CN110353657B - Method and device for screening multiple waveform types based on double-selection mechanism - Google Patents

Abstract

The application relates to a method and a device for screening various waveform types based on a double-selection mechanism, which adopt the double-selection mechanism for classification based on a waveform type classification model-based on CNN neural network classification and KNN cluster analysis method to construct judgment of various waveform types and have the advantages of high screening speed and high accuracy.

Description

Method and device for screening multiple waveform types based on double-selection mechanism

Technical Field

The application belongs to the technical field of electrocardiosignal type identification, and particularly relates to a method and a device for screening multiple waveform types based on a double-selection mechanism.

Background

After the cardiac signal is paced by the sinus node, a series of potential changes are conducted to each part of the heart through the conducting system to form the electrophysiological activity of the cardiac muscle. According to the time sequence of the activation of the heart, the change of the body surface potential is recorded, and the formed continuous curve is the electrocardiogram.

Typical electrocardiograms include P-wave, QRS-wave, T-wave. The P wave reflects the potential change in the atrial depolarization process; the P-R interval represents the time period from activation of the sinoatrial node through the atrioventricular junction to the onset of depolarization of the ventricular muscle; the QRS complex reflects the potential change in the ventricular depolarization process; the T wave represents the potential change during repolarization of the ventricular muscle. The electrocardiogram has strong complexity, and people of different races, sexes and ages have great difference under various pathological conditions, even if the types of electrocardiogram expression of the same person at different moments are different. Doctors are limited by their expertise and experience, often depend too much on the automatic diagnosis results given by machines, and at present, no method for accurately identifying the waveform type of the electrocardiosignals is available, which easily causes errors in interpretation of the electrocardiosignals.

Disclosure of Invention

The technical problem to be solved by the invention is as follows: in order to overcome the defects in the prior art, the method and the device for screening the multiple waveform types based on the double selection mechanism are provided, and the waveform types can be accurately and quickly identified.

The technical scheme adopted by the invention for solving the technical problems is as follows:

a method for screening various waveform types based on a double-selection mechanism,

acquiring unknown multi-lead electrocardiosignals to be detected;

collecting clinical rest multi-lead electrocardiosignals of a plurality of known waveform types, and giving different numbers to the electrocardiosignals of each waveform type;

if the data sampling frequency of the multi-lead electrocardiosignals with unknown waveform types is different from the data sampling frequency of the collected clinical rest multi-lead electrocardiosignals, resampling the data sampling frequency of the multi-lead electrocardiosignals with unknown waveform types and/or the data sampling frequency of the collected clinical rest multi-lead electrocardiosignals to be the same data sampling frequency;

taking the collected clinical rest multi-lead electrocardiosignals with known waveform types as input, taking the serial numbers corresponding to the waveform types corresponding to the multi-lead electrocardiosignals as output, substituting the serial numbers into a convolutional neural network, and training to obtain a waveform type classification model;

the multi-lead electrocardiosignals with unknown waveform types are used as input and substituted into a trained waveform type classification model, and the output is an array P ═ { P ═ P_iI | ═ 0,1,2 …, n-1}, n is the number of waveform types;

respectively randomly extracting the collected clinical rest multi-lead electrocardiosignals of known types according to waveform types, randomly extracting 10% of data of each waveform type to form an auxiliary test set, and adding the obtained multi-lead electrocardiosignals of unknown waveform types to the tail end of the auxiliary test set in a disordered sequence to form a mixed test set;

carrying out unsupervised classification on the mixed test set by adopting a K-means cluster analysis method, setting the number of unsupervised classifications to be n, checking a set which is unsupervised and classified into the same class with the multi-lead electrocardiosignals of unknown waveform types, and calculating the proportion K of the electrocardiosignals of different types in the set to be { K ═ K {_i|i＝0,1…,n-1}；

And sequentially executing the first judgment condition and the second judgment condition, terminating when any judgment condition is met, and outputting the type of the unknown type of the multi-lead electrocardiosignal to be detected:

judging a first condition: if p is_x>0.5 and k_x>A%, the multi-lead electrocardiosignal of unknown waveform type to be detected is considered as the waveform type with the number of x, p_xThe maximum value in the array P is shown, wherein if n is more than or equal to 5, A is 30, and if n is less than 5, A is 60;

and judging a second condition: if p is_y>0.5 and k_y>A%, then recognizeThe multi-lead electrocardiosignal of unknown waveform type to be detected is the waveform type with the serial number y corresponding to p_yThe second largest value in the array P, wherein if n is greater than or equal to 5, A is 30, and if n is less than 5, A is 60;

if the judgment conditions of one and two are not met, or K is equal to K_iIf all the elements in | i ═ 0,1 … n-1} are 0, the waveform type of the multi-lead electrocardiosignal with unknown waveform type cannot be judged.

Preferably, in the multiple waveform type screening method based on the double selection mechanism of the present invention, the waveform type of the electrocardiographic signal at least includes: normal electrocardiogram, ST segment horizontal elevation, ST segment horizontal depression and ST segment arch elevation, wherein n is 4.

Preferably, according to the multiple waveform type screening method based on the double selection mechanism, the difference value of the number of the collected electrocardiosignals of different waveform types in the electrocardiosignals of the known waveform type is less than 10%.

Preferably, in the method for screening multiple waveform types based on the double-selection mechanism, the convolutional neural network consists of layer1-layer9 of 9 layers; the layers 1-7 are composed of a convolution layer and a pooling layer; the convolutional layer in layer1 contains 5 kernels, the sizes of the convolutional kernels are both 29, and the step size and the kernel size in the pooling layer in layer1 are both 2;

the layer2 convolutional layer contains 5 kernels, the sizes of the convolutional kernels are both 15, and the step size and the kernel size in the pooling layer in the layer2 are both 2;

the layer3 convolutional layer comprises 5 kernels, the sizes of the convolutional kernels are all 13, and the step size and the kernel size in the pooling layer in the layer3 are both 2;

the layer4 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are both 5, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the layer5 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are both 5, and the step size and the kernel size in the pooling layer in the layer5 are both 2;

the layer6 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are all 3, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the layer7 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are all 3, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the number of input layer neurons of the fully-connected layer8 is consistent with the number of output characteristics of the layer7, the number of output layer neurons is 360, and the output result of the layer9 is an array P ═ P_i|i＝0,1,2,3}，p_i∈[0,1]And is a floating point number.

Preferably, in the multiple waveform type screening method based on the double selection mechanism, the loss function in the convolutional neural network adopts a coordinated _ cross loss function.

The present application further provides a multiple waveform type screening apparatus based on a dual-selection mechanism, including:

the waveform type classification model is obtained by training a convolution neural network by a plurality of clinical rest multi-lead electrocardiosignals with n kinds of known waveform type electrocardiosignals, the input of the waveform type classification model is the multi-lead electrocardiosignals of unknown waveform type to be detected, and the output is an array P ═ { P ═ P_i|i＝0,1,2…,n-1}；

The mixed test set generation model is used for randomly extracting the collected clinical rest multi-lead electrocardiosignals with known waveform types according to the waveform types, randomly extracting 10% of data of each waveform type to form an auxiliary test set, and adding the acquired multi-lead electrocardiosignals with unknown waveform types to the tail end of the auxiliary test set in a disordered sequence to form a mixed test set;

a KNN clustering model for unsupervised classification of the mixed test set by a K-means clustering analysis method, setting the number of unsupervised classifications to be n, checking the set of the multi-lead electrocardiosignals with unknown waveform types and classified into the same class by unsupervised classification, and calculating the occupation ratio K of the electrocardiosignals with different waveform types in the set to be { K ═ K { (K) }_i|i＝0,1…,n-1}；

And the judging module is used for sequentially executing the first judging condition and the second judging condition, terminating when any judging condition is met, and outputting the type of the multi-lead electrocardiosignal of the unknown type to be detected: judging a first condition: if p is_x>0.5 and k_x>A%, the unknown waveform class to be detected is consideredThe multi-lead electrocardiosignal of the type is a waveform type with the serial number of x corresponding to p_xThe maximum value in the array P is shown, wherein if n is more than or equal to 5, A is 30, and if n is less than 5, A is 60; and judging a second condition: if p is_y>0.5 and k_y>A%, the multi-lead electrocardiosignal of unknown waveform type to be detected is considered as the waveform type with the serial number y corresponding to p_yThe second largest value in the array P, wherein if n is greater than or equal to 5, A is 30, and if n is less than 5, A is 60; if the judgment conditions of one and two are not met, or K is equal to K_iIf all the elements in | i ═ 0,1 … n-1} are 0, the waveform type of the multi-lead electrocardiosignal with unknown waveform type cannot be judged.

Preferably, in the multiple waveform type screening method based on the double selection mechanism of the present invention, the waveform type of the electrocardiographic signal at least includes: normal electrocardiogram, ST segment horizontal elevation, ST segment horizontal depression and ST segment arch elevation, wherein n is 4.

Preferably, according to the multiple waveform type screening method based on the double selection mechanism, the difference value of the number of the collected electrocardiosignals of different waveform types in the electrocardiosignals of the known waveform type is less than 10%.

Preferably, in the method for screening multiple waveform types based on the double-selection mechanism, the convolutional neural network consists of layer1-layer9 of 9 layers; the layers 1-7 are composed of a convolution layer and a pooling layer; the convolutional layer in layer1 contains 5 kernels, the sizes of the convolutional kernels are both 29, and the step size and the kernel size in the pooling layer in layer1 are both 2;

the layer2 convolutional layer contains 5 kernels, the sizes of the convolutional kernels are both 15, and the step size and the kernel size in the pooling layer in the layer2 are both 2;

the layer3 convolutional layer comprises 5 kernels, the sizes of the convolutional kernels are all 13, and the step size and the kernel size in the pooling layer in the layer3 are both 2;

the layer4 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are both 5, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the layer5 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are both 5, and the step size and the kernel size in the pooling layer in the layer5 are both 2;

the layer6 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are all 3, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the layer7 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are all 3, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the number of input layer neurons of the fully-connected layer8 is consistent with the number of output characteristics of the layer7, the number of output layer neurons is 360, and the output result of the layer9 is an array P ═ P_i|i＝0,1,2,3}，p_i∈[0,1]And is a floating point number.

Preferably, in the multiple waveform type screening method based on the double selection mechanism, the loss function in the convolutional neural network adopts a coordinated _ cross loss function.

The invention has the beneficial effects that:

according to the method and the device for screening various waveform types based on the double-selection mechanism, the double-selection mechanism for classification is adopted to judge various waveform types based on the waveform type classification model, the CNN neural network classification and KNN clustering analysis method, and the method and the device have the advantages of high screening speed and high accuracy.

Drawings

The technical solution of the present application is further explained below with reference to the drawings and the embodiments.

Fig. 1 is a schematic flowchart of a multiple waveform type screening method based on a double-selection mechanism according to an embodiment of the present application.

Detailed Description

It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict.

The technical solutions of the present application will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.

Examples

The embodiment provides a method for screening multiple waveform types based on a double-selection mechanism, as shown in fig. 1, which includes the following steps:

s1: acquiring a multi-lead electrocardiosignal of an unknown waveform type to be detected, such as a twelve-lead electrocardiosignal;

s2: the method comprises the steps of collecting a plurality of clinical rest multi-lead electrocardiosignals of known waveform types, and giving different numbers to the electrocardiosignals of each waveform type, wherein the clinical rest twelve-lead electrocardiosignals of the known waveform types usually exceed 2 ten thousand, and the types of the electrocardiosignals at least comprise: the electrocardiogram signal processing method comprises the following steps that four waveform types including a normal electrocardiogram, an ST segment horizontal elevation, an ST segment horizontal depression and an ST segment back elevation are adopted, the difference value of the quantity of electrocardiosignals of different waveform types is less than 10%, the quantity of the electrocardiosignals of each waveform type is not less than 2000, different numbers are given to the electrocardiosignals of each waveform type, for example, the normal electrocardiogram is marked as 0, the ST segment horizontal elevation mark is 1, the ST segment horizontal depression mark is 2, and the ST segment back elevation mark is 3;

s3: the multi-lead electrocardiosignals with unknown waveform types and the collected clinical rest multi-lead electrocardiosignals can be preprocessed, for example, if the data sampling frequency of the multi-lead electrocardiosignals with unknown waveform types and the collected clinical rest multi-lead electrocardiosignals is different, the data sampling frequency of the multi-lead electrocardiosignals with unknown waveform types and/or the collected clinical rest multi-lead electrocardiosignals is resampled to be the same data sampling frequency; for example, the data sampling frequency of the electrocardiosignals is 500Hz, if not 500Hz, the electrocardiosignals can be changed into 500Hz through resampling, and the electrocardiosignals with unknown waveform type and the collected clinical rest electrocardiosignals with multiple leads can be filtered according to the requirements before use, for example, a band-pass filter of [0.5-50] Hz is used for filtering;

s4: taking the collected clinical rest multi-lead electrocardiosignals with known waveform types as input, taking the serial numbers corresponding to the waveform types corresponding to the multi-lead electrocardiosignals as output, substituting the serial numbers into a convolutional neural network, and training to obtain a waveform type classification model;

wherein, the convolutional neural network is composed of 9 layers of networks. The layers 1-7 are composed of a convolution layer and a pooling layer; the convolutional layer in layer1 contains 5 kernels, the sizes of the convolutional kernels are both 29, and the step size and the kernel size in the pooling layer in layer1 are both 2;

the layer2 convolutional layer contains 5 kernels, the sizes of the convolutional kernels are both 15, and the step size and the kernel size in the pooling layer in the layer2 are both 2;

the layer3 convolutional layer comprises 5 kernels, the sizes of the convolutional kernels are all 13, and the step size and the kernel size in the pooling layer in the layer3 are both 2;

the layer4 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are both 5, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the layer5 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are both 5, and the step size and the kernel size in the pooling layer in the layer5 are both 2;

the layer6 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are all 3, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the layer7 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are all 3, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the number of neurons in an input layer of the fully-connected layer8 is consistent with the number of output features of the layer7, and the number of neurons in an output layer is 360, so that 360 features are output after calculation; the 360 features are input into a classifier layer9 as input, layer9 outputs a diagnosis result through calculation, and when the output result of layer9 is an array P, and n is 4, the array P is { P ═ P {_i|i＝0,1,2,3}，p_i∈[0,1]And is a floating point number. (the number i is the same as the number of the type of the ECG signal during training, p_iRespectively represent: normal electrocardiogram, ST segment horizontal elevation, ST segment horizontal depression and ST segment back elevation, namely a probability value. ) The loss function adopts a catagorical _ crosssentryloss function;

s5: the multi-lead electrocardiosignals with unknown waveform types are used as input and substituted into a trained waveform type classification model, and the output is an array P ═ { P ═ P_iI is 0,1,2 …, n-1, and n is the number of the waveform types of the electrocardiosignals;

s6: respectively randomly extracting the collected clinical rest multi-lead electrocardiosignals with known waveform types according to the types, randomly extracting 10% of data of each waveform type to form an auxiliary test set, and adding the obtained multi-lead electrocardiosignals with unknown waveform types to the tail end of the auxiliary test set in a disordered sequence to form a mixed test set; such as: after 10% of electrocardio information with four waveform types is extracted, and when the number of the auxiliary test sets is 800, the obtained twelve-lead electrocardiosignals with unknown waveform types are added to the tail end of the auxiliary test set in a disorderly sequence to form a mixed test set containing 801 electrocardiosignals.

S7: carrying out unsupervised classification on the mixed test set by adopting a K-means cluster analysis method, setting the number of unsupervised classifications to be n, checking a set which is unsupervised and classified into the same class with the multi-lead electrocardiosignals of unknown waveform types, and calculating the proportion K of the electrocardiosignals of different types in the set to be { K ═ K {_i|i＝0,1…,n-1}；

S8: sequentially executing the first judgment condition and the second judgment condition, stopping when any judgment condition is met, and outputting the type of the multi-lead electrocardiosignal of the unknown type to be detected, wherein x and y are both a number in i:

judging a first condition: if p is_x>0.5 and k_x>A%, the multi-lead electrocardiosignal of unknown waveform type to be detected is considered as the waveform type with the number of x, p_xIs the largest value in the array P, wherein if n is greater than or equal to 5, a is 30, and if n is less than 5, a is 60, for example, if n is 4, a may be 60;

and judging a second condition: if p is_y>0.5 and k_y>A%, the multi-lead electrocardiosignal of unknown waveform type to be detected is considered as the waveform type with the serial number y corresponding to p_yThe second largest value in the array P, wherein if n is greater than or equal to 5, A is 30, and if n is less than 5, A is 60;

if the judgment conditions of one and two are not met, or K is equal to K_iIf all the elements in | i ═ 0,1 … n-1} are 0, the waveform type of the multi-lead electrocardiosignal with unknown waveform type cannot be judged.

The present embodiment further provides a device for screening multiple waveform types based on a dual-selection mechanism, including:

a waveform type classification model, wherein the waveform type classification model is obtained by the convolutional neural network training of a plurality of clinical resting multi-lead electrocardiosignals with known waveform types and n types of electrocardiosignals,the input of the waveform type classification model is a multi-lead electrocardiosignal of unknown waveform type to be detected, and the output is an array P ═ { P ═ P_i|i＝0,1,2…,n-1}；

The training method of the waveform type classification model comprises the following steps:

the method comprises the steps of collecting a plurality of clinical rest multi-lead electrocardiosignals of known waveform types, and giving different numbers to the electrocardiosignals of each waveform type, wherein the clinical rest twelve-lead electrocardiosignals of the known waveform types usually exceed 2 ten thousand, and the waveform types of the electrocardiosignals at least comprise: the electrocardiogram signal generator comprises four types, namely a normal electrocardiogram, an ST section horizontal elevation, an ST section horizontal depression and an ST section arch back elevation, wherein n is 4, the difference value of the number of electrocardiosignals of different waveform types is less than 10%, the number of the electrocardiosignals of each waveform type is not less than 2000, and different numbers are given to the electrocardiosignals of each waveform type, for example, the normal electrocardiogram is 0, the ST section horizontal elevation is 1, the ST section horizontal depression is 2 and the ST section arch back elevation is 3; if the electrocardiosignals do not meet the processing conditions, the electrocardiosignals can be preprocessed, the data sampling frequency of the electrocardiosignals is recommended to be 500Hz, if not, the data sampling frequency of the electrocardiosignals can be changed into 500Hz through resampling, and the unknown type of the multi-lead electrocardiosignals and the collected clinical rest multi-lead electrocardiosignals can be filtered according to the requirements before being used, for example, a band-pass filter of [0.5-50] Hz is used for filtering;

taking the collected clinical rest multi-lead electrocardiosignals with known waveform types as input, taking the serial numbers corresponding to the waveform types corresponding to the multi-lead electrocardiosignals as output, substituting the serial numbers into a convolutional neural network, and training to obtain a waveform type classification model;

wherein, the convolutional neural network is composed of 9 layers of networks. The layers 1-7 are composed of a convolution layer and a pooling layer; the convolutional layer in layer1 contains 5 kernels, the sizes of the convolutional kernels are both 29, and the step size and the kernel size in the pooling layer in layer1 are both 2;

the layer2 convolutional layer contains 5 kernels, the sizes of the convolutional kernels are both 15, and the step size and the kernel size in the pooling layer in the layer2 are both 2;

the layer3 convolutional layer comprises 5 kernels, the sizes of the convolutional kernels are all 13, and the step size and the kernel size in the pooling layer in the layer3 are both 2;

the layer4 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are both 5, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the layer5 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are both 5, and the step size and the kernel size in the pooling layer in the layer5 are both 2;

the layer6 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are all 3, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the layer7 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are all 3, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the number of neurons in an input layer of the fully-connected layer8 is consistent with the number of output features of the layer7, and the number of neurons in an output layer is 360, so that 360 features are output after calculation; the 360 features are input into a classifier layer9, layer9 outputs a diagnosis result through calculation, a prediction function is model, prediction _ classes, the output result of layer9 forms an array P, and when n is 4, the array P is { P ═ 4 ═ P {_i|i＝0,1,2,3}，p_i∈[0,1]. (the number i is the same as the number of the type of the ECG signal during training, p_iRespectively represent: normal electrocardiogram, ST segment horizontal elevation, ST segment horizontal depression and ST segment back elevation, namely a probability value. ) The loss function adopts a catagorical _ crosssentryloss function;

the mixed test set generation model is used for randomly extracting the collected clinical rest multi-lead electrocardiosignals with known waveform types according to the waveform types, randomly extracting 10% of data of each waveform type to form an auxiliary test set, and adding the acquired multi-lead electrocardiosignals with unknown waveform types to the tail end of the auxiliary test set in a disordered sequence to form a mixed test set; such as: after 10% of electrocardio information with four waveform types is extracted, and when the number of the auxiliary test sets is 800, the obtained twelve-lead electrocardiosignals with unknown waveform types are added to the tail end of the auxiliary test set in a disorderly sequence to form a mixed test set containing 801 electrocardiosignals.

KNN clustering model for clustering by k-meansThe analysis method carries out unsupervised classification on the mixed test set, sets the number of unsupervised classifications to be n, checks that the multi-lead electrocardiosignals with unknown waveform types are classified into the same set by the unsupervised classification, and calculates the proportion K to K of the electrocardiosignals with different types in the set_i|i＝0,1…,n-1}；

And the judging module is used for sequentially executing the first judging condition and the second judging condition, terminating when any judging condition is met, and outputting the type of the multi-lead electrocardiosignal of the unknown type to be detected: judging a first condition: if p is_x>0.5 and k_x>A%, the multi-lead electrocardiosignal of unknown waveform type to be detected is considered as the waveform type with the number of x, p_xThe maximum value in the array P is shown, wherein if n is more than or equal to 5, A is 30, and if n is less than 5, A is 60; and judging a second condition: if p is_y>0.5 and k_y>A%, the multi-lead electrocardiosignal of unknown waveform type to be detected is considered as the waveform type with the serial number y corresponding to p_yThe second largest value in the array P, wherein if n is greater than or equal to 5, A is 30, and if n is less than 5, A is 60; if the judgment conditions of one and two are not met, or K is equal to K_iIf all the elements in | i | -0, 1 … n-1} are 0, it is considered that the multi-lead electrocardiosignal of unknown type cannot be judged. x and y are each a number from 0 to 3.

The method and the device for screening various waveform types based on the double-selection mechanism adopt the double-selection mechanism for classification based on the waveform type classification model, the CNN neural network classification and KNN cluster analysis method to construct judgment of various waveform types, and have the advantages of high screening speed and high accuracy.

The application judges the electrocardiogram waveform, and the waveform cannot finally determine whether the patient is ill or not, so that a doctor needs to make judgment.

In light of the foregoing description of the preferred embodiments according to the present application, it is to be understood that various changes and modifications may be made without departing from the spirit and scope of the invention. The technical scope of the present application is not limited to the contents of the specification, and must be determined according to the scope of the claims.

As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media having computer-usable program code embodied in the medium.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

Claims (10)

1. A method for screening various waveform types based on a double-selection mechanism is characterized in that,

acquiring unknown multi-lead electrocardiosignals to be detected;

collecting a plurality of clinical rest multi-lead electrocardiosignals with known waveform types, and giving different numbers to the electrocardiosignals with each waveform type;

if the data sampling frequency of the multi-lead electrocardiosignals with unknown waveform types is different from the data sampling frequency of the collected clinical rest multi-lead electrocardiosignals, resampling the data sampling frequency of the multi-lead electrocardiosignals with unknown waveform types and/or the data sampling frequency of the collected clinical rest multi-lead electrocardiosignals to be the same data sampling frequency;

taking the collected clinical rest multi-lead electrocardiosignals with known waveform types as input, taking the serial numbers corresponding to the waveform types corresponding to the multi-lead electrocardiosignals as output, substituting the serial numbers into a convolutional neural network, and training to obtain a waveform type classification model;

the multi-lead electrocardiosignals with unknown waveform types are used as input and substituted into a trained waveform type classification model, and the output is an array P ═ { P ═ P_iI | ═ 0,1,2 …, n-1}, n is the number of waveform types;

respectively randomly extracting the collected clinical rest multi-lead electrocardiosignals of known types according to waveform types, randomly extracting 10% of data of each waveform type to form an auxiliary test set, and adding the obtained multi-lead electrocardiosignals of unknown waveform types to the tail end of the auxiliary test set in a disordered sequence to form a mixed test set;

carrying out unsupervised classification on the mixed test set by adopting a K-means cluster analysis method, setting the number of unsupervised classifications to be n, checking a set which is unsupervised and classified into the same class with the multi-lead electrocardiosignals of unknown waveform types, and calculating the proportion K of the electrocardiosignals of different types in the set to be { K ═ K {_i|i＝0,1…,n-1}；

And sequentially executing the first judgment condition and the second judgment condition, terminating when any judgment condition is met, and outputting the type of the unknown type of the multi-lead electrocardiosignal to be detected:

judging a first condition: if p is_x>0.5 and k_x>A%, the multi-lead electrocardiosignal of unknown waveform type to be detected is considered as the waveform type with the number of x, p_xThe maximum value in the array P is shown, wherein if n is more than or equal to 5, A is 30, and if n is less than 5, A is 60;

and judging a second condition: if p is_y>0.5 and k_y>A%, the multi-lead electrocardiosignal of unknown waveform type to be detected is considered as the waveform type with the serial number y corresponding to p_yThe second largest value in the array P, wherein if n is greater than or equal to 5, A is 30, and if n is less than 5, A is 60;

if the judgment conditions of one and two are not met, or K is equal to K_iIf all the elements in | i ═ 0,1 … n-1} are 0, the waveform type of the multi-lead electrocardiosignal with unknown waveform type cannot be judged.

2. The method for screening multiple waveform types based on the double selection mechanism as claimed in claim 1, wherein the waveform types of the electrocardiographic signal at least comprise: normal electrocardiogram, ST segment horizontal elevation, ST segment horizontal depression and ST segment arch elevation, wherein n is 4.

3. The method according to claim 2, wherein the difference between the number of the collected electrocardiosignals of different waveform types in the electrocardiosignals of known waveform types is less than 10%.

4. The method for screening multiple waveform types based on the double selection mechanism according to claim 3, wherein the convolutional neural network is composed of 9-layer network layers 1-9; the layers 1-7 are composed of a convolution layer and a pooling layer; the convolutional layer in layer1 contains 5 kernels, the sizes of the convolutional kernels are both 29, and the step size and the kernel size in the pooling layer in layer1 are both 2;

the layer2 convolutional layer contains 5 kernels, the sizes of the convolutional kernels are both 15, and the step size and the kernel size in the pooling layer in the layer2 are both 2;

the layer3 convolutional layer comprises 5 kernels, the sizes of the convolutional kernels are all 13, and the step size and the kernel size in the pooling layer in the layer3 are both 2;

the layer4 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are both 5, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the layer5 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are both 5, and the step size and the kernel size in the pooling layer in the layer5 are both 2;

the layer6 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are all 3, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the layer7 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are all 3, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the number of input layer neurons of the fully-connected layer8 is consistent with the number of output characteristics of the layer7, the number of output layer neurons is 360, and the output result of the layer9 is an array P ═ P_i|i＝0,1,2,3}，p_i∈[0,1]And is a floating point number.

5. The method for screening multiple waveform types based on the double selection mechanism according to claim 4, wherein the loss function in the convolutional neural network is a probabilistic _ crosssentryloss function.

6. A multiple waveform type screening device based on a double-selection mechanism is characterized by comprising:

the waveform type classification model is obtained by training a convolution neural network by a plurality of clinical rest multi-lead electrocardiosignals with n kinds of known waveform type electrocardiosignals, the input of the waveform type classification model is the multi-lead electrocardiosignals of unknown waveform type to be detected, and the output is an array P ═ { P ═ P_i|i＝0,1,2…,n-1}；

The mixed test set generation model is used for randomly extracting the collected clinical rest multi-lead electrocardiosignals with known waveform types according to the waveform types, randomly extracting 10% of data of each waveform type to form an auxiliary test set, and adding the acquired multi-lead electrocardiosignals with unknown waveform types to the tail end of the auxiliary test set in a disordered sequence to form a mixed test set;

a KNN clustering model for unsupervised classification of the mixed test set by a K-means clustering analysis method, setting the number of unsupervised classifications to be n, checking the set of the multi-lead electrocardiosignals with unknown waveform types and classified into the same class by unsupervised classification, and calculating the occupation ratio K of the electrocardiosignals with different waveform types in the set to be { K ═ K { (K) }_i|i＝0,1…,n-1}；

And the judging module is used for sequentially executing the first judging condition and the second judging condition, terminating when any judging condition is met, and outputting the type of the multi-lead electrocardiosignal of the unknown type to be detected: judging a first condition: if p is_x>0.5 and k_x>A%, the multi-lead electrocardiosignal of unknown waveform type to be detected is considered as the waveform type with the number of x, p_xThe maximum value in the array P is shown, wherein if n is more than or equal to 5, A is 30, and if n is less than 5, A is 60; and judging a second condition: if p is_y>0.5 and k_y>A%, the multi-lead electrocardiosignal of unknown waveform type to be detected is considered as the waveform type with the serial number y corresponding to p_yThe second largest value in the array P, wherein if n is greater than or equal to 5, A is 30, and if n is less than 5, A is 60; if the judgment conditions of one and two are not met, or K is equal to K_iIf all the elements in | i ═ 0,1 … n-1} are 0, the waveform type of the multi-lead electrocardiosignal with unknown waveform type cannot be judged.

7. The apparatus according to claim 6, wherein the waveform types of the cardiac signal at least include: normal electrocardiogram, ST segment horizontal elevation, ST segment horizontal depression and ST segment arch elevation, wherein n is 4.

8. The dual-selection mechanism-based multi-waveform-type screening device according to claim 7, wherein the difference between the number of the collected electrocardiosignals of different waveform types in the electrocardiosignals of known waveform types is less than 10%.

9. The multiple waveform type screening apparatus based on the double selection mechanism as claimed in claim 8, wherein the convolutional neural network is composed of 9 layer network layers 1-9; the layers 1-7 are composed of a convolution layer and a pooling layer; the convolutional layer in layer1 contains 5 kernels, the sizes of the convolutional kernels are both 29, and the step size and the kernel size in the pooling layer in layer1 are both 2;

the layer2 convolutional layer contains 5 kernels, the sizes of the convolutional kernels are both 15, and the step size and the kernel size in the pooling layer in the layer2 are both 2;

the layer3 convolutional layer comprises 5 kernels, the sizes of the convolutional kernels are all 13, and the step size and the kernel size in the pooling layer in the layer3 are both 2;

the layer4 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are both 5, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the layer5 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are both 5, and the step size and the kernel size in the pooling layer in the layer5 are both 2;

the layer6 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are all 3, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the layer7 convolutional layer contains 10 kernels, the sizes of the convolutional kernels are all 3, and the step size and the kernel size in the pooling layer in the layer4 are both 2;

the number of input layer neurons of the fully-connected layer8 is consistent with the number of output characteristics of the layer7, the number of output layer neurons is 360, and the output result of the layer9 is an array P ═ P_i|i＝0,1,2,3}，p_i∈[0,1]And is a floating point number.

10. The apparatus according to claim 9, wherein the loss function in the convolutional neural network is a probabilistic _ crosstransmit loss function.

Images