CN111481187B

CN111481187B - Method for detecting arrhythmia by artificial intelligence based on arterial pressure wave characteristics

Info

Publication number: CN111481187B
Application number: CN202010460777.1A
Authority: CN
Inventors: 张闻涛; 郑颖; 李祥
Original assignee: Tongxintang Health Technology Beijing Co ltd
Current assignee: Tongxintang Health Technology Beijing Co ltd
Priority date: 2020-05-27
Filing date: 2020-05-27
Publication date: 2023-06-23
Anticipated expiration: 2040-05-27
Also published as: CN111481187A

Abstract

The application relates to an artificial intelligence arrhythmia detection method based on arterial pressure wave characteristics. The method comprises the following steps: acquiring brachial artery pressure wave data of a target object; determining key points in the brachial artery pressure wave according to the brachial artery pressure wave data, and determining characteristic data of the brachial artery pressure wave according to the key points; wherein the characteristic data is used for characterizing the fluctuation state of the brachial artery pressure wave; and detecting the heart rhythm type according to the characteristic data to obtain the heart rhythm type of the target object. The method can be used for detecting the heart rhythm type based on the brachial artery pressure wave data.

Description

Method for detecting arrhythmia by artificial intelligence based on arterial pressure wave characteristics

Technical Field

The application relates to the technical field of heart rhythm detection, in particular to an artificial intelligence arrhythmia detection method based on arterial pressure wave characteristics.

Background

Electrocardiogram (ECG) records the physiological changes of the electrical activity generated by each cardiac cycle of the human heart, and is an important basis for diagnosing cardiovascular diseases. The clinician can judge whether the patient is arrhythmia according to the electrocardiogram so as to treat the arrhythmia.

In the related art, electrocardiographic examination generally requires a plurality of lead lines, and is therefore mainly used in hospitals. In practice, the brachial artery is the place where the centrifugal heart is closest and easiest to collect pressure wave data from the surface after the aorta comes out, the brachial artery pressure wave data can be measured simultaneously with the blood pressure, the pumping function of the heart can be reflected, and the waveform is simpler than the electrocardiogram.

However, methods for arrhythmia detection based on brachial artery pressure wave data are currently lacking.

Disclosure of Invention

In view of the above, it is desirable to provide a method for artificially and intelligently detecting arrhythmia based on arterial pressure wave characteristics, which can detect the type of heart rhythm based on brachial artery pressure wave data.

A method of artificially detecting arrhythmias based on arterial pressure wave characteristics, the method comprising:

acquiring brachial artery pressure wave data of a target object;

determining key points in the brachial artery pressure wave according to the brachial artery pressure wave data, and determining characteristic data of the brachial artery pressure wave according to the key points; the characteristic data are used for representing the fluctuation state of the brachial artery pressure wave;

and detecting the heart rhythm type according to the characteristic data to obtain the heart rhythm type of the target object.

In one embodiment, determining the key point in the brachial artery pressure wave according to the brachial artery pressure wave data and determining the characteristic data of the brachial artery pressure wave according to the key point includes:

converting brachial artery pressure wave data from a time domain to a frequency domain, and carrying out frequency selection according to a preset frequency range to obtain frequency domain data of brachial artery pressure waves;

determining key points in the brachial artery pressure wave according to the amplitude of the frequency domain data of the brachial artery pressure wave;

dividing a preset frequency range into a plurality of frequency intervals, and determining characteristic data corresponding to each frequency interval according to the key points.

In one embodiment, determining the key point in the brachial artery pressure wave according to the amplitude of the frequency domain data of the brachial artery pressure wave includes:

determining whether each piece of frequency domain data meets a preset condition, and determining the frequency domain data meeting the preset condition as a key point;

the preset conditions comprise: the amplitude of the frequency domain data is larger than the amplitude of the previous frequency domain data and the amplitude of the next frequency domain data and larger than a preset threshold value.

In one embodiment, the preset threshold is a preset percentile in the frequency domain data sequence; the frequency domain data sequence is a data sequence obtained by arranging a plurality of frequency domain data in an amplitude from small to large.

In one embodiment, the dividing the preset frequency range into a plurality of frequency intervals, determining the feature data corresponding to each frequency interval according to the key points includes:

dividing a preset frequency range according to the length of a preset interval to obtain a plurality of frequency intervals;

calculating characteristic data corresponding to each frequency interval according to the key points in each frequency interval; the feature data includes the number of keypoints in the frequency interval, the amplitude accumulated value of the keypoints, and the frequency accumulated value of the keypoints.

In one embodiment, the brachial artery pressure wave data includes pressure increasing and decreasing data and constant pressure data; the elevation pressure data comprises systolic pressure data and diastolic pressure data; the constant pressure data and the lifting pressure data are in linear relation.

In one embodiment, the detecting the heart rhythm type according to the feature data to obtain the heart rhythm type of the target object includes:

and inputting the characteristic data into a pre-trained heart rhythm type detection model to obtain the heart rhythm type of the target object output by the heart rhythm type detection model.

In one embodiment, the process of training the heart rhythm type detection model includes:

acquiring sample feature data of a plurality of training objects and gold standards corresponding to the sample feature data; the gold standard corresponding to the sample characteristic data is used for representing the heart rhythm type corresponding to the training object; the heart rhythm type corresponding to the training object is determined according to the diagnosis result of the electrocardiogram;

And training the classification model based on the sample characteristic data of the plurality of training objects and the gold standard corresponding to each sample characteristic data to obtain a heart rhythm type detection model.

In one embodiment, the acquiring sample feature data of the plurality of training objects includes:

acquiring feature data of a plurality of candidate objects;

scoring the characteristic data of each candidate object by adopting a preset scoring function to obtain scoring results corresponding to each candidate object;

and selecting a training object from the plurality of candidate objects according to the scoring result corresponding to each candidate object, and taking the characteristic data of the training object as sample characteristic data.

In one embodiment, the selecting a training object from the plurality of candidate objects according to the scoring result corresponding to each candidate object includes:

sorting the plurality of candidate objects in order of the score from high to low;

determining k candidate objects which are ranked at the front as training objects; wherein k is a positive integer.

In one embodiment, the training of the classification model based on the sample feature data of the plurality of training objects and the gold standard corresponding to each sample feature data to obtain the heart rhythm type detection model includes:

Obtaining a value range and a value step length of model parameters of the classification model;

and determining a target value of the model parameter according to the value range and the value step length of the model parameter and the gold standard corresponding to the characteristic data of each sample to obtain the heart rhythm type detection model.

In one embodiment, the classification model is a multi-classification model, and determining the target value of the model parameter according to the value range and the value step length of the model parameter and the gold standard corresponding to the feature data of each sample to obtain the heart rhythm type detection model includes:

initializing model parameters according to the value range of the model parameters; model parameters include input feature quantity, error tolerance and support vector metrics;

training the initialized classification model based on sample feature data of a plurality of training objects to obtain average accuracy corresponding to the classification model;

updating the value of each model parameter according to the value step length of the model parameter, and repeatedly executing the steps of obtaining the average accuracy corresponding to the classification model until the value of the model parameter is updated, so as to obtain a plurality of average accuracy;

and selecting candidate values of model parameters according to the plurality of average accuracy rates, and performing performance detection on the classification model according to the candidate values of the model parameters and gold standards corresponding to the feature data of each sample to obtain a heart rhythm type detection model with model performance meeting preset detection conditions.

In one embodiment, training the initialized classification model based on the sample feature data corresponding to the plurality of training objects to obtain an average accuracy corresponding to the classification model includes:

dividing the plurality of sample feature data into 10 mutually exclusive subsets;

each time, 9 mutually exclusive subsets are adopted as training sets, the rest subset is adopted as a test set to carry out 10 times of training and testing, and the average value of the accuracy of the 10 times of testing is adopted as average accuracy.

In one embodiment, performing performance detection on the classification model according to the candidate values of the model parameters to obtain a heart rhythm type detection model with model performance meeting a preset detection condition, including:

calculating the model performance of the classification model according to the candidate values of the model parameters; model performance includes accuracy, sensitivity and specificity;

and under the condition that the accuracy rate is determined to be larger than the preset accuracy rate, the sensitivity is larger than the preset sensitivity and the specificity is larger than the preset specificity, determining that the model performance meets the preset detection condition, and determining the candidate value of the model parameter as the target value of the model parameter when the model performance meets the preset detection condition to obtain the heart rhythm type detection model.

In one embodiment, before the characteristic data is input into the pre-trained heart rhythm type detection model to obtain the heart rhythm type of the target object output by the heart rhythm type detection model, the method further includes:

Saving the heart rhythm type detection model and model parameters as model files;

and running the web server, loading the model file and generating a calling interface of the heart rhythm type detection model.

An apparatus for artificial intelligence detection of arrhythmias based on arterial pressure wave characteristics, the apparatus comprising:

the pressure wave data acquisition module is used for acquiring brachial artery pressure wave data of the target object;

the characteristic data determining module is used for determining key points in the brachial artery pressure wave according to the brachial artery pressure wave data and determining characteristic data of the brachial artery pressure wave according to the key points; the characteristic data are used for representing the fluctuation state of the brachial artery pressure wave;

and the heart rhythm type detection module is used for detecting the heart rhythm type according to the characteristic data to obtain the heart rhythm type of the target object.

In one embodiment, the feature data determining module includes:

the data conversion submodule is used for converting brachial artery pressure wave data from a time domain to a frequency domain, and carrying out frequency selection according to a preset frequency range to obtain frequency domain data of the brachial artery pressure wave;

the key point determining submodule is used for determining key points in the brachial artery pressure wave according to the amplitude of the frequency domain data of the brachial artery pressure wave;

The characteristic data determining sub-module is used for dividing the preset frequency range into a plurality of frequency intervals and determining characteristic data corresponding to each frequency interval according to the key points.

In one embodiment, the above-mentioned key point determining submodule is specifically configured to determine whether each frequency domain data meets a preset condition, and determine the frequency domain data meeting the preset condition as a key point;

In one embodiment, the feature data determining submodule is specifically configured to divide a preset frequency range according to a preset interval length to obtain a plurality of frequency intervals; calculating characteristic data corresponding to each frequency interval according to the key points in each frequency interval; the feature data includes the number of keypoints in the frequency interval, the amplitude accumulated value of the keypoints, and the frequency accumulated value of the keypoints.

In one embodiment, the above-mentioned heart rhythm type detection module is specifically configured to input the feature data into a pre-trained heart rhythm type detection model to obtain a heart rhythm type of the target object output by the heart rhythm type detection model.

In one embodiment, the apparatus further comprises:

the training sample acquisition module is used for acquiring sample characteristic data of a plurality of training objects and gold standards corresponding to the sample characteristic data; the gold standard corresponding to the sample characteristic data is used for representing the heart rhythm type corresponding to the training object; the heart rhythm type corresponding to the training object is determined according to the diagnosis result of the electrocardiogram;

and the training module is used for training the classification model based on the sample characteristic data of the plurality of training objects and the gold standard corresponding to each sample characteristic data to obtain a heart rhythm type detection model.

In one embodiment, the training sample acquiring module includes:

the characteristic data acquisition sub-module is used for acquiring characteristic data of a plurality of candidate objects;

The scoring module is used for scoring the characteristic data of each candidate object by adopting a preset scoring function to obtain a scoring result corresponding to each candidate object;

and the training object determining sub-module is used for selecting the training object from the plurality of candidate objects according to the scoring result corresponding to each candidate object, and taking the characteristic data of the training object as sample characteristic data.

In one embodiment, the training object determining submodule is specifically configured to sort the plurality of candidate objects in order of scores from high to low; determining k candidate objects which are ranked at the front as training objects; wherein k is a positive integer.

In one embodiment, the training module includes:

the value obtaining sub-module is used for obtaining the value range and the value step length of the model parameters of the classification model;

and the training sub-module is used for determining the target value of the model parameter according to the value range and the value step length of the model parameter and the gold standard corresponding to the characteristic data of each sample to obtain the heart rhythm type detection model.

In one embodiment, the classification model is a multi-classification model, and the training submodule is specifically configured to initialize model parameters according to a value range of the model parameters; model parameters include input feature quantity, error tolerance and support vector metrics; training the initialized classification model based on sample feature data of a plurality of training objects to obtain average accuracy corresponding to the classification model; updating the value of each model parameter according to the value step length of the model parameter, and repeatedly executing the steps of obtaining the average accuracy corresponding to the classification model until the value of the model parameter is updated, so as to obtain a plurality of average accuracy; and selecting candidate values of model parameters according to the plurality of average accuracy rates, and performing performance detection on the classification model according to the candidate values of the model parameters and gold standards corresponding to the feature data of each sample to obtain a heart rhythm type detection model with model performance meeting preset detection conditions.

In one embodiment, the training submodule is specifically configured to divide the plurality of sample feature data into 10 mutually exclusive subsets; each time, 9 mutually exclusive subsets are adopted as training sets, the rest subset is adopted as a test set to carry out 10 times of training and testing, and the average value of the accuracy of the 10 times of testing is adopted as average accuracy.

In one embodiment, the training submodule is specifically configured to calculate model performance of the classification model according to candidate values of model parameters; model performance includes accuracy, sensitivity and specificity; and under the condition that the accuracy rate is determined to be larger than the preset accuracy rate, the sensitivity is larger than the preset sensitivity and the specificity is larger than the preset specificity, determining that the model performance meets the preset detection condition, and determining the candidate value of the model parameter as the target value of the model parameter when the model performance meets the preset detection condition to obtain the heart rhythm type detection model.

In one embodiment, the apparatus further comprises:

the model file storage module is used for storing the heart rhythm type detection model and model parameters as model files;

and the calling interface generation module is used for running the web server, loading the model file and generating a calling interface of the heart rhythm type detection model.

An electronic device comprising a memory storing a computer program and a processor that when executing the computer program performs the steps of:

acquiring brachial artery pressure wave data of a target object;

A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:

acquiring brachial artery pressure wave data of a target object;

In the method for detecting arrhythmia based on artificial intelligence of arterial pressure wave characteristics, a terminal acquires brachial artery pressure wave data of a target object; determining key points in the brachial artery pressure wave according to the brachial artery pressure wave data, and determining characteristic data of the brachial artery pressure wave according to the key points; and detecting the heart rhythm type according to the characteristic data to obtain the heart rhythm type of the target object. By way of the disclosed embodiments, a way of detecting the type of heart rhythm based on brachial artery pressure wave data is provided. Compared with the prior art that the heart rhythm type is detected according to the electrocardiogram, the data acquisition mode of the embodiment of the disclosure is simpler and easier to realize, and the terminal can be miniaturized and portable, so that the heart rhythm detection is more popular.

Drawings

FIG. 1 is a flow chart of a method of artificially detecting arrhythmias based on characteristics of arterial pressure waves in one embodiment;

FIG. 2 is a flow chart of the steps of determining keypoints in a brachial artery pressure wave and determining characteristic data of the brachial artery pressure wave based on the keypoints in one embodiment;

FIG. 3-1 is a schematic diagram of time domain data of buck-boost data in one embodiment;

FIG. 3-2 is a diagram of frequency domain data of buck-boost data in one embodiment;

3-3 are diagrams of time domain data of constant pressure data in one embodiment;

FIGS. 3-4 are diagrams of frequency domain data of constant pressure data in one embodiment;

FIG. 4 is a flow chart illustrating the steps of training a heart rhythm type detection model according to one embodiment;

FIG. 5 is a flowchart illustrating a training step of performing a classification model based on sample feature data of a plurality of training objects and gold standards corresponding to the sample feature data according to one embodiment;

FIG. 6 is a schematic diagram of average accuracy achieved in one embodiment;

FIG. 7 is a schematic diagram of a step of updating model parameters in one embodiment;

FIG. 8 is a flow chart of a method of artificially detecting arrhythmias based on characteristics of arterial pressure waves in another embodiment;

FIG. 9 is one of the block diagrams of an apparatus for artificial intelligence detection of arrhythmias based on characteristics of arterial pressure waves in one embodiment;

FIG. 10 is a block diagram second of an apparatus for artificial intelligence detection of arrhythmias based on arterial pressure wave characteristics in one embodiment;

FIG. 11 is a block diagram III of an apparatus for artificial intelligence detection of arrhythmias based on characteristics of arterial pressure waves in one embodiment;

fig. 12 is an internal structural diagram of an electronic device in one embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.

In one embodiment, as shown in fig. 1, a method for detecting arrhythmia based on artificial intelligence of arterial pressure wave characteristics is provided, and the embodiment of the disclosure is applied to a terminal for illustration, it is understood that the method can also be applied to a server, and can also be applied to a system including the terminal and the server, and implemented through interaction of the terminal and the server. In an embodiment of the present disclosure, the method includes the steps of:

Step 101, brachial artery pressure wave data of a target object is acquired.

Wherein, brachial artery pressure wave data is data representing brachial artery pressure waves. Optionally, the brachial artery pressure wave data includes pressure elevation data and constant pressure data; the pressure-raising and lowering data comprise systolic pressure data and diastolic pressure data, and the constant pressure data and the pressure-raising and lowering data are in linear relation. For example, the constant pressure data is the sum of one third of the systolic data and two thirds of the diastolic data. Systolic pressure, also known as high pressure, when a person's heart contracts, the pressure in the artery rises, the pressure in the artery is highest in the middle of the heart contraction, and at this time, the pressure of blood on the inner wall of the blood vessel is systolic pressure. Diastolic pressure, also known as low pressure, is the pressure produced when the arterial blood vessel elastically contracts when the human heart is diastolic.

The terminal can acquire brachial artery pressure wave data of a target object through a sphygmomanometer; the brachial artery pressure wave data of the target object may be acquired from a plurality of brachial artery pressure wave data stored in the server. The embodiments of the present disclosure are not limited in this regard.

Step 102, determining key points in the brachial artery pressure wave according to the brachial artery pressure wave data, and determining characteristic data of the brachial artery pressure wave according to the key points.

Wherein the characteristic data is used for characterizing the fluctuation state of the brachial artery pressure wave. The key points may be peaks, troughs, etc. of the brachial artery pressure wave.

After the brachial artery pressure wave data are obtained, the wave peaks and wave troughs in the brachial artery pressure wave can be determined according to the brachial artery pressure data, and then the wave peak number, the wave trough number, the wave peak amplitude, the wave trough pattern, the wave peak frequency, the wave trough frequency and the like in the brachial artery pressure wave are determined, so that the characteristic data of the brachial artery pressure wave are obtained. The embodiment of the present disclosure does not limit the feature data.

And step 103, detecting the heart rhythm type according to the characteristic data to obtain the heart rhythm type of the target object.

Wherein the heart rhythm type includes normal heart rhythm and arrhythmia; arrhythmia includes atrial arrhythmia and ventricular arrhythmia; atrial arrhythmias may include atrial fibrillation, atrial extra-systole, atrial flutter, and the like; ventricular arrhythmias may include ventricular premature beats and the like. The embodiments of the present disclosure are not limited in the type of heart rhythm.

Since the brachial artery pressure wave varies in the fluctuation state between the normal case of the heart rhythm and the abnormal case of the heart rhythm, and there is also a difference in the fluctuation state in each abnormal case of the heart rhythm. Therefore, after obtaining the characteristic data of the brachial artery pressure wave, that is, the fluctuation state of the brachial artery pressure wave, the heart rhythm type can be determined according to the fluctuation state of the brachial artery pressure wave.

In one embodiment, as shown in fig. 2, a step of determining a keypoint in the brachial artery pressure wave from the brachial artery pressure wave data and determining characteristic data of the brachial artery pressure wave from the keypoint is involved. On the basis of the above embodiment, it may include:

step 201, converting the brachial artery pressure wave data from the time domain to the frequency domain, and performing frequency selection according to the preset frequency range to obtain frequency domain data of the brachial artery pressure wave.

The preset frequency range is the effective range of frequency domain data of the brachial artery pressure wave. In practical applications, the preset frequency range may be 0-600Hz.

The acquired brachial artery pressure wave data are time domain data, the brachial artery pressure wave data can be converted from the time domain to the frequency domain by adopting discrete Fourier transform, and then frequency selection is carried out according to a preset frequency range, so that frequency domain data of the brachial artery pressure wave are obtained. The time domain data of the voltage increasing and decreasing data are shown in fig. 3-1, the frequency domain data of the voltage increasing and decreasing data are shown in fig. 3-2, the time domain data of the constant voltage data are shown in fig. 3-3, and the frequency domain data of the constant voltage data are shown in fig. 3-4.

Step 202, determining a key point in the brachial artery pressure wave according to the amplitude of the frequency domain data of the brachial artery pressure wave.

When the peak of the brachial artery pressure wave is taken as the key point of the brachial artery pressure wave, the peak of the brachial artery pressure wave is determined according to the amplitude of the frequency domain data of the brachial artery pressure wave, and the key point of the brachial artery pressure wave can be determined.

In one embodiment, the step of determining a keypoint in the brachial artery pressure wave may comprise: determining whether each piece of frequency domain data meets a preset condition, and determining the frequency domain data meeting the preset condition as a key point. The preset conditions comprise: the amplitude of the frequency domain data is larger than the amplitude of the previous frequency domain data and the amplitude of the next frequency domain data and larger than a preset threshold value.

For example, the preset threshold is h, vi is the amplitude of the ith frequency domain data, vi-1 is the amplitude of the previous frequency domain data, and vi+1 is the amplitude of the subsequent frequency domain data. When Vi > Vi-1, and Vi > Vi+1, and Vi > h, the data point is determined to be a keypoint and labeled 1; otherwise, the data point is not a critical point and is marked as 0.

Where percentile is a statistical term, if a set of data is ordered from small to large and a corresponding cumulative percentile is calculated, the value of the data corresponding to a certain percentile is referred to as the percentile of that percentile. For example, if the preset percentile is the 95 th percentile, the amplitude of the frequency domain data corresponding to the 95 th percentile is the preset threshold value in the frequency domain data sequence obtained by arranging the plurality of frequency domain data from small to large in amplitude. The preset threshold may also be selected empirically, and the embodiment of the present disclosure does not limit the preset threshold.

It can be appreciated that the key points are determined according to the amplitude of the frequency domain data, so that the wave crest in the brachial artery pressure wave can be accurately found, and the fluctuation state of the brachial artery pressure wave can be accurately determined.

Step 203, dividing the preset frequency range into a plurality of frequency intervals, and determining feature data corresponding to each frequency interval according to the key points.

Dividing a preset frequency range according to the length of a preset interval to obtain a plurality of frequency intervals; and calculating the characteristic data corresponding to each frequency interval according to the key points in each frequency interval to obtain a plurality of characteristic data.

The preset interval length can be a fixed length value, such as 50Hz, 75Hz and 100Hz; or may be an empirically set variable length value.

The feature data includes the number of keypoints in the frequency interval, the amplitude accumulated value of the keypoints, and the frequency accumulated value of the keypoints. Namely, determining the number of key points in the frequency interval for each frequency interval; summing the amplitudes of the key points to obtain an amplitude accumulated value of the key points; and carrying out summation calculation on the frequencies of the plurality of key points to obtain a frequency accumulated value of the key points. The number of key points, the amplitude accumulated value of the key points and the frequency accumulated value of the key points are taken as characteristic data of the frequency interval.

It can be appreciated that the more frequency intervals are divided, the more characteristic data are obtained, the more detailed the fluctuation state of the brachial artery pressure wave is described, and the more accurate the subsequent heart rhythm type detection is performed.

In the process of determining the key points in the brachial artery pressure wave according to the brachial artery pressure wave data and determining the characteristic data of the brachial artery pressure wave according to the key points, the terminal converts the brachial artery pressure wave data from a time domain to a frequency domain and selects the frequency domain data of the brachial artery pressure wave in a preset frequency range; then, according to the amplitude of the frequency domain data of the brachial artery pressure wave, the key points in the brachial artery pressure wave are found out; and determining the fluctuation state of the brachial artery pressure wave of each frequency interval in each preset frequency range according to the key points to obtain a plurality of characteristic data. According to the embodiment of the disclosure, the characteristic data representing the fluctuation state of the brachial artery pressure wave is obtained, so that the subsequent accurate heart rhythm type detection according to the characteristic data is facilitated.

In one embodiment, the heart rhythm type detection according to the characteristic data to obtain the heart rhythm type of the target object may include: and inputting the characteristic data into a pre-trained heart rhythm type detection model to obtain the heart rhythm type of the target object output by the heart rhythm type detection model.

The heart rhythm type detection model can be a support vector machine (Support Vector Machine, SVM) multi-classification model or other multi-classification models. The embodiments of the present disclosure are not limited in this regard.

The heart rhythm type detection model is trained in advance, after the feature data corresponding to the target object are obtained, the feature data are input into the heart rhythm type detection model, and the heart rhythm type detection model carries out classification processing according to the feature data and outputs classification results. The classification result is the heart rhythm type of the target object.

In practical application, a heart rhythm type detection model can be deployed at a terminal to detect heart rhythm types; the heart rhythm type detection model may also be deployed at the server, with the terminal performing heart rhythm type detection by providing an invocation interface.

In one embodiment, the heart rhythm type detection model and model parameters are saved as model files; and running the web server, loading the model file and generating a calling interface of the heart rhythm type detection model. The call interface may be a representational state transfer REST (Representational State Transfer) or other call interface.

Taking representational state transfer REST as an example, a heart rhythm type detection model is trained using a scikit-learn framework, and the trained heart rhythm type detection model and model parameters are persisted as model files using jobilib. Then, the flash server is run, invoking the heart rhythm type detection model and model parameters through the jobilib.load () load file. After that, the REST API is opened, and the heart rhythm type detection service can be provided to the terminal.

And in the process of detecting the heart rhythm type according to the characteristic data to obtain the heart rhythm type of the target object, training the heart rhythm type detection model in advance, and inputting the characteristic data into the heart rhythm type detection model after obtaining the characteristic data to obtain the heart rhythm type of the target object output by the heart rhythm type detection model. In the embodiment of the disclosure, the heart rhythm type is detected through a pre-trained heart rhythm type detection model, so that important auxiliary information is provided for diagnosis of a clinician; and is more intelligent than determining the type of heart rhythm from an electrocardiogram.

In one embodiment, as shown in FIG. 4, the process of training the heart rhythm type detection model may include:

step 301, obtaining sample feature data of a plurality of training objects and gold standards corresponding to the sample feature data.

The gold standard corresponding to the sample characteristic data is used for representing the heart rhythm type corresponding to the training object; the heart rhythm type corresponding to the training object is determined according to the diagnosis result of the electrocardiogram.

Acquiring feature data of a plurality of candidate objects; scoring the characteristic data of each candidate object by adopting a preset scoring function to obtain scoring results corresponding to each candidate object; and selecting a training object from the plurality of candidate objects according to the scoring result corresponding to each candidate object, and taking the characteristic data of the training object as sample characteristic data.

In one embodiment, acquiring the feature data of the plurality of candidates may include: the brachial artery pressure wave data of the candidate object is collected, and the characteristic data of the candidate object is determined according to the modes of the steps 301 to 302.

And acquiring electrocardiographic data of the candidate object while acquiring brachial artery pressure wave data of the candidate object, and determining the heart rhythm type corresponding to the candidate object according to the electrocardiographic data. After the candidate object is determined as the training object, the heart rhythm type corresponding to the candidate object is determined as a gold standard corresponding to the sample characteristic data of the training object, namely, the heart rhythm type corresponding to the training object is determined according to the diagnosis result of the electrocardiogram.

In one embodiment, the predetermined scoring function is a Fisher scoring function, such as equation (1):

wherein p represents the number of categories, μ represents the mean value of the characteristic data f, n _i Represents the number of samples of the i (i=1, 2, … … p) th class, μ ⁱ Sum sigma ⁱ The mean and variance of the feature data f in the i-th class of samples are represented.

It will be appreciated that the higher the score of Fisher, the better the characterization feature data.

In one embodiment, selecting a training object from a plurality of candidate objects according to a scoring result corresponding to each candidate object includes: sorting the plurality of candidate objects in order of the score from high to low; determining k candidate objects which are ranked at the front as training objects; wherein k is a positive integer.

For example, k is 10, and after the 15 candidates are arranged in order of the score from high to low, the top 10 candidates are taken as training objects.

Step 302, training the classification model based on the sample feature data of the plurality of training objects and the gold standard corresponding to each sample feature data to obtain a heart rhythm type detection model.

The classification model may be a support vector machine (Support Vector Machine, SVM) multi-classification model, or may be other classification models. The number of input features of the SVM multi-classification model is k, and a Gaussian radial basis function (Radial Basis Function, RBF) is adopted as a kernel function.

Obtaining a value range and a value step length of model parameters of the classification model; and determining a target value of the model parameter according to the value range and the value step length of the model parameter and the gold standard corresponding to the characteristic data of each sample to obtain the heart rhythm type detection model.

The value range of the model parameter defines the minimum value and the maximum value of the model parameter, and the value step length of the model parameter determines the value length of each updating of the model parameter. The value range and the value step length of the model parameters can be preset according to the experience value.

It can be understood that the sample feature data of a plurality of training objects and the gold standard corresponding to each sample feature data are obtained, and the classification model is trained based on the sample feature data of the plurality of training objects and the gold standard corresponding to each sample feature data to obtain the heart rhythm type detection model, so that a more intelligent detection method is provided for heart rhythm type detection, and the heart rhythm type detection can be popularized more.

In one embodiment, as shown in FIG. 5, step 302 may include:

step 401, initializing the model parameters according to the value ranges of the model parameters.

The classification model is a multi-classification model, and model parameters comprise the number of input features, error tolerance and support vector metrics. The larger the value of the error tolerance, the less tolerant the error, the easier the over-fitting, and conversely, the easier the under-fitting. The support vector metric is a parameter of the radial basis function, implicitly determining the distribution of the data after mapping to the new feature space, the larger the support vector metric, the smaller the support vector metric, and the more support vectors.

The method comprises the steps of presetting the number k of model parameter input characteristics, error tolerance c and the range of support vector measurement g to be [ k_begin, k_end ], [ c_begin, c_end ] and [ g_begin, g_end ], and initializing an SVM multi-classification model by adopting k_begin, c_begin and g_begin.

And step 402, training the initialized classification model based on sample feature data of a plurality of training objects to obtain the average accuracy corresponding to the classification model.

And obtaining the average accuracy corresponding to the classification model by adopting a 10-fold Cross Validation (Cross Validation) mode. As shown in fig. 6, the plurality of sample feature data is divided into 10 mutually exclusive subsets; and each time, 9 mutually exclusive subsets are adopted as training sets, the rest subsets are adopted as test sets to perform 10 times of training and testing, the average value of the accuracy of the 10 times of testing is adopted as average accuracy, and the average accuracy is stored in a data set P.

Wherein, the calculation formula of the average accuracy ac is as formula (2):

wherein ac _i The accuracy of the ith test, i.e., the ratio of the number of sample feature data divided into pairs to the total number of all sample feature data, is expressed.

And step 403, updating the value of each model parameter according to the value step length of the model parameter, and repeatedly executing the step of obtaining the average accuracy corresponding to the classification model until the value of the model parameter is updated, so as to obtain a plurality of average accuracy.

The value step sizes of the three model parameters are respectively set to be k_step, c_step and g_step. After the average accuracy is obtained according to the initialized k, c and g, updating the value of the model parameters by adopting a Grid Search mode. As shown in fig. 7, it is determined whether g is equal to g_end, and if not, g is updated to g+g_step according to the value step. Then, step 502 is repeatedly executed to obtain an average accuracy corresponding to the classification model, and the average accuracy is saved in the data set P.

If g is equal to g_end, it is determined whether c is equal to c_end. If c is not equal to c_end, c is updated to c+c_step according to the value step, then, the determination of g as g_begin and step 502 are repeatedly performed to obtain an average accuracy, and the average accuracy is stored in the data set P.

If c is equal to c_end, it is determined whether k is equal to k_end. If k is not equal to k_end, k is updated to k+k_step according to the value step, then setting g to g_begin, setting c to c_begin, and step 502 are repeatedly performed to obtain an average accuracy, and the average accuracy is stored in the data set P.

After the values of the model parameters are updated, a plurality of average accuracy rates can be obtained.

And step 404, selecting candidate values of model parameters according to a plurality of average accuracy rates, and performing performance detection on the classified model according to the candidate values of the model parameters and gold standards corresponding to the feature data of each sample to obtain a heart rhythm type detection model with the model performance meeting preset detection conditions.

And storing a plurality of average accuracy rates from the data set P, finding out the maximum value of the average accuracy rate, and determining the values of the model parameters k, c and g corresponding to the maximum value of the average accuracy rate as candidate values of the model parameters. Then, calculating the model performance of the classification model according to the candidate values of the model parameters and the gold standards corresponding to the feature data of each sample; the model performance comprises accuracy, sensitivity and specificity. And under the condition that the accuracy rate is determined to be larger than the preset accuracy rate, the sensitivity is larger than the preset sensitivity and the specificity is larger than the preset specificity, determining that the model performance meets the preset detection condition, and determining the candidate value of the model parameter as the target value of the model parameter when the model performance meets the preset detection condition to obtain the heart rhythm type detection model.

It can be appreciated that high sensitivity indicates that missed diagnosis is not easy, and can be used for screening, primary screening and exclusion diagnosis; the high specificity indicates that misdiagnosis is not easy, and can be used for carrying out accurate diagnosis.

In one embodiment, calculating the model performance of the classification model according to the candidate values of the model parameters and the gold standards corresponding to the feature data of each sample may include: adopting a binary Confusion Matrix (fusion Matrix), and assuming that a gold standard is only of two types, namely positive (positive) and negative (negative), wherein TP (True positives) in the Confusion Matrix is positive and is divided into the number of samples of positive by a classification model; FP (False positives) is the number of samples in the negative example but divided into positive examples by the classification model; FN (False negatives) is the number of samples in the positive example but divided into the negative example by the classification model; TN (True negatives) is the counterexample and is divided into the counterexample number of samples by the classification model. Wherein, the calculation formula of the total number of samples is as formula (3):

total number of samples = tp+fp+fn+tn— is- (3)

The calculation formula of the accuracy is formula (4):

ac＝(TP+TN)/(TP+FP+FN+TN)---------------------(4)

the calculation formula of the sensitivity is formula (5):

sn＝TP/(TP+FN)------------------------------(5)

the calculation formula of the specificity is formula (6):

sp＝TN/(FP+TN)---------------------------------(6)

in the process of training the classification model based on the sample feature data of the plurality of training objects and the gold standard corresponding to each sample feature data to obtain the heart rhythm type detection model, the grid search mode and the 10-fold cross validation mode are adopted to determine the value of the model parameters of the classification model, and the two classification confusion matrix is adopted to detect the model performance, so that the trained heart rhythm type detection model has the advantages of high accuracy, high sensitivity, high specificity and the like, and further the problems of missed diagnosis, misdiagnosis and the like of heart rhythm type detection results can be avoided, and therefore, the heart rhythm type detection model is adopted for detection, and the reliability is high.

In one embodiment, as shown in fig. 8, a method for detecting arrhythmia based on artificial intelligence of arterial pressure wave characteristics is provided, which may include:

step 501, brachial artery pressure wave data of a target object is acquired.

Step 502, converting the brachial artery pressure wave data from the time domain to the frequency domain, and performing frequency selection according to the preset frequency range to obtain frequency domain data of the brachial artery pressure wave.

In step 503, a key point in the brachial artery pressure wave is determined according to the amplitude of the frequency domain data of the brachial artery pressure wave.

In one embodiment, determining whether each piece of frequency domain data meets a preset condition, and determining the frequency domain data meeting the preset condition as a key point; the preset conditions comprise: the amplitude of the frequency domain data is larger than the amplitude of the previous frequency domain data and the amplitude of the next frequency domain data and larger than a preset threshold value.

Step 504, dividing the preset frequency range into a plurality of frequency intervals, and determining feature data corresponding to each frequency interval according to the key points.

In one embodiment, dividing a preset frequency range according to a preset interval length to obtain a plurality of frequency intervals; calculating characteristic data corresponding to each frequency interval according to the key points in each frequency interval; the feature data includes the number of keypoints in the frequency interval, the amplitude accumulated value of the keypoints, and the frequency accumulated value of the keypoints.

Step 505, inputting the characteristic data into a pre-trained heart rhythm type detection model to obtain the heart rhythm type of the target object output by the heart rhythm type detection model.

In the method for detecting arrhythmia based on artificial intelligence of the arterial pressure wave characteristics, a terminal acquires brachial artery pressure wave data of a target object, performs data conversion on the brachial artery pressure wave data, and selects effective frequency domain data of the brachial artery pressure wave; then, determining key points in the brachial artery pressure wave, and determining characteristic data corresponding to each frequency interval in a preset frequency range according to the key points; and finally, inputting the characteristic data into a pre-trained heart rhythm type detection model to obtain the heart rhythm type of the target object. The embodiment of the disclosure provides a method for detecting the type of heart rhythm based on brachial artery pressure wave data, and the method for detecting arrhythmia based on artificial intelligence of arterial pressure wave characteristics is applicable to various scenes and is easier to popularize because the brachial artery pressure wave data are easy to collect.

It should be understood that, although the steps in the flowcharts of fig. 1 to 8 are shown in order as indicated by the arrows, these steps are not necessarily performed in order as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least a portion of the steps of fig. 1-8 may include multiple steps or stages that are not necessarily performed at the same time, but may be performed at different times, nor does the order in which the steps or stages are performed necessarily occur sequentially, but may be performed alternately or alternately with other steps or at least a portion of the steps or stages in other steps.

In one embodiment, as shown in FIG. 9, there is provided an apparatus for artificial intelligence detection of arrhythmias based on arterial pressure wave characteristics, comprising:

a pressure wave data acquisition module 601 for acquiring brachial artery pressure wave data of a target object;

the characteristic data determining module 602 is configured to determine a key point in the brachial artery pressure wave according to the brachial artery pressure wave data, and determine characteristic data of the brachial artery pressure wave according to the key point; the characteristic data are used for representing the fluctuation state of the brachial artery pressure wave;

the heart rhythm type detection module 603 is configured to perform heart rhythm type detection according to the feature data to obtain a heart rhythm type of the target object.

In one embodiment, the feature data determining module 602 includes:

In one embodiment, the above-mentioned heart rhythm type detection module 603 is specifically configured to input the feature data into a pre-trained heart rhythm type detection model to obtain a heart rhythm type of the target object output by the heart rhythm type detection model.

In one embodiment, as shown in fig. 10, the apparatus further comprises:

a training sample acquiring module 604, configured to acquire sample feature data of a plurality of training objects and gold standards corresponding to the sample feature data; the gold standard corresponding to the sample characteristic data is used for representing the heart rhythm type corresponding to the training object;

the training module 605 is configured to perform training of the classification model based on sample feature data of a plurality of training objects and gold standards corresponding to the sample feature data, and obtain a heart rhythm type detection model.

In one embodiment, the training sample acquiring module 704 includes:

In one embodiment, the training module 605 includes:

In one embodiment, the classification model is a support vector machine SVM multi-classification model, and the training submodule is specifically configured to initialize model parameters according to a value range of the model parameters; model parameters include input feature quantity, error tolerance and support vector metrics; training the initialized classification model based on sample feature data of a plurality of training objects to obtain average accuracy corresponding to the classification model; updating the value of each model parameter according to the value step length of the model parameter, and repeatedly executing the steps of obtaining the average accuracy corresponding to the classification model until the value of the model parameter is updated, so as to obtain a plurality of average accuracy; and selecting candidate values of model parameters according to the plurality of average accuracy rates, and performing performance detection on the classification model according to the candidate values of the model parameters and gold standards corresponding to the feature data of each sample to obtain a heart rhythm type detection model with model performance meeting preset detection conditions.

In one embodiment, as shown in fig. 11, the apparatus further comprises:

a model file saving module 606 for saving the heart rhythm type detection model and the model parameters as a model file;

the calling interface generating module 607 is used for running the web server, loading the model file and generating the calling interface of the heart rhythm type detection model.

Specific limitations regarding the means for artificially and intelligently detecting arrhythmias based on arterial pressure wave characteristics may be found in the above limitations regarding the method for artificially and intelligently detecting arrhythmias based on arterial pressure wave characteristics, and will not be described in detail herein. The various modules in the above-described device for artificial intelligence based on arterial pressure wave characteristics for detecting arrhythmias may be implemented in whole or in part in software, hardware, and combinations thereof. The above modules may be embedded in hardware or independent of a processor in the electronic device, or may be stored in software in a memory in the electronic device, so that the processor may call and execute operations corresponding to the above modules.

In one embodiment, an electronic device is provided, which may be a terminal, and an internal structure diagram thereof may be as shown in fig. 12. The electronic device includes a processor, a memory, a communication interface, a display screen, and an input device connected by a system bus. Wherein the processor of the electronic device is configured to provide computing and control capabilities. The memory of the electronic device includes a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The communication interface of the electronic device is used for carrying out wired or wireless communication with an external terminal, and the wireless mode can be realized through WIFI, an operator network, NFC (near field communication) or other technologies. The computer program is executed by a processor to implement a method of artificially detecting arrhythmias based on arterial pressure wave characteristics. The display screen of the electronic equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic equipment can be a touch layer covered on the display screen, can also be keys, a track ball or a touch pad arranged on the shell of the electronic equipment, and can also be an external keyboard, a touch pad or a mouse and the like.

It will be appreciated by those skilled in the art that the structure shown in fig. 12 is merely a block diagram of a portion of the structure associated with the present application and is not limiting of the electronic device to which the present application is applied, and that a particular electronic device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.

In one embodiment, an electronic device is provided that includes a memory having a computer program stored therein and a processor that when executing the computer program performs the steps of:

acquiring brachial artery pressure wave data of a target object;

In one embodiment, the processor when executing the computer program further performs the steps of:

In one embodiment, the brachial artery pressure wave data includes up-down pressure data and constant pressure data; the elevation pressure data comprises systolic pressure data and diastolic pressure data; the constant pressure data and the lifting pressure data are in linear relation.

acquiring feature data of a plurality of candidate objects;

In one embodiment, the classification model is a multi-classification model, and the processor when executing the computer program further performs the following steps:

In one embodiment, a computer readable storage medium is provided having a computer program stored thereon, which when executed by a processor, performs the steps of:

acquiring brachial artery pressure wave data of a target object;

In one embodiment, the computer program when executed by the processor further performs the steps of:

acquiring feature data of a plurality of candidate objects;

In one embodiment, the classification model is a multi-classification model, and the computer program when executed by the processor further performs the steps of:

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, or the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory. By way of illustration, and not limitation, RAM can be in the form of a variety of forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), and the like.

The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.

The above examples merely represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the invention. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application is to be determined by the claims appended hereto.

Claims

1. An apparatus for artificially and intelligently detecting arrhythmias based on characteristics of arterial pressure waves, the apparatus comprising:

the characteristic data determining module is used for converting brachial artery pressure wave data from a time domain to a frequency domain, and carrying out frequency selection according to a preset frequency range to obtain frequency domain data of the brachial artery pressure wave; determining whether each piece of frequency domain data meets a preset condition, and determining the frequency domain data meeting the preset condition as a key point; dividing the preset frequency range according to the length of a preset interval to obtain a plurality of frequency intervals; calculating characteristic data corresponding to each frequency interval according to the key points in each frequency interval; the characteristic data comprises the number of key points in the frequency interval, the amplitude accumulated value of the key points and the frequency accumulated value of the key points; wherein the characteristic data is used for characterizing the fluctuation state of the brachial artery pressure wave; the preset conditions comprise: the amplitude of the frequency domain data is larger than the amplitude of the previous frequency domain data and the amplitude of the next frequency domain data and larger than a preset threshold value;

2. The apparatus of claim 1, wherein the predetermined threshold is a predetermined percentile in a sequence of frequency domain data; the frequency domain data sequence is a data sequence obtained by arranging a plurality of frequency domain data according to the amplitude from small to large.

3. The apparatus of claim 1, wherein the device comprises a plurality of sensors,

the brachial artery pressure wave data comprise pressure increasing and decreasing data and constant pressure data; the elevation pressure data comprises systolic pressure data and diastolic pressure data; the constant pressure data and the lifting pressure data are in linear relation.

4. The device according to any one of claim 1 to 3, wherein,

the heart rhythm type detection module is further used for inputting the characteristic data into a pre-trained heart rhythm type detection model to obtain the heart rhythm type of the target object output by the heart rhythm type detection model.

5. The apparatus of claim 4, wherein the apparatus further comprises:

And the training module is used for training the classification model based on the sample characteristic data of the plurality of training objects and the gold standard corresponding to each sample characteristic data to obtain the heart rhythm type detection model.

6. The apparatus of claim 5, wherein the training sample acquisition module comprises:

and the training object determining submodule is used for selecting the training object from the plurality of candidate objects according to the scoring result corresponding to each candidate object, and taking the characteristic data of the training object as the sample characteristic data.

7. The apparatus of claim 6, wherein the device comprises a plurality of sensors,

the training object determining submodule is further used for sequencing the plurality of candidate objects according to the order of the scores from high to low; determining k candidate objects which are ranked at the front as the training objects; wherein k is a positive integer.

8. The apparatus of claim 5, wherein the training module comprises:

and the training sub-module is used for determining a target value of the model parameter according to the value range and the value step length of the model parameter and gold standards corresponding to the sample characteristic data to obtain the heart rhythm type detection model.

9. The apparatus of claim 8, wherein the device comprises a plurality of sensors,

the training submodule is used for initializing the model parameters according to the value range of the model parameters; the model parameters comprise input feature quantity, error tolerance and support vector metrics; training the initialized classification model based on the sample feature data of the plurality of training objects to obtain the average accuracy corresponding to the classification model; updating the value of each model parameter according to the value step length of the model parameter, and repeatedly executing the steps of obtaining the average accuracy corresponding to the classification model until the value of the model parameter is updated, so as to obtain a plurality of average accuracy; and selecting candidate values of the model parameters according to the average accuracy rates, and performing performance detection on the classification model according to the candidate values of the model parameters and gold standards corresponding to the sample characteristic data to obtain the heart rhythm type detection model with model performance meeting preset detection conditions.

10. The apparatus of claim 9, wherein the device comprises a plurality of sensors,

the training submodule is further used for dividing the sample characteristic data into 10 mutually exclusive subsets; and each time, 9 mutually exclusive subsets are adopted as training sets, the rest subsets are adopted as test sets to perform 10 times of training and testing, and the average value of the accuracy of 10 times of testing is adopted as the average accuracy.

11. The apparatus of claim 9, wherein the device comprises a plurality of sensors,

the training submodule is further used for calculating the model performance of the classification model according to the candidate value of the model parameter; the model performance comprises accuracy, sensitivity and specificity; and under the condition that the accuracy is larger than the preset accuracy, the sensitivity is larger than the preset sensitivity and the specificity is larger than the preset specificity, determining that the model performance meets the preset detection condition, and determining candidate values of model parameters as target values of the model parameters when the model performance meets the preset detection condition to obtain the heart rhythm type detection model.

12. The apparatus of claim 4, wherein the apparatus further comprises:

the model file storage module is used for storing the heart rhythm type detection model and the model parameters as model files;

And the calling interface generating module is used for running a web server, loading the model file and generating a calling interface of the heart rhythm type detection model.