CN1111121A - Self-adaptation analytical method and apparatus for electrocardiac and pulse signal - Google Patents
Self-adaptation analytical method and apparatus for electrocardiac and pulse signal Download PDFInfo
- Publication number
- CN1111121A CN1111121A CN 94112295 CN94112295A CN1111121A CN 1111121 A CN1111121 A CN 1111121A CN 94112295 CN94112295 CN 94112295 CN 94112295 A CN94112295 A CN 94112295A CN 1111121 A CN1111121 A CN 1111121A
- Authority
- CN
- China
- Prior art keywords
- signal
- pulse signal
- input
- frequency
- wavelet
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The said apparatus consists of detecting electrodes, sensor, signal collector, microcomputer system and utility software package. On the base of conventional digital signal processing technology, self-adaptable nerval network technology and wavelet technology is used and flexible data record is provided. The present invention can adaptably operate, i.e., it can rapidly recognize learned object and adapt to new object, and it has excellent noise immunity and thinking ability.
Description
The present invention relates to medical domain, particularly a kind of electrocardio and pulse signal adaptive analysis method and device thereof.
There are every year 1000 liang of million peoples to die from cardiovascular disease in the world, account for 1/4th of human annual death toll.Cardiovascular disease is more common in middle age and old people, if can make things convenient for, diagnose in early days, exactly this type of disease, will protect the health of people, prolong the average life span and play significant role for defeating cardiovascular disease.People are of long duration to the concern of cardiovascular disease, and sampling and the analysis of painstaking effort tubing physiological signal such as electrocardiosignal EGG and pulse signal PLG has been subjected to attention widely, and various instruments constantly occur.There is the dynamic electrocardiogram instrument to go on the market in the seventies, i.e. Holter, its volume is less, is carried by the user, with one day 24 hours electrocardiosignal of magnetic tape recording mode recording user, as the foundation of diagnosis.It provides huge help for correct some ND disease of diagnosis.But it has only played the effect of record Holter, and user's health status will treat that the doctor reads record data with specialized equipment and could judge, goes still far away mutually for the requirement of " convenient, early stage ".
In recent years, Japanese Casio company had released wrist formula sphygomanometers.As long as the user just can know the pressure value of oneself pick off gently on the touch-surface.This table has three minutes continuous measurement functions, 30 groups of daily high pressure, low pressure, pulse storage and Presentation Functions, the effect of having played diagnosis to a certain extent and having reported to the police.But it requires frequent own measurement data of user and own analysis result, and its reliability also has been subjected to influence on the other hand.
Also have some to adopt computer on the market, the specialty ECG analysis system that function ratio is more powerful, as the comprehensive instrument of Marquatte MAC-15 electrocardial vector produced in USA, the HBD-II type electrocardio multiple domain information automatic analyzer of Hong Kong WEX company, the comprehensive automatic analyzer of SR-1000 type electrocardio of Boai, Zhongshan City, Guangdong Province medical electric factory.They generally all have automatic collection, analysis, the diagnostic function of electrocardiosignal, the medical record management function, and the software advanced person, but they select valency all very expensive, thereby only be fit to bigger hospital's use, it is impossible spreading to family and individual.Its typical case sees also IJCNN:International Joint Conference on Neural Networks, 1989, Vol II, P69-74.
The object of the present invention is to provide a kind ofly can make things convenient for, in early days, electrocardio and the pulse signal adaptive analysis method and the device thereof of diagnosis of cardiovascular diseases exactly.
Purpose of the present invention is finished by giving technical scheme: electrocardio and pulse signal adaptive analysis method, comprise the obtaining of human ecg signal, pulse signal, mould/number conversion and Digital Signal Processing, adopted self adaptation artificial neural network and wavelet technology in the processing.
Its detailed process is to carry out correct detection and input with human ecg signal with through the pulse signal of piezoelectric transduction; Signal is extracted from noise, carry out data pick-up, compress, carry out data pretreatment such as smoothing processing, baseline correction, digital filtering and related operation, this is an important step in the waveform analysis, be the precondition and guarantee of correctly discerning, and satisfy the needs of follow-up working software.
Adopt wavelet transform, signal decomposition is become the multiple yardstick composition of weave in, and the time domain or the spatial domain sampling step length composition two-dimensional phase plane of the yardstick composition that varies in size being adopted corresponding thickness, wavelet transform has been represented in the input signal content of every kind of wavelet in the contained wavelet family, this transformed value is passed to artificial neural network as input, being that varying signal is original is input as wavelet transform feature input, because wavelet transform has symmetry and has amplified difference between pattern, then the node of neutral net and interconnecting number can reduce in a large number, learn the also just shortening greatly of required time, the more important thing is that this method can reduce the error that causes owing to the neutral net fault-tolerance, make identification have higher reliability.
Use self organizing neural network and multi-layer perception network: self organizing neural network carries out self-organizing, the extensive parallel processing of self-stabilization to many and complicated arbitrarily arbitrarily two-dimensional model, finish real-time learning, set up health model, the nonequilibrium environment of self adaptation, the object of having learnt is had stable quick identification ability, adapt to the not new object of study simultaneously again rapidly.Fault-tolerance and associative ability with the multi-layer perception network, the intuitionistic fuzzy diagnostic function of simulation doctor in clinical diagnosis, construct the three-layer network processing cardioelectric signals of one 150 input nodes, 50 hidden nodes, 6 output nodes respectively, with one 50 input nodes, 10 hidden nodes, the three-layer network of 6 output nodes is handled pulse signal, chooses 30 kinds of electrocardiosignaies (being divided into 6 classes) and pulse signal (falling into 5 types) respectively as training set, and nonlinear function is selected for use
Net in the formula
jBe j neuronic input value, θ
jBe threshold value, θ
0Be the amount for the adjustment function shape, after training error was less than 0.01, two kinds of networks had all had classification capacity preferably.After the random noise that adds to training set to signal amplitude 10%, network still can correctly be discerned.
Scheme provides convenient, flexible data record, the user can write down electrocardio, the pulse signal of instant a certain length with it, also can write down a segment signal that begins from the predetermined a certain moment with it, as write down a segment signal in the sleep, it can also store one group of detected abnormal signal automatically, after being filled with, if the user does not keep, abnormal signal record afterwards will cover former record automatically.
The device of a kind of as the aforementioned electrocardio of the present invention and pulse signal adaptive analysis method defined comprises hardware configuration part and software configuration part that detecting electrode, pick off, signal acquiring board, microsystem and application software are formed.
Wherein the major part of hardware is:
The pick device of electrocardiosignal; With the Zn-Cu electrode is that skin electrode is measured the electromotive force that is produced by cardiac electrical activity.
The piezoelectric transducer device of pulse signal adopts the PT14M3 physiological pressure transducer as fluctuation sensor.
The small-signal amplifying device, ecg signal amplifier adopts three grades of amplifications, prime zero about 50 microvolts of input noise, high input impedance, the common mode rejection ratio that 80db is above, frequency response 0.2-200Hz.The pulse signal amplifier adopts two-stage to amplify, and preamplifier adopts differential ratio to amplify, and post-amplifier adopts in-phase proportion to amplify, and corresponding temperature-compensating is arranged, and the amplification of entire circuit is adjustable at 500-5000.
The low pass of noise reduction, high pass and bandreject filtering device.The frequency range of the main component of electrocardiosignal for suppressing noise and convenient back level work, after preamplifier, has designed band resistance, low pass and high pass filter at 1-100Hz.The mid frequency of band elimination filter is 50Hz, and notch depth 40db, Q-value are 0.75.The cut-off frequency of low pass filter is 150Hz, and 150Hz has the decay of 6db/ octave with upper frequency.The cut-off frequency of high pass filter is 1Hz, and the low side frequency has the decay of 5db/ octave.Its main constituent of pulse signal concentrates on 0.5-10Hz, and the low pass filter cutoff frequency of being followed after putting before it is elected 20Hz as.
To signal collected sampling maintenance and mould/analog-to-digital conversion apparatus, because sampling rate is lower, a sampling hold circuit was arranged before mould/number conversion, adopt to keep chip LF398, pull-in time is 25 microseconds; Mould/number conversion chip is selected the ADC0809 of 8 passage variable connectors for use, and its clock Frequency Design is adjustable at 10K-1.5MHz.
The hardware slave part of electrocardio and pulse signal adaptive analysis device is:
The checkout gear that counter electrode comes off and pick off breaks;
The reference voltage device is provided;
Realize human computer conversation's device;
Report to the police and receiving system through outer emission:, designed infrared emission, received alarm device for special circumstances such as user monitoring, alarming and the disabled patient's of some aphasis the monitoring, alarming etc. in sleep relatively in using.Its infrared launcher order is made up of coding circuit, carrier frequency oscillation circuit, infrared emission circuit; Receiving system is made up of infrared receiving circuit, amplification demodulator circuit, decoding circuit, speech chip, speaker; Wherein the carrier frequency frequency is elected 38KHz as, and infrared amplification and demodulation function are made up of a slice C1490HA chip and a small amount of outward element.
The software configuration of electrocardio and pulse signal adaptive analysis device partly is:
Peripheral equipment management: this part software is to the self check of system and initialization, various control and managements to program control amplification, data acquisition, mould/number conversion, screen display, infrared emission etc., the interrupt requests of response peripheral hardware, and finish the human computer conversation, more peripheral hardware is combined into an organic whole.
The data pretreatment software mainly is data pick-up and compression, smoothing processing, baseline correction, digital filtering, probability distribution, related operation software.
Waveshape learning and monitoring of software, this part are the soul places of whole software system, the signal that peripheral hardware is gathered, the pretreatment of carrying out previously be for this part used, the final goal place that this a part of recognition result is a whole system.For new user, learn, understand, store user's signal waveform feature by this a part of software, and set up user's fitness mode; For the old user who stores fitness mode, the present invention passes through the electrocardio of the continuous monitor user ' of this part, the variation of pulse signal, and compares with the fitness mode of being stored, and notes abnormalities and in time reports to the police.
The conventional analysis software of electrocardiosignal, pulse signal, this is can be used for the medical worker for the waveshape signal that will be write down and result thereof, and and compares with the result of artificial neural network and wavelet technical finesse.This a part of conventional analysis software is the software module simulation of fourier transform, one to three order derivative method, interval mobile method, regression analysis, nine parameter discriminant equation methods and electrocardio, pulse signal.
Artificial neural network and wavelet technology: this a part of software is used for the extraction and analysis and the storage of signal characteristic and the packed record of data with artificial neural network technology; The wavelet technology is used for jump signal detects the frequency analysis of instantaneous signal; The wavelet transform value is passed to neutral net as input, varying signal is original to be input as the input of wavelet transform eigenvalue, artificial neural network and wavelet technology are organically combined, the node number and the interconnecting number of neutral net have been reduced, its learning time also shortens greatly, has improved the reliability of identification.
Data logging software, this a part of software provides convenient, flexible data record, can write down the electrocardio or the pulse signal of instant a certain length, or record is from being scheduled to the segment signal that a certain moment begins; As the segment signal in the sleep; Can also store one group of detected abnormal signal automatically, after being filled with, if the user does not keep, abnormal signal record afterwards will cover former record automatically.
The present invention has following beneficial effect:
1. adaptivity: the design philosophy that adaptivity of the present invention at first is reflected as data acquisition modes is adaptive.It does not stick to medically general acquisition mode.The present invention learns used data, can pick up from different users and any different position, and Monitoring Data is also taken from same user's same position after needing only.This just brings great convenience to the user.Adaptivity of the present invention also is reflected on the processing mode of data.Because the analyzing and processing of data is mainly finished by neutral net, neutral net of the present invention again can be according to the data pattern structure, and operation adaptively can be finished real-time unsupervised learning, and can adapt to the environment of non-stationary.Like this, the present invention just can adapt to the not new object of study simultaneously again rapidly to the object quick identification of having learnt.
2. anti-interference: the present invention has good capacity of resisting disturbance.In the experiment of multi-layer perception network of the present invention, after input added 10% random noise into signal amplitude, the present invention still can operate as normal.The interference that factors such as the baseline drift during experiment is used, mould/number conversion error, circuit noise cause is lower than this level, thereby capacity of resisting disturbance of the present invention is very strong.
3. intelligent: the present invention is according to the structure auto-action of neutral net, utilizes embedding people's wherein experience and intelligence not resembling in many traditional analysis programmes.In other words, the present invention is that " oneself " thought deeply, analyzed, do not have hard row to remove,, can make things convenient for, carry out in early days, exactly electrocardio and the pulse signal adaptive analysis and the diagnosis of cardiovascular disease so be more suitable for popularizing the requirement of using according to general discrimination standard.
Description of drawings of the present invention is as follows:
Fig. 1 is human body information self-adapting handling principle figure.
Fig. 2 is the hardware block diagram of complete machine.
Fig. 3 is the complete machine software architecture diagram.
Fig. 4 is the analog to digital conversion circuit sketch map.
Fig. 5 is the electrocardiosignal of smoothing processing not.
Fig. 6 is the electrocardiosignal after the smoothing processing.
Fig. 7 is the self organizing artificial neural network sketch map.
Fig. 8 is self organizing artificial neural network (ART) and the bonded sketch map of multi-layer perception network (BP).
Fig. 9 is the structural representation of multi-layer perception network (BP).
Figure 10 is multi-layer perception network (BP) algorithm block diagram.
Figure 11 is the amplitude spectrum characteristic of wavelet transform.
Figure 12 is the phase frequency spectrum characteristic of wavelet transform.
Figure 13 carries out the signal (1) that son becomes one group of pulse signal of conversion.
Figure 14 carries out the signal (2) that son becomes one group of pulse signal of conversion.
Figure 15 carries out the signal (3) that son becomes one group of pulse signal of conversion.
The wavelet transform result of pulse signal when Figure 16 is a=1 (1).
The wavelet transform result of pulse signal when Figure 17 is a=2 (1).
The wavelet transform result of pulse signal when Figure 18 is a=4 (1).
The wavelet transform result of pulse signal when Figure 19 is a=8 (1).
The wavelet transform result of pulse signal when Figure 20 is a=1 (2).
The wavelet transform result of pulse signal when Figure 21 is a=2 (2).
The wavelet transform result of pulse signal when Figure 22 is a=4 (2).
The wavelet transform result of pulse signal when Figure 23 is a=8 (2).
The wavelet transform result of pulse signal when Figure 24 is a=1 (3).
The wavelet transform result of pulse signal when Figure 25 is a=2 (3).
The wavelet transform result of pulse signal when Figure 26 is a=4 (3).
The wavelet transform result of pulse signal when Figure 27 is a=8 (3).
Figure 28 is wavelet (WT) and the bonded block diagram of neutral net (BP).
Figure 29 is the complete machine workflow diagram.
Below in conjunction with accompanying drawing the invention process is further elaborated:
The function of each organ of human body, each system interknits and coordinates, make human body become unified integral body, simultaneously, the internal and external environment of human body is all constantly changing, when extraneous and intrinsic factor (comprising pathologic and physiological) when changing in the body each function all need to adjust accordingly, to adapt to this variation.When for example inflammation appearred in human body, body temperature can raise, and the heart beating meeting is accelerated, and therefore, by the monitoring of enough detailed human body signal such as electrocardiosignal and pulse signal, just may know whole human body health status roughly by inference fully.But human body signal is very complicated, and a large amount of noises is always mixing, artificial neural network adopts MPP and distributed storage, has very strong fault-tolerance, associative ability and learning capacity, for identification provides new approach, utilize the adaptive diagnosis of artificial neuron style, judge the health status of human body, the self adaptation here mainly is to handle the modal distortion that is caused by noise, can adjust processing procedure adaptively according to the change of environment, and can carry out self-organizing adaptively to the internal structure of memorizer and pattern classifier according to the mode data structure.See also Fig. 1, system of the present invention carried out training study 2 to artificial neural network earlier before using, and once study, network is just insensitive to the reaction of the microvariations of its input, and this is so-called fault-tolerance; And can extract the substitutive characteristics of input set, this promptly is so-called self-organizing, adaptivity; Training relief network is set up different health model 3 adaptively for different users.In training, allow the user to participate in 4 or do not participate in, the user is training network targetedly, and network has obtained a large amount of human body signals and has been stored in the neuronic interconnection.Then human body signal input 1 is constantly gathered by system, by Processing with Neural Network, for some little distortion and interference, network is abandoned automatically, to some bigger variations that causes because of motion, posture etc., because neutral net is learnt fully in training thereby also can be made accurate judgment; And, just can cause the variation of network output for the essential change of some healthy reasons, with health model 3 and relatively 7 contrasts of model, in case exceed the threshold value 5 that sets in the study, the result that promptly enters a judgement 6 is for unusual, and has adaptive diagnostic function.
The present inventor recommends following embodiment: present embodiment comprises hardware designs and software configuration, sees also Fig. 2, and the major part of complete machine hardware is:
1. the detecting electrode 9 of electrocardiosignal;
2. the detection of pulse signal and piezoelectricity conversion-fluctuation sensor 14;
3. programmable amplifier 10 and 15 is finished the amplification of gathering small-signal, and the latch 11 and 18 of signal after amplifying;
4. low pass filter 12 and 16, high-pass filtering and the bandreject filtering of noise reduction purpose;
5. realize sampling maintenance and mould/number conversion 13 to acquired signal;
6. the interface of peripheral hardware and microcomputer and handling part 20.
The slave part of complete machine hardware is:
1. come off testing circuit 8 and pick off break detection circuit 19 of electrode;
2. provide reference voltage, zero correction and span to proofread and correct 17;
3. realize interactive Interface 22;
4. infrared emission is reported to the police and the alarm device 21 that receives.
After tested, the main performance of hardware is as follows:
1. power supply: ± 10 volts, direct current;
2. sample frequency: 100-1000 is conspicuous;
3. sampling resolution: 8;
4. baseline drift:＜2 millivolts;
5. input impedance:＞5 megaohms;
6. common mode rejection ratio: 100db;
7. sensitivity: 10 millivolts ± 5%
8. disturb and suppress: disturb inhibition＞30db 50 hertz;
9. complete machine gain: 60-80db.
See also Fig. 3, the software configuration of complete machine comprises softwares such as peripheral equipment management 23, data pretreatment 24, waveshape learning and monitoring 25, conventional analysis 26, neutral net and wavelet technology 27, data record 28.
Complete machine is to implement according to aforesaid total technical scheme, below the major part of complete machine is carried out division or is carried out details with respect to aforesaid total technical scheme and replenish.
1. the electrocardio detecting electrode 9: select the Zn-Cu electrode for use from practicality, its cost is lower, uses easily, is particularly suitable for the experiment of motion conditions.Although it is kind of a polarizing electrode,, can work well fully as long as contact surface is enough big.When the sufficiently high words of the input impedance of preamplifier, actual electrode impedance has just receded into the background.
2. fluctuation sensor 14, adopt PT14M3 type physiological pressure transducer, and it is the diffuse si force sensing resistance bridge-type pressure transducer that adopts the integrated new technique of a complete set of micro mechanical structure to make.It has solved non-linear interior external compensation and overvoltage protection problem that quasiconductor expands silicon pressure sensor; stable performance, sensitivity and resolution are higher, and frequency response is also very wide, energy antidetonation, anti-drop; almost there is not leakage current, very suitable for the application that the present invention is in the motion.
3. ecg signal amplifier and pulse signal amplifier are illustrated in global schema.
4. Filter Design is also set forth in total technical scheme.
5. analog to digital conversion circuit 13 has been introduced in global schema, and its sketch map sees also Fig. 4.
6. infrared warning device 21, introduce in global schema.
Other slave parts of the microcomputer interface of relevant complete machine and complete machine and software configuration, the applied in any combination that relates to prior art that has, no longer repeat to set forth, only the application and the wavelet The Application of Technology of the Signal Pretreatment in the signal processing, artificial neural network are done to replenish introduction below.
Simply introduce earlier several main preprocess methods:
1. the software of institute's image data is proofreaied and correct:
The ADC0809 chip adopts the rate conversion mode to carry out analog digital conversion, has reduced the requirement to reference voltage, has eliminated very most source of error.But because complete machine is bigger, consider the ground interference of noise, and the factors such as instability that power supply may occur can influence mould/number precision, so the data of gathering have been carried out the software correction, from the unnecessary passage of ADC809, select two-way respectively ground connection and+5 volts of reference voltages, measure respectively with the conversion value D of+5 volts of voltages
Gno, D
+ 5V, the data point conversion income value that note is being measured is D
X, corrected value is D
X ', then have:
D
x′/255＝D
X/（D
+5V-D
gnd）
Be D
X '=255 * D
X/ (D
+ 5V-D
Gnd)
2. smoothing processing:
In order to reduce the burr effect, eliminate random disturbances, adopt the three-point weight moving average method to carry out smoothing processing to data:
Y（n）＝（1/4）[X（n-1）+2X（n）+X（n+1）]，
X(n wherein) being n sample input, is that n sample exported Y(n).Fig. 5 and Fig. 6 show the situation of electrocardiosignal before and after smoothing processing.
3. elimination trend term, the frequency content of cycle greater than record length is called trend term.If do not eliminate trend term in the data, in relevant and power spectrumanalysis very big distortion can appear.Spectrum when the trend term in the data can make low frequency estimates to lose fully verity.The frequency of electrocardio, pulse signal is all very low, and the work of eliminating trend term just seems particularly important.Adopt method of least square to eliminate trend term, can eliminate the baseline that is linear condition and move, also can eliminate trend term with higher order polynomial.
The application of following sponsor's artificial neural networks:
1. the application of self organizing neural network (ART):
See also Fig. 7, the basic structure of ART is a two-tier network, lower floor is the input node, the upper strata is an output node, connection weights from top to bottom are Tij, and connection weights from bottom to up are bij, utilize this structure, adopt the Euclidean distance between computer input pattern and sample, differentiate the classification situation of input pattern with the mode of warning parameter comparison.In the application of electrocardio and pulse signal differentiation, this network is used for learning the input sample, and the result is stored among the interconnection weight, and capacity is decided by internodal number of interconnections, this network also can obtain enough knowledge in study after, be used for monitoring the abnormal conditions of electrocardio, pulse signal.Maximum difficulty is to guard against the setting of parameter in the practical application, too big then insensitive to the variation of signal, too little then by mistake the police can not avoid, but it and multi-layer perception (BP) network combine practicality, just can maximize favourable factors and minimize unfavourable ones, obtain thicker classification results quickly, be used for instructing the training and the identification of BP network with bigger warning value of consult volume, the training time of BP network shortens by means of the guidance of ART network, and the identification ability of ART improves owing to the adding of BP network.The bonded sketch map of ART and BP network sees also Fig. 8.
2. the application of multi-layer perception (BP) network, the structural representation of multi-layer perception BP network sees also Fig. 9, and its algorithm block diagram is referring to Figure 10.This complete machine is constructed one 150 input nodes respectively, and 50 hidden nodes, the processing cardioelectric signals of the three-layer network of 6 output nodes and one 50 input node, 10 are node layer by layer surely; The three-layer network of 6 output nodes is handled pulse signal, chooses 30 kinds of electrocardios (being divided into 6 classes) and pulse (falling into 5 types) signal respectively as training set, and nonlinear function is selected for use
f＝l/[l+e-（net
j+θ
j）/θ
0]，
Net wherein
jBe j neuronic input value, θ
jBe threshold value, θ
0Be the amount of introducing for the adjustment function shape, after training error was less than 0.01, two kinds of networks had all had classification capacity preferably.After the random noise that adds to training set to signal amplitude 10%, network still can correctly be discerned.
Should being used as simply of antithetical phrase wave technology described again:
Wavelet transform (WT) can resolve into the multiple yardstick composition of weave in signal or image, and adopts the time domain or the spatial domain sampling step length of corresponding thickness for the yardstick composition that varies in size, thereby can constantly focus on any minor detail of object.
The basic function of WT is obtained through change of scale and translation by a prototype function Ψ (X):
, for the signal S(x of a finite energy), its continuous WT may be defined as:
φ（a、b）＝∫Ψ
a（x-b）S（x）dx，
Also can be designated as φ (a, b)=∫ Ψ
a(x) dx S(x+b),
Electrocardio, pulse signal are carried out having selected following mother and sons' ripple for use in the WT research:
φ（x）＝ (sin（x）)/(x) exp（j2x），
The phase characteristic of its amplitude characteristic is shown in respectively among Figure 11, Figure 12.
By this mother and sons' ripple, can obtain following wavelet collection through simple change:
ψa（x）＝ 1/(a) （ (sin（x/a）)/(x/a) ）exp（j2x/a），
Its FT characteristic is:
The establishment of following formula can be proved by following formula:
＝ 1/(a) [ (sin(x/a))/(x/a) ]exp( (2jx)/(a) )
Under study for action, a
0=2
0, i.e. (a
0, a
1, a
2, a
3)=(2
0, 2
1, 2
2, 2
3).
One group of pulse signal choosing shown in Figure 13,14,15 carries out wavelet transform, and it the results are shown in Figure 16,17,18,19; 20,21,22,23; 24, in 25,26,27.
WT has represented in the input signal content of every kind of wavelet in the contained wavelet family, and more once the transformation results of each pulse signal correspondence among Figure 16 to Figure 27 can find that they have tangible difference.Signal 1 is all inequality with four groups of WT of signal 2, and especially when a=8, difference maximum, the situation of signal 3 also are so, and this is because the structure at their peak causes.Can see the difference of the WT of generation equally for the structure difference of electrocardiosignal.
If this transformed value as input, pass to neutral net, varying signal is original to be input as the input of WT eigenvalue, as shown in figure 28, because WT has symmetry and amplified difference between pattern, then the node number of neutral net and interconnecting number can reduce in a large number, learns the also just shortening greatly of required time, this method can reduce the error that causes owing to the neutral net fault-tolerance, makes identification have higher reliability.And complete machine is to utilize soft, hardware as preceding introduction to carry out work according to flow process shown in Figure 29, and it has finished the collection and the analysis task of data effectively.
Claims (2)
1, a kind of electrocardio and pulse signal adaptive analysis method comprise the obtaining of human ecg signal, pulse signal, mould/number conversion and Digital Signal Processing, it is characterized in that having adopted self adaptation artificial neural network and wavelet technology, specifically:
1-1. signal input: carry out correct detection and input with human ecg signal with through the pulse signal of piezoelectric transduction;
1-2. data pretreatment: signal is extracted from noise, carry out the data pick-up compression, carry out smoothing processing, baseline correction, digital filtering and related operation;
1-3. wavelet transform: the multiple yardstick composition that signal decomposition is become weave in, and the time domain or the spatial domain sampling step length composition two-dimensional phase plane of the yardstick composition that varies in size being adopted corresponding thickness, wavelet transform has been represented in the input signal content of every kind of wavelet in the contained wavelet family, this transformed value is passed to neutral net as input, and promptly varying signal is original is input as wavelet transform eigenvalue input;
1-4. self organizing neural network and multi-layer perception network application: self organizing neural network carries out self-organizing to many and complicated arbitrarily arbitrarily two-dimensional model, the extensive parallel processing of self-stabilization, finish real-time learning, set up health model, the environment of self adaptation non-stationary, the object of having learnt had stable quick identification ability, adapt to the not new object of study simultaneously again rapidly, fault-tolerance and associative ability with the multi-layer perception network, intuition and the fuzzy diagnosis function of simulation doctor in clinical diagnosis, construct one 150 input nodes respectively, 50 hidden nodes, the three-layer network processing cardioelectric signals of 6 output nodes, one 50 input nodes, 10 hidden nodes, the three-layer network of 6 output nodes is handled pulse signal, chooses 30 kinds of electrocardiosignaies (being divided into 6 classes) and pulse signal (falling into 5 types) respectively as training set, and nonlinear function is selected for use:
Net in the formula
jBe j neuronic input value, θ
jBe threshold value, θ
OBe in order to regulate the amount of letter function shape, after training error was less than 0.01, two kinds of networks had all had classification capacity preferably;
1-5. data record: convenient, flexible data record is provided, can write down instant electrocardio or pulse signal, or record is from being scheduled to the segment signal that a certain moment begins, as write down a segment signal in the sleep, or store one group of detected abnormal signal automatically, do not keep as the user after being filled with, abnormal signal record afterwards will cover former record automatically.
2, the device of a kind of electrocardio according to claim 1 and pulse signal adaptive analysis method defined, comprise hardware configuration part and software configuration part that detecting electrode, pick off, signal acquiring board, microsystem and application software are formed, it is characterized in that:
2-1. the major part of hardware is:
2-1-1. the pick device of electrocardiosignal;
With the Zn-Cu electrode is that skin electrode is measured the electromotive force that is produced by cardiac electrical activity;
2-1-2. pulse signal is the piezo-electric conversion equipment;
Adopt the PT14M3 physiological pressure transducer as fluctuation sensor;
2-1-3. small-signal amplifying device, ecg signal amplifier adopts three grades of amplifications, the prestage zero about 50 μ V of input noise, high input impedance, the common mode rejection ratio that 80db is above, 0.2-200Hz frequency response, the pulse signal amplifier adopts two-stage to amplify, and preamplifier adopts differential ratio to amplify, and post-amplifier adopts in-phase proportion to amplify, corresponding temperature-compensating is arranged, and the amplification of entire circuit is adjustable at 500-5000;
2-1-4. the low pass of noise reduction, high pass and bandreject filtering device.
The frequency range of the Main Ingredients and Appearance of electrocardiosignal for suppressing noise and convenient back level work, after preamplifier, has designed band resistance, low pass and high pass filter at 1-100Hz, and the mid frequency of band elimination filter is 50Hz, and notch depth 40db, Q-value are 0.75; The cut-off frequency of low pass filter is 150Hz, and 150Hz has the decay of 6db/ octave with upper frequency; The cut-off frequency of high pass filter is 1Hz, and the low side frequency has the decay of 5db/ octave;
Its main constituent of pulse signal concentrates on 0.5-10Hz, and the low pass filter cutoff frequency of being followed after putting before it is elected 20Hz as;
2-1-5. to signal collected sampling maintenance and analog-digital commutator;
Because sampling rate is lower, and a sampling hold circuit was arranged before mould/number conversion, adopt to keep chip LF398, pull-in time is 25 microseconds; Mould/number conversion chip is selected the ADC0809 of 8 passage variable connectors for use, and its clock Frequency Design is adjustable at 10K-1.5MHz;
2-2. the slave part of hardware has:
2-2-1. counter electrode comes off and the checkout gear of pick off broken string;
2-2-2. the reference voltage device is provided;
2-2-3. realize human computer conversation's device;
2-2-4. infrared emission is reported to the police and receiving system;
Compare special circumstances such as the monitoring, alarming of user in sleep and some aphasis patient's monitoring, alarming in using, designed infrared sending-receiving alarm, its infrared launcher order is made up of coding circuit, carrier frequency oscillation circuit, infrared emission circuit, and receiving system is made up of infrared receiving circuit, amplification demodulator circuit, decoding circuit, speech chip, speaker; Wherein the carrier frequency frequency is elected 38KHz, infrared amplification and demodulation function as and is made up of a slice C1490HA chip and a small amount of outward element;
2-3. software configuration partly has:
2-3-1. peripheral equipment management: this part software is to the self check of system and initialization, various control and managements to program control amplification, data acquisition, mould/number conversion, screen display, infrared emission etc., the interrupt requests of response peripheral hardware, and finish the human computer conversation, more peripheral hardware is combined into an organic whole;
2-3-2. the data pretreatment mainly contains data pick-up and compression, smoothing processing, baseline correction, digital filtering, probability distribution, related operation;
2-3-3. waveshape learning and monitoring: this part software is used for learning, understanding user's signal waveform feature, and set up the fitness mode of user's health, to storing the old user of fitness mode, install the electrocardio of continuous monitor user ', the variation of pulse signal, and compare with the fitness mode of being stored, note abnormalities and in time report to the police;
2-3-4. the conventional analysis software of electrocardiosignal, pulse signal, this part software provides the conventional analysis to electrocardiosignal, pulse signal, as fourier transform, one to three order derivative method, interval mobile method, regression analysis, the software simulations of nine parameter discriminant equation methods and electrocardio, pulse signal etc. compare with the result who obtains with artificial neural network and wavelet technical finesse;
2-3-5. artificial neural network and wavelet technology are used for the extraction and analysis and the storage of signal characteristic, the packed record of data with artificial neural network technology; The wavelet technology is used for jump signal detects the frequency analysis of instantaneous signal; The wavelet transform value is passed to neutral net as input, and varying signal is original to be input as the input of wavelet transform eigenvalue, has reduced the node number and the interconnecting number of neutral net, and learning time shortens greatly, has improved the reliability of identification;
2-3-6. data record, this part software provides convenient, flexible data record, can write down the electrocardio or the pulse signal of instant a certain length, or record is from being scheduled to the segment signal that a certain moment begins; Or store one group of detected abnormal signal automatically.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 94112295 CN1111121A (en) | 1994-08-30 | 1994-08-30 | Self-adaptation analytical method and apparatus for electrocardiac and pulse signal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 94112295 CN1111121A (en) | 1994-08-30 | 1994-08-30 | Self-adaptation analytical method and apparatus for electrocardiac and pulse signal |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN1111121A true CN1111121A (en) | 1995-11-08 |
Family
ID=5036080
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 94112295 Pending CN1111121A (en) | 1994-08-30 | 1994-08-30 | Self-adaptation analytical method and apparatus for electrocardiac and pulse signal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1111121A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100349544C (en) * | 2004-08-16 | 2007-11-21 | 深圳迈瑞生物医疗电子股份有限公司 | Improved electronic blood pressure-detecting method and device |
| CN100518631C (en) * | 2008-04-15 | 2009-07-29 | 中国医学科学院生物医学工程研究所 | Three-dimensional dynamic pulse manifestation testing system |
| CN101194273B (en) * | 2005-03-16 | 2010-06-16 | 皇家创新有限公司 | Spatio-temporal self organising map |
| CN101785670A (en) * | 2009-01-22 | 2010-07-28 | 陈跃军 | Intelligent blurry electrocardiogram on-line analyzer system |
| CN101799974A (en) * | 2010-03-12 | 2010-08-11 | 上海交通大学 | Electrocardio signal transmission method based on self-adaptive codebook |
| CN102525431A (en) * | 2010-12-10 | 2012-07-04 | 李祥 | Cardiovascular physiology signal detection device and method |
| CN102685233A (en) * | 2012-05-11 | 2012-09-19 | 上海交通大学 | Mobile medical care remote monitoring system and data transmission method thereof |
| CN102824166A (en) * | 2012-07-05 | 2012-12-19 | 华东师范大学 | Device for sorting treatment of pulse wave |
| CN103271726A (en) * | 2012-12-10 | 2013-09-04 | 中国人民解放军第一五二中心医院 | Wireless portable vital sign monitoring device |
| CN105125338A (en) * | 2015-08-06 | 2015-12-09 | 成都康拓邦科技有限公司 | Medical instrument for relieving eyeground vasculopathy and control method of medical instrument |
| CN105590011A (en) * | 2014-10-20 | 2016-05-18 | 深圳市迈迪加科技发展有限公司 | Pulse regression model-based electrocardiography data correction method and system |
| CN105595992A (en) * | 2016-01-29 | 2016-05-25 | 安徽华米信息科技有限公司 | Method and device for collecting electrocardiosignals and wearable equipment |
| CN107103182A (en) * | 2017-03-28 | 2017-08-29 | 南京医科大学 | A kind of heart disease Warning System and method based on deep learning algorithm |
| CN107802247A (en) * | 2017-12-12 | 2018-03-16 | 吉林大学 | A kind of automatic inflation type pulse condition detection device |
| CN109982737A (en) * | 2016-11-30 | 2019-07-05 | 索尼公司 | Output-controlling device, output control method and program |
| CN112203582A (en) * | 2018-06-01 | 2021-01-08 | 深圳市长桑技术有限公司 | Pulse propagation time determination method and system |
| CN113702317A (en) * | 2021-08-30 | 2021-11-26 | 中国农业科学院农业信息研究所 | Drainage basin non-point source pollution component sensor, monitoring system and method |
| CN116058814A (en) * | 2021-11-01 | 2023-05-05 | 北京荣耀终端有限公司 | Heart rate detection method and electronic equipment |
-
1994
- 1994-08-30 CN CN 94112295 patent/CN1111121A/en active Pending
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100349544C (en) * | 2004-08-16 | 2007-11-21 | 深圳迈瑞生物医疗电子股份有限公司 | Improved electronic blood pressure-detecting method and device |
| CN101194273B (en) * | 2005-03-16 | 2010-06-16 | 皇家创新有限公司 | Spatio-temporal self organising map |
| CN100518631C (en) * | 2008-04-15 | 2009-07-29 | 中国医学科学院生物医学工程研究所 | Three-dimensional dynamic pulse manifestation testing system |
| CN101785670A (en) * | 2009-01-22 | 2010-07-28 | 陈跃军 | Intelligent blurry electrocardiogram on-line analyzer system |
| CN101799974A (en) * | 2010-03-12 | 2010-08-11 | 上海交通大学 | Electrocardio signal transmission method based on self-adaptive codebook |
| CN102525431B (en) * | 2010-12-10 | 2014-01-08 | 李祥 | Cardiovascular physiology signal detection device and method |
| CN102525431A (en) * | 2010-12-10 | 2012-07-04 | 李祥 | Cardiovascular physiology signal detection device and method |
| CN102685233A (en) * | 2012-05-11 | 2012-09-19 | 上海交通大学 | Mobile medical care remote monitoring system and data transmission method thereof |
| CN102824166A (en) * | 2012-07-05 | 2012-12-19 | 华东师范大学 | Device for sorting treatment of pulse wave |
| CN103271726A (en) * | 2012-12-10 | 2013-09-04 | 中国人民解放军第一五二中心医院 | Wireless portable vital sign monitoring device |
| CN105590011B (en) * | 2014-10-20 | 2019-04-30 | 深圳市迈迪加科技发展有限公司 | A kind of ecg signal data modification method and system based on pulse regression model |
| CN105590011A (en) * | 2014-10-20 | 2016-05-18 | 深圳市迈迪加科技发展有限公司 | Pulse regression model-based electrocardiography data correction method and system |
| CN105125338A (en) * | 2015-08-06 | 2015-12-09 | 成都康拓邦科技有限公司 | Medical instrument for relieving eyeground vasculopathy and control method of medical instrument |
| CN105595992A (en) * | 2016-01-29 | 2016-05-25 | 安徽华米信息科技有限公司 | Method and device for collecting electrocardiosignals and wearable equipment |
| US11116433B2 (en) | 2016-01-29 | 2021-09-14 | Hefei Huami Micro-Electronics Co., Ltd. | ECG signal collection and wearable apparatus |
| CN109982737A (en) * | 2016-11-30 | 2019-07-05 | 索尼公司 | Output-controlling device, output control method and program |
| CN109982737B (en) * | 2016-11-30 | 2022-06-28 | 索尼公司 | Output control device, output control method, and program |
| CN107103182A (en) * | 2017-03-28 | 2017-08-29 | 南京医科大学 | A kind of heart disease Warning System and method based on deep learning algorithm |
| CN107802247A (en) * | 2017-12-12 | 2018-03-16 | 吉林大学 | A kind of automatic inflation type pulse condition detection device |
| CN112203582A (en) * | 2018-06-01 | 2021-01-08 | 深圳市长桑技术有限公司 | Pulse propagation time determination method and system |
| CN112203582B (en) * | 2018-06-01 | 2024-04-16 | 深圳市长桑技术有限公司 | Pulse propagation time determination method and system |
| CN113702317A (en) * | 2021-08-30 | 2021-11-26 | 中国农业科学院农业信息研究所 | Drainage basin non-point source pollution component sensor, monitoring system and method |
| CN113702317B (en) * | 2021-08-30 | 2023-10-27 | 中国农业科学院农业信息研究所 | River basin non-point source pollution component sensor, monitoring system and method |
| CN116058814A (en) * | 2021-11-01 | 2023-05-05 | 北京荣耀终端有限公司 | Heart rate detection method and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN1111121A (en) | Self-adaptation analytical method and apparatus for electrocardiac and pulse signal | |
| CN110236518B (en) | Electrocardio and heart-shock signal combined classification method and device based on neural network | |
| Li et al. | Classification of ECG signals based on 1D convolution neural network | |
| Jin et al. | Predicting cardiovascular disease from real-time electrocardiographic monitoring: An adaptive machine learning approach on a cell phone | |
| CN104523266B (en) | A kind of electrocardiosignal automatic classification method | |
| CN102283670B (en) | Method and device for quantitatively analyzing heart sound signals | |
| Wang et al. | Classification of surface EMG signals using optimal wavelet packet method based on Davies-Bouldin criterion | |
| CN106725428A (en) | A kind of electrocardiosignal sorting technique and device | |
| CN109758145B (en) | Automatic sleep staging method based on electroencephalogram causal relationship | |
| CN109620262A (en) | A kind of Emotion identification system and method based on wearable bracelet | |
| CN105125206B (en) | A kind of intelligent cardiac monitoring method and device | |
| CN204931634U (en) | Based on the depression evaluating system of physiologic information | |
| JPH03207342A (en) | Analyzing assemblage for physiological wave form | |
| CN1909831A (en) | Cardiac monitoring method | |
| CN107861628A (en) | A kind of hand gestures identifying system based on human body surface myoelectric signal | |
| CN113317794B (en) | Vital sign analysis method and system | |
| CN114052744B (en) | Electrocardiosignal classification method based on impulse neural network | |
| Portet et al. | Evaluation of real-time QRS detection algorithms in variable contexts | |
| CN113855053A (en) | Wearable muscle threshold monitoring system based on myoelectricity | |
| KR20070087787A (en) | Emotion classification system of real time bio signals and emotion classification method | |
| CN109288515A (en) | Periodical monitoring method and device based on premature beat signal in wearable ECG signal | |
| CN113243890A (en) | Sleep apnea syndrome recognition device | |
| CN109394203A (en) | The monitoring of phrenoblabia convalescence mood and interference method | |
| CN108143412A (en) | A kind of control method of children's brain electricity mood analysis, apparatus and system | |
| Phinyomark et al. | Applications of variance fractal dimension: A survey |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C06 | Publication | ||
| PB01 | Publication | ||
| C01 | Deemed withdrawal of patent application (patent law 1993) |