CN108968951B

CN108968951B - Electrocardiogram detection method, device and system

Info

Publication number: CN108968951B
Application number: CN201810928372.9A
Authority: CN
Inventors: 张玮; 朱涛; 罗伟; 李毅
Original assignee: Wuhan Zoncare Bio Medical Electronics Co ltd
Current assignee: Wuhan Zoncare Bio Medical Electronics Co ltd
Priority date: 2018-08-15
Filing date: 2018-08-15
Publication date: 2021-06-22
Anticipated expiration: 2038-08-15
Also published as: CN108968951A

Abstract

The application provides an electrocardiogram detection method, device and system, wherein a received electrocardiogram to be detected is input into an SVM model, and output signal distribution and signal labels of the electrocardiogram to be detected are obtained; and when the signal label is an abnormal signal label, obtaining a target image according to the abnormal signal label and the output signal distribution, and inputting the target image into the large category detection model to obtain the target symptom category of the electrocardiogram to be detected. And detecting the syndrome subcategories of the electrocardiograms to be detected respectively through the first subcategory detection model and the second subcategory detection model, and comprehensively detecting the electrocardiograms to be detected according to the two obtained detection results and the target syndrome categories so as to finally determine the syndrome subcategories of the electrocardiograms to be detected. Therefore, on one hand, parameters do not need to be adjusted manually, and on the other hand, the detection accuracy is improved.

Description

Electrocardiogram detection method, device and system

Technical Field

The application relates to the field of medical image processing, in particular to an electrocardiogram detection method, device and system.

Background

The electrocardiogram is mainly used for reflecting the electric excitation process of the heart and is an important clinical means for doctors to perform heart examination and diagnosis. The electrocardiogram has strong complexity, and people of different ethnicity, sex and age have great difference under various pathological conditions. In actual clinical diagnosis, usually, a doctor judges and recognizes a electrocardiogram with the aid of a machine detection result in combination with his own clinical experience. In this case, due to the lack of knowledge expertise and experience accumulation of doctors, the detection results given by the machine are often too dependent, the accuracy is limited, and erroneous judgment of abnormal electrocardiograms is easily caused.

In the prior art, a neural network is generally adopted to classify electrocardiosignals so as to find different feature information contained in the electrocardiosignals in a time domain and a frequency domain, and the more kinds of feature extraction, the higher the classification accuracy. Common learning methods are based on cardiac beat type identification marked in an MIT arrhythmia database, normal abnormal signal identification or identification and classification aiming at a special pathological electrocardiosignal, however, the method is limited by the number of samples of the electrocardiosignal, and the universality of the method is poor.

In the prior art, feature information is usually extracted manually, and the type of the manually extracted feature information is limited, which can affect the accuracy of classification of electrocardiosignals.

Disclosure of Invention

In view of the above, it is an object of the present application to provide an electrocardiogram detection method, apparatus and system, which at least partially improve the above-mentioned problems.

In order to achieve the above purpose, the embodiment of the present application adopts the following technical solutions:

in a first aspect, an embodiment of the present application provides an electrocardiogram detection method, where the method includes:

inputting the received electrocardiogram to be detected into an SVM model to obtain the output signal distribution and signal labels of the electrocardiogram to be detected;

when the signal label is an abnormal signal label, obtaining a target image according to the abnormal signal label and the output signal distribution, and inputting the target image into a large-class detection model to identify a target symptom class corresponding to the electrocardiogram to be detected, wherein the large-class detection model is a CNN model;

respectively inputting target lead signals of the electrocardiogram to be detected to a first subcategory detection model and a second subcategory detection model to respectively obtain two probability groups, wherein each probability group comprises the probability that the electrocardiogram to be detected belongs to each syndrome subcategory, the first subcategory detection model is a CNN model, and the second subcategory detection model is an LSTM model;

and on the basis of a comprehensive detection model, jointly judging the symptom subcategory to which the electrocardiogram to be detected belongs according to the two probability groups and the target symptom category, wherein the comprehensive detection model is an Attention model.

In a second aspect, an embodiment of the present application further provides an electrocardiogram detection apparatus, which includes:

the abnormality judgment module is used for inputting the received electrocardiogram to be detected into an SVM model to obtain the output signal distribution and signal labels of the electrocardiogram to be detected;

the large-category detection module is used for obtaining a target image according to the abnormal signal mark and the output signal distribution when the signal label is the abnormal signal mark, and inputting the target image into a large-category detection model to identify a target symptom category corresponding to the electrocardiogram to be detected, wherein the large-category detection model is a CNN model;

the subcategory detection module is used for respectively inputting target lead signals of the electrocardiogram to be detected to a first subcategory detection model and a second subcategory detection model to respectively obtain two probability groups, wherein each probability group comprises the probability that the electrocardiogram to be detected belongs to each symptom subcategory, the first subcategory detection model is a CNN model, and the second subcategory detection model is an LSTM model;

and the comprehensive detection module is used for jointly judging the symptom subcategory to which the electrocardiogram to be detected belongs according to the two probability groups and the target symptom category based on a comprehensive detection model, wherein the comprehensive detection model is an Attention model.

In a third aspect, an embodiment of the present application further provides an electrocardiogram detection system, where the system includes a data server, a deep learning server, and an application server, which are communicatively connected to each other;

the application server is communicated with a user terminal through a network so as to forward the electrocardiogram to be detected, which is sent by the user terminal, to the deep learning server;

the deep learning server comprises a processor and a machine-readable storage medium, wherein machine-readable instructions are stored on the machine-readable storage medium and when executed, the machine-readable instructions cause the processor to implement the electrocardiogram detection method provided by the embodiment of the application to detect the electrocardiogram to be detected sent by the application server;

the data server is used for providing training samples for the deep learning server so as to enable the deep learning server to carry out deep learning.

Compared with the prior art, the embodiment of the application has the following beneficial effects:

according to the electrocardiogram detection method, device and system, the received electrocardiogram to be detected is input into an SVM model, and output signal distribution and signal labels of the electrocardiogram to be detected are obtained; and when the signal label is an abnormal signal label, obtaining a target image according to the abnormal signal label and the output signal distribution, and inputting the target image into the large category detection model to obtain the target symptom category of the electrocardiogram to be detected. And detecting the syndrome subcategories of the electrocardiograms to be detected respectively through the first subcategory detection model and the second subcategory detection model, and comprehensively detecting the electrocardiograms to be detected according to the two obtained detection results and the target syndrome categories so as to finally determine the syndrome subcategories of the electrocardiograms to be detected. Therefore, on one hand, parameters do not need to be adjusted manually, and on the other hand, the detection accuracy is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained from the drawings without inventive effort.

Fig. 1 is a schematic connection diagram of an electrocardiogram detection system according to an embodiment of the present application;

fig. 2 is a schematic block diagram of a deep learning server according to an embodiment of the present disclosure;

fig. 3 is a schematic flowchart of an electrocardiogram detection method according to an embodiment of the present application;

fig. 4 is a schematic functional block diagram of an electrocardiogram detection apparatus according to an embodiment of the present application.

Icon: 10-an electrocardiogram detection system; 11-a deep learning server; 111-a memory; 112-a processor; 113-a communication unit; 12-a data server; 13-an application server; 20-a user terminal; 30-an electrocardiogram detection device; 31-an abnormality judgment module; 32-large category detection module; 33-subcategory detection module; 34-comprehensive detection module.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. The components of the embodiments of the present application, generally described and illustrated in the figures herein, can be arranged and designed in a wide variety of different configurations.

Thus, the following detailed description of the embodiments of the present application, presented in the accompanying drawings, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined and explained in subsequent figures.

Fig. 1 is a schematic connection diagram of an electrocardiogram detection system 10 according to an embodiment of the present application. The electrocardiogram detection system 10 comprises a deep learning server 11, a data server 12 and an application server 13 which are connected with each other in a communication way, wherein the deep learning server 11, the data server 12 and the application server 13 can be positioned in the same local area network.

The application server 13 may be a web application server, which serves as an interface for the electrocardiographic detection system 10 to communicate with external devices, and may be communicatively connected to the user terminal 20 through an ethernet network. In the embodiment of the present application, the user terminal 20 may be any device having a data processing function and a communication function, such as a Personal Computer (PC), a smart phone, and a tablet computer.

In implementation, a user may send an electrocardiogram to be detected to the application server 13 through the user terminal 20, the application server 13 sends the electrocardiogram to be detected to the deep learning server 11, and the deep learning server 11 is configured to detect the electrocardiogram to be detected by the electrocardiogram detection method provided in the embodiment of the present application. The data server 12 is used for providing training samples to the deep learning server 11 for deep learning.

Optionally, in this embodiment of the application, the data server 12 may be a single server, or may be a data server cluster formed by multiple data servers. Correspondingly, the deep learning server 12 may be a single server, or may be a deep learning server cluster composed of a plurality of deep learning servers. The data server cluster and the deep learning server cluster can be realized in a distributed cluster mode to improve data access speed and perform redundant backup.

Referring to fig. 2, fig. 2 is a block diagram of a deep learning server 11 according to an embodiment of the present disclosure. The deep learning server 11 includes an electrocardiogram detection apparatus 30, a memory 111, a processor 112, and a communication unit 113.

The memory 111, the processor 112 and the communication unit 113 are electrically connected to each other directly or indirectly to realize data transmission or interaction. For example, the components may be electrically connected to each other via one or more communication buses or signal lines. The ecg detecting device 30 includes at least one software function module which can be stored in the memory 111 in the form of software or firmware (firmware) or is fixed in an Operating System (OS) of the deep learning server 11. The processor 112 is used for executing executable modules stored in the memory 111, such as software functional modules and computer programs included in the electrocardiogram detection apparatus 30.

The Memory 111 may be, but is not limited to, a Random Access Memory (RAM), a Read Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Read-Only Memory (EPROM), an electrically Erasable Read-Only Memory (EEPROM), and the like.

The processor 112 may be an integrated circuit chip having signal processing capabilities. The Processor may be a general-purpose Processor, including a Central Processing Unit (CPU), a Network Processor (NP), and the like; but may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components. The various methods, steps, and logic blocks disclosed in the embodiments of the present application may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The communication unit 113 is configured to establish a communication connection between the deep learning server 11 and another device, for example, a communication connection between the data server 12 and the application server 13 may be established through a local area network, and for example, a communication connection between the user terminal 20 may be established through an ethernet network.

It should be understood that the structure shown in fig. 2 is only an illustration, and the deep learning server 11 may also include more or less components than those shown in fig. 2, or have a completely different configuration from that shown in fig. 2. Further, the components shown in FIG. 2 may be implemented in software, hardware, or a combination thereof.

It should be noted that, in the embodiment of the present application, components included in the data server 12, the application server 13, the user terminal 20, and the like, and connection relationships between the components may be similar to those of the deep learning server 11, and are not described herein again.

Referring to fig. 3, fig. 3 is a flowchart illustrating an electrocardiogram detection method according to an embodiment of the present application, where the method can be applied to the deep learning server 11 shown in fig. 2, and the method includes various steps which will be described in detail below.

And step S31, inputting the received electrocardiogram to be detected into an SVM model to obtain the output signal distribution and signal labels of the electrocardiogram to be detected.

The electrocardiogram to be detected is converted into a digitized electrocardiosignal to be detected, and then is input into an SVM (support vector machine) model, and the SVM model is used for judging whether the input electrocardiosignal is an abnormal signal.

Wherein, the electrocardiosignal of waiting to detect includes 12 lead signals, 12 lead signals are respectively: I. II, III, aVR, aVL, aVF, V1, V2, V3, V4, V5 and V6. In implementation, 12 lead signals may be input as one picture, or the 12 lead signals may be processed into 12 pictures respectively, and then the 12 pictures are taken together as input signals; or 6 of the 12 lead signals are stored as an image, the other 6 lead signals are stored as an image, and then the two images are taken as input signals; or a 3 × 4 mode is adopted, that is, every 3 lead signals are stored as one picture to obtain 4 pictures, and then the 4 pictures are used as input signals. The present embodiment does not limit this.

In practical applications, the waveform of the normal electrocardiogram of people of different sexes and ages is different, for example: infants between 0 and 3 years old, children between 3 and 6 years old, and children between 6 and 12 years old have electrocardiograms that are significantly different from those of adults due to their cardiac victory anatomical features. In the conventional normal electrocardiogram recognition standard, comprehensive judgment is carried out according to different forms and variation trends of heart rate, P-R interval, QRS time limit, QTc interval, QRS wave electric axis, P wave, QRS wave and T wave in 12 lead signals so as to determine whether the electrocardiogram is abnormal or not. Correspondingly, in the embodiment of the present application, it is first determined whether the electrocardiosignal to be detected is an abnormal signal.

During implementation, a series of calculation operations are performed on the electrocardiosignal to be detected so as to map the electrocardiosignal to an output space from an input space, and through a series of processing such as projection, transformation and mapping, continuous iteration, back propagation calculation and correction are performed on the calculation result of the output space and the output label, so that a series of data distribution combinations representing different output labels are finally obtained. By adopting the SVM model, the minimization of risk and confidence range can be realized, thereby achieving the purpose of obtaining good statistical rules under the condition of less statistical sample quantity.

In this embodiment, the SVM model maps the data of the electrocardiosignal to be detected to a high-dimensional space through a kernel function K, so as to solve the problem that the linearity is not separable in the original space. The objective function of the SVM model may be:

where phi denotes the feature space mapping, i.e. the dual form transformation, of the input signals x to f. In this embodiment, the kernel function needs to satisfy: k (x, z) ═ K<φ(x)·φ(z)>Performing inner product calculation in the feature space, which may adopt a gaussian kernel function, specifically as follows:

in this embodiment, the value of the signal label output by the SVM model may be 1 or-1, where 1 is a normal signal label and-1 is an abnormal signal label. In other words, when the output signal label is 1, the electrocardiosignal to be detected is a normal signal label; and when the output signal label is-1, the electrocardiosignal to be detected is represented as an abnormal signal.

Step S32, when the signal label is an abnormal signal label, obtaining a target image according to the abnormal signal label and the output signal distribution, and inputting the target image into a large-class detection model to identify the target symptom class corresponding to the electrocardiogram to be detected, wherein the large-class detection model is a CNN model.

Wherein the target image can be obtained by the following process: and performing pixel value normalization processing on the abnormal signal marks and the output signal distribution, specifically normalizing to 0-255, and splicing the obtained normalized data above a standard electrocardiogram to obtain the target image with the size of 200 x 1000.

In practical applications, there are usually six disease categories of normal electrocardio, arrhythmia, indoor conduction block, atrioventricular hypertrophy, ST segment abnormality and myocardial infarction, wherein each disease category comprises a plurality of disease sub-categories. In this embodiment, for each symptom category, at least 10 symptom subcategories with the highest frequency of occurrence in the symptom category are determined, for each determined symptom subcategory, at least 10000 adult electrocardiograms of the symptom subcategories are obtained, and waveform signals in each electrocardiogram are converted into corresponding digital signals, so as to obtain corresponding electrocardiosignal samples. Thus, the electrocardiographic signal samples corresponding to the electrocardiographic reports form a training sample library. Thus, the training sample library includes 51 syndrome categories, each 10000 data.

In this embodiment, the SVM model, the large class detection model, and other subsequent models may be trained through the training sample library.

In the digital processing process, assuming that each disease symptom category in the training sample library includes 10 disease symptom subcategories, the categories can be numbered, wherein the number corresponding to the normal electrocardiogram is 1000, and the disease symptom subcategories in the remaining five disease symptom categories are 2001-. Correspondingly, the gender may be represented by 0 and 1, e.g., 1 for male and 0 for female. The age can be entered directly as a numerical value.

Optionally, in this embodiment, the step of training the large category detection model according to the training sample library may include the following sub-steps:

firstly, aiming at each training sample, when the abnormal signal mark and the output signal distribution of the training sample are obtained through the SVM model, normalization processing is carried out on the abnormal signal mark and the output signal distribution, and the obtained normalized data are spliced above a standard electrocardiogram to obtain a first image; and

secondly, acquiring patient information of the training sample, and combining an electrocardiosignal sample obtained according to the training sample with the patient information to generate a second image, wherein the electrocardiosignal sample comprises 12 lead signals of the training sample.

Wherein the size of the second image may be 1000 × 1000 pixels.

Thirdly, training the large-category detection model through the first image and the second image, and adjusting parameters of the large-category detection model in a training process to enable the large-category detection model to accurately identify the target symptom category of each training sample.

In this embodiment, the large class detection model may include a plurality of combinations of convolutional layers and downsampling layers, the plurality of combinations being connected in sequence, and an output of a downsampling layer in a previous combination being an input of a convolutional layer in a next combination. Wherein the plurality of combinations are respectively a first combination, a second combination, a third combination, a fourth combination and a fifth combination.

Considering the parameter sharing property of the CNN (Convolutional Neural Network) model and the specificity that the richer the eigenvalues are, the higher the accuracy is observed, the Convolutional layer C1 in the first combination may include 64 Convolutional kernels, which may employ the Sobel operator, and each Convolutional kernel has a size of 9 × 9. Wherein, the edge characteristic of the image can be enhanced by using the sobel operator. By convolutional layer C1, 200 different features of the input image can be obtained, for a total of 64 feature maps, each feature map having a size of 1192 × 992.

Alternatively, in convolutional layer C1, there may be 100 connection weight parameters per neuron.

Alternatively, the convolutional layer C1 may use a linear correction unit PRuLU function as the activation function, which is:

in this case, due to the negative half-axis slope

The PReLU convergence speed is faster because of the non-constant state, and the calculation process is as follows:

the downsampling layer S1 in the first combination may downsample the feature map output by the convolutional layer C1 in a 4 × 4 average pooling manner, and the size of the feature map may be reduced to 298 × 248 by the downsampling layer S1.

Optionally, in this embodiment, the convolutional layer C2 in the second combination may include 128 convolution kernels using sobel operators, each convolution kernel has a size of 5 × 5, and the convolutional layer C2 may use a tanh function as the activation function. Correspondingly, the down-sampling slice S2 in the second combination may be used to perform 2 × 2 down-sampling, and a feature map with a size of 147 × 122 may be obtained through the down-sampling slice S2.

Optionally, in this embodiment, the convolutional layer C3 in the third combination may be used to perform convolution processing on the output of the down-sampling layer S2, where the convolutional layer C3 includes 256 convolutional kernels with a size of 6 × 5, and the size of the feature map output by the convolutional layer C3 is 142 × 118. The downsampled slices S3 in the third set may be used to perform 2 × 2 average pooling, with an output signature size of 71 × 59.

Optionally, in this embodiment, the convolutional layer C4 in the fourth combination is used to convolve the downsampled layer S3, the convolutional layer C4 includes 512 convolution kernels with a size of 6 × 6, and the feature map output by the convolutional layer C4 has a size of 66 × 54. The down-sampled slice S4 in the fourth combination was used to perform an average pooling of 2 × 2 to output a signature of size 33 × 27.

Optionally, in this embodiment, the convolutional layer C5 in the fifth combination includes 512 convolutional kernels with size 4 × 4 to output a feature map with size 30 × 24; the downsampled slice S5 in the fifth set was used to perform 2 × 2 average pooling to output a signature of size 15 × 12. Wherein the fifth downsampling layer S5 (i.e., the last downsampling layer in the plurality of combinations) is sequentially connected with a full connection layer and an output layer, wherein the full connection layer may be activated by a sigmoid function, and the output layer may be a softmax layer.

In this embodiment, the output layer may include a plurality of output nodes, each output node corresponding to a category of medical conditions. In one embodiment, the output layer may include 7 output nodes corresponding to 7 cases of normal cardiac electricity, arrhythmia, indoor conduction block, atrioventricular hypertrophy, ST segment abnormalities, myocardial infarction, and others.

In this embodiment, the output result of the large category detection module may be input to two types of deep learning networks, namely CNN and LSTM (long short-term memory network), respectively, so as to detect the symptom subcategory.

Step S33, respectively inputting the target lead signals of the electrocardiogram to be detected to a first subcategory detection model and a second subcategory detection model to respectively obtain two probability groups, wherein each probability group comprises the probability that the electrocardiogram to be detected belongs to each symptom subcategory, the first subcategory detection model is a CNN model, and the second subcategory detection model is an LSTM model.

In one embodiment, the target lead signal may be a 12 lead signal. In yet another embodiment, since the III, aVL and aVF leads among the 12 lead signals are a linear combination of I, II leads, only 8 lead signals of I, II, V1, V2, V3, V4, V5 and V6 may be input as target lead signals into the above two subcategory detection models.

Optionally, step S33 may include the following sub-steps:

taking the target lead signal as an input signal, and sampling the input signal for a preset time length t at a sampling rate of 500;

combining the obtained sampling data and the patient information into t × 5000 target data, and inputting the target data into the first sub-category detection model and the second sub-category detection model respectively.

The preset time period t may be 10 seconds(s), and the unit of the sampling rate is Hz.

In the present embodiment, since the signal of each lead reflects a physiological electrical signal of a different part of the heart, the conduction potential of the heartbeat at the same time point is a highly correlated signal, and therefore, the electrocardiographic signal can be subjected to feature extraction by using a convolution kernel of 10 × 10 size.

Specifically, the first sub-category detection model may employ 3 convolutional layers, 3 pooling layers, 1 output layer, and 1 fully-connected layer output including 1024 neurons, and may finally output the probability of each sub-category (syndrome sub-category). The specific structures of the 3 convolutional layers and the 3 pooling layers are shown in the following table:

layer name	Convolution kernel size	Number of convolution kernels	Convolution step size
				Convolutional layer 1	10×10	64	1×2
Pooling layer 1	1×1	-	-
				Convolutional layer 2	1×10	128	1×5
Pooling layer 2	1×4	-	-
				Convolutional layer 3	10×10	512	1×10
Pooling layer 3	1×2	-	-

Regarding the second sub-category detection model, considering that the cardiac electrical signals are in a lead signal, points between each cardiac signal have time continuity, which reflects the comprehensive potential output of the heart during one beat, the potential is transmitted from the sinus node output to other parts of the atrium and the ventricle, and therefore, the cardiac electrical signals have continuous correlation. Based on this, the LSTM model can be used to implement the second sub-category detection model, because the LSTM model is a special rnn (recurrentneurnetwork) network, the long-term dependence problem can be solved well.

In this embodiment, the LSTM model may include 4 layers of structures, including 10 steps, the sequence length in a single data is 5000 points, the hidden layer includes 200 neurons, and 100000 training times are required.

In this embodiment, the LSTM model mainly includes three gates and one memory unit (cell), wherein the three gates are forgetgate, inputgate and outputgate, respectively. The forgetgate may be used to add and delete information to a cell through a gating cell. With the forgetgate, it is possible to choose whether or not to pass the information, which includes the sigmoid neural network layer and a pair-wise multiplication operation, whose output is a value between 0 and 1, which is used to indicate how much information is allowed to pass, where 0 indicates no passing at all and 1 indicates passing at all. The inputgate decides which values are used for updating through sigmoid, and the tanh layer is used for generating a new candidate value Ct which is possibly added to the cell as a candidate value generated by the current layer, and the generated values can be combined for updating. The Outputgate is used for determining the output of the LSTM model, and firstly obtains an initial output through a sigmoid layer, then uses a tanh function to scale the Ct value between-1 and 1, and then multiplies the Ct value by the output obtained by the sigmoid one by one to obtain the output of the model.

Step S34, based on a comprehensive detection model, judging the symptom subcategory to which the electrocardiogram to be detected belongs together according to the two probability groups and the target symptom category, wherein the comprehensive detection model is an Attention model.

After the syndrome subcategories are detected through a CNN (first subcategory detection model) and an LSTM (second subcategory detection model), multi-layer projection transformation and calculation are carried out on electrocardiosignals to be detected, and output probabilities corresponding to each syndrome subcategory are obtained. However, it is found through research that the number of disease symptom subcategories is large, and the finally calculated maximum probability value may be very small, which is not enough to prove that the disease symptom subcategories are a certain subcategory, so in the embodiment of the present application, the probabilities output by the first subcategory detection model and the second subcategory detection model are combined through a comprehensive detection model, and the target disease symptom category output by the large category detection model is used as prior knowledge, and the target lead signal jointly detects the disease symptom subcategories to which the electrocardiographic signals to be detected belong.

Under the condition of adding the outputs of the plurality of models, the electrocardiosignals to be detected are combined, so that the accuracy of the final classification result can be greatly improved, and the weights among the neurons of the comprehensive detection model can be well solved.

Fig. 4 is a functional block diagram of an electrocardiogram detection apparatus 30 according to an embodiment of the present application. The electrocardiogram detection apparatus 30 includes an abnormality determination module 31, a large category detection module 32, a sub-category detection module 33, and a comprehensive detection module 34.

The abnormality determining module 31 is configured to input the received electrocardiogram to be detected into an SVM model, so as to obtain output signal distribution and signal labels of the electrocardiogram to be detected.

In the present embodiment, the description of the abnormality determination module 31 may refer to the detailed description of step S31 shown in fig. 3, that is, step S31 may be executed by the abnormality determination module 31.

The large-category detection module 32 is configured to, when the signal tag is an abnormal signal tag, obtain a target image according to the abnormal signal tag and the output signal distribution, and input the target image into a large-category detection model to identify a target symptom category corresponding to the electrocardiogram to be detected, where the large-category detection model is a CNN model.

In the present embodiment, the description of the large category detecting module 32 may refer to the detailed description of step S32 shown in fig. 3, that is, step S32 may be executed by the large category detecting module 32.

The sub-category detection module 33 is configured to input the target lead signals of the electrocardiogram to be detected to a first sub-category detection model and a second sub-category detection model respectively, so as to obtain two probability groups, where each probability group includes a probability that the electrocardiogram to be detected belongs to each of the sub-categories of symptoms, where the first sub-category detection model is a CNN model, and the second sub-category detection model is an LSTM model.

In the present embodiment, the description of the sub-category detecting module 33 may refer to the detailed description of step S33 shown in fig. 3, that is, step S33 may be executed by the sub-category detecting module 33.

The comprehensive detection module 34 is configured to jointly determine a syndrome sub-category to which the electrocardiogram to be detected belongs according to the two probability groups and the target syndrome category based on a comprehensive detection model, where the comprehensive detection model is an Attention model.

In the present embodiment, the description of the comprehensive detection module 34 may specifically refer to the detailed description of step S34 shown in fig. 3, that is, step S34 may be executed by the comprehensive detection module 34.

In summary, according to the electrocardiogram detection method, device and system provided by the embodiment of the present application, the received electrocardiogram to be detected is input into the SVM model, so as to obtain the output signal distribution and signal label of the electrocardiogram to be detected; and when the signal label is an abnormal signal label, obtaining a target image according to the abnormal signal label and the output signal distribution, and inputting the target image into the large category detection model to obtain the target symptom category of the electrocardiogram to be detected. And detecting the syndrome subcategories of the electrocardiograms to be detected respectively through the first subcategory detection model and the second subcategory detection model, and comprehensively detecting the electrocardiograms to be detected according to the two obtained detection results and the target syndrome categories so as to finally determine the syndrome subcategories of the electrocardiograms to be detected. Therefore, on one hand, parameters do not need to be adjusted manually, and on the other hand, the detection accuracy is improved.

In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. The apparatus embodiments described above are merely illustrative, and for example, the flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

In addition, functional modules in the embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.

The above description is only for the specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present application, and shall be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. An electrocardiogram detection apparatus, characterized in that it comprises:

2. The electrocardiographic detection device according to claim 1, wherein the large category detection module is specifically configured to:

and carrying out pixel value normalization processing on the abnormal signal marks and the output signal distribution, and splicing the obtained normalized data above a standard electrocardiogram to obtain the target image.

3. The electrocardiographic detection device according to claim 1 or 2, characterized in that the device further comprises:

the training module is used for establishing a training sample library, wherein the training sample library comprises six disease symptom categories of normal electrocardio, arrhythmia, indoor conduction block, atrioventricular hypertrophy, ST segment abnormity and myocardial infarction, each disease symptom category comprises at least 10 disease symptom sub-categories with the highest frequency of occurrence, and each disease symptom sub-category comprises at least 10000 adult electrocardiograms;

and training the SVM model, the large category detection model, the first sub-category detection model, the second sub-category detection model and the comprehensive detection model according to the training sample library.

4. The electrocardiographic detection device according to claim 3, wherein the training module is specifically configured to:

for each training sample, when the abnormal signal mark and the output signal distribution of the training sample are obtained through the SVM model, normalization processing is carried out on the abnormal signal mark and the output signal distribution, and the obtained normalized data are spliced above a standard electrocardiogram to obtain a first image; and

acquiring patient information of the training sample, and generating a second image by combining an electrocardiosignal sample obtained according to the training sample and the patient information, wherein the electrocardiosignal sample comprises 12 lead signals of the training sample;

training the large-category detection model through the first image and the second image, and adjusting parameters of the large-category detection model in the training process to enable the large-category detection model to accurately identify the target symptom category of each training sample.

5. The electrocardiogram detection apparatus according to claim 1 or 2, wherein,

the objective function of the SVM model is as follows:

the kernel function of the SVM model is a Gaussian kernel function:

6. the electrocardiographic detection device according to claim 1 or 2, wherein the large class detection model comprises a plurality of combinations of convolutional layers and downsampling layers, the plurality of combinations are connected in sequence, the output of the downsampling layer in the former combination is the input of the convolutional layer in the next combination, the last downsampling layer is connected with a full connection layer and an output layer in sequence, the output layer comprises a plurality of output nodes, and each output node corresponds to a kind of symptom class; the target symptom categories are identified by:

acquiring a plurality of probability values respectively output by the plurality of output nodes;

and taking the symptom category corresponding to the maximum value in the probability values as the target symptom category.

7. The electrocardiographic detecting apparatus according to claim 4, wherein the target lead signals include 8 lead signals of I, II, V1, V2, V3, V4, V5 and V6 in the electrocardiogram to be detected.

8. The electrocardiographic detection device of claim 7 wherein the subcategory detection module is further specifically configured to:

9. An electrocardiogram detection system is characterized by comprising a data server, a deep learning server and an application server which are mutually connected in a communication way;

the deep learning server comprises a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-readable instructions which when executed cause the processor to implement the electrocardiogram detection apparatus according to any one of claims 1 to 8, so as to detect the electrocardiogram to be detected sent by the application server;