be classified in an unsupervised learning manner. For example, the self-organizing mapping network may be trained in advance from a training dataset containing electrocardiographic data of different sign types. The trained self-organizing mapping network corresponds to a plurality of regions, each region is formed by mapping the electrocardiogram data with the same sign type, namely each region corresponds to one sign type.

The training data set may be, for example, an ICBEB data set or a PhysioNet data set. The sign types can include a normal sign type and a plurality of abnormal sign types, the normal sign type is used for representing that the heart state corresponding to the electrocardiographic data is not abnormal, and the abnormal sign type is used for representing the disease type when the heart state corresponding to the electrocardiographic data is abnormal. For example, in the case that 8 abnormal sign types are included, the 8 abnormal sign types may be: atrial fibrillation (abbreviated as AF), primary atrioventricular conduction block (abbreviated as LBBB), Right bundle branch conduction block (abbreviated as RBBB), anterior Atrial contraction (abbreviated as PAC), anterior ventricular contraction (abbreviated as PVC), ST-segment depression (abbreviated as ST-segment depression), and ST-segment elevation (abbreviated as ST-segment elevation).

Then, after the target electrocardiogram feature vector is determined, the target electrocardiogram feature vector is input as a preset classification algorithm, the classification algorithm determines a region corresponding to the target electrocardiogram feature vector from a plurality of regions corresponding to the trained self-organizing mapping network, and the sign type corresponding to the region is used as the target sign type corresponding to the target electrocardiogram data. The classification algorithm may be understood as a classifier, for example, when the classifier adopts a KNN (k-nearest neighbor, chinese) classifier, the KNN classifier may use a plurality of regions corresponding to the self-organizing map network as classified targets, then determine target electrocardiogram feature vectors, distances to each of a first number of regions corresponding to the self-organizing map network, and use a sign type corresponding to the target region as a target sign type. The first number is determined when the self-organizing map network is trained according to the training data set (namely the number of the areas corresponding to the self-organizing map network after training is finished), and the target area is the area with the minimum distance from the target electrocardio feature vector.

It should be noted that the present disclosure can be applied to determine not only the type of the physical sign corresponding to the electrocardiographic data, but also any other type of the physical sign corresponding to the biological data. The biological data may include bioelectrical data (electroencephalogram signal, electromyogram signal, electro-oculogram signal, etc.) and bioelectrical data (e.g., blood pressure, blood oxygen, tension, pressure, body temperature, etc.), among others.

In summary, according to the present disclosure, target electrocardiographic data to be processed is obtained, a pre-trained self-organizing mapping network is used to determine a target electrocardiographic feature vector capable of representing the target electrocardiographic data according to the target electrocardiographic data, and a target sign type corresponding to the target electrocardiographic data is determined according to a preset classification algorithm according to the target electrocardiographic feature vector and the pre-trained self-organizing mapping network, wherein the self-organizing mapping network is obtained by training according to a preset training data set. The method and the device can intelligently determine the target sign type corresponding to the target electrocardiogram data through the self-encoder and the self-organizing mapping network in combination with a classification algorithm, do not need manual participation, and can improve the processing efficiency and accuracy of the electrocardiogram data.

Fig. 2 is a flow chart illustrating one step 102 of the embodiment shown in fig. 1. As shown in fig. 2, step 102 may include the steps of:

and step 1021, filtering the target electrocardio data, and performing peak detection on the filtered target electrocardio data to extract a second number of heartbeat data.

In one scenario, high-frequency interference, myoelectric interference, power frequency interference, baseline drift and the like may exist in target electrocardiogram data, and interference for determining the type of a target sign may be affected. Therefore, the target electrocardiographic data may be filtered first, and for example, the target electrocardiographic data may be filtered by using a butterworth band-pass filter (the bandwidth of the filter may be, for example, 0.25Hz to 40 Hz). In order to enable the subsequent classification model to process data with equal length, peak detection can be performed on the filtered target electrocardiographic data to extract a second number of heartbeat data with equal time length. Specifically, a third party's biospy kit from Python can be used to extract heart beats from target electrocardiographic data: firstly, identifying all R wave peak values in target electrocardio data by a BioSppy tool kit, and then cutting in a specified time period before each R wave peak value and a specified time period after the R wave peak value to obtain a second number of continuous time periods containing complete P-QRS-T waves, wherein the data corresponding to each continuous time period can be understood as heartbeat data corresponding to one complete heartbeat. For example, a time period of 0.2 seconds before each R peak and 0.6 seconds after the R peak may be selected for slicing, so as to obtain a second number of heartbeat data, i.e. one heartbeat data corresponds to a sampling time duration of 0.8 seconds.

And step 1022, performing dimensionality reduction processing on a third quantity of heartbeat data in the second quantity of heartbeat data to obtain a heartbeat data sequence, where the third quantity is less than or equal to the second quantity.

And step 1023, inputting the heartbeat data sequence into a self-encoder to obtain the target electrocardio characteristic vector.

In this step, a third number of heartbeat data may be selected from the second number of heartbeat data to determine the target sign type. For example, 1 piece of heartbeat data may be arbitrarily selected from the second number of heartbeat data to determine the target sign type, 4 pieces of heartbeat data may also be arbitrarily selected to determine the target sign type, and 3 pieces of consecutive heartbeat data may also be selected to determine the target sign type, which is not specifically limited by the present disclosure. The target sign type is determined by selecting the third number of heartbeat data from the second number of heartbeat data, so that the data volume used for determining the target sign type can be reduced, and the speed of determining the target sign type is improved.

Then, in order to reduce the influence of redundant information and noise information contained in the third number of heartbeat data on the determination of the target sign type, the accuracy of determining the target sign type is improved. For example, each heartbeat data in the third amount of heartbeat data may be subjected to dimension reduction processing by using Principal Component Analysis (PCA), and a heartbeat data sequence is formed by each heartbeat data subjected to dimension reduction processing. And finally, inputting the heartbeat data sequence into a self-encoder, and encoding the heartbeat data sequence by the self-encoder to obtain the target electrocardio characteristic vector.

Optionally, step 1023 may be implemented by:

step A, dividing the heartbeat data sequence into a plurality of heartbeat data segments according to the specified duration.

And B, inputting each heartbeat data segment into a self-encoder to obtain a sub-feature vector capable of representing the heartbeat data segment.

And step C, splicing the sub-feature vectors corresponding to each heartbeat data segment to obtain the target electrocardio feature vector.

For example, the accuracy of determining the target sign type can be further improved by generating a more distinguishable target electrocardiogram feature vector. Specifically, the heartbeat data sequence may be first divided into a plurality of heartbeat data segments according to a specified duration. For example, in the case where the heartbeat data sequence includes 300 sampling points, the heartbeat data sequence may be divided into 6 heartbeat data segments (i.e., one heartbeat data segment with 50 sampling points) with 50 sampling periods as a specified duration. Then, each heartbeat data segment can be input into a self-encoder, each heartbeat data segment is encoded by the self-encoder to obtain a sub-feature vector of each heartbeat data segment, and the sub-feature vectors corresponding to each heartbeat data segment are spliced to obtain a target electrocardiogram feature vector. The sub-feature vectors of each heartbeat data segment can reflect the corresponding main features of the heartbeat data segment, and the target electrocardio feature vectors with higher distinguishability can be obtained by splicing the sub-feature vectors of different heartbeat data segments. Meanwhile, by dividing the heartbeat data sequence, the dimensionality of data input into the self-encoder can be reduced, the encoding efficiency of the self-encoder is improved, and the speed of determining the type of the target physical sign is further improved.

FIG. 3 is a flow diagram illustrating a manner of training an ad hoc mapping network in accordance with an exemplary embodiment. As shown in fig. 3, the self-organizing map network can be trained by:

step 201, a training data set is obtained, wherein the training data set comprises a plurality of training electrocardiograph data.

For example, when training the self-organizing map network, a training data set including a plurality of training electrocardiographic data may be obtained first, and each training electrocardiographic data corresponds to a training sign type. Wherein, the training data set can adopt an ICBEB 2018 data set or a PhysioNet Computing in Cardiology Challenge 2017 data set.

In the case where the ICBEB 2018 data set is used as the training data set, the training data set is obtained using twelve leads for 6 to 60 seconds of data acquisition at a sampling frequency of 500 Hz. The training data set comprises 6877 training electrocardiograph data, the training data set corresponds to 9 training sign types, the 9 training sign types comprise 1 Normal sign type (namely, Normal in table 1) and 8 abnormal sign types (namely, AF, I-AVB, LBBB, RBBB, PAC, PVC, STD and STE in table 1), and the recording quantity of each training sign type is shown in table 1.

TABLE 1

Training type of physical sign	Number of records
		Normal	918
Atrial fibrillation(AF)	1098
		First-degree atrioventricular block(I-AVB)	704
Left bundle branch block(LBBB)	207
		Right bundle branch block(RBBB)	1695
Premature atrial contraction(PAC)	556
		Premature ventricular contraction(PVC)	672
ST-segment depression(STD)	825
		ST-segment elevated(STE)	202

In the case where the training data set is the PhysioNet Computing in Cardiology Challenge 2017 data set, the training data set is obtained by performing data acquisition for 9 to 60 seconds using a single lead link at a sampling frequency of 300 Hz. The training data set comprises 8528 training electrocardiograph data, the training data set corresponds to 4 training sign types, the 4 training sign types comprise 1 Normal sign type (namely, Normal in table 2) and 3 abnormal sign types (namely, AF, Other rhythm and Noisy in table 2), and the record quantity of each training sign type is shown in table 2.

TABLE 2

Training type of physical sign	Number of records
		Normal	5154
Atrial fibrillation(AF)	771
		Other rhythm	2557
Noisy	46

Where Other rhythm is Other rhythms and noise is noise.

Step 202, according to any training electrocardiogram data, determining the best matching node matched with the training electrocardiogram data in each node included in the network to be trained.

Specifically, before training the SOM (Self-organizing mapping network, Chinese), the training data set needs to be processed to be in a format required by the Self-organizing mapping network. For example, each training electrocardiographic data may be extracted from a file (e.g., Matlab file) storing a training dataset in advance, and each training electrocardiographic data may be filtered using a butterworth band-pass filter. Secondly, each training electrocardiogram data after being filtered can be converted into a Numpy array, and all the training electrocardiogram data can be inserted into a CSV (Comma-Separated Values in English). And then carrying out peak detection on each training electrocardiogram data in the CSV file to extract each heartbeat data included in the training electrocardiogram data, taking all heartbeat data included in the training electrocardiogram data as a training sample, and inserting all the training samples into the CSV file as independent lines. Then, dimension reduction processing may be performed on each training sample to obtain each training sample after the dimension reduction processing.

When the SOM is trained, the weight of each node (i.e., neuron) of the network to be trained may be initialized, and then a training sample Xi is randomly selected from all the training samples subjected to the dimensionality reduction processing, and the Xi is input to the input layer of the network to be trained. Traversing each node of the output layer of the network to be trained by the network to be trained, calculating the Euclidean distance between Xi and each node, and selecting the node with the minimum distance as the Best Matching node (BMU) matched with the training electrocardiogram data corresponding to the training sample.

In step 203, neighborhood nodes in the topological neighborhood of the best matching node are determined.

And 204, updating the weights of the neighborhood nodes and the optimal matching nodes according to the distances between the neighborhood nodes and the optimal matching nodes.

And repeating the step 202 to the step 204 until the network to be trained meets a preset training end condition to obtain the self-organizing mapping network.

For example, after determining the best matching node, the neighborhood nodes included in the topological neighborhood of the best matching node may be determined according to the neighborhood radius σ of the best matching node. And then, updating the weights of the neighborhood nodes and the optimal matching nodes by using a preset weight updating formula according to the distance between the neighborhood nodes and the optimal matching nodes so as to finish one-time training of the network to be trained. Then, a training sample can be selected again, and the above steps are repeatedly executed to train the network to be trained continuously until the network to be trained meets the preset training end condition, so as to obtain the trained SOM. The training end condition may be that training of the network to be trained is determined to be completed after each training sample in all the training samples is used for training the network to be trained.

The weight update formula includes:

wherein the content of the first and second substances,

t is the number of times of training the network to be trained, w _ ij (t) is the weight vector of the node at (i, j) in the network to be trained after the network to be trained is trained for the t time, i is the row coordinate of the network to be trained, j is the column coordinate of the network to be trained, d is the distance between the neighborhood node and the best matching node, x (t) is the input vector corresponding to the training sample, σ (t) is the neighborhood radius function, α (t) is the learning rate, n is the number of the training samples, and σ (t) is the number of the training samples₀Is the initial neighborhood radius, r₀Is the initial radius.

Further, the self-encoder may be trained by: firstly, a specified number of training samples are selected, then, each training sample in the specified number of training samples is used as the input of an auto-encoder in sequence, the training sample is used as the output of the auto-encoder, and the auto-encoder is trained. And simultaneously, optimizing the trained self-encoder by using a preset loss function, and obtaining the trained self-encoder when the loss function reaches the minimum value. In addition, since a single self-encoder operates with the best performance of the architecture with a single hidden layer, the decoder part in the self-encoder can be discarded after the self-encoder is trained, and only the encoder part is reserved, so as to improve the performance of the self-encoder.

In summary, according to the present disclosure, target electrocardiographic data to be processed is obtained, a pre-trained self-organizing mapping network is used to determine a target electrocardiographic feature vector capable of representing the target electrocardiographic data according to the target electrocardiographic data, and a target sign type corresponding to the target electrocardiographic data is determined according to a preset classification algorithm according to the target electrocardiographic feature vector and the pre-trained self-organizing mapping network, wherein the self-organizing mapping network is obtained by training according to a preset training data set. The method and the device can intelligently determine the target sign type corresponding to the target electrocardiogram data through the self-encoder and the self-organizing mapping network in combination with a classification algorithm, do not need manual participation, and can improve the processing efficiency and accuracy of the electrocardiogram data.

Fig. 4 is a block diagram illustrating an apparatus for processing electrocardiographic data according to an exemplary embodiment. As shown in fig. 4, the apparatus 300 may include:

the obtaining module 301 is configured to obtain target electrocardiographic data to be processed.

And the processing module 302 is configured to determine, according to the target electrocardiographic data, a target electrocardiographic feature vector capable of representing the target electrocardiographic data through a pre-trained self-encoder.

The determining module 303 is configured to determine, according to the target electrocardiogram feature vector and a pre-trained self-organized mapping network, a target sign type corresponding to the target electrocardiogram data through a preset classification algorithm, where the self-organized mapping network is obtained by training according to a preset training data set.

Fig. 5 is a block diagram of one type of processing module shown in the embodiment of fig. 4. As shown in fig. 5, the processing module 302 may include:

the extracting sub-module 3021 is configured to filter the target electrocardiographic data and perform peak detection on the filtered target electrocardiographic data to extract a second amount of heartbeat data.

The dimension reduction submodule 3022 is configured to perform dimension reduction processing on a third number of heartbeat data in the second number of heartbeat data to obtain a heartbeat data sequence, where the third number is smaller than or equal to the second number.

And the encoding submodule 3023 is configured to input the heartbeat data sequence into the self-encoder to obtain the target electrocardiogram feature vector.

Optionally, the encoding submodule 3023 is configured to:

and dividing the heartbeat data sequence into a plurality of heartbeat data segments according to the specified time length.

Each heartbeat data segment is input into the self-encoder to obtain a sub-feature vector capable of representing the heartbeat data segment.

And splicing the sub-feature vectors corresponding to each heartbeat data segment to obtain the target electrocardio feature vector.

FIG. 6 is a block diagram of one type of determination module shown in the embodiment shown in FIG. 4. As shown in fig. 6, the determining module 303 includes:

the first determining submodule 3031 is configured to determine a distance from the target electrocardiogram feature vector to each of a first number of regions corresponding to the self-organizing map network, where the first number is determined when the self-organizing map network is trained according to the training data set.

The second determining submodule 3032 is configured to use the sign type corresponding to the target region as the target sign type, where the target region is a region with a minimum distance from the target electrocardiogram feature vector.

Optionally, the self-organizing map network is trained by:

a training data set is obtained, wherein the training data set comprises a plurality of training electrocardiogram data.

And determining the best matching node matched with the training electrocardiogram data in each node included in the network to be trained according to any training electrocardiogram data.

Neighborhood nodes in the topological neighborhood of the best matching node are determined.

And updating the weights of the neighborhood nodes and the optimal matching nodes according to the distances between the neighborhood nodes and the optimal matching nodes.

And repeatedly executing the steps of determining the optimal matching node matched with the training electrocardiogram data in each node included in the network to be trained according to any training electrocardiogram data, and updating the weights of the neighborhood nodes and the optimal matching node according to the distances between the neighborhood nodes and the optimal matching node until the network to be trained meets the preset training end condition so as to obtain the self-organization mapping network.

With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.

In summary, according to the present disclosure, target electrocardiographic data to be processed is obtained, a pre-trained self-organizing mapping network is used to determine a target electrocardiographic feature vector capable of representing the target electrocardiographic data according to the target electrocardiographic data, and a target sign type corresponding to the target electrocardiographic data is determined according to a preset classification algorithm according to the target electrocardiographic feature vector and the pre-trained self-organizing mapping network, wherein the self-organizing mapping network is obtained by training according to a preset training data set. The method and the device can intelligently determine the target sign type corresponding to the target electrocardiogram data through the self-encoder and the self-organizing mapping network in combination with a classification algorithm, do not need manual participation, and can improve the processing efficiency and accuracy of the electrocardiogram data.

Fig. 7 is a block diagram illustrating an electronic device 700 in accordance with an example embodiment. As shown in fig. 7, the electronic device 700 may include: a processor 701 and a memory 702. The electronic device 700 may also include one or more of a multimedia component 703, an input/output (I/O) interface 704, and a communication component 705.

The processor 701 is configured to control the overall operation of the electronic device 700, so as to complete all or part of the steps in the above-mentioned method for processing electrocardiographic data. The memory 702 is used to store various types of data to support operation at the electronic device 700, such as instructions for any application or method operating on the electronic device 700 and application-related data, such as contact data, transmitted and received messages, pictures, audio, video, and the like. The Memory 702 may be implemented by any type of volatile or non-volatile Memory device or combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic Memory, flash Memory, magnetic disk, or optical disk. The multimedia components 703 may include screen and audio components. Wherein the screen may be, for example, a touch screen and the audio component is used for outputting and/or inputting audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may further be stored in the memory 702 or transmitted through the communication component 705. The audio assembly also includes at least one speaker for outputting audio signals. The I/O interface 704 provides an interface between the processor 701 and other interface modules, such as a keyboard, mouse, buttons, etc. These buttons may be virtual buttons or physical buttons. The communication component 705 is used for wired or wireless communication between the electronic device 700 and other devices. Wireless Communication, such as Wi-Fi, bluetooth, Near Field Communication (NFC), 2G, 3G, 4G, NB-IOT, eMTC, or other 5G, etc., or a combination of one or more of them, which is not limited herein. The corresponding communication component 705 may thus include: Wi-Fi module, Bluetooth module, NFC module, etc.

In an exemplary embodiment, the electronic Device 700 may be implemented by one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components, for performing the above-mentioned method for Processing electrocardiographic data.

In another exemplary embodiment, a computer readable storage medium is also provided, which comprises program instructions, which when executed by a processor, implement the steps of the above-mentioned method for processing electrocardiographic data. For example, the computer readable storage medium may be the memory 702 including the program instructions, which are executable by the processor 701 of the electronic device 700 to perform the above-mentioned processing method of electrocardiographic data.

In another exemplary embodiment, a computer program product is also provided, which comprises a computer program executable by a programmable apparatus, the computer program having code portions for performing the above-mentioned method of processing electrocardiographic data when executed by the programmable apparatus.

The preferred embodiments of the present disclosure are described in detail with reference to the accompanying drawings, however, the present disclosure is not limited to the specific details of the above embodiments, and various simple modifications may be made to the technical solution of the present disclosure within the technical idea of the present disclosure, and these simple modifications all belong to the protection scope of the present disclosure.

It should be noted that, in the foregoing embodiments, various features described in the above embodiments may be combined in any suitable manner, and in order to avoid unnecessary repetition, various combinations that are possible in the present disclosure are not described again.

In addition, any combination of various embodiments of the present disclosure may be made, and the same should be considered as the disclosure of the present disclosure, as long as it does not depart from the spirit of the present disclosure.