CN110974211A - ST-segment classification neural network of high-order polynomial activation function and application thereof - Google Patents
ST-segment classification neural network of high-order polynomial activation function and application thereof Download PDFInfo
- Publication number
- CN110974211A CN110974211A CN201911250657.2A CN201911250657A CN110974211A CN 110974211 A CN110974211 A CN 110974211A CN 201911250657 A CN201911250657 A CN 201911250657A CN 110974211 A CN110974211 A CN 110974211A
- Authority
- CN
- China
- Prior art keywords
- output
- dimensional
- layer
- lead
- size
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Animal Behavior & Ethology (AREA)
- Veterinary Medicine (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Signal Processing (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Pathology (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Psychiatry (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Cardiology (AREA)
- Power Engineering (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The application relates to an ST-segment classification neural network of a high-order polynomial activation function, which has good effect of removing random noise through mean filtering, removes a base line through wavelet filtering, and finally removes noise in signals. And based on the combination of the convolutional neural network and the high-order polynomial activation function, the complexity of the model can be directly increased by the great divergence of the high-order polynomial function, the problem of super-parameter selection in the regularization process is avoided, and the generalization capability of the neural network is obviously improved.
Description
Technical Field
The application belongs to the technical field of electrocardiogram processing, and particularly relates to an ST-segment classification neural network of a high-order polynomial activation function and application thereof.
Background
The ST segment is a segment from the end of the QRS complex at the J point to the start of the T wave, and is an important component of the electrocardiogram. The normal ST segment is a shallow upwind with the erection of the T wave. The change of the ST segment includes: ST elevation, ST depression, ST elongation and ST shortening. In combination with clinical symptoms, changes in the ST segment can be an important basis for diagnosing myocardial infarction. The importance of the ST segment lies in whether the ST segment is depressed or raised, and the ST segment has the characteristics of low frequency and small amplitude, and the form of the ST segment is easy to be interfered by external noise to change in the electrocardio detection process, thereby increasing the analysis difficulty. Therefore, accurate localization of the ST segment and quantitative analysis of the waveform are key to the rapid diagnosis of the corresponding heart disease. The current electrocardiograph detection algorithm is mainly applied to classification and identification of electrocardiograph signals, and the maturity of the automatic diagnosis algorithm for ST-segment waveforms is low. When the activation function of the main flow is selected as the activation function of the model, the regularization parameter selection is difficult, and the generalization capability of the model is poor.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: in order to solve the defects in the prior art, the ST-segment classification neural network of the high-order polynomial activation function and the application thereof are provided.
The technical scheme adopted by the invention for solving the technical problems is as follows:
an ST-segment classification neural network of higher order polynomial activation functions, comprising:
a plurality of groups of coiling layers, pooling layers, a flat layer and a full-connection layer;
collecting a plurality of known clinical rest multi-lead electrocardiograms, wherein the multi-lead electrocardiograms comprise t types including normal electrocardiograms and a plurality of ST section abnormal types, the sampling frequencies of the multi-lead electrocardiograms are the same or are preprocessed to be the same, the multi-lead electrocardiograms are filtered by the same mean filter and then filtered by a wavelet filtering method, and different lead electrocardiosignals in the multi-lead electrocardiosignals are spliced to form new long-chain electrocardiosignals as elements of an electrocardiosignal training set; calibrating tag vectors of the electrocardiosignals according to types, wherein the tag vectors of different types are different and are (a1, a2, … …, at), only one of a1, a2 and … …, and the rest are 0;
the fully-connected layer comprises a hidden layer and an output layer, wherein the excitation function in the hidden layer is a high-order polynomial activation function f (x) anxn+an-1xn-1+…+a1x+a0N is a natural number, and soft is adopted as an excitation function in an output layerA max function; the loss function in the fully connected layer is categoratic _ cross;
training the convolutional neural network by using the long-chain electrocardiosignal data of the training set as input and the corresponding label vector as output, and determining the parameters of each convolutional layer, each pooling layer, each flat layer and each full-connection layer;
the outputs of the fully-connected layers are the tag vectors (a1, a2, … …, at), only one of a1, a2, … …, at is 1, and the rest are 0.
Preferably, the ST-segment classification neural network of the higher-order polynomial activation function of the present invention, where n is 5, and a5＝0.001，a3＝-0.0060，a1＝0.2003，a0＝0.5000。
Preferably, the ST-segment classification neural network of the higher-order polynomial activation function of the present invention,
the convolutional Layer and the pooling Layer have 18 layers of Layer1-Layer 18;
layer1 is a one-dimensional convolution Layer, wherein 5 convolution kernels with the size of 37 are in total, and the output is 5 one-dimensional vectors;
layer2 is a one-dimensional maximum pooling Layer, wherein the size and step length of the kernel are both 2, and the output is 5 one-dimensional vectors;
layer3 is a one-dimensional convolution Layer, wherein, 5 convolution kernels with the size of 31 are in total, and the output is 5 one-dimensional vectors;
layer4 is a one-dimensional maximum pooling Layer, wherein the size and step length of the kernel are both 2, and the output is 5 one-dimensional vectors;
layer5 is a one-dimensional convolution Layer, wherein 5 convolution kernels with the size of 29 are totally arranged, and 5 one-dimensional vectors with the output of 14948 are output;
layer6 is a one-dimensional maximum pooling Layer, wherein the size and step length of the kernel are both 2, and the output is 5 one-dimensional vectors;
layer7 is a one-dimensional convolution Layer, wherein, 10 convolution kernels with the size of 23 are in total, and the output is 10 one-dimensional vectors;
layer8 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer9 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 15 are totally arranged, and the output is 10 one-dimensional vectors;
layer10 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer11 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 9 are totally arranged, and the output is 10 one-dimensional vectors;
layer12 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer13 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 5 are in total, and the output is 10 one-dimensional vectors;
layer14 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer15 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 1 are in total, and the output is 10 one-dimensional vectors;
layer16 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer17 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 1 are in total, and the output is 10 one-dimensional vectors;
layer18 is a one-dimensional maximum pooling Layer where the kernel size and step size are both 2 and the output is 10 one-dimensional vectors.
Preferably, the present invention is a ST-segment classification neural network of higher-order polynomial activation functions, and the convolutional layer and the pooling layer have a ReLU function as an activation function.
Preferably, the ST-segment classified neural network of the high-order polynomial activation function of the present invention splices the electrocardiographic signals of different leads in the multi-lead electrocardiographic signals to form a new long-chain electrocardiographic signal, arranges and combines the electrocardiographic signals according to the sequence of different leads to obtain long-chain electrocardiographic signals in all arrangement sequences, and uses the long-chain electrocardiographic signals in all arrangement sequences as a training set.
Preferably, in the ST-segment classified neural network with a high-order polynomial activation function of the present invention, when different lead electrocardiographic signals in the multi-lead electrocardiographic signals are spliced to form a new long-chain electrocardiographic signal, the electrocardiographic signals are arranged and combined according to different lead sequences to obtain long-chain electrocardiographic signals in all arrangement sequences, and the long-chain electrocardiographic signals in each different sequence form a training set, and a plurality of different convolutional neural networks are obtained through training in different training sets.
Preferably, the ST-segment classification neural network of the high-order polynomial activation function of the invention comprises ST-segment horizontal elevation, ST-segment horizontal depression and ST-segment arch elevation, and the ST-segment classification neural network and the ST-segment horizontal depression form 4 types together with a normal electrocardiogram.
Preferably, the ST-segment classification neural network of the high-order polynomial activation function of the present invention adopts a training algorithm as follows: a random gradient descent algorithm, an Adam algorithm, a RMSProp algorithm, an adagard algorithm, an adapelta algorithm, an Adamax algorithm.
The invention also provides an application method of the ST-segment classification neural network of the high-order polynomial activation function, which comprises the following steps:
s1: acquiring multi-lead electrocardiosignals, wherein the sampling frequencies of the multi-lead electrocardiosignals are the same or are preprocessed to be the same, the multi-lead electrocardiosignals are filtered by the same mean filter and then filtered by a wavelet filtering method, and different lead electrocardiosignals in the multi-lead electrocardiosignals are spliced to form new long-chain electrocardiosignals;
s2: inputting a new long-chain electrocardiosignal into the ST-segment classification neural network of the high-order polynomial activation function;
s3: and determining the type of the multi-lead electrocardiogram according to the label vector output by the ST-segment classification neural network of the high-order polynomial activation function.
The invention also provides an application method of the ST-segment classification neural network of the high-order polynomial activation function, which uses the ST-segment classification neural network of the high-order polynomial activation function and comprises the following steps:
s1: acquiring multi-lead electrocardiosignals, wherein the sampling frequencies of the multi-lead electrocardiosignals are the same or are preprocessed to be the same, the multi-lead electrocardiosignals are filtered by the same mean filter and then filtered by a wavelet filtering method, different lead electrocardiosignals in the multi-lead electrocardiosignals are spliced to form new long-chain electrocardiosignals, and the long-chain electrocardiosignals in all arrangement sequences are acquired by arranging and combining according to different lead sequences;
s2: respectively inputting the long-chain electrocardiosignals in different sequences into the convolutional neural network trained by the training set in the corresponding sequence;
s3: and if a certain label vector output by all the convolutional neural networks accounts for more than 80% of the total output number of the convolutional neural networks, determining the type of the multi-lead electrocardiogram by using the label vector.
The invention has the beneficial effects that:
the ST-segment classification neural network of the high-order polynomial activation function has good effect of removing random noise through mean filtering, and finally removes noise in signals through wavelet filtering. And based on the combination of the convolutional neural network and the high-order polynomial activation function, the complexity of the model can be directly increased by the great divergence of the high-order polynomial function, the problem of super-parameter selection in the regularization process is avoided, and the generalization capability of the neural network is obviously improved.
Drawings
The technical solution of the present application is further explained below with reference to the drawings and the embodiments.
FIG. 1 is a diagram illustrating an ST-segment classification neural network structure of a higher-order polynomial activation function according to an embodiment of the present application;
fig. 2 is a graph comparing the effects of using a higher order polynomial function and a Sigmoid function.
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.
Example 1
The present embodiment provides an ST-segment classification neural network of a higher-order polynomial activation function, as shown in fig. 1, including:
a plurality of groups of coiling layers, pooling layers, a flat layer and a full-connection layer;
collecting a plurality of known clinical rest multi-lead electrocardiograms (such as 12 leads) of which the sampling frequency is the same or is preprocessed to be the same, wherein the multi-lead electrocardiograms comprise t types including a normal electrocardiogram and a plurality of ST segment abnormal types (such as ST segment abnormal types including ST segment horizontal elevation, ST segment horizontal depression and ST segment dorsal elevation, and the t type is formed by the normal electrocardiogram together with the T type to be 4 types, and at least 15000 pieces of data of each type are generally included to achieve the sufficient plurality of clinical rest multi-lead electrocardiograms, and the multi-lead electrocardiogram is filtered by the same mean filter and then filtered by a wavelet filtering method (the parameters of the mean filter can be 5, the parameters of the wavelet filtering can be 'db 4', and the level can be 8), splicing different lead electrocardiosignals in the multi-lead electrocardiosignals to form a new long-chain electrocardiosignal serving as an element of an electrocardiosignal training set; calibrating tag vectors of the electrocardiosignals according to types, wherein the tag vectors of different types are different and are (a1, a2, … …, at), only one of a1, a2 and … …, and the rest are 0; (for example, labeling the corresponding normal electrocardiogram, ST-segment horizontal elevation, ST-segment horizontal depression, and ST-segment arch-back elevation as (1,0,0,0), (0,1,0,0), (0,0,1,0), (0,0,0,1))
The fully-connected layer comprises a hidden layer and an output layer, wherein the excitation function in the hidden layer is a high-order polynomial activation function f (x) anxn+an-1xn-1+…+a1x+a0N is a natural number, and the excitation function in the output layer adopts a softmax function; the loss function in the fully connected layer is categoratic _ cross;
training the convolutional neural network by using the long-chain electrocardiosignal data of the training set as input and the corresponding label vector as output, and determining the parameters of each convolutional layer, each pooling layer, each flat layer and each full-connection layer;
the outputs of the fully-connected layers are the tag vectors (a1, a2, … …, at), only one of a1, a2, … …, at is 1, and the rest are 0.
Preferably, n is 5 in the higher-order polynomial activation function, and a5＝0.001，a3＝-0.0060，a1＝0.2003，a0＝0.5000，a4And a2 Is 0.
The convolutional Layer and the pooling Layer have 18 layers of Layer1-Layer 18;
layer1 is a one-dimensional convolution Layer, wherein 5 convolution kernels with the size of 37 are in total, and the output is 5 one-dimensional vectors;
layer2 is a one-dimensional maximum pooling Layer, wherein the size and step length of the kernel are both 2, and the output is 5 one-dimensional vectors;
layer3 is a one-dimensional convolution Layer, wherein, 5 convolution kernels with the size of 31 are in total, and the output is 5 one-dimensional vectors;
layer4 is a one-dimensional maximum pooling Layer, wherein the size and step length of the kernel are both 2, and the output is 5 one-dimensional vectors;
layer5 is a one-dimensional convolution Layer, wherein 5 convolution kernels with the size of 29 are totally arranged, and 5 one-dimensional vectors with the output of 14948 are output;
layer6 is a one-dimensional maximum pooling Layer, wherein the size and step length of the kernel are both 2, and the output is 5 one-dimensional vectors;
layer7 is a one-dimensional convolution Layer, wherein, 10 convolution kernels with the size of 23 are in total, and the output is 10 one-dimensional vectors;
layer8 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer9 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 15 are totally arranged, and the output is 10 one-dimensional vectors;
layer10 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer11 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 9 are totally arranged, and the output is 10 one-dimensional vectors;
layer12 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer13 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 5 are in total, and the output is 10 one-dimensional vectors;
layer14 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer15 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 1 are in total, and the output is 10 one-dimensional vectors;
layer16 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer17 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 1 are in total, and the output is 10 one-dimensional vectors;
layer18 is a one-dimensional maximum pooling Layer where the kernel size and step size are both 2 and the output is 10 one-dimensional vectors. The convolutional and pooling layers preferably have a ReLU function as the activation function.
As an improvement, different lead electrocardiosignals in the multi-lead electrocardiosignals are spliced to form a new long-chain electrocardiosignal, the electrocardiosignals are arranged and combined according to different lead sequences to obtain long-chain electrocardiosignals in all arrangement sequences, and the long-chain electrocardiosignals in all arrangement sequences are used as a training set together. I.e., the last training set can be found to be 12! The long-chain electrocardiosignal data of (factorial) times can also be taken as a training set, wherein the long-chain electrocardiosignals of which parts are arranged in sequence.
Example 2
The present embodiment provides an ST-segment classification neural network of a higher-order polynomial activation function, including: the device is composed of a plurality of neural networks; the number of the neural networks is multiplied by the number of the leads;
the constitution of each neural network is the same as that in embodiment 1;
the training set for each neural network is derived from long chain electrocardiosignals in an arrangement order.
Twelve leads, i.e., I, II, III, avL, avF, avR, v1, v2, v3, v4, v5, v 6; the sequence of the links of the long chain cardiac signals may be:
1.I、II、III、avL、avF、avR、v1、v2、v3、v4、v5、v6
2.II、I、III、avL、avF、avR、v1、v2、v3、v4、v5、v6
3.III、II、I、avL、avF、avR、v1、v2、v3、v4、v5、v6
and so on to finally be 12! (where | means factorial) of the order method, 12! Training long-chain electrocardiosignal training sets, and obtaining 12!by training each training set! A neural network.
Example 3
The present embodiment provides an application method of an ST-segment classification neural network of a higher-order polynomial activation function, including the following steps:
s1: acquiring multi-lead electrocardiosignals, wherein the sampling frequencies of the multi-lead electrocardiosignals are the same or are preprocessed to be the same, the multi-lead electrocardiosignals are filtered through the same mean filter, and then are filtered through a wavelet filtering method (the filtering method is the same as that adopted in the ST-segment classification neural network training of a high-order polynomial activation function), and different lead electrocardiosignals in the multi-lead electrocardiosignals are spliced to form new long-chain electrocardiosignals;
s2: inputting a new long-chain electrocardiosignal into an ST-segment classification neural network of a high-order polynomial activation function as described in example 1;
s3: the type of the multi-lead electrocardiogram is determined according to the type represented by the label vector. (for example, when the four types of the corresponding normal electrocardiogram, ST segment horizontal elevation, ST segment horizontal depression and ST segment arch back elevation are labeled as (1,0,0,0), (0,1,0,0), (0,0,1,0), (0,0,1,0) respectively during training, the output is (1,0,0,0), (0,1,0,0), (0,0,1,0), (0,0, 1) respectively representing the normal electrocardiogram, ST segment horizontal elevation, ST segment horizontal depression and ST segment arch back elevation).
Example 4
The present embodiment provides a method for applying an ST-segment neural network using a higher-order polynomial activation function, where the ST-segment neural network using a higher-order polynomial activation function as described in embodiment 2 includes the following steps:
s1: acquiring multi-lead electrocardiosignals, wherein the sampling frequencies of the multi-lead electrocardiosignals are the same or are preprocessed to be the same, the multi-lead electrocardiosignals are filtered by the same mean filter and then filtered by a wavelet filtering method, different lead electrocardiosignals in the multi-lead electrocardiosignals are spliced to form new long-chain electrocardiosignals, and the long-chain electrocardiosignals (to be identified) in all arrangement sequences are obtained by arranging and combining according to different lead sequences;
s2: respectively inputting long-chain electrocardiosignals (to be identified) in different sequences into a convolutional neural network trained by a training set in a corresponding sequence as in example 2; that is, inputting the signals into different convolutional neural networks according to different sequences, and inputting a training set of which lead arrangement sequence is adopted during training, wherein long-chain electrocardiosignals (to be identified) obtained in the corresponding sequence are input;
s3: and if a certain label vector output by all the convolutional neural networks accounts for more than 80% of the total output number of the convolutional neural networks, determining the type of the multi-lead electrocardiogram by using the label vector.
Such as: 12 leads with 12! (here! is the factorial meaning) a neural network, which is formed 12! Inputting all the long-chain electrocardiosignals into the corresponding neural network to obtain 12! And (3) judging the type of the multi-lead electrocardiogram by using the label vectors, wherein whether a certain label vector exceeds more than 80% of the total number needs to be judged, the label vectors exceeding 80% are used as the type result of the multi-lead electrocardiogram, and if the label vectors do not exceed 80%, an alarm is prompted, so that accurate identification cannot be realized.
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 (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), 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. An ST-segment classification neural network of a higher order polynomial activation function, comprising:
a plurality of groups of coiling layers, pooling layers, a flat layer and a full-connection layer;
collecting a plurality of known clinical rest multi-lead electrocardiograms, wherein the multi-lead electrocardiograms comprise t types including normal electrocardiograms and a plurality of ST section abnormal types, the sampling frequencies of the multi-lead electrocardiograms are the same or are preprocessed to be the same, the multi-lead electrocardiograms are filtered by the same mean filter and then filtered by a wavelet filtering method, and different lead electrocardiosignals in the multi-lead electrocardiosignals are spliced to form new long-chain electrocardiosignals as elements of an electrocardiosignal training set; calibrating tag vectors of the electrocardiosignals according to types, wherein the tag vectors of different types are different and are (a1, a2, … …, at), only one of a1, a2 and … …, and the rest are 0;
the fully-connected layer comprises a hidden layer and an output layer, wherein the excitation function in the hidden layer is a high-order polynomial activation function f (x) anxn+an-1xn-1+…+a1x+a0N is a natural number, and the excitation function in the output layer adopts a softmax function; the loss function in the fully connected layer is categoratic _ cross;
training the convolutional neural network by using the long-chain electrocardiosignal data of the training set as input and the corresponding label vector as output, and determining the parameters of each convolutional layer, each pooling layer, each flat layer and each full-connection layer;
the outputs of the fully-connected layers are the tag vectors (a1, a2, … …, at), only one of a1, a2, … …, at is 1, and the rest are 0.
2. The ST-segment classification neural network of higher-order polynomial activation functions of claim 1, wherein n-5 in higher-order polynomial activation function, and a5＝0.001，a3＝-0.0060，a1＝0.2003，a0＝0.5000。
3. The ST-segment classification neural network of higher-order polynomial activation functions of claim 1 or 2,
the convolutional Layer and the pooling Layer have 18 layers of Layer1-Layer 18;
layer1 is a one-dimensional convolution Layer, wherein 5 convolution kernels with the size of 37 are in total, and the output is 5 one-dimensional vectors;
layer2 is a one-dimensional maximum pooling Layer, wherein the size and step length of the kernel are both 2, and the output is 5 one-dimensional vectors;
layer3 is a one-dimensional convolution Layer, wherein, 5 convolution kernels with the size of 31 are in total, and the output is 5 one-dimensional vectors;
layer4 is a one-dimensional maximum pooling Layer, wherein the size and step length of the kernel are both 2, and the output is 5 one-dimensional vectors;
layer5 is a one-dimensional convolution Layer, wherein 5 convolution kernels with the size of 29 are totally arranged, and 5 one-dimensional vectors with the output of 14948 are output;
layer6 is a one-dimensional maximum pooling Layer, wherein the size and step length of the kernel are both 2, and the output is 5 one-dimensional vectors;
layer7 is a one-dimensional convolution Layer, wherein, 10 convolution kernels with the size of 23 are in total, and the output is 10 one-dimensional vectors;
layer8 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer9 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 15 are totally arranged, and the output is 10 one-dimensional vectors;
layer10 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer11 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 9 are totally arranged, and the output is 10 one-dimensional vectors;
layer12 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer13 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 5 are in total, and the output is 10 one-dimensional vectors;
layer14 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer15 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 1 are in total, and the output is 10 one-dimensional vectors;
layer16 is a one-dimensional maximum pooling Layer, wherein the size and the step length of the kernel are both 2, and the output is 10 one-dimensional vectors;
layer17 is a one-dimensional convolution Layer, wherein 10 convolution kernels with the size of 1 are in total, and the output is 10 one-dimensional vectors;
layer18 is a one-dimensional maximum pooling Layer where the kernel size and step size are both 2 and the output is 10 one-dimensional vectors.
4. The ST-segment neural network of higher-order polynomial activation functions of claim 3, wherein the convolutional and pooling layers have ReLU functions as activation functions.
5. The ST-segment neural network with higher-order polynomial activation functions according to any one of claims 1-4, wherein different lead electrocardiographic signals in the multi-lead electrocardiographic signals are spliced to form a new long-chain electrocardiographic signal, the electrocardiographic signals are arranged and combined according to the sequence of different leads to obtain long-chain electrocardiographic signals in all arrangement sequences, and the long-chain electrocardiographic signals in all arrangement sequences are used as a training set together.
6. The ST-segment neural network with higher-order polynomial activation functions according to any one of claims 1-5, wherein when splicing different lead electrocardiographic signals in a multi-lead electrocardiographic signal to form a new long-chain electrocardiographic signal, the electrocardiographic signals are arranged and combined according to the sequence of different leads to obtain long-chain electrocardiographic signals in all arrangement sequences, the long-chain electrocardiographic signals in each different sequence form a training set respectively, and a plurality of different convolutional neural networks are obtained through training of different training sets.
7. The ST-segment neural network of higher-order polynomial activation functions of any one of claims 1 to 6, wherein the ST-segment anomaly types include ST-segment horizontal elevation, ST-segment horizontal depression, and ST-segment dorsum elevation, and constitute 4 types together with a normal electrocardiogram.
8. The ST-segment classification neural network of higher-order polynomial activation functions according to any one of claims 1-5, wherein the training algorithm used in training is: a random gradient descent algorithm, an Adam algorithm, a RMSProp algorithm, an adagard algorithm, an adapelta algorithm, an Adamax algorithm.
9. A method for applying an ST-segment classification neural network of a high-order polynomial activation function is characterized by comprising the following steps:
s1: acquiring multi-lead electrocardiosignals, wherein the sampling frequencies of the multi-lead electrocardiosignals are the same or are preprocessed to be the same, the multi-lead electrocardiosignals are filtered by the same mean filter and then filtered by a wavelet filtering method, and different lead electrocardiosignals in the multi-lead electrocardiosignals are spliced to form new long-chain electrocardiosignals;
s2: inputting a new long chain electrocardiographic signal into an ST-segment classification neural network of a higher order polynomial activation function as claimed in any one of claims 1-8;
s3: and determining the type of the multi-lead electrocardiogram according to the label vector output by the ST-segment classification neural network of the high-order polynomial activation function.
10. A method for applying an ST-segment neural network of higher order polynomial activation functions, comprising the steps of:
s1: acquiring multi-lead electrocardiosignals, wherein the sampling frequencies of the multi-lead electrocardiosignals are the same or are preprocessed to be the same, the multi-lead electrocardiosignals are filtered by the same mean filter and then filtered by a wavelet filtering method, different lead electrocardiosignals in the multi-lead electrocardiosignals are spliced to form new long-chain electrocardiosignals, and the long-chain electrocardiosignals in all arrangement sequences are acquired by arranging and combining according to different lead sequences;
s2: respectively inputting the long-chain electrocardiosignals in different sequences into a convolutional neural network trained by a training set in a corresponding sequence as claimed in claim 1;
s3: and if a certain label vector output by all the convolutional neural networks accounts for more than 80% of the total output number of the convolutional neural networks, determining the type of the multi-lead electrocardiogram by using the label vector.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911250657.2A CN110974211A (en) | 2019-12-09 | 2019-12-09 | ST-segment classification neural network of high-order polynomial activation function and application thereof |
| PCT/CN2020/109292 WO2021114704A1 (en) | 2019-12-09 | 2020-08-14 | St segment classification neural network of high-order polynomial activation function, and application thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911250657.2A CN110974211A (en) | 2019-12-09 | 2019-12-09 | ST-segment classification neural network of high-order polynomial activation function and application thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110974211A true CN110974211A (en) | 2020-04-10 |
Family
ID=70091307
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911250657.2A Pending CN110974211A (en) | 2019-12-09 | 2019-12-09 | ST-segment classification neural network of high-order polynomial activation function and application thereof |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN110974211A (en) |
| WO (1) | WO2021114704A1 (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112883803A (en) * | 2021-01-20 | 2021-06-01 | 武汉中旗生物医疗电子有限公司 | Deep learning-based electrocardiosignal classification method and device and storage medium |
| WO2021114704A1 (en) * | 2019-12-09 | 2021-06-17 | 上海数创医疗科技有限公司 | St segment classification neural network of high-order polynomial activation function, and application thereof |
| CN114662678A (en) * | 2022-05-17 | 2022-06-24 | 华中科技大学 | Variable activation function convolution neural network and training method thereof |
| CN111666865B (en) * | 2020-06-02 | 2023-05-23 | 上海数创医疗科技有限公司 | Multi-lead electrocardiosignal convolutional neural network classification method and application method thereof |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102961129A (en) * | 2012-10-26 | 2013-03-13 | 上海交通大学无锡研究院 | Method for analyzing abnormal electrocardiogram tension for remote medical care |
| CN105997055A (en) * | 2016-07-11 | 2016-10-12 | 吉林大学 | Automatic classification method, system and device of electrocardiosignal ST band |
| CN107007279A (en) * | 2017-03-17 | 2017-08-04 | 浙江大学 | A kind of noninvasive intracardiac exciting independent positioning method of exception based on stacking-type self-encoding encoder |
| CN107714023A (en) * | 2017-11-27 | 2018-02-23 | 乐普（北京）医疗器械股份有限公司 | Static ecg analysis method and apparatus based on artificial intelligence self study |
| CN109643397A (en) * | 2016-09-06 | 2019-04-16 | 日本电信电话株式会社 | Time series data characteristic amount extraction device, time series data Characteristic Extraction method and time series data Characteristic Extraction program |
| CN110151168A (en) * | 2019-06-12 | 2019-08-23 | 上海移视网络科技有限公司 | More disease identifications and warning algorithm |
| CN110414565A (en) * | 2019-05-06 | 2019-11-05 | 北京邮电大学 | A kind of neural network method of cutting out based on Group Lasso for power amplifier |
| CN110543901A (en) * | 2019-08-22 | 2019-12-06 | 阿里巴巴集团控股有限公司 | image recognition method, device and equipment |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10426364B2 (en) * | 2015-10-27 | 2019-10-01 | Cardiologs Technologies Sas | Automatic method to delineate or categorize an electrocardiogram |
| CN110141219B (en) * | 2019-06-20 | 2022-03-15 | 鲁东大学 | Lead fusion deep neural network-based myocardial infarction automatic detection method |
| CN110522440B (en) * | 2019-08-12 | 2021-04-13 | 广州视源电子科技股份有限公司 | Electrocardiosignal recognition device based on grouping convolution neural network |
| CN110974211A (en) * | 2019-12-09 | 2020-04-10 | 上海数创医疗科技有限公司 | ST-segment classification neural network of high-order polynomial activation function and application thereof |
-
2019
- 2019-12-09 CN CN201911250657.2A patent/CN110974211A/en active Pending
-
2020
- 2020-08-14 WO PCT/CN2020/109292 patent/WO2021114704A1/en active Application Filing
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102961129A (en) * | 2012-10-26 | 2013-03-13 | 上海交通大学无锡研究院 | Method for analyzing abnormal electrocardiogram tension for remote medical care |
| CN105997055A (en) * | 2016-07-11 | 2016-10-12 | 吉林大学 | Automatic classification method, system and device of electrocardiosignal ST band |
| CN109643397A (en) * | 2016-09-06 | 2019-04-16 | 日本电信电话株式会社 | Time series data characteristic amount extraction device, time series data Characteristic Extraction method and time series data Characteristic Extraction program |
| CN107007279A (en) * | 2017-03-17 | 2017-08-04 | 浙江大学 | A kind of noninvasive intracardiac exciting independent positioning method of exception based on stacking-type self-encoding encoder |
| CN107714023A (en) * | 2017-11-27 | 2018-02-23 | 乐普（北京）医疗器械股份有限公司 | Static ecg analysis method and apparatus based on artificial intelligence self study |
| CN110414565A (en) * | 2019-05-06 | 2019-11-05 | 北京邮电大学 | A kind of neural network method of cutting out based on Group Lasso for power amplifier |
| CN110151168A (en) * | 2019-06-12 | 2019-08-23 | 上海移视网络科技有限公司 | More disease identifications and warning algorithm |
| CN110543901A (en) * | 2019-08-22 | 2019-12-06 | 阿里巴巴集团控股有限公司 | image recognition method, device and equipment |
Non-Patent Citations (1)
| Title |
|---|
| 任晓霞: "基于Dropout 深度卷积神经网络的ST 段波形分类算法", 《传感技术学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021114704A1 (en) * | 2019-12-09 | 2021-06-17 | 上海数创医疗科技有限公司 | St segment classification neural network of high-order polynomial activation function, and application thereof |
| CN111666865B (en) * | 2020-06-02 | 2023-05-23 | 上海数创医疗科技有限公司 | Multi-lead electrocardiosignal convolutional neural network classification method and application method thereof |
| CN112883803A (en) * | 2021-01-20 | 2021-06-01 | 武汉中旗生物医疗电子有限公司 | Deep learning-based electrocardiosignal classification method and device and storage medium |
| CN112883803B (en) * | 2021-01-20 | 2023-09-01 | 武汉中旗生物医疗电子有限公司 | Electrocardiogram signal classification method, device and storage medium based on deep learning |
| CN114662678A (en) * | 2022-05-17 | 2022-06-24 | 华中科技大学 | Variable activation function convolution neural network and training method thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2021114704A1 (en) | 2021-06-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110974211A (en) | ST-segment classification neural network of high-order polynomial activation function and application thereof | |
| US9949714B2 (en) | Method, electronic apparatus, and computer readable medium of constructing classifier for disease detection | |
| KR102451795B1 (en) | ECG signal detection method | |
| CN110717415B (en) | ST-segment classification convolutional neural network based on feature selection and using method thereof | |
| CN108647565B (en) | Data preprocessing method for classifying electrocardiosignals based on deep learning model | |
| KR102463764B1 (en) | Arrhythmia detection method, device, electronic device and computer storage medium | |
| JP6986724B2 (en) | ECG interference identification method based on artificial intelligence | |
| CN110236520B (en) | Electrocardiogram type recognition device based on double convolution neural network | |
| CN110717416B (en) | Neural network training method for ST segment classification recognition based on feature selection | |
| CN110046604B (en) | Single-lead ECG arrhythmia detection and classification method based on residual error network | |
| CN108113666A (en) | Recognition methods, device and the equipment of atrial fibrillation signal | |
| Tadesse et al. | Cardiovascular disease diagnosis using cross-domain transfer learning | |
| CN110037682B (en) | Method for recognizing heart rhythm type based on improved convolutional neural network | |
| CN110037683B (en) | Improved convolutional neural network for recognizing heart rhythm type and training method thereof | |
| CN111666865A (en) | Multi-lead electrocardiosignal convolution neural network classification model and use method thereof | |
| CN116211322A (en) | Depression recognition method and system based on machine learning electroencephalogram signals | |
| Srivastava et al. | Channel self-attention deep learning framework for multi-cardiac abnormality diagnosis from varied-lead ECG signals | |
| CN115130509A (en) | Electrocardiosignal generation method based on conditional variational self-encoder | |
| CN112603324B (en) | Neural network training method based on improved loss function | |
| CN115177260A (en) | Intelligent electrocardio-heart sound diagnosis method and device based on artificial neural network | |
| CN112022140B (en) | Automatic diagnosis method and system for diagnosis conclusion of electrocardiogram | |
| CN112587152B (en) | 12-lead T wave extraction method fusing U-net network and filtering method | |
| CN112587149A (en) | Atrial premature beat target detection device based on convolutional neural network | |
| CN111657914A (en) | Electrocardiogram form recognition method and device based on deep learning | |
| CN110897630A (en) | Ventricular tachycardia and heart rhythm identification method and device based on transfer learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200410 |