CN115067967B

CN115067967B - Method for determining heart beat signal datum point, method and device for identifying heart beat type

Info

Publication number: CN115067967B
Application number: CN202110260893.3A
Authority: CN
Inventors: 赵巍; 李振齐; 胡静
Original assignee: Guangzhou Xicoo Medical Technology Co ltd; Guangzhou Shiyuan Electronics Thecnology Co Ltd; Guangzhou Shiyuan Artificial Intelligence Innovation Research Institute Co Ltd
Current assignee: Guangzhou Xicoo Medical Technology Co ltd; Guangzhou Shiyuan Electronics Thecnology Co Ltd; Guangzhou Shiyuan Artificial Intelligence Innovation Research Institute Co Ltd
Priority date: 2021-03-10
Filing date: 2021-03-10
Publication date: 2024-10-18
Anticipated expiration: 2041-03-10
Also published as: CN115067967A

Abstract

The invention discloses a heart beat signal reference point determining method, a heart beat type identifying method and a heart beat signal reference point identifying device. After the electrocardiosignals are acquired, feature extraction is carried out on the electrocardiosignals, feature graphs of the electrocardiosignals on a plurality of different scales are obtained, probability graphs of the QRS complexes are determined based on the feature graphs, local maximum points with probability values larger than a probability threshold in the probability graphs represent candidate points of the position information of the QRS complexes, for each QRS complex, candidate points with the maximum probability values are screened out from the plurality of probability graphs to serve as target points representing the position information of the QRS complexes, the feature graphs corresponding to the target points serve as target feature graphs, and the reference points of the heart beat signals of the QRS complexes are determined based on the regional feature graphs on the target feature graphs. The detection is not required based on the preset anchor points and anchor frames, and only the reference points of the heart beat signals to which the QRS complex belongs are determined based on the regional feature images on the target feature images, so that the calculation resources are saved, and the detection efficiency of the reference points is improved.

Description

Method for determining heart beat signal datum point, method and device for identifying heart beat type

Technical Field

The embodiment of the invention relates to the technical field of electrocardiograms, in particular to a heart beat signal datum point determining method, a heart beat type identifying method and a heart beat type identifying device.

Background

With the aging development of society, the incidence and hazard of cardiovascular diseases are continuously increased. At present, the number of patients with cardiovascular diseases in China is about 2.6 hundred million, the death rate of the cardiovascular diseases caused by the death is the first leading component of urban and rural resident disease death, and the number of patients is still continuously increasing.

Electrocardiogram is taken as a routine physical examination item and has important significance for diagnosing and monitoring cardiovascular diseases. In clinical medicine, doctors conduct electrocardiographic analysis based mainly on electrocardiographic waveforms and measuring certain key parameters to determine patient conditions. Therefore, there is a need for an electrocardiographic analysis method that can quickly and accurately extract waveform information of an electrocardiogram and measure key parameters to assist doctors in working.

In the prior art, a mode of adding an anchor frame at an anchor point is generally adopted to detect a heart beat signal or a characteristic waveform. Namely, each point on the feature map is used as an anchor point, a plurality of anchor frames with different scales centering on the anchor point are established, features in the anchor frames are detected, and a beat signal or a feature waveform is determined. According to the scheme, all points on the feature map are required to be traversed, a plurality of anchor frames with different scales are established by taking each point as an anchor point, then target detection is carried out on the features in each anchor frame, the calculated amount is very large, and the detection efficiency is low.

Disclosure of Invention

The invention provides a heart beat signal datum point determining method, a heart beat type identifying method and a heart beat type identifying device, so as to improve datum point detection efficiency.

In a first aspect, an embodiment of the present invention provides a method for determining a reference point of a heart beat signal, including:

acquiring an electrocardiosignal, wherein the electrocardiosignal comprises a plurality of heart beat signals;

extracting features of the electrocardiosignals to obtain feature graphs of the electrocardiosignals in a plurality of different scales;

Determining a probability map of the QRS complex based on the characteristic map, wherein local maximum points with probability values larger than a probability threshold value in the probability map represent candidate points of the position information of the QRS complex;

For each QRS complex, screening candidate points with the maximum probability value from a plurality of probability maps to be used as target points for representing the position information of the QRS complex, and taking a feature map corresponding to the target points as a target feature map;

and determining a reference point of the heart beat signal to which the QRS complex belongs based on a regional characteristic diagram on the target characteristic diagram, wherein the regional characteristic diagram is a characteristic in a preset range taking the target point as the center on the target characteristic diagram.

In a second aspect, an embodiment of the present invention further provides a method for identifying a beat type, where the method for determining a beat signal reference point based on the first aspect of the present invention determines a beat signal reference point, where the beat signal reference point includes a start point and an end point of the beat signal, and the method for identifying a beat type includes:

inputting the heart beat feature images in the range of the starting point and the ending point of the heart beat signals on the target feature image into a trained heart beat type identification model for processing to obtain probability values of the heart beat signals belonging to each heart beat type;

And taking the heart beat type corresponding to the maximum value in the probability value as the heart beat type to which the heart beat signal belongs.

In a third aspect, an embodiment of the present invention further provides a device for determining a reference point of a heart beat signal, including:

the electrocardiosignal acquisition module is used for acquiring electrocardiosignals, and the electrocardiosignals comprise a plurality of heart beat signals;

the characteristic diagram extracting module is used for extracting characteristics of the electrocardiosignals to obtain characteristic diagrams of the electrocardiosignals in a plurality of different scales;

A probability map determining module, configured to determine a probability map of a QRS complex based on the feature map, where a local maximum point in the probability map, where the probability value is greater than a probability threshold, represents a candidate point of the location information of the QRS complex;

A target determining module, configured to, for each QRS complex, screen out candidate points with the largest probability values from a plurality of the probability maps as target points representing location information of the QRS complex, and use feature maps corresponding to the target points as target feature maps;

the reference point determining module is used for determining the reference point of the heart beat signal to which the QRS complex belongs based on a regional characteristic diagram on the target characteristic diagram, wherein the regional characteristic diagram is a characteristic in a preset range with the target point as the center on the target characteristic diagram.

In a fourth aspect, an embodiment of the present invention provides a beat type recognition device, where the beat signal reference point determining device provided in the third aspect of the present invention determines a reference point of a beat signal, where the reference point of the beat signal includes a start point and an end point of the beat signal, and the beat type recognition device includes:

the probability value acquisition module is used for inputting the heart beat characteristic images in the range of the starting point and the end point of the heart beat signals on the target characteristic image into a trained heart beat type identification model for processing to obtain probability values of the heart beat signals belonging to each heart beat type;

And the heart beat type determining module is used for taking the heart beat type corresponding to the maximum value in the probability value as the heart beat type to which the heart beat signal belongs.

In a fifth aspect, an embodiment of the present invention further provides a computer apparatus, including:

One or more processors;

a storage means for storing one or more programs;

The one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method for determining a beat signal reference point as provided in the first aspect of the present invention, or to implement the method for identifying a beat type as provided in the second aspect of the present invention.

In a sixth aspect, an embodiment of the present invention further provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a method for determining a beat signal reference point as provided in the first aspect of the present invention, or implements a method for identifying a beat type as provided in the second aspect of the present invention.

According to the heart beat signal reference point determining method provided by the embodiment of the invention, after the electrocardiosignal is acquired, feature extraction is carried out on the electrocardiosignal, feature graphs of the electrocardiosignal on a plurality of different scales are obtained, probability graphs of the QRS complex are determined based on the feature graphs, local maximum value points with probability values larger than a probability threshold in the probability graphs represent candidate points of the position information of the QRS complex, for each QRS complex, candidate points with the maximum probability values are screened out from the plurality of probability graphs and serve as target points for representing the position information of the QRS complex, the feature graph corresponding to the target points serves as a target feature graph, and the reference point of the heart beat signal of the QRS complex is determined based on the regional feature graph on the target feature graph. And the process of traversing all points on the target feature map as anchor points, establishing a plurality of anchor frames with different scales by taking the anchor points as the centers and detecting each anchor frame is not required, and only the reference point of the heart beat signal to which the QRS complex belongs is determined based on the regional feature map on the target feature map, so that the calculation resources are saved, and the detection efficiency of the reference point is improved.

Drawings

FIG. 1A is a flowchart of a method for determining a beat signal reference point according to an embodiment of the present invention;

FIG. 1B is a schematic block diagram of a heartbeat signal provided by an embodiment of the present invention;

Fig. 2A and 2B are diagrams illustrating a method for determining a reference point of a heart beat signal according to a second embodiment of the present invention;

FIG. 2C is a diagram of the original electrocardiographic signal before noise reduction filtering;

FIG. 2D is a diagram of the electrocardiograph signal after noise reduction and filtering;

FIG. 2E is a schematic diagram of a network structure of a feature pyramid according to an embodiment of the present invention;

FIG. 2F is a schematic diagram of a first convolution block according to an embodiment of the present disclosure;

FIG. 2G is a schematic diagram of a second convolution block according to an embodiment of the present disclosure;

FIG. 2H is a flow chart of determining target points from feature maps according to an embodiment of the invention;

FIG. 2I is a diagram of an electrocardiograph signal according to an embodiment of the present invention;

FIG. 2J is a probability map after non-maximum suppression processing according to an embodiment of the present invention;

FIG. 2K is a flowchart illustrating the detection of the offset and the width of the beat signal according to the embodiment of the present invention;

FIG. 2L is a schematic diagram of a heart beat datum point detected according to an embodiment of the present invention;

FIG. 3A is a flowchart of a method for identifying a beat type according to a third embodiment of the present invention;

FIG. 3B is a schematic diagram of a heart beat type recognition model according to an embodiment of the present invention;

fig. 4 is a schematic structural diagram of a heart beat signal reference point determining device according to a fourth embodiment of the present invention;

Fig. 5 is a schematic structural diagram of a beat type recognition device according to a fifth embodiment of the present invention;

fig. 6 is a schematic structural diagram of a computer device according to a seventh embodiment of the present invention.

Detailed Description

The invention is described in further detail below with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting thereof. It should be further noted that, for convenience of description, only some, but not all of the structures related to the present invention are shown in the drawings.

Example 1

Fig. 1A is a flowchart of a method for determining a beat signal reference point according to an embodiment of the present invention, where the method may be implemented by a beat signal reference point determining device according to an embodiment of the present invention, and the device may be implemented by software and/or hardware, and is generally configured in a computer device, as shown in fig. 1A, and the method specifically includes the following steps:

s101, acquiring electrocardiosignals which comprise a plurality of heart beat signals.

Specifically, the heart is excited by the myocardium before and after the beat. During myocardial activation, a weak bioelectric current is generated. Thus, each cardiac cycle of the heart is accompanied by a bioelectrical change. Such bioelectrical changes may be communicated to various parts of the body surface. Because of the different tissues of each part of the body, the distance from the heart is different, and the potential of the electrocardiosignal at different parts of the body is also different. For a normal heart, the direction, frequency, intensity of such bioelectrical changes are regular. If the electric signals of different parts of the body surface are detected by the electrodes, the electric signals are amplified by an amplifier and are recorded by a recorder, and an Electrocardiogram (ECG) can be obtained.

An electrocardiogram is a graph in which the heart is excited successively by a pacing site, an atrium, and a ventricle in each cardiac cycle, and various potential changes are extracted from the body surface by an electrocardiograph along with changes in bioelectricity. Since each beat of the heart is regular, the waveform pattern in the electrocardiogram is also regular. The waveform diagram in an electrocardiogram, which may entirely represent one cardiac cycle of the heart, is referred to as a beat signal. An ECG recording typically contains up to one hundred thousand heart beat signals, and in an embodiment of the invention at least one continuous heart beat is intercepted from the ECG recording as an electrocardiographic signal. Specifically, an electrocardiogram may be obtained from the electrocardiographic examination results of the user.

Fig. 1B is a schematic diagram of a heart beat signal according to an embodiment of the present invention, and referring to fig. 1B, it can be seen that the heart beat signal includes a plurality of characteristic wave signals, such as P wave, QRS wave group, and T wave, and sometimes U wave. Wherein the horizontal axis represents the time axis, the time (ms) is the unit, the vertical axis represents the intensity of electrocardiosignals, and the intensity is characterized by voltage (mV). In clinic, a time axis is usually obtained by extending a straight section (TS section) between T wave and S wave in electrocardiosignals.

S102, extracting features of the electrocardiosignals to obtain feature graphs of the electrocardiosignals in a plurality of different scales.

In the embodiment of the invention, the characteristic extraction is carried out on the electrocardiosignal by adopting a deep neural network. In the deep neural network, the resolution of the feature map output by the low-level network is higher, and the feature map comprises more position and detail information, but the semanteme is lower and the noise is more because of less convolution. The feature map output by the high-level network has stronger semantic information, but has very low resolution and poorer perception capability on details. In the embodiment of the invention, a plurality of feature maps with different scales are extracted from the electrocardiosignal and used for detecting the subsequent datum point, the position, the detail information and the semantic information are considered, and the detection accuracy and the detection efficiency of the datum point are improved.

S103, determining a probability map of the QRS complex based on the feature map, wherein local maximum points with probability values larger than a probability threshold in the probability map represent candidate points of the position information of the QRS complex.

In the embodiment of the invention, the probability map of the QRS complex is obtained by processing a plurality of feature maps with different scales, such as convolution, activation and the like, wherein the probability map is a probability graph with time as an abscissa axis and a probability value as an ordinate axis, and reflects the probability that each moment (i.e. a certain point on the abscissa) in the electrocardiosignal represents the position information of the QRS complex. In the embodiment of the present invention, in order to save calculation resources, a local maximum point in the probability map, where the probability value is greater than the probability threshold (the probability threshold is set to 0.5, for example), is used as a candidate point representing the position information of the QRS complex.

And S104, screening candidate points with the maximum probability value from the multiple probability maps for each QRS complex as target points for representing the position information of the QRS complex, and taking the feature map corresponding to the target points as a target feature map.

In the embodiment of the invention, for each QRS complex, a candidate point with the highest probability value is selected from a plurality of probability maps to be used as a target point for representing the position information of the QRS complex, and then a feature map corresponding to the target point is used as a target feature map.

For each probability map, the probability value corresponding to the peak position of the R wave in one QRS complex is the maximum value of the probability values corresponding to all points in the current heart beat signal, so that the point representing the position information of the current QRS complex in each probability map can be determined, and then the probability values of the points representing the position information of the current QRS complex in each probability map are compared to obtain the point with the maximum probability value as the target point representing the position information of the current QRS complex. Then, the feature map corresponding to the target point is set as a target feature map (the target point is determined on the probability map of the target feature map).

In some embodiments of the present invention, the target point may be used as a reference point for locating the beat signal, and then the heart rate is calculated according to the time interval between two adjacent reference points; in other embodiments of the present invention, the beat type may also be identified based on the features of the reference point in the preset area on the corresponding target feature map, and embodiments of the present invention are not described in detail herein.

S105, determining a reference point of the heart beat signal to which the QRS complex belongs based on a regional feature map on the target feature map, wherein the regional feature map is a feature in a preset range taking the target point as the center on the target feature map.

In the embodiment of the invention, the regional characteristic diagram is a characteristic in a preset range with a target point as a center on the target characteristic diagram. For example, the preset range may be set based on clinical prior information, and the fiducial points for different signature waveforms on the heart beat signal may be determined based on different preset ranges. For example, based on the width of the clinical heart beat signal, the width of the preset range is set to be 1s, and the region feature map with the target point as the center and the width of 1s on the target feature map is obtained. For another example, based on the width of the clinical QRS complex, the width of the preset range is set to be 0.3s, and a region feature map with the width of 0.3s on the target feature map with the target point as the center is obtained. The resulting regional signature is then processed, e.g., convolved, pooled, etc., to detect therefrom a heart beat signal or QRS complex and to determine the start and end points of the heart beat signal or QRS complex. The start point and the end point of the heart beat signal and/or the start point and the end point of the QRS complex are used as datum points of the heart beat signal.

In the embodiment of the invention, the probability map of the QRS complex corresponding to the feature map is obtained by processing the feature map, then, for each QRS complex, candidate points with the largest probability value are screened out from a plurality of probability maps to serve as target points for representing the position information of the QRS complex, the feature map corresponding to the target points serves as a target feature map, and then, the reference point of the heart beat signal to which the QRS complex belongs (namely, the heart beat signal or the QRS complex is detected) is determined based on the regional feature map on the target feature map.

Example two

Fig. 2A and 2B are diagrams showing a method for determining a reference point of a heart beat signal according to a second embodiment of the present invention, where the method is refined based on the first embodiment, and detailed processes of steps in the foregoing embodiment are described in detail, as shown in fig. 2A and 2B, and the method includes:

S201, acquiring electrocardiosignals which comprise a plurality of heart beat signals.

The waveform diagram in an electrocardiogram, which may entirely represent one cardiac cycle of the heart, is referred to as a beat signal. An ECG recording typically contains up to one hundred thousand heart beat signals, and in an embodiment of the present invention, at least one continuous heart beat or a preset time period (e.g., 10 s) is taken from the ECG recording as an electrocardiographic signal. Specifically, an electrocardiogram may be obtained from the electrocardiographic examination results of the user.

In some embodiments of the present invention, to improve the quality of the electrocardiographic signals, the electrocardiographic signals (hereinafter referred to as raw electrocardiographic signals for convenience of distinction) intercepted from the ECG recordings may be preprocessed to obtain the desired electrocardiographic signals.

In the embodiment of the invention, the noise reduction and filtering processing can be performed on the original electrocardiosignal, so that the noise in the electrocardiosignal is reduced, and the signal to noise ratio is improved. Specifically, the noise reduction filtering process may include the steps of:

1. Slope suppression.

Specifically, a plurality of sample points are equidistantly arranged on an original electrocardiosignal in advance, the original electrocardiosignal is subjected to differential processing to obtain a slope between two adjacent sample points, then the slope between the two adjacent sample points is compared with a preset slope threshold value, the smaller value of the slope and the slope threshold value is taken to obtain a new slope, and finally the new slope is added with the slope sequentially from the first sample point.

2. Baseline drift was filtered out.

Specifically, the output signal of the step 1 is filtered by using an average filter (window is 1 s) to obtain a baseline, the baseline is filtered by using a low-pass filter with cutoff frequency of 0.5Hz, and finally the baseline is filtered from the output signal of the step 1.

3. High frequency noise is filtered out.

Specifically, the output signal of the step 2 is subjected to low-pass filtering by using a filter with a cut-off frequency of 35Hz, and high-frequency noise is filtered.

4. And (5) standardization.

Specifically, the output signal of step 3 may be processed into an "zero-mean, one-variance" electrocardiograph signal by using z-score normalization, where the specific formula of z-score is as follows:

wherein, the signal intensity in the electrocardiosignal is the average value of the intensity of the electrocardiosignal, and sigma is the standard deviation of the intensity of the electrocardiosignal.

Fig. 2C is an original electrocardiograph signal diagram before the noise reduction filtering process, and fig. 2D is an electrocardiograph signal diagram after the noise reduction filtering process, and as can be seen from fig. 2C and fig. 2D, after the noise reduction filtering process, baseline drift and noise signals in the original electrocardiograph signal are filtered, so that the signals have better intelligibility.

In the embodiment of the invention, the electrocardiosignal can be a multi-lead signal, namely, the multi-lead electrocardiosignal is obtained by collecting the electrical signals of a plurality of parts of the body. After noise reduction and filtering processing is carried out on the original electrocardiosignal, quality screening can be carried out on the multi-lead signal, and 2 leads with the best signal quality are selected, wherein the specific flow is as follows:

And carrying out Fourier transformation on the electrocardiosignal subjected to noise reduction and filtering, converting the electrocardiosignal from a time domain to a frequency domain, and calculating the duty ratio of the energy of a 15-30Hz frequency band (corresponding to the QRS complex) in the total signal energy. Then the leads are discarded if the ratio is less than 50% in order from the top. The priority is selected according to the leads (the priority is ordered as II leads, V1 leads, V5 leads, I leads, V2 leads, V3 leads, V4 leads, V6 leads) and the energy ratio of the QRS complex frequency band is 2 leads. If the energy of the QRS complex band of the high priority lead is 0.9 times higher than the low priority, the high priority lead is selected, otherwise the low priority lead is selected. If only one lead is satisfactory, the lead is duplicated. If all the leads do not meet the requirements, a prompt message is sent out and the analysis algorithm is exited.

According to the embodiment of the invention, the electrocardiosignals with the plurality of leads are screened, and 2 leads with the best signal quality are screened out from the electrocardiosignals, so that the electrocardiosignals with the plurality of leads can be adapted, and the detection accuracy of the accuracy of heart beat signals is improved.

S202, inputting the electrocardiosignal into a pre-trained feature pyramid network to obtain a feature map with three scales.

Specifically, a feature pyramid network (Feature Pyramid Networks) is adopted to fuse the features output by the high-level network with the features output by the low-level network to obtain three-scale (sampling rate) feature images, namely a first feature image, a second feature image and a third feature image, wherein the first feature image, the second feature image and the third feature image are respectively 1 time, 0.5 time and 0.25 time of the length of an electrocardiosignal. In the embodiment of the invention, a plurality of feature maps with different scales are extracted from the electrocardiosignal and used for detecting the subsequent datum point, the position, the detail information and the semantic information are considered, and the detection accuracy and the detection efficiency of the datum point are improved.

Fig. 2E is a schematic diagram of a network structure of a feature pyramid according to an embodiment of the present invention, and as shown in fig. 2E, in an embodiment of the present invention, a feature pyramid network includes a first convolution block C0, a second convolution block C1, a first downsampling layer, a third convolution block C2, a second downsampling layer, a fourth convolution block C4, a first upsampling layer, a second upsampling layer, three first convolution layers, and two second convolution layers. The processing process of the characteristic pyramid network on the electrocardiosignals is as follows:

1. and inputting the electrocardiosignal into a first convolution block for processing to obtain a first characteristic.

Fig. 2F is a schematic structural diagram of a first convolution block according to an embodiment of the present disclosure, and as shown in fig. 2F, exemplary first convolution block C0 includes a convolution layer, a normalization layer, and an activation layer according to an embodiment of the present disclosure. In a specific embodiment of the invention, the sampling frequency of the electrocardiosignal is 250Hz, namely, 250 sample points are acquired for 1 s. Correspondingly, the convolution kernel length (k) of the convolution layer is 15 sample points, the convolution step length is 1 sample point, the filling size (p) is 7 sample points, and the channel number is 32. Wherein the length (k) and the filling size (p) of the convolution kernel satisfy (k-1)/2=p to ensure that the lengths of the electrocardiographic signals before and after convolution are consistent. The convolution layer carries out convolution processing on the input electrocardiosignal, extracts the characteristic representing the characteristic attribute of the electrocardiosignal, normalizes the characteristic output by the convolution layer through normalization processing of the normalization layer, finally, inputs the activation layer, and activates by Relu functions in the activation layer to obtain a first characteristic. Specifically, the Relu function has the expression:

f(x)＝max(0,x)

when the input is negative, then the Relu function dies out without activation at all. The Relu function output is either 0 or a positive number. The ReLU can overcome the problem of gradient disappearance and accelerate the training speed. It should be noted that, in other embodiments of the present invention, the activation function in the activation layer may be other activation functions, for example, sigmoid function or Tanh function, which are not limited herein.

2. And inputting the first characteristic into a second convolution block for processing to obtain a second characteristic.

Fig. 2G is a schematic structural diagram of a second convolution block according to an embodiment of the present disclosure, and as shown in fig. 2G, in an embodiment of the present disclosure, a second convolution block C1 includes a convolution layer 1, a normalization layer 1, an activation layer 1, a convolution layer 2, a normalization layer 2, and an activation layer 2. The convolution kernel lengths of the convolution layer 1 and the convolution layer 2 are 15 sample points, the convolution step sizes are 1 sample point, the filling size (p) is 7 sample points, and the channel number is 32. Wherein the length (k) and the filling size (p) of the convolution kernel satisfy (k-1)/2=p to ensure consistent lengths of features before and after convolution. The convolution layer 1 carries out convolution processing on the first input feature, outputs the result to the normalization layer 1, carries out normalization processing on the result output by the convolution layer 1 through the normalization layer 1, inputs the result to the activation layer 1, is activated by a Relu function in the activation layer 1, carries out convolution processing on the structure output by the activation layer 1, outputs the result to the normalization layer 2, carries out normalization processing on the result output by the convolution layer 2 through the normalization layer 2, adds the result with the input (namely the first feature) of the second convolution block C1, inputs the result to the activation layer 2, and is activated by a Relu function in the activation layer 2 to obtain the second feature.

3. And carrying out downsampling processing on the second feature to obtain a first downsampled feature.

In an exemplary embodiment of the present invention, the second feature is input into the first downsampling layer to perform downsampling processing, so as to obtain a first downsampled feature having a resolution half of that of the electrocardiograph signal. For example, the length of the electrocardiograph signal is 10s, the resolution is unchanged after the electrocardiograph signal is processed by the first convolution block C0 and the second convolution block C1, still is 10s, the length of the obtained first downsampling characteristic is 5s after the downsampling process of the first downsampling layer, and the resolution is half of that of the electrocardiograph signal.

4. And inputting the first downsampling characteristic into a third convolution block for processing to obtain a third characteristic.

And inputting the first downsampling characteristic into a third convolution block for processing to obtain a third characteristic. The structure of the third convolution block C2 is similar to that of the second convolution block C1, except that in the third convolution block C2, the number of channels of the convolution layer is 64, and by increasing the number of channels, the dimension of the feature is increased, so that the extracted feature has stronger semantic information. Specifically, the processing procedure of the third convolution block C2 may refer to the processing procedure of the second convolution block C1, which is not described herein.

5. And carrying out downsampling processing on the third feature to obtain a second downsampled feature.

Illustratively, in an embodiment of the present invention, the first downsampled feature is processed by the third convolution block C2 to obtain a third feature having a constant resolution. And after the third feature is subjected to downsampling treatment by the second downsampling layer, obtaining a second downsampled feature with the resolution of one fourth of that of the electrocardiosignal. For example, after two downsampling processes, the length of the resulting second downsampled feature is 2.5s, and the resolution becomes one fourth of the electrocardiographic signal. Exemplary downsampling embodiments may include random sampling, convolution, pooling, and the like, and embodiments of the invention are not limited in this regard.

6. And inputting the second downsampling characteristic into a fourth convolution block for processing to obtain a fourth characteristic.

And inputting the second downsampled feature into a fourth convolution block C3 for processing to obtain a fourth feature. The structure of the fourth convolution block C3 is similar to that of the second convolution block C1, except that in the fourth convolution block C3, the number of channels of the convolution layer is 128, and by increasing the number of channels, the dimension of the feature is increased, so that the extracted feature has stronger semantic information. Specifically, the processing procedure of the fourth convolution block C3 may refer to the processing procedure of the second convolution block C1, which is not described herein.

7. And carrying out convolution processing on the fourth feature to obtain a third feature map.

The fourth feature is input into a first convolution layer for convolution processing, and a third feature map T3 is obtained. In the embodiment of the invention, the length of the convolution kernels of the three first convolution layers is 1 sample point, the convolution step length is 1 sample point, the filling size is zero, and the channel numbers are 64. The resolution of the third profile T3 is one fourth of the electrocardiographic signal.

8. And carrying out up-sampling treatment on the third feature map, and carrying out feature fusion on the third feature map and the feature obtained after the convolution treatment on the third feature map to obtain a fifth feature.

The third feature map is input into a first upsampling layer for upsampling to obtain a first upsampled feature with half the resolution of the electrocardiograph signal, and the first upsampled feature and the feature of the third feature subjected to convolution processing by the first convolution layer are added to obtain a fifth feature.

9. And carrying out convolution processing on the fifth feature to obtain a second feature map.

Illustratively, the fifth feature is input into a second convolution layer for processing to obtain a second feature map. In the embodiment of the invention, the length of the convolution kernels of the two second convolution layers is 1 sample point, the convolution step length is 1 sample point, the filling size is zero, and the channel number is 64. The resolution of the second feature map T2 is half of the electrocardiographic signal.

10. And carrying out up-sampling treatment on the fifth feature, and carrying out feature fusion on the fifth feature and the feature obtained after the convolution treatment on the fifth feature to obtain a sixth feature.

The fifth feature map is input into a second upsampling layer for upsampling to obtain a second upsampled feature with the same resolution as the electrocardiosignal, and the second upsampled feature and the feature of the second feature subjected to convolution processing by the first convolution layer are added to obtain a sixth feature. By way of example, specific ways of upsampling may include linear interpolation, transpose convolution (also referred to as deconvolution, fractional step convolution), pooling, and the like, and embodiments of the present invention are not limited in this regard.

11. And carrying out convolution processing on the sixth feature to obtain a first feature map.

Illustratively, the sixth feature is input into a second convolution layer for processing, so as to obtain a first feature map T1, where the resolution of the first feature map T1 is the same as that of the electrocardiographic signal.

And (3) determining probability maps of the QRS complex corresponding to each feature map based on the three feature maps with different scales after obtaining the three feature maps with different scales, and specifically referring to step S203-step S205.

S203, carrying out convolution processing on the feature map to obtain a first sub-feature map.

Fig. 2H is a flowchart of determining a target point from feature maps according to an embodiment of the present invention, and exemplary, as shown in fig. 2H, for each feature map, the feature map is input into a convolution layer for convolution processing, so as to obtain a first sub-feature map. The convolution kernel of the convolution layer has a length of 15 sample points, a filling size of 7 sample points, a convolution step length of 1 sample point, and a channel number of 1.

S204, activating the first sub-feature map to obtain a second sub-feature map.

The first sub-feature map is input into an activation layer for processing, and a second sub-feature map is obtained. The activation function in the activation layer is also a Sigmoid function. The Sigmoid function maps the input variable (i.e., the value in the first sub-feature map) to a value between (0, 1). Specifically, the mathematical expression of the Sigmoid function is:

wherein x is a numerical value in the first sub-feature map.

S205, performing smoothing processing on the second sub-feature map to obtain a probability map of the QRS complex.

Illustratively, the second sub-feature map is input into a smoothing layer for smoothing. In the embodiment of the invention, the smoothing processing is essentially to use a convolution check with 15 sample points, 1 sample point in convolution step length and 7 sample points in filling size to carry out convolution processing on the second sub-feature map, so that the second sub-feature map becomes smooth, and a probability map of the QRS complex corresponding to each feature map is obtained.

The probability map is a probability graph taking time as an abscissa axis and taking a probability value as an ordinate axis, and reflects the probability that each moment (namely a certain point on the abscissa) in the electrocardiosignal represents the position information of the QRS complex. In the embodiment of the present invention, a local maximum point in the probability map, the probability value of which is greater than the probability threshold (the probability threshold is set to 0.5, for example), is used as a candidate point representing the position information of the QRS complex.

In the embodiment of the invention, local maximum points with probability values larger than a probability threshold (the probability threshold is set to 0.5 as an example) in the probability map are taken as candidate points representing the position information of the QRS complex, and the maximum value is screened out of the candidate points.

Fig. 2I is an electrocardiograph signal diagram provided by the embodiment of the present invention, fig. 2J is a non-maximal value suppression processed probability diagram provided by the embodiment of the present invention, as shown in fig. 2J, in the probability diagram, a cross symbol represents an abscissa position of a local maximal value point whose probability value is smaller than a threshold value, a solid dot is reserved after the non-maximal value suppression processing, and an abscissa position of a local maximal value point whose probability value is greater than the threshold value, that is, a QRS complex position detected on a feature diagram corresponding to the probability diagram, where the probability value is the largest, and the abscissa position where the maximum value is located is the QRS complex position determined based on the feature diagram corresponding to the probability diagram, which is a potential target point.

In the embodiment of the present invention, the probability threshold may be adjusted according to different signal conditions, and for example, when the motion interference is strong and it is difficult to detect the target point of the QRS complex, the probability threshold may be set to 0.05.

Next, a candidate point with the maximum probability value is selected from the multiple probability maps by adopting a non-maximum suppression method as a target point for representing the position information of the QRS complex, and a feature map corresponding to the target point is taken as a target feature map, and specific reference is made to step S206-step S210.

S206, sorting all candidate points according to the descending order of the probability values, and generating a checking list.

And arranging all the candidate points obtained in the steps in descending order according to the size of the probability value from large to small to generate a checking list.

S207, taking the candidate point with the largest probability value in the checking list as a target point.

And taking the candidate point with the maximum probability value in the examination list as a target point, wherein the target point is used for representing the position information of the QRS complex where the target point is located.

S208, taking the feature map corresponding to the target point as a target feature map.

Specifically, after determining the target point, determining a feature map corresponding to the probability map where the target point is located, and taking the feature map corresponding to the target point as a target feature map.

S209, deleting all candidate points within a preset distance range of the target point.

Specifically, the target point and all candidate points within the preset distance range of the target point are deleted, so that only one corresponding target point and one feature map are ensured for each QRS complex.

The preset distance range may be adjusted according to different subjects, and for example, since the maximum heart rate of an adult is generally not more than 300bpm, the preset distance range may be set to 0.2s. The maximum heart rate of the infant does not exceed 350bpm, so the preset distance range can be adjusted to 0.15s.

S210, judging whether the checking list is empty.

Specifically, whether the current examination list is empty is determined, if not, S207 is executed again, and the candidate point with the largest probability value in the examination list is used as the target point, so the above process is repeated until the examination list is empty, step S211 is executed, and at this time, we determine the target point representing the position information of each QRS complex on the electrocardiographic signal and the target feature map corresponding to the target point.

S211, carrying out convolution processing on the first region feature map on the target feature map to obtain a third sub-feature map.

In the embodiment of the invention, the first region feature map is a feature in a first preset range with the target point as the center on the target feature map. Illustratively, the first preset range is a width (e.g., 1 s) of the heart beat signal determined based on clinical prior information, the first preset range being used to detect the heart beat signal. Fig. 2K is a flowchart of detecting the offset and the width of the beat signal in the embodiment of the present invention, and as shown in fig. 2K, an exemplary embodiment inputs a first region feature map with a width of 1s on the target feature map with the target point as the center into a convolution layer for convolution processing, so as to obtain a third sub feature map. The convolution kernel of the convolution layer has a length of 1 sample point, the convolution step length is 1 sample point, and the channel number is 2.

S212, carrying out global average pooling processing on the third sub-feature map to obtain a two-dimensional vector, wherein elements in the two-dimensional vector respectively represent a first offset of the middle point of the heart beat signal relative to the target point and the width of the heart beat signal.

For example, the third sub-feature map is input into a pooling layer, global average pooling is performed on the third sub-feature map, so as to obtain a two-dimensional vector [ ctr _offset1,width₁ ], an element ctr _offset1 in the two-dimensional vector represents a first offset of a midpoint of the heart beat signal to which the target point belongs relative to the target point, and width ₁ represents a width of the heart beat signal.

S213, determining a start point and an end point of the heart beat signal based on the first offset and the width of the heart beat signal, wherein the reference point of the heart beat signal comprises the start point and the end point of the heart beat signal.

Illustratively, in the embodiment of the present invention, the process of determining the start point and the end point of the beat signal based on the first offset and the width of the beat signal is as follows:

1. a first value is calculated based on a natural number e and an index of half the width of the heart beat signal.

2. And calculating the sum of the position of the target point and the first offset to obtain a second value.

3. And calculating the difference between the second value and the first value to obtain the position of the starting point of the heart beat signal.

4. And calculating the sum of the position of the starting point of the heart beat signal and the first numerical value to obtain the position of the end point of the heart beat signal.

The mathematical expression of the above determination process for determining the start and end points of the heart beat signal is as follows:

onset₁＝qrs_loc+ctr_offset1-exp(width₁)/2

offset₁＝onset₁+exp(width₁)/2

Where onset ₁ is the position of the start of the beat signal, offset ₁ is the position of the end of the beat signal, qrs _loc is the position of the target point (position on the abscissa).

S214, carrying out convolution processing on the second region feature map on the target feature map to obtain a fourth sub-feature map.

In the embodiment of the present invention, the second preset range may also be set for detecting QRS complexes. The second region feature map is a feature in a second preset range with the target point as the center on the target feature map. Illustratively, the second preset range is the width (e.g., 0.3 s) of the QRS complex determined based on clinical prior information. For the process of detecting the offset and width of the QRS complex, reference may be made to fig. 2K. And (3) inputting the second regional characteristic diagram with the width of 0.3s on the target characteristic diagram by taking the target point as the center into a convolution layer for convolution processing to obtain a fourth sub-characteristic diagram. The convolution kernel of the convolution layer has a length of 1 sample point, the convolution step length is 1 sample point, and the channel number is 2.

And S215, carrying out global average pooling processing on the fourth sub-feature map to obtain a two-dimensional vector, wherein elements in the two-dimensional vector respectively represent a second offset of the middle point of the QRS complex relative to the target point and the width of the QRS complex.

Illustratively, global average pooling is performed on the fourth sub-feature map to obtain a two-dimensional vector [ ctr _offset2,width₂ ], where the element ctr _offset2 in the two-dimensional vector represents a second offset of the midpoint of the QRS complex to which the target point belongs relative to the target point, and width ₂ represents the width of the QRS complex.

S216, determining a start point and an end point of the QRS complex based on the second offset and the width of the QRS complex, the fiducial point of the heart beat signal including the start point and the end point of the QRS complex.

Illustratively, in an embodiment of the present invention, the process of determining the start and end points of the QRS complex based on the second offset and the width of the QRS complex is as follows:

1. A third value is calculated based on the natural number e and based on half the width of the QRS complex as an index.

2. And calculating the sum of the position of the target point and the second offset to obtain a fourth value.

3. And calculating the difference between the fourth value and the third value to obtain the position of the starting point of the QRS complex.

4. And calculating the sum of the position of the starting point of the QRS complex and the third numerical value to obtain the position of the end point of the QRS complex.

The mathematical expression of the above-described determination process for determining the start and end points of the QRS complex is as follows:

onset₂＝qrs_loc+ctr_offset2-exp(width₂)/2

offset₂＝onset₂+exp(width₂)/2

Where offset ₂ is the position of the start of the beat signal and offset ₂ is the position of the end of the beat signal.

S217, taking half of the sum of the position of the start point of the heart beat signal and the position of the start point of the QRS complex as the position of the end point of the P wave.

Illustratively, in an embodiment of the present invention, after determining the start point of the heart beat signal and the start point of the QRS complex, half of the sum of the position of the start point of the heart beat signal and the position of the start point of the QRS complex is taken as the position of the end point of the P-wave.

Fig. 2L is a schematic diagram of a heart beat reference point detected according to an embodiment of the present invention, and as shown in fig. 2L, a P-wave start point (i.e. a start point of a heart beat signal), a P-wave end point, a QRS complex start point, a QRS complex end point, and a T-wave end point (i.e. an end point of a heart beat signal) are sequentially from left to right.

According to the heart beat signal reference point determining method provided by the embodiment of the invention, after the electrocardiosignal is acquired, the characteristic extraction is carried out on the electrocardiosignal, so that the characteristic graphs of the electrocardiosignal on a plurality of different scales are obtained, the probability graph of the QRS complex is determined based on the characteristic graph, the local maximum value point with the probability value larger than the probability threshold in the probability graph represents the candidate point of the position information of the QRS complex, the candidate point with the maximum probability value is screened out from the plurality of probability graphs in a non-maximum value suppression mode to serve as the target point for representing the position information of the QRS complex, the characteristic graph corresponding to the target point serves as the target characteristic graph, and the reference point of the heart beat signal to which the QRS complex belongs is determined based on the regional characteristic graph on the target characteristic graph, so that the calculation resource is saved, and the detection efficiency is improved. In addition, a plurality of feature maps with different scales are extracted from the electrocardiosignal and used for detecting the subsequent datum points, so that the position, the detail information and the semantic information are considered, and the detection accuracy and the detection efficiency of the datum points are improved. In addition, by detecting the heart beat signal and the QRS complex respectively for the regional feature diagrams in the preset range of 2 different scales, compared with a method based on pixel point segmentation and smoothing, the method has higher speed and better robustness, and the condition that the datum point cannot be detected can not occur.

Example III

Fig. 3A is a flowchart of a method for identifying a beat type according to a third embodiment of the present invention, where the method is based on the method for determining a beat signal reference point according to any of the above embodiments of the present invention, and the beat signal reference point includes a start point and an end point of the beat signal. The method can be implemented by the heart beat type identification device provided by the embodiment of the invention, and the device can be implemented by software and/or hardware, and is usually configured in computer equipment. As shown in fig. 3A, the method specifically includes the following steps:

S301, inputting the heart beat feature images in the range of the start point and the end point of the heart beat signals on the target feature image into a trained heart beat type identification model for processing, and obtaining probability values of the heart beat signals belonging to each heart beat type.

Illustratively, after the start point and the end point of the heartbeat signal are detected in the foregoing embodiment, the heartbeat feature map within the range of the start point and the end point of the heartbeat signal on the target feature map is input into the trained heartbeat type recognition model for processing, so as to obtain the probability value of the heartbeat signal belonging to each heartbeat type.

Fig. 3B is a schematic structural diagram of a heart beat type recognition model according to an embodiment of the present invention, and exemplary, as shown in fig. 3B, the heart beat type recognition model includes a convolution layer, a pooling layer and a full connection layer, and a processing procedure of the heart beat type recognition model on an input heart beat feature map is as follows:

1. And carrying out convolution processing on the heart beat feature map to obtain heart beat features.

Illustratively, the heart beat feature map is input into a convolution layer for convolution processing, so as to obtain heart beat features. Illustratively, the convolution kernel of the convolution layer has a length of 15 sample points, a convolution step of 1 sample point, a fill size of 7 sample points, and a convolution channel of 5.

2. And carrying out global average pooling treatment on the heart beat characteristics to obtain characteristic vectors.

Illustratively, the heart beat features are input into a pooling layer for global average pooling processing to obtain feature vectors.

3. And splicing (Concatenate) the characteristic vector and the prior vector to obtain a spliced vector, wherein the prior vector comprises characteristic attributes of heart beat signals obtained based on clinical prior information.

And splicing (dimension increasing) the feature vector and the prior vector to obtain a spliced vector, enhancing the influence of clinical prior information on the identification result and improving the identification accuracy. The prior vector comprises characteristic attributes of heart beat signals obtained based on clinical prior information, and specifically, the prior vector is a 5-dimensional vector [ rri _a,rri_b, pr_int, qrs_dur, qt_int ], and the prior vector elements respectively represent the ratio of the distance between the current heart beat QRS complex and the previous heart beat QRS complex to the average QRS complex (rri _a), the ratio of the distance between the current heart beat QRS complex and the next heart beat QRS complex to the average QRS complex (rri _b), the time from the current heart beat P-wave start to the QRS complex start (pr_int), the time from the QRS complex start to the QRS complex end (qrs_dur), and the time from the QRS complex start to the T-wave end (qt_int).

4. And mapping the spliced vector into probability values of the heart beat signals belonging to each heart beat type.

Illustratively, the spliced vector is input into a full connection layer for processing, the full connection layer performs characteristic weighting on the spliced vector, and the spliced vector is mapped into probability values of heart beat signals belonging to each heart beat type. In the embodiment of the invention, the full connection layer outputs a 5-dimensional vector, and elements in the vector respectively represent probability values of the current heart beat belonging to 5 heart beat types. The 5 beat types are exemplified by sinus beat (N), supraventricular beat (S), ventricular beat (V), sinus and ventricular fusion beat (F), and other types of beats (Q), respectively. The 5-dimensional vector of the full link layer output is then noted as [ N, S, V, F, Q ].

In some embodiments of the invention, the beat type recognition model may also be a softmax function layer for normalizing the plurality of probability values output by the full connection layer.

S302, taking the heart beat type corresponding to the maximum value in the probability value as the heart beat type to which the heart beat signal belongs.

For example, the heart beat type corresponding to the maximum value in the obtained plurality of probability values is taken as the heart beat type to which the heart beat signal belongs. For example, if the 5-dimensional vector output by the full connection layer is [0.05,0.02,0.8,0.1,0.03], the heart beat type corresponding to 0.8, namely, ventricular heart beat (V), is taken as the heart beat type identified by the heart beat type identification model.

In the above embodiment, the heart beat type recognition model includes a convolution layer, a pooling layer and a full connection layer as examples, and the heart beat type recognition method in the embodiment of the present invention is described.

According to the heart beat type identification method provided by the embodiment of the invention, based on the heart beat signal reference point determination method provided by any embodiment of the invention, the reference point of the heart beat signal is determined, the reference point of the heart beat signal comprises the starting point and the end point of the heart beat signal, the heart beat characteristic images in the range of the starting point and the end point of the heart beat signal on the target characteristic image are input into a trained heart beat type identification model for processing, the probability value of the heart beat signal belonging to each heart beat type is obtained, and the heart beat type corresponding to the maximum value in the probability value is used as the heart beat type to which the heart beat signal belongs. In the process of detecting the reference point, all points on the target feature map do not need to be traversed to serve as anchor points, a plurality of anchor frames with different scales do not need to be established by taking the anchor points as centers, and in the process of detecting each anchor frame, only the reference point of the heart beat signal to which the QRS complex belongs is determined on the basis of the regional feature map on the target feature map, so that calculation resources are saved, and the detection efficiency is improved. In addition, the accuracy of heart beat type recognition is improved by splicing the characteristic feature vector extracted by the heart beat type recognition model with the prior vector which is obtained based on clinical prior information and reflects the characteristic attribute of the heart beat signal.

Example IV

Fig. 4 is a schematic structural diagram of a heart beat signal reference point determining device according to a fourth embodiment of the present invention, as shown in fig. 4, the device includes:

An electrocardiograph signal acquisition module 401, configured to acquire an electrocardiograph signal, where the electrocardiograph signal includes a plurality of cardiac beat signals;

the feature map extracting module 402 is configured to perform feature extraction on the electrocardiograph signal to obtain feature maps of the electrocardiograph signal at a plurality of different scales;

A probability map determining module 403, configured to determine a probability map of a QRS complex based on the feature map, where a local maximum point in the probability map having a probability value greater than a probability threshold represents a candidate point of the location information of the QRS complex;

A target determining module 404, configured to, for each QRS complex, screen out candidate points with the largest probability values from the multiple probability maps as target points representing the location information of the QRS complex, and a feature map corresponding to the target points as a target feature map;

the reference point determining module 405 is configured to determine a reference point of a heart beat signal to which the QRS complex belongs based on a region feature map on the target feature map, where the region feature map is a feature in a preset range on the target feature map centered on the target point.

In some embodiments of the present invention, the feature map extraction module 402 is configured to:

Inputting the electrocardiosignal into a pre-trained characteristic pyramid network to obtain three-scale characteristic graphs, wherein the three-scale characteristic graphs are a first characteristic graph, a second characteristic graph and a third characteristic graph respectively, and the first characteristic graph, the second characteristic graph and the third characteristic graph are 1 time, 0.5 time and 0.25 time of the length of the electrocardiosignal respectively.

In some embodiments of the present invention, the feature map extraction module 402 includes:

the first feature extraction unit is used for inputting the electrocardiosignal into a first convolution block for processing to obtain a first feature;

The second feature extraction unit is used for inputting the first feature into a second convolution block for processing to obtain a second feature;

The first downsampling feature extraction unit is used for downsampling the second features to obtain first downsampling features;

the third feature extraction unit is used for inputting the first downsampling feature into a third convolution block for processing to obtain a third feature;

a second downsampling feature extraction unit, configured to perform downsampling processing on the third feature to obtain a second downsampled feature;

A fourth feature extraction unit, configured to input the second downsampled feature into a fourth convolution block for processing, to obtain a fourth feature;

a third feature map extracting unit, configured to perform convolution processing on the fourth feature to obtain the third feature map;

A fifth feature extraction unit, configured to perform upsampling processing on the third feature map, and perform feature fusion with features obtained by convolving the third feature to obtain a fifth feature;

The second feature map extracting unit is used for carrying out convolution processing on the fifth feature to obtain a second feature map;

A sixth feature extraction unit, configured to perform upsampling processing on the fifth feature, and perform feature fusion with a feature obtained by performing convolution processing on the second feature, to obtain a sixth feature;

And the first feature map extracting unit is used for carrying out convolution processing on the sixth feature to obtain the first feature map.

In some embodiments of the present invention, probability map determination module 403 includes:

The first sub-feature map extraction unit is used for carrying out convolution processing on the feature map to obtain a first sub-feature map;

the second sub-feature image extraction unit is used for activating the first sub-feature image to obtain a second sub-feature image;

and the probability map determining unit is used for carrying out smoothing processing on the second sub-feature map to obtain a probability map of the QRS complex.

In some embodiments of the invention, the targeting module 404 includes:

the checking list generating unit is used for ordering all candidate points in descending order according to the size of the probability value to generate a checking list;

A target point determination unit configured to take a candidate point with the largest probability value in the inspection list as a target point;

A target feature map determining unit, configured to take a feature map corresponding to the target point as a target feature map;

a deleting unit, configured to delete the target point and all candidate points within a preset distance range of the target point;

and the return execution unit is used for returning to the step of taking the candidate point with the maximum probability value in the checking list as the target point until the checking list is empty.

In some embodiments of the present invention, the fiducial point determination module 405 includes:

The third sub-feature extraction unit is used for carrying out convolution processing on a first area feature map on the target feature map to obtain a third sub-feature map, wherein the first area feature map is a feature in a first preset range with the target point as the center on the target feature map;

The first two-dimensional vector determining unit is used for carrying out global average pooling processing on the third sub-feature map to obtain two-dimensional vectors, wherein elements in the two-dimensional vectors respectively represent a first offset of the middle point of the heart beat signal relative to the target point and the width of the heart beat signal;

A first reference point determining unit for determining a start point and an end point of the heart beat signal based on the first offset and the width of the heart beat signal, wherein the reference point of the heart beat signal includes the start point and the end point of the heart beat signal.

In some embodiments of the present invention, the first reference point determination unit includes:

a first numerical calculation subunit, configured to calculate a first numerical value based on a natural number e and based on a half of a width of the heart beat signal as an index;

A second numerical value calculating subunit, configured to calculate a sum of the position of the target point and the first offset to obtain a second numerical value;

a heart beat signal start point determining subunit, configured to calculate a difference between the second value and the first value, and obtain a position of a start point of the heart beat signal;

And the heart beat signal end point determining subunit is used for calculating the sum of the position of the start point of the heart beat signal and the first numerical value to obtain the position of the end point of the heart beat signal.

In some embodiments of the present invention, the fiducial point determination module 405 further includes:

A fourth sub-feature map extracting unit, configured to perform convolution processing on a second region feature map on the target feature map to obtain a fourth sub-feature map, where the second region feature map is a feature in a second preset range on the target feature map with the target point as a center;

A second two-dimensional vector determining unit, configured to perform global average pooling processing on the fourth sub-feature map to obtain two-dimensional vectors, where elements in the two-dimensional vectors respectively represent a second offset of a midpoint of the QRS complex relative to the target point, and a width of the QRS complex;

A second fiducial point determination unit for determining a start point and an end point of the QRS complex based on the second offset and the width of the QRS complex, the fiducial point of the heart beat signal including the start point and the end point of the QRS complex.

In some embodiments of the present invention, the second reference point determination unit includes:

A third numerical value calculation subunit, configured to calculate a third numerical value based on a natural number e and based on a half of the width of the QRS complex as an index;

A fourth numerical value calculating subunit, configured to calculate a sum of the position of the target point and the second offset to obtain a fourth numerical value;

A QRS complex origin determining subunit configured to calculate a difference between the fourth value and the third value, and obtain a location of the origin of the QRS complex;

and the QRS complex end point determining subunit is used for calculating the sum of the position of the starting point of the QRS complex and the third numerical value to obtain the position of the end point of the QRS complex.

In some embodiments of the present invention, the fiducial points comprise a start point and an end point of a heart beat signal, a start point and an end point of a QRS complex, an end point of a P-wave, the heart beat signal fiducial point determining device further comprises:

and the P-wave end point determining module is used for taking half of the sum of the position of the starting point of the heart beat signal and the position of the starting point of the QRS complex as the position of the end point of the P wave.

The heart beat signal reference point determining device can execute the heart beat signal reference point determining method provided by any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the executing method.

Example five

Fig. 5 is a schematic structural diagram of a beat type recognition device according to a fifth embodiment of the present invention, where the beat signal reference point determining device according to the foregoing embodiment of the present invention determines a reference point of a beat signal, where the reference point of the beat signal includes a start point and an end point of the beat signal, and as shown in fig. 5, the device includes:

The probability value obtaining module 501 is configured to input a beat feature map within a range of a start point and an end point of the beat signal on the target feature map into a trained beat type recognition model for processing, so as to obtain a probability value of each beat type of the beat signal;

and a beat type determining module 502, configured to use a beat type corresponding to the maximum value in the probability value as a beat type to which the beat signal belongs.

In some embodiments of the present invention, the probability value acquisition module 501 includes:

The heart beat feature extraction unit is used for carrying out convolution processing on the heart beat feature map to obtain heart beat features;

the feature vector extraction unit is used for carrying out global average pooling treatment on the heart beat features to obtain feature vectors;

the splicing unit is used for splicing the characteristic vector and the prior vector to obtain a spliced vector, and the prior vector comprises characteristic attributes of heart beat signals obtained based on clinical prior information;

and the probability value determining unit is used for mapping the spliced vector into probability values of the heart beat signals belonging to each heart beat type.

The heart beat type identification device can execute the heart beat type identification method provided by any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

Example six

Fig. 6 is a schematic structural diagram of a computer device according to a seventh embodiment of the present invention, as shown in fig. 6, the computer device includes a processor 601, a memory 602, a communication module 603, an input device 604 and an output device 605; the number of processors 601 in the computer device may be one or more, one processor 601 being taken as an example in fig. 6; the processor 601, memory 602, communication module 603, input means 604 and output means 605 in the computer device may be connected by a bus or other means, in fig. 6 by way of example. The processor 601, the memory 602, the communication module 603, the input means 604 and the output means 605 described above may be integrated on the control motherboard of the computer device.

The memory 602 is a computer readable storage medium, and can be used to store a software program, a computer executable program, and a module, such as a module corresponding to the beat signal reference point determination method or the beat type identification method in the present embodiment. The processor 601 executes various functional applications of the computer device and data processing by executing software programs, instructions and modules stored in the memory 602, that is, implements the beat signal reference point determination method or the beat type identification method provided by the above-described embodiments.

The memory 602 may include primarily a program storage area and a data storage area, wherein the program storage area may store an operating system, at least one application program required for functionality; the storage data area may store data created according to the use of the computer device, etc. In addition, the memory 602 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some examples, memory 602 may further include memory located remotely from processor 601, which may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The communication module 603 is configured to establish a connection with an external device (e.g. an intelligent terminal), and implement data interaction with the external device. The input means 604 may be used to receive entered numeric or character information and to generate key signal inputs related to user settings and function control of the computer device.

The computer device provided in this embodiment may execute the method for determining the beat signal reference point or the method for identifying the beat type provided in any of the above embodiments of the present invention, and specifically correspond to the functions and the beneficial effects.

Example seven

A seventh embodiment of the present invention provides a storage medium containing computer-executable instructions, on which a computer program is stored, which when executed by a processor implements a beat signal reference point determining method or a beat type identifying method as provided in any of the above embodiments of the present invention.

The heart beat signal datum point determining method comprises the following steps:

The heart beat type identification method determines a reference point of a heart beat signal based on the heart beat signal reference point determination method provided by the foregoing embodiment of the present invention, where the reference point of the heart beat signal includes a start point and an end point of the heart beat signal, and the heart beat type identification method includes:

Of course, the storage medium containing the computer executable instructions provided in the embodiments of the present invention is not limited to the above-described method operations, and may also perform the related operations in the beat signal reference point determination method or the beat type identification method provided in the embodiments of the present invention.

It should be noted that, for the apparatus, device and storage medium embodiments, since they are substantially similar to the method embodiments, the description is relatively simple, and reference should be made to the description of the method embodiments for relevant points.

From the above description of embodiments, it will be clear to a person skilled in the art that the present invention may be implemented by means of software and necessary general purpose hardware, but of course also by means of hardware, although in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as a floppy disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a FLASH Memory (FLASH), a hard disk, or an optical disk of a computer, where the instructions include a plurality of instructions for causing a computer device (which may be a robot, a personal computer, a server, or a network device, etc.) to perform the beat signal reference point determination method or the beat type identification method according to any embodiment of the present invention.

It should be noted that, in the above apparatus, each module, unit and sub-unit included is only divided according to the functional logic, but is not limited to the above division, so long as the corresponding function can be implemented; in addition, the specific names of the functional units are also only for distinguishing from each other, and are not used to limit the protection scope of the present invention.

It is to be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution device. For example, if implemented in hardware, as in another embodiment, may be implemented using any one or combination of the following techniques, as is well known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.

In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Note that the above is only a preferred embodiment of the present invention and the technical principle applied. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, while the invention has been described in connection with the above embodiments, the invention is not limited to the embodiments, but may be embodied in many other equivalent forms without departing from the spirit or scope of the invention, which is set forth in the following claims.

Claims

1. A method for determining a beat signal reference point, comprising:

Determining a reference point of a heart beat signal to which the QRS complex belongs based on a regional feature map on the target feature map, wherein the regional feature map is a feature in a preset range taking the target point as the center on the target feature map;

The determining, based on the regional feature map on the target feature map, a reference point of the heart beat signal to which the QRS complex belongs, including: and carrying out rolling and pooling operation on the regional feature map on the target feature map so as to determine the starting point and the ending point of the heart beat signal, wherein the reference point of the heart beat signal comprises the starting point and the ending point of the heart beat signal.

2. The method for determining a cardiac beat signal reference point according to claim 1, wherein the feature extraction is performed on the cardiac signal to obtain feature maps of the cardiac signal at a plurality of different scales, including:

3. The method for determining a heart beat signal reference point according to claim 2, wherein inputting the electrocardiographic signal into a pre-trained feature pyramid network to obtain a three-scale feature map comprises:

inputting the electrocardiosignal into a first convolution block for processing to obtain a first characteristic;

inputting the first feature into a second convolution block for processing to obtain a second feature;

Performing downsampling processing on the second feature to obtain a first downsampled feature;

Inputting the first downsampling feature into a third convolution block for processing to obtain a third feature;

performing downsampling processing on the third feature to obtain a second downsampled feature;

Inputting the second downsampling characteristic into a fourth convolution block for processing to obtain a fourth characteristic;

Performing convolution processing on the fourth feature to obtain the third feature map;

Performing up-sampling processing on the third feature map, and performing feature fusion with features obtained by performing convolution processing on the third feature map to obtain fifth features;

Performing convolution processing on the fifth feature to obtain the second feature map;

Performing up-sampling processing on the fifth feature, and performing feature fusion on the fifth feature and the feature obtained by performing convolution processing on the fifth feature and the second feature to obtain a sixth feature;

And carrying out convolution processing on the sixth feature to obtain the first feature map.

4. The method of claim 1, wherein determining a probability map of QRS complexes based on the feature map comprises:

Carrying out convolution processing on the feature map to obtain a first sub-feature map;

activating the first sub-feature map to obtain a second sub-feature map;

And carrying out smoothing treatment on the second sub-feature map to obtain a probability map of the QRS complex.

5. The heart beat signal reference point determination method according to claim 1, wherein for each QRS complex, selecting a candidate point having a largest probability value from a plurality of the probability maps as a target point representing position information of the QRS complex, and a feature map corresponding to the target point as a target feature map, comprises:

Ordering all candidate points in descending order according to the size of the probability value to generate a checking list;

taking a candidate point with the maximum probability value in the checking list as a target point;

Taking the feature map corresponding to the target point as a target feature map;

deleting all candidate points within a preset distance range of the target point;

And returning to the step of taking the candidate point with the maximum probability value in the checking list as the target point until the checking list is empty.

6. The method of claim 1, wherein performing a convolution and pooling operation on the region feature map on the target feature map to determine a start point and an end point of the heart beat signal comprises:

Performing convolution processing on a first region feature map on the target feature map to obtain a third sub-feature map, wherein the first region feature map is a feature in a first preset range on the target feature map with the target point as the center;

Carrying out global average pooling processing on the third sub-feature map to obtain a two-dimensional vector, wherein elements in the two-dimensional vector respectively represent a first offset of the middle point of the heart beat signal relative to the target point and the width of the heart beat signal;

a start point and an end point of the heartbeat signal are determined based on the first offset and a width of the heartbeat signal.

7. The method of claim 6, wherein determining a start point and an end point of the heart beat signal based on the first offset and a width of the heart beat signal comprises:

Calculating a first numerical value taking a natural number e as a base number and taking half of the width of the heart beat signal as an index;

calculating the sum of the position of the target point and the first offset to obtain a second value;

Calculating the difference between the second value and the first value to obtain the position of the starting point of the heart beat signal;

And calculating the sum of the position of the starting point of the heart beat signal and the first numerical value to obtain the position of the end point of the heart beat signal.

8. The method according to claim 1, wherein calculating the reference point of the heart beat signal to which the QRS complex belongs based on the region feature map on the target feature map, comprises:

performing convolution processing on a second region feature map on the target feature map to obtain a fourth sub-feature map, wherein the second region feature map is a feature in a second preset range on the target feature map with the target point as the center;

Carrying out global average pooling processing on the fourth sub-feature map to obtain a two-dimensional vector, wherein elements in the two-dimensional vector respectively represent a second offset of the midpoint of the QRS complex relative to the target point and the width of the QRS complex;

Determining a start point and an end point of the QRS complex based on the second offset and the width of the QRS complex, the fiducial point of the heart beat signal including the start point and the end point of the QRS complex.

9. The method of claim 8, wherein determining the start and end of the QRS complex based on the second offset and the width of the QRS complex comprises:

calculating a third numerical value based on a natural number e and based on half of the width of the QRS complex as an index;

calculating the sum of the position of the target point and the second offset to obtain a fourth value;

Calculating the difference between the fourth value and the third value to obtain the position of the start point of the QRS complex;

And calculating the sum of the position of the starting point of the QRS complex and the third numerical value to obtain the position of the ending point of the QRS complex.

10. The method of any one of claims 1-9, wherein the fiducial points comprise a start and an end of a heart beat signal, a start and an end of a QRS complex, and an end of a P-wave, the method further comprising:

taking half of the sum of the position of the start point of the heart beat signal and the position of the start point of the QRS complex as the position of the end point of the P wave.

11. A heart beat type recognition method characterized by determining a reference point of a heart beat signal based on the heart beat signal reference point determination method of any one of claims 1 to 10, the reference point of the heart beat signal including a start point and an end point of the heart beat signal, the heart beat type recognition method comprising:

12. The method for identifying a heart beat type according to claim 11, wherein inputting the heart beat feature map within the range of the start point and the end point of the heart beat signal on the target feature map into a trained heart beat type identification model for processing to obtain a probability value of the heart beat signal belonging to each heart beat type, comprises:

performing convolution processing on the heart beat feature map to obtain heart beat features;

Carrying out global average pooling treatment on the heart beat characteristics to obtain characteristic vectors;

splicing the characteristic vector and the prior vector to obtain a spliced vector, wherein the prior vector comprises characteristic attributes of heart beat signals obtained based on clinical prior information;

and mapping the spliced vector into probability values of the heart beat signals belonging to each heart beat type.

13. A heart beat signal reference point determining device, comprising:

the reference point determining module is used for determining a reference point of the heart beat signal to which the QRS complex belongs based on a regional feature map on the target feature map, wherein the regional feature map is a feature in a preset range with the target point as the center on the target feature map;

14. A beat type recognition device, characterized in that a beat signal reference point including a start point and an end point of the beat signal is determined based on the beat signal reference point determination device of claim 13, the beat type recognition device comprising: