Specific implementation mode
Below by drawings and examples, technical scheme of the present invention will be described in further detail.
Dynamic monitor system proposed by the present invention carries out multi-parameter sign monitoring towards non-in-patient.Different from clinic
Monitoring, the real-time of the data transmission involved by dynamic monitor system of the invention, data transmission, storage and exception report and
Response mechanism all has essential distinction with clinical monitoring.
In addition, by existing patient monitor research it was found that the electrocardio of detection, blood pressure, blood oxygen, pulse, breathing,
In the vital signs index such as body temperature, cardiac electrical monitoring is to be different from other parameters, is believed by the electrocardio that sensor obtains
It number needs just to extract effective information therein by the algorithm calculating of a series of complex, be processed for other opposite signals
The more complicated difficulty of journey, and it is susceptible to the link of detection mistake.
Electrocardiosignal is the weak current that the electrical activity of cardiac muscle cell reflects in body surface, is remembered by external electrode and amplification
Recording system is recorded.Other non-cardiogenic electric signals, such as skeletal muscle activity can also be recorded simultaneously in recording process
Electromyography signal interference brought etc..Therefore it is considered that needing to carry out effective disturbance ecology and exclusion to electrocardiosignal,
It can effectively reduce because being reported by mistake caused by interference signal.
In addition, electrocardiosignal is the embodiment of myocardial electrical activity process, therefore electrocardiosignal is in addition to that can be used for detecting heart rate
In addition, the information of a large amount of heart state can also be embodied.When heart state goes wrong, electrocardiosignal will appear
It is corresponding to change.It was found that existing monitoring is set during to existing multi-parameter monitoring equipment is studied in the industry
It is standby to carry out very limited analysis and alarm to electrocardiosignal.In this regard, knowing in addition to carrying out effective interference to electrocardiosignal
It not and excludes, to reduce because except wrong report caused by interference signal, it is believed that can also be improved from the following:
First, it needs to accurately identify P waves, T waves in heartbeat feature extraction, the more inspections that can be detected to avoid heartbeat
And missing inspection, such as to some special ECG signals, such as the letter of the tall and big T waves of the slower patient of the rhythm of the heart or T wave hypertrophys
Number more inspections.
Second, more careful division is carried out to the classification of heartbeat, and cannot only rest on sinus property, supraventricular and room property this
Three kinds of classification require to meet the complicated comprehensive analysis of electrocardiogram doctor.
Third accurately identifies room and flutters atrial fibrillation and ST-T changes, changes ST sections and T waves to the heart so as to help to provide
The help of myocardial ischemia analysis.
4th, heartbeat and cardiac electrical event are accurately identified.
In the present invention, we are directed to above-mentioned several points, by the analysis calculating to electrocardiogram (ECG) data, especially introduce artificial intelligence
Energy (AI) technology is fluttered and is trembleed to the digital signal of acquisition carries out arrhythmia analysis, long pause stops fighting, and block is early
It fights and escape beat, bradycardia is aroused in interest too fast, and ST sections change detection, the analyses and classification of cardiac electrical event, and accurate report is generated to reach
The purpose of alert signal, to effectively carry out the monitoring of patient vital signs.
There is data to suggest that 90% or more heart disease burst all occurs except medical institutions, therefore for having
The crowd of heart disease history records and monitors the heart condition under daily state, is necessary.
Based on above-mentioned discovery, the present invention proposes a kind of user oriented sign information dynamic monitor method, can apply
In the user oriented monitoring outside medical institutions.It may be implemented in various dynamic monitor equipment, including wearable device.Under
Face combines the flow chart of user oriented sign information dynamic monitor method shown in FIG. 1, to the user oriented body of the present invention
Reference breath dynamic monitor method is described in detail.In the sign information dynamic monitor of the present invention, the monitoring to electrocardiosignal is again
Most important one.
As shown in Figure 1, the user oriented sign information dynamic monitor method of the present invention mainly includes the following steps:
Step 110, the monitoring criteria data that dynamic monitor equipment receives user's input or server issues;
Specifically, dynamic monitor equipment can be specifically the multi-parameter for including single lead or multi-lead electrocardio monitoring function
Monitoring device, every dynamic monitor equipment have unique device id.When dynamic monitor equipment is assigned to a use to be monitored
Family in use, can be according to the user the case where, corresponding monitoring criteria data are configured in dynamic monitor equipment.Therefore, it needs
The measurand information of measurand is determined first.
For cardioelectric monitor, monitoring criteria data can be understood as indicating monitored user's electrocardiosignal
The reference data or information for whether needing to generate alarm whether normal, for different users, the setting of monitoring criteria data
Can be different, it can specifically believe by way of configuring input in dynamic monitor equipment or by server according to user
Breath is configured and is issued to the mode of dynamic monitor equipment and obtained.In the present embodiment, monitoring criteria data may include having
Measurand information and the anomalous ecg event information set.Anomalous ecg event information includes generation anomalous ecg in need
The information of the various anomalous ecg events of alarm is acquired ECG data, analyzes etc. and is a series of in dynamic monitor equipment
Processing can be by determining whether the anomalous ecg event is electrocardio when obtaining the anomalous ecg event of ECG data instruction
Event specified in abnormal events information and determine whether generate alarm.
For other physical sign parameters, such as pulse data, blood pressure data, breath data, blood oxygen saturation data and body temperature
Data, monitoring criteria data can be the corresponding parameter thresholds set.The parameter threshold of each parameter can have different more
Group parameter threshold, can select according to the actual conditions of monitored person.Preferably, in the present invention it is possible to before monitoring
Monitoring criteria data are determined previously according to measurand, then determine corresponding given threshold according to monitoring criteria data.
In this example, monitoring criteria data include at least measurand information and anomalous ecg event information.
Step 120, dynamic monitor equipment carries out monitoring data acquisition to measurand, obtains the sign monitoring of measurand
Data;
Specifically, dynamic monitor equipment has the sign signal acquisitions such as electrode, probe, the cuff being in contact with measurand
Device is acquired the sign of measurand by sign signal acquisition device, and obtains sign monitoring by digitized processing
Data.Sign monitoring data includes at least ECG data, it is also possible to including pulse data described above, blood pressure data, exhale
Inhale data, blood oxygen saturation data and temperature data etc..There is sign monitoring data time attribute information, each data point to have
Corresponding data acquisition time, this time are time attribute information.While carrying out data acquisition, the acquisition of this data
Time also is recorded, and the time attribute information as sign monitoring data is stored.
To more fully understand the intent of the present invention and realization method, below to the acquisition method of all kinds of sign monitoring datas and
Principle carries out briefly introducing explanation:
ECG data:Guard of honor heart cell bioelectrical activity is recorded by the ecg signal acquiring of noninvasive ECG examination
The signal of generation is acquired record in the form of single lead or multi-lead.
Pulse data:Pulse be arteries with heart relax contracting and the phenomenon that periodical pollex, pulse include intravascular pressure,
The variation of a variety of physical quantitys such as volume, displacement and wall tension.We preferably use photoelectricity positive displacement pulses measure, sensor
It is made of, is could be sandwiched on the finger tip or auricle of measured light source and photoelectric transformer two parts.Light source is selected to oxygen in arterial blood
The selective wavelength of hemoglobin is closed, for example uses spectrum in the light emitting diode of 700-900nm.This Shu Guang is through outside human body
All blood vessels change the light transmittance of this Shu Guang when arterial hyperemia volume changes, by photoelectric transformer receive through tissue transmission or
The light of reflection is changed into electric signal and amplifier is sent to amplify and export, and thus reflects arterial vascular volume variation.Pulse is to follow one's inclinations
Dirty pollex and periodically variable signal, arteries volume also change periodically, the signal intensity week of photoelectric transformer
Phase is exactly pulse frequency, i.e. pulse data.
Blood pressure data:The maximum pressure reached when heart contraction is known as systolic pressure, and blood is advanced to aorta by it, and
Maintain systemic circulation.The minimum pressure reached when cardiac enlargement is known as diastolic pressure, it enables blood to flow back into atrium dextrum.Blood pressure
Integral divided by heart cycle T of the waveform in one week are known as mean pressure.The measurement of blood pressure data can be realized there are many method, specifically may be used
It is divided into invasive measurement and non-invasive measurement.We preferably use Korotkoff's Sound method and the noninvasive survey of two class of vibration measuring method in multi-parameter monitor
Amount method.Korotkoff's Sound method is to detect the Korotkoff's Sound (pulse sound) under cuff to measure blood pressure, Korotkoff's Sound non-invasive blood pressure monitoring system
System includes cuff inflation system, cuff, korotkoff sound sensor, audio is amplified and automatic gain adjustment circuit, A/D converter, micro-
Processor and display unit grade.Vibration measuring method is to detect the oscillation wave of gas in gas sleeve, and oscillation wave is derived from the beating of vascular wall, measures
The reference point of oscillation wave can measure blood pressure data, including systolic pressure (PS), diastolic pressure (PD) and mean pressure (PM).Vibration measuring method obtains
The method for obtaining pulsatile motion wave can obtain pulsatile motion wave by measurement and obtain blood pressure number by microphone and pressure sensor
According to.For under some special application scenarios, blood pressure data can also be obtained by way of invasive measurement.Such as weight
Some patients in the ward disease intensive care group (ICU) are monitored, so that it may by being directly intubated in artery, by the another of intubation
One end is connected to the real-time acquisition that blood pressure data is realized in the pressure detecting system for filling liquid sterilized.This invasive monitoring
The advantages of method includes:Blood pressure size can be shown in real time, and can show continuous blood pressure waveform;In low blood pressure
State can have accurate reading;The patient comfort of long-term record gets a promotion, and avoids inflation deflation for a long time in non-invasive measurement
Caused wound;More information can be extracted, include that can extrapolate Vascular capacitance etc. from the form of blood pressure waveform.
Breath data:Respiration measurement is the pith of lung kinetic energy inspection.Patient monitor is exhaled by measuring respiratory wave to measure
Frequency (beat/min) is inhaled to get to breath data.The measurement of respiratory rate can directly measure respiratory air flow by thermistor
This variation is transformed into voltage signal by temperature change by bridge circuit;Impedance method can also be used to measure respiratory rate, because
For respiratory movement when, wall of the chest muscle alternation tension and relaxation, thorax alternately deforms, and the electrical impedance of injected organism tissue also alternately changes therewith.It surveys
The various ways such as bridge method, modulation method, constant pressure source method and Method of constant flow source can be used in the variation of amount respiratory impedance value.In patient monitor
In, respiratory impedance electrode can also be shared with electrocardioelectrode, can detect respiratory impedance variation and breathing when detecting electrocardiosignal simultaneously
Frequency.
Blood oxygen saturation data:Blood oxygen saturation is to weigh the important parameter for the ability that blood of human body carries oxygen.Blood oxygen is full
Transmission beam method (or bounce technique) dual wavelength (feux rouges R and infrared light IR) photoelectric detecting technology may be used in measurement with degree, and detection is red
The ratio between alternating component caused by the light absorption of light and infrared light by arterial blood and non-pulsating tissue (epidermis, muscle, venous blood
Deng) stabilization component (direct current) value for causing light absorption, by can be calculated oximetry value SpO2, i.e. blood oxygen saturation number
According to.Since the pulsation rule of photosignal is consistent with the rule of heartbeat, so also can be simultaneously according to the period of detecting signal
Determine pulse data.
Temperature data:Body temperature is the important indicator for understanding life state.The measurement of body temperature uses the temperature-sensitive of negative temperature coefficient
Resistance is as temperature sensor, using electric bridge as detection circuit.Integrated thermometric may be used in we in specific application
Circuit measures to obtain temperature data.Also the temperature measurement circuit that twice or more can be used, measures the temperature difference pair of two different parts
Measured value is modified.Body surface probe and body cavity probe can also be used, guards body surface and cavity temperature respectively.It is special at some
Application in, in order to avoid cross infection, using infrared Infrared Technique can also put question to the monitoring of data.
In patient monitor, we set temperature measurement accuracy at 0.1 DEG C, to there is faster thermometric to respond.
In the present invention, we can by the above method to measurand carry out sign monitoring data acquisition, obtain by
Survey the sign monitoring data of object.
In front it has been noted that cardiac electrical monitoring is relative to vital signs such as other blood pressures, blood oxygen, pulse, breathing, body temperature
The monitoring of index is increasingly complex, therefore ECG data is used different from other sign monitoring datas in the present invention
Processing method, the specific identification for using the electrocardiogram automatic analysis method based on artificial intelligence self study to carry out ECG data,
Processing and abnormal judgement.
Sign information dynamic monitor is carried out in following flows and for mainly for the processing procedure of ECG data
The explanation of method.Other sign datas can be used as reference information, is combined with the handling result of ECG data, to carry out quilt
The judgement of human observer sign state.
Step 130, wave group feature recognition is carried out to ECG data, the characteristic signal of ECG data is obtained, according to spy
Reference number carries out beat classification to ECG data, and beat classification information is obtained in conjunction with electrocardiogram basic law reference data, and
Generate ECG events data;
Specifically, in this example, the processing to ECG data can execute in dynamic monitor equipment, can also be
It executes in the server.Wherein, dynamic monitor equipment can be attached by wired mode or wireless network and server into
Row data transmission.Data transmission is preferably realized using wireless network when carrying out real-time Data Transmission.It is mentioned here wireless
Network includes wireless but is not limited to the WLAN (WIFI) based on IEEE 802.11b standards, bluetooth, 3G/4G/5G movements
Communication network, the modes such as Internet of Things.
After dynamic monitor monitoring of equipment obtains ECG data, ECG data can be stored in dynamic monitor equipment
It is local, it can also be to send data in server to be preserved, dynamic monitor equipment be loaded in the data of transmission
The information of device id.The information of monitored person is obtained so as to corresponding accordingly.
The processing procedure of the ECG data of the present invention, uses the electrocardiogram based on artificial intelligence self study and automatically analyzes
Method is realized based on artificial intelligence convolutional neural networks (CNN) model.CNN models are the supervision in deep learning
Learning method is exactly multi-layer network (hidden layer hidden layer) connection structure of a simulative neural network, input signal
Each hidden layer is passed sequentially through, carries out Mathematical treatment (Convolution convolution, the ponds Pooling of a series of complex wherein
Change Regularization regularizations, prevents over-fitting, Dropout from temporarily abandoning, Activation activation, generally uses
Relu activation primitives), some features of object to be identified are successively automatically taken out, are then passed again using these features as input
It is delivered to higher leveled hidden layer to be calculated, the entire letter of to the last several layers of full articulamentum (Full Connection) reconstruct
Number, it carries out logic (logistics) using Softmax functions and returns, reach the classification of multiple target.
CNN belongs to the supervised learning method in artificial intelligence, and in the training stage, input signal is by multiple hidden layers
Reason reaches last full articulamentum, and the classification results that softmax logistic regressions obtain, (label is marked with known classification results
Label) between can be there are one error, a core concept of deep learning is exactly by a large amount of sample iteration come constantly minimum
Change this error, to which the parameter for connecting each hidden layer neuron be calculated.This process generally requires construction one especially
Loss function (cost function), utilize the gradient descent algorithm and error backpropagation algorithm of nonlinear optimization
(backpropagation algorithm, BP), fast and effeciently minimization entire depth (number of plies of hidden layer) and range
All Connecting quantities in (dimension of feature) all sufficiently complex neural network structure.
The data that needs identify are input to training pattern by deep learning, by the first hidden layer, the second hidden layer, third
Hidden layer is finally output recognition result.
In the present invention, wave group feature recognition, disturbance ecology, beat classification etc. are carried out to ECG data and is all based on people
The training pattern of work intelligent self-learning is exported as a result, analyze speed is fast, and order of accuarcy is high.
The processing procedure to ECG data is either executed in dynamic monitor equipment or in the server, it is specific
Processing procedure can refer to flow shown in Fig. 3, carry out as steps described below,
Step 131, the data format of ECG data is converted into preset standard data format by resampling, and to turning
The ECG data of preset standard data format after changing carries out first and is filtered;
Specifically, the format adaptation of ECG data is read, there is different readings to realize different equipment, after reading,
Need adjust baseline, according to gain conversions at millivolt data.By data resampling, converting data to whole process can be handled
Sample frequency.Then high frequency is removed by filtering, it is accurate to improve artificial intelligence analysis for the noise jamming and baseline drift of low frequency
Rate.By treated, ECG data is preserved with preset standard data format.
It is solved by this step different using different leads, the difference of sample frequency and transmission data format, Yi Jitong
Cross digital signal filter removal high frequency, the noise jamming and baseline drift of low frequency.
High-pass filter, low-pass filter and medium filtering can be respectively adopted in digital signal filter, Hz noise, flesh
Electrical interference and baseline drift interference are eliminated, and the influence to subsequent analysis is avoided.
More specifically, low pass, the progress zero phase-shift filtering of high pass Butterworth filter may be used, to remove baseline drift
And High-frequency Interference, retain effective electrocardiosignal;Medium filtering can then utilize data point electricity in the sliding window of preset duration
The median of pressure amplitude value substitutes the amplitude of window center sequence.The baseline drift of low frequency can be removed.
Step 132, the ECG data after being filtered to first carries out heartbeat detection process, identifies ECG data packet
The multiple heartbeat data included;
Specifically, each heartbeat data correspond to a cardiac cycle, including the amplitude of corresponding P waves, QRS complex, T waves and
Beginning and ending time data.Heartbeat detection in this step is made of two processes, when signal processing, from first filtering
The characteristic spectra of QRS complex is extracted in treated ECG data;Second is that determining QRS complex by the way that rational threshold value is arranged
Time of origin.Generally can include P waves, QRS complex, T wave components and noise contribution in electrocardiogram.The frequency of general QRS complex
Rate range, between 20Hz, can propose QRS wave group congruences 5 by a bandpass filter within this range.However P
Wave, the frequency range of the frequency range of T waves and noise and QRS complex frequency range overlap, therefore not by the method for signal processing
The signal of non-QRS complex can be completely removed.Therefore it needs to extract QRS complex from signal envelope by the way that rational threshold value is arranged
Position.Specific detection process is a kind of process based on peak detection.Threshold value is carried out for each peak value sequence in signal
Judge, enters QRS complex when more than threshold value and judge flow, carry out the detection of more features, such as phase, form etc. between RR.
Multi-parameter monitor often carry out non-volatile recording, during heartbeat signal amplitude and frequency at every moment
All changing, and under morbid state, this characteristic can show stronger.When carrying out threshold value setting, need according to data
Feature dynamically carries out adjusting thresholds in the situation of change of time domain.In order to improve the accuracy rate and positive rate of detection, QRS complex inspection
It surveys the big mode for mostly using double width degree threshold value binding time threshold value to carry out, there is high threshold higher positive rate, Low threshold to have
Higher Sensitivity rate, the phase is more than certain time threshold value between RR, is detected using Low threshold, and missing inspection situation is reduced.And low threshold
Value is easy to be influenced by T waves, myoelectricity noise since threshold value is relatively low, be easy to cause more inspections, therefore preferentially carried out using high threshold
Detection.
All there is lead parameter for the heartbeat data of different leads, to characterize the heart which lead is the heartbeat data be
It fights data.Therefore also the information that source determines its lead can be transmitted according to it while obtaining ECG data, it will
Lead parameter of this information as heartbeat data.
Step 133, the detection confidence level of each heartbeat is determined according to heartbeat data;
Specifically, confidence calculations module is during heartbeat detects, according in the phase between the amplitude and RR of QRS complex
The amplitude proportional of noise signal can provide the estimated value that confidence level is detected for QRS complex.
Step 134, disturbance ecology is carried out to heartbeat data according to two disaggregated model of disturbance ecology, whether obtains heartbeat data
There are interfering noises, and a probability value for judging interfering noise;
Because multi-parameter monitor is easily influenced interference phenomenon occur during non-volatile recording by a variety of, lead to acquisition
Heartbeat data invalid or inaccuracy cannot correctly reflect the situation of testee, while also increase diagnosis difficulty and workload;
And interference data are also the principal element for causing intellectual analysis tool not work effectively.Therefore, outer signals are interfered and is dropped
It is particularly important to minimum.
This step has that precision is high based on using deep learning algorithm as the end-to-end two Classification and Identifications model of core, extensive
The strong feature of performance, can efficiently solve that electrode slice, which falls off, the main interference source such as motion artifacts and electrostatic interference generates disturbs
Dynamic problem, overcome traditional algorithm because interference data variation it is various it is irregular caused by recognition effect difference problem.
It can specifically realize by the following method:
Step A carries out disturbance ecology to heartbeat data using two disaggregated model of disturbance ecology;
Step B identifies in heartbeat data, heartbeat interval be more than or equal to it is default between phase decision threshold data slot;
Step C, the data slot that phase decision threshold between presetting is more than or equal to heartbeat interval carry out abnormal signal judgement, really
Whether fixed is abnormal signal;
Wherein, the identification of abnormal signal mainly include whether to fall off for electrode slice, low-voltage situations such as.
Step D, then with preset time width, is determined sliding in data slot if not abnormal signal according to setting duration
The initial data point and termination data point of dynamic sampling, and sliding sampling is carried out to data slot by initial data point, until end
Only until data point, multiple sampled data sections are obtained;
Step E carries out disturbance ecology to each sampled data section.
Above-mentioned steps A-E is illustrated with a specific example.That the heart rate of each lead is set according to this
One data volume carries out cutting sampling, is then separately input to two disaggregated model of disturbance ecology and classifies, and obtains disturbance ecology knot
A probability value for fruit and corresponding result;It is more than or equal to 2 seconds heartbeat data to heartbeat interval, first judges whether it is that signal overflows
Go out, low-voltage, electrode delamination;If not the above situation, just according to the first data volume, since the heartbeat of the left side, to fight continuity
It is not overlapped sliding sampling with the first data volume, is identified.
Input can be the first data volume heartbeat data of any lead, then two disaggregated model of disturbance ecology be used to carry out
Whether classification, directly output are the classification results interfered, and acquisition result is fast, and accuracy is high, and stability is good, can be carried for subsequent analysis
For more effective good data.
Because data is interfered often as caused by the effect of external disturbance factor, mainly to have electrode slice to fall off, low electricity
Situations such as pressure, electrostatic interference and motion artifacts, the interference data that not only different disturbing sources generate are different, and identical disturbing source produces
Raw interference data are also varied;Simultaneously in view of although interference data diversity cloth is wider, the difference with normal data
It is different very big, so being also to ensure diversity as far as possible, while moving window being taken to slide when collecting the training data of interference
Sampling increases the diversity of interference data as far as possible, so that model is more robust to interference data, even if following interference data
Any interference different from the past, but compared to normal data, can also be more than normal data with the similarity of interference, to make
Model Identification interferes the ability enhancing of data.
Two disaggregated model of disturbance ecology used in this step can with as shown in figure 3, network first use level 2 volume lamination,
Convolution kernel size is 1x5, and a maximum value pond is added after every layer.Convolution kernel number often passes through primary maximum pond since 128
Change layer, convolution kernel number is double.It is two full articulamentums and a softmax grader after convolutional layer.Due to the model
Number of classifying is 2, so there are two output units by softmax, respective classes is corresponding in turn to, using cross entropy as loss function.
Training for the model, we use the data slot accurately marked from 300,000 patients nearly 4,000,000.Mark
Note is divided into two classes:Normal ECG signal either has the ECG signal segment significantly interfered with.We pass through customized development
Tool carries out segment mark, then preserves interference fragment information with self-defined standard data format.
In training process, carries out tens repeating queries using two GPU servers and train.In a specific example, adopt
Sample rate is 200Hz, and data length is a segment D [300] of 300 ecg voltage values (millivolt), and input data is:
InputData (i, j), wherein i is i-th of lead, and j is j-th of segment D of lead i.Input data all by breaing up at random
Just start to train, ensure that training process restrains, meanwhile, too many sample is collected in control from the ECG data of the same patient
This, improves the generalization ability of model, both the accuracy rate under real scene.After training convergence, 1,000,000 independent test datas are used
It is tested, accuracy rate can reach 99.3%.Separately there is specific test data such as the following table 1.
|
Interference |
Normally |
Sensitivity rate (Sensitivity) |
99.14% |
99.32% |
Positive prediction rate (Positive Predicitivity) |
96.44% |
99.84% |
Table 1
Step 135, the validity of heartbeat data is determined according to detection confidence level, also, according to the effective heart rate of determination
According to lead parameter and heartbeat data, result based on disturbance ecology and time rule, which merge, generates heart beat time sequence data,
And heartbeat is generated according to heart beat time sequence data and analyzes data;
Specifically, due to the complexity of ECG signal and each lead may by different degrees of interference effect,
There can be the case where more inspections and missing inspection by single lead detection heartbeat, different leads detect the time representation number of heartbeat result
According to not being aligned, so needing to merge the heartbeat data of all leads according to disturbance ecology result and time rule, life
At a complete heart beat time sequence data, the time representation data of unified all lead heartbeat data.Wherein, time representation
Data are for indicating temporal information of each data point on electrocardiographic data signals time shaft.When according to this unified heartbeat
Between sequence data can use the threshold values that pre-sets when subsequent analysis calculates, each lead heartbeat data are cut
It cuts, the heartbeat to generate each lead that concrete analysis needs analyzes data.
The heartbeat data of above-mentioned each lead need to be determined according to the detection confidence level obtained in step 133 before merging
The validity of heartbeat data.
Specifically, the heartbeat data merging process that lead heartbeat merging module executes is as follows:According to electrocardiogram basic law
The refractory period of reference data obtains the time representation data combination of different lead heartbeat data, abandons the larger heartbeat of its large deviations
Data, voting for generating to the combination of above-mentioned time representation data merges heart beat locations, when will merge heart beat locations addition merging heartbeat
Between sequence, be moved to next group of pending heartbeat data, cycle executes until completing the merging of all heartbeat data.
Wherein, electrocardiographic activity refractory period can be preferably between 200 milliseconds to 280 milliseconds.The different lead hearts of acquisition
The time representation data combination for data of fighting should meet the following conditions:Each lead is most in the time representation data combination of heartbeat data
Include the time representation data of a heartbeat data more.When voting the time representation data combination of heartbeat data, make
Occupy the percentage of effect lead number with the lead number of detection heartbeat data to determine;If the time representation data of heartbeat data correspond to
Think that the lead is invalid lead to this heartbeat data when being low-voltage section, interference section and electrode delamination in the position of lead.
When calculating merging heartbeat specific location, the time representation statistical average that heartbeat data may be used obtains.In merging process,
This method avoids wrong merging provided with a refractory period.
In this step, a unified heart beat time sequence data is exported by union operation.The step simultaneously can
Reduce the more inspection rates and omission factor of heartbeat, the effective susceptibility and positive prediction rate for improving heartbeat detection.
Step 136, the feature that according to beat classification model heartbeat analysis data are carried out with amplitude and time representation data carries
It takes and analyzes, obtain a classification information of heartbeat analysis data;
Different cardioelectric monitor equipment difference existing for signal measurement, acquisition or leads of output etc., because
This can use simple single lead sorting technique or multi-lead sorting technique as the case may be.Multi-lead classification side
Method includes two kinds of lead ballot Decision Classfication method and lead synchronization association sorting technique again.Lead ballot Decision Classfication method be
Heartbeat analysis data based on each lead carry out lead independent sorting, then result is voted for merging and determines that the ballot of classification results is determined
Plan method;Lead synchronization association sorting technique is then using the side that the heartbeat analysis data of each lead are synchronized with association analysis
Method.Single lead sorting technique is exactly to analyze data to the heartbeat of single lead equipment, is directly classified using Correspondence lead model,
It does not vote decision process.Several sorting techniques described above are illustrated respectively below.
Single lead sorting technique includes:
According to heart beat time sequence data, single lead heartbeat data are subjected to the heartbeat analysis number that cutting generates single lead
According to, and the beat classification model for being input to the correspondence lead that training obtains carries out the feature extraction of amplitude and time representation data
And analysis, obtain a classification information of single lead.
Lead ballot Decision Classfication method can specifically include:
The first step, according to heart beat time sequence data, each lead heartbeat data are cut, to generate each lead
Heartbeat analyzes data;
Second step, according to the obtained corresponding beat classification model of each lead of training to the heartbeat of each lead analyze data into
The feature extraction and analysis of row amplitude and time representation data, obtain the classification information of each lead;
Third step carries out classification ballot decision calculating according to the classification information and lead weighted value of each lead with reference to coefficient,
Obtain a classification information.Specifically, lead weighted value with reference to coefficient is obtained based on electrocardiogram (ECG) data bayesian statistical analysis
To each lead to the ballot weight coefficient of different beat classifications.
Lead synchronization association sorting technique can specifically include:
According to heart beat time sequence data, each lead heartbeat data are cut, to generate the heartbeat point of each lead
Analyse data;Then the multi-lead synchronization association disaggregated model obtained according to training synchronizes the heartbeat analysis data of each lead
The feature extraction and analysis of amplitude and time representation data obtain a classification information of heartbeat analysis data.
The synchronization association sorting technique input of heartbeat data is all leads of Holter equipment, according to heartbeat point
The unified heartbeat site of data is analysed, the data point of same position and certain length in each lead is intercepted, synchronous transport is given by instructing
Experienced artificial intelligence deep learning model carries out calculating analysis, and output is that each heart beat locations point has considered all lead hearts
The accurate beat classification of electrical picture signal feature and the heartbeat rhythm of the heart feature of forward-backward correlation in time.
This method has fully considered that electrocardiogram difference leads are actually to measure cardiac electric signals different
The information flow that cardiac electric axis vector direction is transmitted, the various dimensions numerical characteristic that ECG signal is transmitted over time and space carry out
Comprehensive analysis significantly improves conventional method and relies solely on single lead and independently analyze, result is then summarized progress
Statistical ballot mode and be easier the defect of classification error obtained, greatly improve the accuracy rate of beat classification.
The beat classification model used in this step can be with as shown in figure 4, be specifically as follows based on artificial intelligence depth
The end-to-end multi-tag disaggregated model that the models such as the convolutional neural networks AlexNet, VGG16, Inception of habit inspire.Specifically
Say, the network of the model is one 7 layers of convolutional network, and an activation primitive is closely followed after each convolution.First layer is two
The convolutional layer of a different scale is six convolutional layers later.The convolution kernel of seven layers of convolution is 96,256,256,384,384 respectively,
384,256.In addition to first layer convolution kernel is 5 and 11 respectively there are two scale, other layer of convolution kernel scale is 5.Third, five, six,
It is pond layer after seven layers of convolutional layer.Finally follow two full articulamentums.
Beat classification model in this step, we use training set include 300,000 patients 17,000,000 data samples into
Row training.These samples are that the requirement diagnosed according to ambulatory ECG analysis carries out accurately mark generation, mark to data
Primarily directed to common arrhythmia cordis, block and ST sections and the change of T waves can meet the model instruction of different application scene
Practice.The information of mark is specifically preserved with preset standard data format.In the pretreatment of training data, to increase the extensive of model
Ability has done small size sliding for the less classification of sample size and has carried out amplification data, has been exactly using each heartbeat as base specifically
Plinth, it is 2 times mobile according to a fixed step size (such as 10-50 data point), it can thus increase by 2 times of data, improve to this
The recognition accuracy of the fewer classification samples of a little data volumes.It is verified by actual result, generalization ability is also improved.
Having used two GPU servers to carry out tens repeating queries training after training convergence in a hands-on process makes
It is tested with 5,000,000 independent test datas, accuracy rate can reach 91.92%.
Wherein, the length of the interception of training data can be 1 second to 10 seconds.For example sample rate is 200Hz, is with 2.5s
Sampling length, the data length of acquirement are a segment D [500] of 500 ecg voltage values (millivolt), and input data is:
InputData (i, j), wherein i is i-th of lead, and j is j-th of segment D of lead i.Input data all by breaing up at random
Just start to train, ensure that training process restrains, meanwhile, too many sample is collected in control from the ECG data of the same patient
This, improves the generalization ability of model, both the accuracy rate under real scene.It is synchronous to input all corresponding of leads when training
Segment data D, according to the multichannel analysis method of image analysis, to multiple Spatial Dimensions (different cardiac electric axis of each time location
Vector) leads synchronize study, to obtain a classification results more more accurate than conventional algorithm.
Step 137, ST sections are input to the heartbeat of the specific heartbeat in classification information result analysis data and T waves changes
Varying model is identified, and determines ST sections and T wave evaluation informations;
ST sections are specially the corresponding lead position to change with T waves ST section of heartbeat analysis data with T wave evaluation informations
Information.Because clinical diagnosis requires to navigate to specific lead for the change of ST sections and T waves.
Wherein, the specific heartbeat data of a classification information refer to that other may change comprising sinus property heartbeat (N) and comprising ST
The heartbeat of the heart beat type of change analyzes data.
ST sections and T waves change lead locating module by the specific heartbeat data of a classification information, according to each lead according to
It is secondary to be input to the artificial intelligence deep learning training patterns that one is ST sections of identification and T waves change, calculating analysis is carried out, output
As a result illustrate whether the feature of lead segment meets the conclusion of ST sections and the change of T waves, be assured that ST sections and the change of T waves in this way
The information in those specific leads occurred, i.e. ST sections and T wave evaluation informations.Specific method can be:A classification information
In the result is that each lead heartbeat of sinus property heartbeat analyzes data, input to ST section and T waves and change model, to sinus property heartbeat analysis number
Judge according to identification one by one is carried out, to determine sinus property heartbeat analysis data with the presence or absence of ST section and T waves change feature and generation
Specific lead location information, determines ST sections and T wave evaluation informations.
The ST sections and T waves used in this step changes model can be with as shown in figure 5, be specifically as follows based on artificial intelligence depth
Spend the end-to-end disaggregated model that the models such as the convolutional neural networks AlexNet and VGG16 of study inspire.Concretely, the model
It is one 7 layers of network, model contains 7 convolution, 5 pondizations and 2 full connections.The convolution kernel that convolution uses is 1x5,
The number of filter of every layer of convolution is different.It is 96 that level 1 volume, which accumulates number of filter,;Level 2 volume accumulates and the 3rd layer of convolution connects
With number of filter 256;The 5th layer of convolution of 4th layer of convolution sum is used in conjunction, number of filter 384;6th layer of convolution filter
Number is 384;7th layer of convolution filter number is 256;It is pond after 1st, 3,5,6,7 layer of convolutional layer.Two are followed by entirely to connect
It connects, result is also finally divided by two classes using Softmax graders.In order to increase the non-linear of model, data more higher-dimension is extracted
The feature of degree, therefore the pattern being used in conjunction using two convolution.
Because accounting of the heartbeat in all heartbeats with ST section and the change of T waves is relatively low, training data in order to balance
The harmony of diversity and each categorical data amount chooses the training number for changing without ST sections and T waves and having ST sections and the change of T waves
It is about 2 according to ratio:1, it ensure that model good generalization ability and does not occur more to training data accounting in assorting process
A kind of tendentiousness.Since the form of heartbeat is varied, the form of Different Individual performance is not quite similar, therefore, for model
More preferably estimate the distribution of each classification, can effectively extract feature, training sample is from all ages and classes, weight, gender and residence
Individual is collected;In addition, because ECG data of the single individual within the same period is often that height is similar, in order to
Overlearning is avoided, when obtaining the data of single individual, a small amount of sample in different time periods is randomly selected from all data;
Finally, due to that there are inter-individual differences is big for the heartbeat form of patient, and the feature that a internal similarity is high, thus instructed dividing
Practice, test set when, different patients is assigned to different data sets, avoid the data of same individual and meanwhile appear in training set and
In test set, gained model test results ensure that the reliability and universality of model closest to true application scenarios as a result,.
Step 138, according to heart beat time sequence data, P waves is carried out to heartbeat analysis data and T wave characteristics detect, are determined
The detailed features information of P waves and T waves in each heartbeat;
Specifically, detailed features information includes amplitude, direction, form and the data of beginning and ending time;To heartbeat signal
In analysis, the various features in P waves, T waves and QRS wave are also the important evidence in ecg analysis.
In P waves and T wave characteristic detection modules, by calculating cutting for cut-off position and P waves and T waves in QRS complex
Branch position, to extract the various features in P waves, T waves and QRS complex.Can be detected respectively by QRS complex cut-off,
Single lead PT detection algorithms and multi-lead PT detection algorithms are realized.
QRS complex cut-off detects:According to the QRS complex section power maximum point of QRS complex detection algorithm offer and rise
Stop finds the R points of QRS complex in single lead, R ' points, S points and S ' points.When there are multi-lead data, calculate each
The median of cut-off is as last cut-off position.
Single P wave in lead, T wave detection algorithms:P waves and T waves are low with respect to QRS complex amplitude, signal is gentle, are easy to be submerged in low
It is the difficult point in detection in frequency noise.This method is according to QRS complex detection as a result, eliminating QRS complex to low frequency band
After influence, third filtering is carried out to signal using low-pass filter, PT wave relative amplitudes is made to increase.Pass through peak detection later
Method finds T waves between two QRS complexes.Because T waves are the wave groups that ventricular bipolar generates, therefore between T waves and QRS complex
Relationship when having specific lock.On the basis of the QRS complex detected, in each QRS complex takes to the phase between next QRS complex
Point (for example being limited in the range after first QRS complex between 400ms to 600ms) detects end point as T waves, in this section
It is interior to choose maximum peak as T waves.The maximum peak of selecting range is P waves in remaining peak value again.Simultaneously also according to P waves and T
The peak value and position data of wave determine direction and the morphological feature of P waves and T waves.Preferably, the cutoff frequency setting of low-pass filtering
Between 10-30Hz.
Multi-lead P waves, T wave detection algorithms:In the case of multi-lead, due to the generation time phase of each wave in heartbeat
Together, spatial distribution is different, and the distribution of the time and space of noise is different, can carry out the detection of P, T wave by tracing to the source algorithm.It is first
QRS complex Processing for removing first is carried out to signal and third filtering is carried out to remove interference to signal using low-pass filter.Later
Each independent element in original waveform is calculated by independent composition analysis algorithm.In each independent element isolated, according to
Distribution characteristics according to its peak value and QRS complex position choose corresponding ingredient as P waves and T wave signals, while determining P waves
Direction with T waves and morphological feature.
Step 139, data are analyzed under a classification information according to electrocardiogram basic law reference data, P waves to heartbeat
Secondary classification processing is carried out with T wave evaluation informations with the detailed features information and ST section of T waves, obtains beat classification information;It is right
Beat classification information carries out analysis matching, generates ECG events data.
Specifically, electrocardiogram basic law reference data is followed in authoritative electrocardiogram textbook to cardiac muscle cell's electro physiology
The description of the primitive rule of activity and electrocardiogram clinical diagnosis generates, such as time interval minimum between two heartbeats, P waves with
The minimum interval etc. of R waves, for a classification information after beat classification to be finely divided again;Main basis is RR between heartbeat
Between the medicine conspicuousness of phase and different heartbeat signal in each lead;Heartbeat auditing module is referred to according to electrocardiogram basic law
Data combine the centainly Classification and Identification of continuous multiple heartbeats analysis data and the detailed features information of P waves and T waves by the room property heart
Classification of fighting splits thinner beat classification, including:Ventricular premature beat (V), room escape (VE), acceleration ventricular premature beat (VT), will
Supraventricular class heartbeat is subdivided into supraventricular premature beat (S), atrial escape (SE), junctional escape beat (JE) and room acceleration premature beat
(AT) etc..
In addition, handled by secondary classification, can also correct occur in a subseries do not meet electrocardiogram basic law
The wrong Classification and Identification of reference data.Beat classification after subdivision is subjected to pattern according to electrocardiogram basic law reference data
Match, find the Classification and Identification for not meeting electrocardiogram basic law reference data, is corrected as according to phase between RR and front and back class indication
Rational classification.
Specifically, being handled by secondary classification, a variety of beat classifications can be exported, such as:It is normal sinus heartbeat (N), complete
Full property right bundle branch block (N_CRB), completeness left bundle branch block (N_CLB), intraventricular block (N_VB), first degree A-V block
(N_B1), it is pre- swash (N_PS), ventricular premature beat (V), room escape (VE), acceleration ventricular premature beat (VT), supraventricular premature beat (S),
The classification knots such as atrial fibrillation (AF), artifact (A) are flutterred in atrial escape (SE), junctional escape beat (JE), acceleration atrial premature beats (AT), room
Fruit.
By this step, the calculating of basal heart rate parameter can also be completed.The hrv parameter of wherein basic calculation includes:RR
Between the parameters such as phase, heart rate, QT times, QTc times.
It then, can be with according to heartbeat secondary classification as a result, carry out pattern match according to electrocardiogram basic law reference data
These the following typical ECG events of classification corresponding to ECG events data are obtained, including but not limited to:
Supraventricular premature beat
Supraventricular premature beat is pairs of
Supraventricular premature beat bigeminy
Supraventricular premature beat trigeminy
Atrial escape
The atrial escape rhythm of the heart
Junctional escape beat
The junctional escape beat rhythm of the heart
Non- paroxysmal supraventricular tachycardia
Most fast supraventricular tachycardia
Longest supraventricular tachycardia
Supraventricular tachycardia
Short battle array supraventricular tachycardia
Auricular flutter-auricular fibrillation
Ventricular premature beat
Ventricular premature beat is pairs of
Ventricular premature beat bigeminy
Ventricular premature beat trigeminy
Room escape
Ventricular escape rhythm
Accelerated idioventricular rhythm
Most fast Ventricular Tachycardia
Longest Ventricular Tachycardia
Ventricular Tachycardia
Burst ventricular tachycardia
Two degree of I type sinoatrial blocks
Two degree of II type sinoatrial blocks
First degree A-V block
Two degree of I type atrioventricular blocks
Two degree of II type atrioventricular blocks
Two degree of II types (2:1) atrioventricular block
Advanced A-V block
Completeness left bundle branch block
Complete right bundle branch block
Intraventricular block
Pre-excitation syndrome
ST sections and the change of T waves
Most Long RR interval
Heartbeat analysis data are generated into ECG events according to beat classification information and electrocardiogram basic law reference data
Data.Cardiac electrical event data include the equipment id information of dynamic monitor equipment.
Step 140, corresponding ECG events information is determined in real time according to ECG events data, and determine electrocardiogram thing
Whether part information is anomalous ecg event information;
Specifically, after obtaining ECG events data, the ECG events that can be learnt by artificial intelligence
The correspondence of data and electrocardiogram time information, correspondence obtains corresponding cardiac electrical event information, for example, ECG events data
Corresponding cardiac electrical event information is sinus property heartbeat event, ventricular premature beat event etc..Only part generates alarm for needs among these
Anomalous ecg event.
Above-mentioned data handling procedure is real-time, therefore dynamic monitor equipment can constantly have electrocardiogram event information
It generates.In practical applications can also reasonable set ECG events information output gap, not only reduce data operation quantity, but also keep away
The case where exempting from missing inspection.
After obtaining ECG events information, with the anomalous ecg event information progress recorded in dynamic monitor equipment
Match, when for anomalous ecg event information, executes step 150, otherwise continue to execute step 120, measurand is continued
Monitoring data acquires.
In the preferred scheme, regardless of whether monitoring that anomalous ecg event occurs, according to ECG data, electrocardiogram thing
Number of packages is according to the monitoring report that can be generated according to preset rules for monitored user.
For example, can be according to prefixed time interval, such as summarize a data in every 24 hours, it generates in the corresponding period and is tested
The monitoring report data of object.
Step 150, it generates and warning message is exported by dynamic monitor equipment;
Specifically, when it is anomalous ecg event information to determine ECG events information, if aforementioned to ECG data
The process for carrying out analyzing processing executes in dynamic monitor equipment, then is directly exported according to electrocardio by dynamic monitor equipment
Abnormal events information generates corresponding warning message.
If it is aforementioned to ECG data carry out analyzing processing process execute in the server, server according to
Anomalous ecg event information generates corresponding warning message, and is set warning message real time down to dynamic monitor according to device id
It is standby, it is exported by dynamic monitor equipment.It is of course also possible to anomalous ecg synchronizing event information be sent, to pass through dynamic monitor
Equipment exports more information.
The output of warning message is that its monitoring data of monitored person is prompted to be abnormal so that monitored person is timely
Solve self-condition, while the other staff near monitored person can also understand monitored person that monitoring data have occurred is different in time
Often, to realize real-time abnormality alarming.
More preferably, warning message includes:Equipment id information, measurand information and anomalous ecg event information, from
And it is able to form complete information output.
It, can be full in pulse data, blood pressure data, breath data, blood oxygen when considering other sign monitoring datas at the same time
When there is the abnormal data beyond corresponding given threshold with one or more of degrees of data and temperature data, and combine electrocardio
Figure event data generates sign and monitors abnormal events information, to determine the output of warning message.
Step 160, the operating mode of dynamic monitor equipment is obtained;
In specific application scenarios, dynamic monitor equipment of the invention can be configured as different operating modes, i.e.,
Different services is provided according to different operating modes.Operating mode at least may include processing locality pattern and background process mould
Formula.
Under processing locality pattern, abnormal events information and related data only local store, and are not reported to background process.
And under background process pattern, abnormal events information and related data are reported to server, and carry out at assignment
Reason so that the abnormal conditions of measurand can be disposed timely and effectively, and better medical services are obtained.
Configuration can locally be completed in dynamic monitor equipment, can also be assigned to user in dynamic monitor equipment and be used
When, it is completed by consistency operation.Certainly, the service for selecting different operating modes i.e. different, charge method can also accordingly not
Together.
Step 170, when for processing locality pattern, dynamic monitor equipment carries out ECG events data and warning message
Record storage;
Step 180, when for background process pattern, server is according to warning message or according to receiving dynamic monitor equipment
Based on the alarm signal of warning message triggering, the User ID of the corresponding responsibility user of measurand information is determined, and generate notice
Information is sent to the user equipment of responsibility user and/or the user equipment of preset related organization;
Specifically, generating warning message if it is dynamic monitor equipment, then dynamic monitor equipment can send out warning message
Server is given, alternatively, dynamic monitor device-to-server sends alarm signal, has warning message generation with notification server,
Certainly equipment id information and measurand information are carried in alarm signal.At this point, server is according to warning message or report
The measurand information carried in alert signal determines responsibility user associated with measurand or the information of related organization.
If warning message is generated in server, server can according to the measurand information in warning message
Determine responsibility user associated with measurand or the information of related organization.
What needs to be explained here is that being previously stored with responsibility user associated with monitored user or pre- in the server
If related organization information.Responsibility user can be guardian, relatives, the family doctor etc. of monitored user, responsibility user
User equipment can be the installation of above-mentioned crowd and operation have the smart mobile phone of respective application, tablet computer or other at least
The equipment with display function is received with information;Related organization can be the medical institutions etc. that monitored user specifies, likewise,
The user equipment of relational structure can also be install and run the smart mobile phone of respective application, tablet computer or other at least
The equipment with display function is received with information.
Server determines the location information of dynamic monitor equipment according to equipment id information, so that it is determined that residing for monitored target
Position.
Server generates notification information according to warning message or alarm signal, at least carried in notification information by
Survey object information, and according to device id determine monitored target present position location information so that recipient, that is, responsibility
User can at least learn the object for being abnormal alarm, so as to quickly be contacted with the person of being monitored.Specifically answering
In, measurand information may include that the phone number etc. of measurand is carried with facilitating contact measurand to measurand
For helping.
In the preferred scheme, further include having anomalous ecg event information in notification information so that the reception of notification information
Person can in advance be judged according to anomalous ecg temporal information, to be carried out to abnormal conditions within first time
Solution, can take corresponding medical rescue measure according to its severity.Feedback letter is sent including to dynamic monitor equipment
Breath, prompts to be monitored person and is abnormal and interim counter-measure, for example drug administration, sits quietly, sees a doctor rapidly, waiting for medical institutions
It makes house calls.
Specifically, after step 160, server receives the feedback information that the user equipment of responsibility user is sent, and root
Feedback information is sent to dynamic monitor equipment according to equipment id information;Feedback information is exported by dynamic monitor equipment.
The implementation procedure of the above-mentioned sign information dynamic monitor method to the embodiment of the present invention has been described in detail, for the ease of
Understand, below we illustrated by taking a specific example of practical application as an example.
In practical application, we can further subdivision service on the basis of processing locality pattern and background process pattern
Pattern can be based on same custodial care facility, provide user's monitoring service more customized in this way.
In a specific example, following three kinds of service modes can be set:
Complete service pattern, emergency services pattern and local service pattern.
In complete service pattern, the sign monitoring data that monitoring obtains is uploaded to service by dynamic monitor equipment in real time
Device, and it is distributed to by server the terminal device of related organization in real time, realize 24 hours round-the-clock monitoring services, also, root
According to prefixed time interval, such as every 24 hours, summarizes a sign monitoring data, generate the monitoring of measurand in the corresponding period
Data reporting.
When being abnormal generation warning message, different processing sides can be used according to the severity of warning message
Formula.
For example, if what is occurred is the alarm of the anomalous ecg event of general severity, feedback information can be passed through
Mode carries out remote guide service;If what is occurred is the higher alarm of severity, medical aid is carried out by server
Task distributes, and provides medical services of visiting by related medical structure, or be monitored person by relevant healthcare institution arrangement and see a doctor.
Under emergency services pattern, dynamic monitor equipment only reports server when generating warning message, it is same to service
Device can use different processing modes according to the severity of warning message.When not being abnormal alarm, dynamic monitor
Equipment only local carries out data monitoring and display.
Under local service pattern, dynamic monitor equipment only carries out local monitor, when generating warning message also only at this
Ground exports.Monitored person can report server after obtaining warning message in such a way that manual triggering sends alarm signal.
In addition, the dynamic monitor equipment of the present invention, additionally it is possible to provide user feel untimely active event record and
The function of reporting, preferably to provide medical services and monitoring to the user.
It can be provided with the switch of key triggering in dynamic monitor equipment, after user's trigger switch, start monitoring and set
Standby monitoring users input, and generate alarm record information.Audiomonitor can include but is not limited to:Microphone, camera,
Touch screen, dummy keyboard etc..User can carry out feeling that the symptom of exception is retouched by modes such as video, voice, word inputs
It states.
Further, can start simultaneously to the ECG data and pulse data, blood pressure before and after user's trigger switch
The interception of data, breath data, blood oxygen saturation data and temperature data generates alert event together with information input by user
Record information.
Then, alarm record information is sent to background server, and is distributed to responsibility user and/or preset pass
The user equipment of online structure, to be disposed in time.
Fig. 6 is a kind of structural schematic diagram of dynamic monitor system provided in an embodiment of the present invention, which includes one
A or multiple dynamic monitor equipment and server.Server and dynamic monitor equipment respectively include:Processor and memory.Storage
Device can be connect by bus with processor.Memory can be nonvolatile storage, such as hard disk drive and flash memory, memory
In be stored with software program and device driver.Software program is able to carry out each of the above method provided in an embodiment of the present invention
Kind function;Device driver can be network and interface drive program.Processor is for executing software program, the software program
It is performed, can realize method provided in an embodiment of the present invention.
User oriented sign information dynamic monitor method and dynamic monitor system provided in an embodiment of the present invention, using number
According to pretreatment, the detection of heartbeat feature, interferer signal detection and beat classification based on deep learning method merge with lead, the heart
The analysis of the audit fought, ECG events and parameter calculates, and one of final automatic output cardiac electrical event result data is complete fast
The automation ECG detecting of fast flow is analyzed, and can be based on the output alarm of ECG detecting analysis result, or combines blood pressure, blood
Oxygen, pulse, breathing, temperature data generate alarm, and the processing of the response based on alarm, by being based on warning message into row information
Distribution processor, including be distributed to medical institutions or be distributed to the terminal device of the association user of monitored person so that it is monitored
Person obtains effective, timely medical services.The user oriented sign information dynamic monitor method and dynamic of the present invention is supervised
Protecting system carries out effective sign monitoring towards non-inpatients, and provides more effective medical treatment to the user based on sign monitoring
Ensure service.
Professional should further appreciate that, described in conjunction with the examples disclosed in the embodiments of the present disclosure
Unit and algorithm steps, can be realized with electronic hardware, computer software, or a combination of the two, hard in order to clearly demonstrate
The interchangeability of part and software generally describes each exemplary composition and step according to function in the above description.
These functions are implemented in hardware or software actually, depend on the specific application and design constraint of technical solution.
Professional technician can use different methods to achieve the described function each specific application, but this realization
It should not be considered as beyond the scope of the present invention.
The step of method described in conjunction with the examples disclosed in this document or algorithm, can use hardware, processor to execute
The combination of software module or the two is implemented.Software module can be placed in random access memory (RAM), memory, read-only memory
(ROM), electrically programmable ROM, electrically erasable ROM, register, hard disk, moveable magnetic disc, CD-ROM or technical field
In any other form of storage medium well known to interior.
Above-described specific implementation mode has carried out further the purpose of the present invention, technical solution and advantageous effect
It is described in detail, it should be understood that the foregoing is merely the specific implementation mode of the present invention, is not intended to limit the present invention
Protection domain, all within the spirits and principles of the present invention, any modification, equivalent substitution, improvement and etc. done should all include
Within protection scope of the present invention.