CN107582037A - Method based on pulse wave design medical product - Google Patents
Method based on pulse wave design medical product Download PDFInfo
- Publication number
- CN107582037A CN107582037A CN201710916372.2A CN201710916372A CN107582037A CN 107582037 A CN107582037 A CN 107582037A CN 201710916372 A CN201710916372 A CN 201710916372A CN 107582037 A CN107582037 A CN 107582037A
- Authority
- CN
- China
- Prior art keywords
- physiological characteristic
- pulse wave
- model
- data
- index
- 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
Landscapes
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
A kind of medical product design method based on pulse wave, step include:1) pulse wave is collected using pulse wave harvester;2) the Pulse wave parameters computing module of deisgn product, for calculating and extracting all kinds of characteristic parameters of pulse wave;3) Numerical model of deisgn product or classification forecast model structure module, model is built for all kinds of characteristic parameters based on pulse wave, numerical prediction is carried out to the physiological characteristic for being used as target variable by model realization or classification is predicted, draws the corresponding physiological status of testee.The design method of the present invention, can effectively improve portability, the accuracy of medical product, while reduce the cost of medical product, and reduce the use threshold of medical product.
Description
Technical field
The invention belongs to biomedical engineering application field, the specifically method based on pulse wave design medical product.
Background technology
With the continuous improvement of our people's living standard, our people also increasingly carries to the attention degree of own health
Height, it is also increasingly urgent to the demand of medical product.At present on the market Major medical product all there is cost is too high, portability compared with
The problems such as requirement low, to operator is higher.
A kind of pulse wave is that the changes such as the diameter of biological peripheral circulation system, volume and pressure are adopted using biology sensor
The bio signal of collection, these signals contain physiological and biochemical property very abundant in organism, and these physiological and biochemical properties are total
Body is divided into following several classes:1st, the glucose content in the oxygen content of blood, blood in chemical signal, such as blood, swashing in blood
Cellulose content etc.;2nd, physical signalling, such as blood vessel hardness, blood viscosity, blood vessel elasticity etc.;3rd, bio signal, such as nervous system
In activity, the function of human body main organs etc..
The content of the invention
The present invention is the correlated characteristic by handling pulse wave, carries out the category filter of big data model to it, can obtain
Go out the physiological status of tested personnel instantly.As a kind of noninvasive, physiological signal for easily gathering, the collection of pulse wave have into
This is low, it is easy to operate the advantages that.Pulse wave harvester has the advantages of small volume.Therefore These characteristics are based on, by pulse wave
Processing be a kind of low cost, easy to operate, medical product design method that can be portable.
The present invention the medical product design method based on pulse wave, can effectively improve medical product portability,
Accuracy, while the cost of medical product is reduced, and reduce the use threshold of medical product.
The present invention design philosophy be:
Pulse wave signal is collected first with pulse wave harvester, transfers to signal processing module to carry out signal transacting;
Then the parameter of pulse wave is calculated;
Then model construction is carried out using Pulse wave parameters, by numerical prediction of the model realization to target variable or divided
Class, draw the corresponding physiological status of testee.
Further, correlation model and data can also be uploaded to high in the clouds and carry out respective handling in above-mentioned steps.
Specifically, the step of medical product design method of the invention based on pulse wave is as follows:
Step 1 is the loading of pulse wave signal:
Pulse wave signal is obtained by pulse wave harvester;
Step 2 is the legitimacy detection of pulse wave signal:
Computing detection is carried out to pulse wave signal by the legitimacy detection module of product, judges whether this pulse wave is legal;
Step 3 is the denoising and conversion of pulse wave:
Denoising and conversion are carried out to pulse wave by the digital filtering module of product;
Step 4 is the extraction of Pulse wave parameters:
Handled to obtain Pulse wave parameters by the Pulse wave parameters computing module of product, and export to big data modeling and
Sample database in determination module, specimen support is provided for the model construction of next step.Export simultaneously to Numerical model
Or classification forecast model is predicted or judged;
Step 5 is the numerical prediction or classification prediction of physiological characteristic (target variable);
Big data in the big data modeling of product and determination module models submodule to the pulse wave in sample database
Parameter carries out feature extraction, builds Numerical model or classification forecast model, model is used using big data decision sub-module
Realize the accurate judgement and assessment to testee's human body physiological characteristics;
Step 6 is Remote data processing:By networks such as product utilization GPRS network, 3G network, 4G networks or WIFI networks
Connected mode accesses internet, and data transfer caused by above step and storage are arrived into remote server, real by remote server
Now data are classified, inquired about, statistics, the function such as analysis and management.
Step 7 is that data are shown:
The display module of product realizes the result of the data, services function of providing remote server with form or figure
Mode is shown on screen.
The medical product based on pulse wave that above-mentioned design method obtains, its job step include:
1) pulse wave is collected using pulse wave harvester;
2) calculate and extract all kinds of characteristic parameters of pulse wave;
3) all kinds of characteristic parameters structure model based on pulse wave, by model realization to physiological characteristic (i.e. target variable)
Numerical prediction or classification prediction, draw the corresponding physiological status of testee.
4), step 1~3) in obtained model and/or data be uploaded to high in the clouds, remotely carry out data processing.
In the step 2), before all kinds of characteristic parameters for calculating and extracting pulse wave, in addition to step:
Computing detection first is carried out to pulse wave signal, judges whether this pulse wave is legal;
Denoising and conversion are carried out to legal pulse wave again:
Finally pulse wave is obtained to above-mentioned steps to calculate, obtain all kinds of characteristic parameters of pulse wave, and export to sample
Database, the model construction of Numerical model or forecast model of classifying for step 3) provide specimen support;Export simultaneously
It is predicted or judges to Numerical model or classification forecast model;
In the step 3), the characteristic parameter of the pulse wave obtained to step 2) extracts, and builds Numerical model
Or classification forecast model, utilize accurate judgement and assessment of the model to the human body physiological characteristics of testee.)
In the step 4), internet is accessed using GPRS network, 3G network, 4G networks and/or WIFI network, by step
1) data transfer and storage~3) obtained arrives long-range cloud server, data are classified by server, inquires about, count,
Analysis and management;
The result for the data, services function that the display module the reception server of product provides is shown in a manner of form or figure
Show on the screen of product.
Brief description of the drawings
Fig. 1 designs to obtain the system of medical product and workflow diagram for this method;
Fig. 2 is big data modeling and determination module submodule workflow diagram;
Fig. 3 is pulse wave legitimacy overhaul flow chart;
Fig. 4 is digital filtering flow chart form 1;
Fig. 5 is digital filtering flow chart form 2;
Fig. 6 is that physiology index classification predicts flow chart.
Embodiment
A kind of medical product design method based on pulse wave, step include:
1) pulse wave is collected using pulse wave harvester;
2) the Pulse wave parameters computing module of deisgn product, for calculating and extracting all kinds of characteristic parameters of pulse wave;
3) Numerical model of deisgn product or classification forecast model structure module, for all kinds of spies based on pulse wave
Parameter structure model is levied, numerical prediction is carried out to the physiological characteristic for being used as target variable by model realization or classification is predicted, is obtained
Go out the corresponding physiological status of testee;
4) long-range cloud server is designed, step 1~3) in obtained model and/or data be uploaded to high in the clouds, remotely
Carry out data processing.
In the step 2), before the Pulse wave parameters computing module of deisgn product, in addition to step:
The legitimacy detection module of first deisgn product, for carrying out computing detection to pulse wave signal, judge this pulse wave
It is whether legal;
The digital filtering module of product is redesigned, for carrying out denoising and conversion to legal pulse wave:
Pulse wave is obtained to above-mentioned steps to calculate, obtain each of pulse wave by the Pulse wave parameters computing module of product
Category feature parameter, and export to sample database, it is the model construction of the Numerical model of step 3) or forecast model of classifying
Specimen support is provided;Output simultaneously is predicted or judged to Numerical model or classification forecast model;
In the step 3), the big data modeling of deisgn product and determination module, the spy of the pulse wave obtained to step 2)
Sign parameter is extracted, and builds Numerical model or classification forecast model, special to the Human Physiology of testee using model
Sign precisely judge and assess.
In the step 4), internet is accessed using GPRS network, 3G network, 4G networks and/or WIFI network, by step
1) data transfer and storage~3) obtained arrives long-range cloud server, data are classified by server, inquires about, count,
Analysis and management;
The result for the data, services function that the display module the reception server of product provides is shown in a manner of form or figure
Show on the screen of product.
In the step 2), the characteristic parameter of pulse wave is divided into three classes:Pulse wave time domain class parameter, pulse wave frequency domain class ginseng
Number, pulse wave statistics class parameter;
The pulse wave time domain class parameter is that the position to the characteristic point of pulse wave sample sequence in time series is instant
Between the parameter that draws of the combined factors analysis such as the area that surrounds of point, amplitude and each point;Here characteristic point includes pulse wave
Crest, trough, flex point, dicrotic pulse point and pip;
The pulse wave frequency domain class parameter is position i.e. frequency, the width to the characteristic point of pulse wave frequency spectrum on frequency domain sequence
The parameter that the analysis of the combined factors such as the area that degree and each point surround is drawn;Here characteristic point includes maximal peak point, each humorous
Crest value point and specific frequency;
The pulse wave statistics class parameter is the pulse wave time domain class parameter and pulse wave frequency domain to multiple pulse wave signals
Class parametric statistics constructs obtained parameter.
Specific to this example, such as Fig. 1, a kind of method that personal medical product is designed based on pulse wave, gathered using pulse wave
Device obtains the pulse wave of measured, is delivered to the legitimacy of signal processing module detection pulse wave, and removes and be delivered to pulse
The noises and pseudomorphism of ripple signal, Pulse wave parameters are calculated by parameter calculating module and are delivered to big data analysis determination module,
The specific physiological status of measured is drawn, finally by data back to high in the clouds data module, is analyzed as sample lifting big data
The accuracy and robustness of determination module.
Step 1 is the loading of pulse wave signal:
Pulse wave signal is passed in the system designed by pulse wave harvester this method;
Step 2 is the legitimacy detection of pulse wave signal:
Computing detection is carried out to pulse wave signal by the computing module of product, judges whether pulse wave is legal;
Step 3 is the denoising and conversion of pulse wave:
Line translation is entered to pulse wave by the digital filtering module of product, obtains effective pulse wave;
Mode 1 (wavelet transformation mode):Wavelet decomposition is carried out to the pulse wave signal collected, obtains senior middle school's low frequency letter
Number, dynamic threshold method is selected, is filtered respectively, removes invalid signals;
Mode 2 (Fourier's mode):Classical filter device processing is carried out to the pulse wave signal collected, passes through trap respectively
Wave filter, high-pass filter, low pass filter, bandpass filter remove invalid signals;
Step 4 is the extraction of Pulse wave parameters:
Handled to obtain Pulse wave parameters by the Pulse wave parameters computing module of product, and exported to the sample of Fig. 2 descriptions
Database, specimen support is provided for the model construction of next step.Export simultaneously to Numerical model or classification forecast model
It is predicted or judges;
Step 5 is structure and the use of the Numerical model of physiological characteristic or forecast model of classifying;
Big data in the big data modeling of product and determination module models pulse wave of the submodule according to sample database
Parameter carries out feature extraction, builds Numerical model or classification forecast model, and utilize mould using big data decision sub-module
Type realizes the accurate judgement and assessment to human body physiological state;
Step 6 is Remote data processing:By networks such as product utilization GPRS network, 3G network, 4G networks or WIFI networks
Connected mode accesses internet, and data transfer caused by above step and storage are arrived into remote server, real by remote server
Now data are classified, inquired about, statistics, the function such as analysis and management.
Step 7 is that data are shown:
The display module of product realizes the result of the data, services function of providing remote server with form or figure
Mode is shown on screen.
Specifically, the medical product that this method designs to obtain is as shown in Figure 1:
Step 1 is the loading of pulse wave signal:
Pulse wave signal is passed in the product designed by pulse wave harvester this method;
Step 2 is the legitimacy detection (as shown in Figure 3) of pulse wave signal:
Pulse wave signal is first loaded into, pulse wave signal length is counted, judges whether its length meets minimum want
Ask, if being unsatisfactory for requiring, the signal is illegal;It is 2048 points that length is carried out to signal, and step-length is the cutting of 300 points,
Respectively and then the value length of signal 0 is calculated, if length is not in target zone.Then the signal is illegal;Fast Fourier change is carried out again
Change, calculate the maximum of frequency spectrum whether in target zone, if not illegal in target zone, the signal.Calculate waveform pole
Difference, if extreme difference in target zone, output signal;
Step 3 is the denoising and conversion of pulse wave:
Mode 1:It is 2048 points that legal signal is carried out to length to signal, and step-length is the cutting of 300 points, is carried out respectively
Smothing filtering, medium filtering, then carry out wavelet decomposition (as shown in Figure 4), and wavelet decomposition formula is:
Wherein, X is the signal after conversion, and a is the time, and b is yardstick, and x (t) is original signal, and ψ is wavelet mother function, when t is
Between independent variable
The use of sym8 small echos by signal decomposition is high frequency, intermediate frequency, low frequency three parts, be from sym small echos herein because
It has regularity, orthogonality, has larger bearing length and supports wavelet transform simultaneously, meanwhile, become with reference to small echo
Change itself has stronger protectiveness to bio signal, can retain the specificity information of pulse wave carrying to greatest extent, make
Three parts signal is rebuild respectively with the wavelet coefficient of decomposition, and the signal less than threshold value is given up.Filter result is recombinated, and is obtained
To useful signal.
Mode 2:Obtained legal signal is subjected to classical filter device conversion (as shown in Figure 5), classical filter method is more
Maturation, such a method have operand small, and arithmetic speed is fast, more thorough to the noise filtering of fixed frequency band, pulse waveform
The advantages that phse conversion can be predicted, respectively by notch filter, high-pass filter, low pass filter, bandpass filter, its
Formula is:
Or:
Wherein:X be wave filter input signal, y be wave filter output signal, a0…aN, b0…bNFor coefficient, N, P, Q
For filter order.
It is useful signal by the signal after wave filter.
Step 4 is the extraction of Pulse wave parameters:
Pulse wave parameters are divided into three classes:Pulse wave time domain class parameter, pulse wave frequency domain class parameter, pulse wave statistics class ginseng
Number.This three classes parameter depicts the cardiovascular function and nervous function of human body from different aspect, and description content covers people
The operation information of body main organs, and reference is provided to the quantitative description of human body items physical signs.
Pulse wave time domain class parameter:
Such shape parameter is characteristic point (such as crest, trough, flex point, dicrotic pulse point, the pip to pulse wave sample sequence
Deng) position (time point) in time series, amplitude, and the ginseng that the combined factors analysis such as area for surrounding of each point is drawn
Number;
The computational methods of pulse wave time domain class parameter:
First, the pulse wave signal being loaded into after step 3 processing;Secondly, each characteristic point is positioned;Again, each feature is calculated
Position (time point), amplitude and the area of point;Pulse wave time domain class parameter is calculated finally by the combinations of attributes of each characteristic point.
Such as:Pulse wave translation time (PWTT) computational methods are:
PWTT=abs (x1-x2)
Wherein x1 is position (time point) corresponding to the pip of positioning, and x2 is the pulse wave acceleration maximum point pair of positioning
The position (time point) answered.
First, the pulse wave signal after loading processing;Then, to above-mentioned signal carry out second differential, positioning first and
Second crest, first crest correspond to x2, and second crest corresponds to x1, completes the positioning of x1 and x2 characteristic points;Finally by
Pulse wave time domain class parameter PWTT is calculated in the poor absolute value of above-mentioned x1 and x2 positions.
The computational methods of pulse wave reflectance factor (RI) are:
Wherein, b is the amplitude that the non-differential pulse wave of dicrotic wave position correspondence is obtained under the first subdifferential, and a is non-differential arteries and veins
Fight the amplitude of ripple maximum point.
First, the pulse wave signal after loading processing;Then, processing is ranked up to pulse wave sample sequence, positions arteries and veins
Fight ripple maximum point;Then, once differentiation is carried out to pulse wave, positions pulse wave dicrotic pulse point;Finally by pulse wave maximum point width
The ratio and normalized for spending a and pulse wave dicrotic pulse point amplitude b draw pulse wave time domain class parameter RI.
Pulse wave pulse cycle total time, t computational methods were:
T=x3-x0
Wherein, x3 is the final wave trough position of pulse wave, and x0 is that pulse wave originates wave trough position.
First, the pulse wave signal after loading processing;Then, processing is ranked up to pulse wave sample sequence, positions arteries and veins
Ripple of fighting originates trough and final trough;Poor finally by the final trough x3 of pulse wave and starting trough x0 position (time point)
Go out pulse wave time domain class parameter t.
Pulse wave rises phase index T1 computational methods:
T1=t1/t*100%
Wherein t is pulse cycle total time, and t1 is time pulse cycle rising stage.
First, the pulse wave signal after loading processing;Then, processing is ranked up to pulse wave sample sequence, positions arteries and veins
Fight ripple maximum, starting trough and final trough;Then, the position (time point) by final trough x3 and starting trough x0 is poor
Draw t;T1 is drawn by the position of pulse wave maximum and position (time point) difference of starting trough again;Finally by pulse week
Rising stage phase time t1 and pulse cycle total time t ratio simultaneously are normalized to obtain T1.
Pulse wave declines phase index T2 computational methods:
T2=t2/t*100%
Wherein t is pulse cycle total time, and t2 is that pulse cycle declines time phase.
First, the pulse wave signal after loading processing;Then, processing is ranked up to pulse wave sample sequence, positions arteries and veins
Fight ripple maximum, starting trough and final trough;Then, the position (time point) by final trough x3 and starting trough x0 is poor
Draw t;T2 is drawn by the position of pulse wave maximum and position (time point) difference of final trough again;Finally by pulse week
Rising stage phase time t1 and pulse cycle total time t ratio simultaneously are normalized to obtain T2.
Above-mentioned parameter and the sample database exported to big data modeling and determination module are calculated, is the model structure of next step
Offer specimen support is provided.Output simultaneously is predicted or judged to Numerical model or classification forecast model;
2nd, pulse wave frequency domain class parameter.
Such shape parameter is characteristic point (such as maximal peak point, each harmonic spike point, the specific frequency to pulse wave frequency spectrum
Deng) position (frequency) on frequency domain sequence, amplitude, and the parameter that the combined factors analysis such as area for surrounding of each point is drawn;
Pulse wave frequency domain parameter computational methods:
First, the pulse wave signal being loaded into after step 3 processing;Then, using modern spectrum analysis means, (such as Fourier becomes
Change, AR Power estimations etc.) frequency spectrum of pulse wave is obtained, again, calculate position (frequency), amplitude and the area of each characteristic point;Finally
Pulse wave frequency domain class parameter is calculated by the combinations of attributes of each characteristic point.
Such as:Pulse wave first harmonic ratio F1.The parameter is pulse wave second harmonic height and first harmonic height
Ratio, circular are:First, the pulse wave signal being loaded into after step 3 processing;Then, modern spectrum analysis means are utilized
(such as Fourier transformation, AR Power estimations) obtains the frequency spectrum of pulse wave, then processing is ranked up to frequency spectrum and navigates to pulse wave
One harmonic wave maximum point and second harmonic maximum point;It is maximum finally by the height and first harmonic of pulse wave second harmonic maximum point
The radiometer of the height of point calculates pulse wave frequency domain class parameter F1.
Pulse wave second harmonic ratio F2, the parameter are the ratio of pulse wave third harmonic height and second harmonic height.
Circular is:First, the pulse wave signal being loaded into after step 3 processing;Secondly, modern spectrum analysis means (such as Fu is utilized
In leaf transformation, AR Power estimations etc.) obtain the frequency spectrum of pulse wave, then processing is ranked up to frequency spectrum and navigates to pulse wave second harmonic
Maximum point and third harmonic maximum point;Finally by the height and the height of second harmonic maximum point of pulse wave third harmonic maximum point
The radiometer of degree calculates pulse wave frequency domain class parameter F2.
Pulse wave first harmonic power P 1, the parameter are pulse wave power spectrum first harmonic wave band general power.It is specific to calculate
Method is:First, the pulse wave signal being loaded into after step 3 processing;Secondly, using modern spectrum analysis means, (such as Fourier becomes
Change, AR Power estimations etc.) power spectrum of pulse wave is obtained, then processing is ranked up to power spectrum and navigates to pulse wave first harmonic most
A little bigger and adjacent two both sides of first harmonic smallest point;Finally by pulse wave first harmonic maximum point and first harmonic two
The area that the adjacent smallest point in both sides surrounds obtains pulse wave frequency domain class parameter P1.
3rd, pulse wave statistics class parameter.
Such shape parameter is the pulse wave time domain class parameter and pulse wave to multiple pulse wave signals using modern statistics
The parameter that frequency domain class parametric configuration obtains.
Pulse wave counts class calculation method of parameters:
Such as:The standard deviation SDNN computational methods of whole pulse cycles are:
SDNN=std (RR)
Wherein:RR is whole pulse cycle sequences, the sequence of the pulse wave time domain class parameter t compositions of specially one section pulse wave
Row, std are calculating standard deviation formula.
The difference standard deviation SDSD computational methods of whole pulse cycles are:
SDSD=std (diff (RR))
Wherein:RR is whole pulse cycle sequences, the sequence of the pulse wave time domain class parameter t compositions of specially one section pulse wave
Row, diff are that differential process is carried out to sequence, and std is calculating standard deviation formula.
Pulse first harmonic ratio standard deviation VF1 computational methods are:
VF1=std (FF)
Wherein:FF is the sequence of the pulse wave frequency domain class parameter F1 compositions of individual different time measure, and std is calculating
Standard deviation formula.
Step 5 is the numerical prediction or classification prediction (as shown in Figure 6) of physiological characteristic:
First, data prediction
The Pulse wave parameters data in sample database are pre-processed using big data modeling submodule.Step is such as
Under:
First, invalid value present in data is rejected, cleans exceptional value, interpolation is carried out to missing values.
Second, data conversion, such as normalization, discretization, exponential transform, logarithmic transformation.
The higher Pulse wave parameters data of quality can be obtained by two above step.Afterwards, Feature Engineering is used,
Choose suitable Pulse wave parameters generation modeling index.
2nd, Feature Engineering
Index generation is carried out to Pulse wave parameters caused by data prediction, mode is as follows:
Mode one, Data Dimensionality Reduction.The dimensionality reduction technologies such as PCA, LDA are taken to carry out Data Dimensionality Reduction to Pulse wave parameters, acquisition is built
Modular character.
Mode two, feature selecting.Mainly there is the feature selecting scheme of three types.
The feature selecting of 1.Filter schemes.The program mainly considers Pulse wave parameters (independent variable) and physiological characteristic (mesh
Mark variable) between association.Main method has related-coefficient test, Chi-square Test, information gain and mutual information etc., so as to raw
Into modeling index.
The feature selecting of 2.Wrapper schemes.The program is mainly decided whether a pulse wave by object function
Parameter is added in model.Main method has search, heuristic search and random search etc. completely, refers to so as to generate modeling
Mark.
The feature selecting of 3.Embedded schemes.The program is mainly special by machine learning progress automatic to Pulse wave parameters
Sign selection.Main method has regularization, entropy and information gain etc., so as to generate modeling index.
Mode three, variable derive.Pulse wave parameters (initial parameter) are processed, generate significant variable.Mainly
Pass through the means such as functional transformation.
By three of the above mode, can obtain needing the index for establishing model.After obtaining modeling index, to modeling
The relation of index and physiological characteristic establishes corresponding model.
3rd, Numerical model or classification forecast model are constructed.
Prediction is broadly divided into two kinds of scenes, and the first scene is to carry out numerical prediction to physiological characteristic;Second is to physiology
Feature is classified.According to different application scenarios, Numerical model or classification forecast model are constructed.
1. physiological characteristic numerical prediction.
When physiological characteristic is numeric type variable, according to different application scenarios and to modeling index and physiological characteristic relation
Understanding it is different, mainly have following three kinds of modeling methods.
1.1 Application of Parametric Model Forecasting physiological characteristic numerical value.The clear and definite scene of relation for modeling index and physiological characteristic, I
Model built using parameter model method realize the numerical prediction to physiological characteristic.Comprise the following steps that:First, by data
(being made up of modeling index and physiological characteristic) is divided into training set and test set.Second, on the basis of training set, tested using intersection
Card method, by homing method come the relation being fitted between modeling index and physiological characteristic, main computational methods have a most young waiter in a wineshop or an inn
Multiply method, Maximum-likelihood estimation and quantile estimate etc., by above method computation model parameter, determine model, so as to real
Now to the accurate prediction of physiological characteristic numerical value.
1.2 nonparametric models predict physiological characteristic numerical value.For the indefinite field of relation of modeling index and physiological characteristic
Scape, we build model to realize the numerical prediction to physiological characteristic using Non-parameter modeling method.Comprise the following steps that:The
One, data (being made up of modeling index and physiological characteristic) are divided into training set and test set.Second, on the basis of training set,
Using cross validation method, nonparametric is built by KERNEL FUNCTION METHOD, arest neighbors function method, splines method, wavelet function method etc.
Model.The accurate prediction to physiological characteristic numerical value is realized by the nonparametric model of above-mentioned structure.
1.3 semi-parameter models predict physiological characteristic numerical value.Clear and definite, the portion for the relation of part modeling index and physiological characteristic
The indefinite scene of relation of point modeling index and physiological characteristic, we are realized pair using half parameter model method structure model
The numerical prediction of physiological characteristic.Comprise the following steps that:First, data are divided into training set and test set.Second, in training set
On the basis of, using cross validation method, by the way that the nonparametric model of the parameter model of a part of index and a part of index is had
Machine combined structure semi-parameter model.The accurate prediction to physiological characteristic numerical value is realized by the semi-parameter model of above-mentioned structure.
2. physiological characteristic classification prediction
When physiological characteristic is classification type variable, physiological characteristic is classified, structure classification forecast model.Classification prediction
The method of model can be divided into two kinds of single classifier and integrated classifier.
2.1. method one:Single classifier
Single classifier mainly has decision tree classifier, neural network classifier, support vector machine classifier, Bayes's classification
Device and logistic regression grader etc..
A. decision tree classifier
In the case where modeling index and physiological characteristic have more obvious linear relationship, decision tree classifier can be used
Physiological characteristic is classified.
It is special according to existing modeling index and its corresponding physiological characteristic combination producing physiological characteristic data, selected part physiology
Data are levied as training set, based on training set, construct decision tree.Comprise the following steps that:First, information gain maximize or
Under the maximized principle of Gini coefficient instructs, selective goal;Second, predictive pruning or rear Pruning strategy can be used, trims decision tree,
It is final to meet that loss function minimizes this qualifications, form the decision tree that may finally classify to data.3rd, by certainly
Plan tree is realized to classification type physiological characteristic exact classification.
B. neural network classifier
In the case where modeling data amount is larger, physiological characteristic can be classified using neural network classifier.
It is special according to existing modeling index and its corresponding physiological characteristic combination producing physiological characteristic data, selected part physiology
Data are levied as training set, based on training set, construct artificial neural network.According to practical application scene, neutral net is determined
The number of the number of plies and each hidden layer neuron;It is determined that after the number of plies and neuron number, principle is declined etc. based on gradient
The parameter of neutral net is adjusted, constantly updates parameter, until the condition for meeting to terminate, finally meets that loss function minimizes this
Qualifications, so as to form the neutral net that may finally classify to data (being made up of modeling index and physiological characteristic).For
The scene of mass data collection, high dimensional feature index, based on systems such as deep learning construction convolutional neural networks, Recognition with Recurrent Neural Network
Row neural network model (artificial intelligence) can realize efficient, the accurate differentiation to physiological characteristic.
C. support vector machine classifier
It is higher in modeling index dimension, in the case that physiological characteristic data amount is less, support vector machine classifier can be used
Physiological characteristic is classified.
It is special according to existing modeling index and its corresponding physiological characteristic combination producing physiological characteristic data, selected part physiology
Data are levied as training set, based on training set, construct SVMs.Suitable kernel function is chosen for different scenes, with
And punishment parameter, supporting vector is finally determined, to meet that loss function minimizes this qualifications, formation may finally be to life
Manage the SVMs of tagsort.
D. Bayes classifier
It is smaller in the correlation of modeling index and in the case that data target dimension is higher, Bayes classifier pair can be used
Physiological characteristic is classified.
It is special according to existing modeling index and its corresponding physiological characteristic combination producing physiological characteristic data, selected part physiology
Data are levied as training set, based on training set, construct Bayes classifier.By analyzing and training collection, physiological characteristic data is obtained
The prior probability of distribution, under the guidance that posterior probability maximizes criterion, build the pattra leaves of physiological characteristic data exact classification
This grader.
E. logistic regression grader
It is index related weaker modeling, and to the required precision of classification it is not high in the case of, can be using logistic regression point
Base grader of the class device as integrated classifier, the final exact classification realized to physiological characteristic.
It is special according to existing modeling index and its corresponding physiological characteristic combination producing physiological characteristic data, selected part physiology
Data are levied as training set, based on training set, constitutive logic returns grader.Under conditions of meeting that loss function minimizes,
Calculating logic returns the parameter of grader, forms the logistic regression grader that may finally classify to data.
2.2 methods two:Integrated classifier
In the case where single classifier is bad to physiological characteristic classifying quality, the essence of judgement is improved using integrated classifier
Degree.Integrated classifier, by the combination to single grader, forms strong classifier based on single grader, final to realize
To the exact classification of data
Integrated classifier can use Boosting classes and bagging classes etc..
Boosting integrated classifiers mainly have Adaboost, Xgboost etc..
C.Adaboost graders
Situation for can not voluntarily carry out characteristic index selection, feature work can be carried out automatically using Adaboost graders
Journey, and complete to model, realize the accurate differentiation of physiological characteristic.Comprise the following steps that:
The physiological characteristic data combined according to existing modeling index and its corresponding physiological characteristic, selected part physiology
Characteristic, based on training set, constructs Adaboost graders as training set.According to Adaboost algorithm, instruction is constantly updated
Practice the sample weights of collection and the weight of base grader, ultimately form the linear combination of base grader, may finally be right so as to be formed
The Adaboost graders of physiological characteristic data classification.
D.Xgboost graders
Situation about classifying for the physiological characteristic for needing voluntarily to construct loss function and discriminant criterion, can use Xgboost
Grader is precisely differentiated to physiological characteristic.It is as follows to implement step:
The physiological characteristic data combined according to existing modeling index and its corresponding physiological characteristic, selected part physiology
Characteristic, based on training set, constructs Xgboost graders as training set.Iterations t is set, builds t grader.
The initial value of the grader of each round is the residual error of last round of predicted value and actual value.Xgboost parameter is adjusted, it is final to obtain
Strong classifier, so as to realize the exact classification to physiological characteristic data.
The main of bagging algorithms includes random forest scheduling algorithm.It is larger for physiological characteristic data amount, characteristic index compared with
It is more, the situation of feature selecting can not be voluntarily carried out, using random forest grader.The step of building random forest grader is such as
Under:
The physiological characteristic data combined according to existing modeling index and its corresponding physiological characteristic, selected part physiology
Characteristic, based on training set, constructs random forest grader as training set.A tree m for setting tree, from original training set
Extract a part of data, m new training datas of construction;Extract part modeling index from original index, construction m is new
Index set.Combine new training data and new index set are corresponding, construct base grader, most base classifiers combination at last, press
The mode chosen according to ballot, strong classifier is formed, so as to realize the exact classification to physiological characteristic data.
4th, Numerical model or classification forecast model are assessed.
Predicted and required according to different physiological characteristics, build Numerical model or classification forecast model.Model completes it
Afterwards, on test set evaluation model quality.Numerical model is mainly with the error mean and error of predicted value and actual value
Standard deviation is as evaluation index.Classify forecast model with a series of fingers such as the accuracy rate of concrete class and prediction classification, recall rates
It is denoted as evaluation index.If model quality can not meet application request, the step for backout feature engineering, regenerate
New physical signs and new model.
5th, determine and use model.
It is determined that the model of generation meets application request, the model is effectively stored, and it is special to novel physiological using the model
Sign data are predicted.
6th, Numerical model and classification forecast model example
1st, Numerical model example
Numerical prediction example is blood pressure forecast model.
The first step, blood pressure data pretreatment.Data cleansing, logarithm value type index normalization, to class are carried out to blood pressure data
Other type index factor, these indexs are PWTT (pulse wave translation time), height (height), weight (body weight), age
(age), gender (sex).Data are divided into training set and test set.
Second step, blood pressure characteristics engineering.Using related-coefficient test selection physical signs PWTT, height and weight.
3rd step, blood pressure model construction.Because index PWTT passes through theoretical proof, and there is linear relationship in blood pressure, and its
Remaining index and the relation of blood pressure are not especially clear and definite, therefore select semi-parametric regression model being trained as blood pressure forecast model
Collection constructs semi-parametric regression model by cross validation.
4th step, blood pressure model evaluation.Blood pressure predicted value and blood pressure actual value error mean on test set are 4.72
Millimetres of mercury, the error to standard deviation of blood pressure actual value and blood pressure predicted value is 7.3 millimetress of mercury, meets national standard error mean
Less than 5 millimetress of mercury, error to standard deviation is less than the requirement of 8 millimetress of mercury.
5th step, the determination and use of model.Because semi-parameter model performance meets application requirement, have to the model
Effect stores and carries out blood pressure prediction to new physiological characteristic data using the model.
2nd, classification forecast model example.
Classification prediction example is drug abuse decision model.
The first step, drug abuse data prediction.Logarithm value index normalizes, to classification index factor, to target variable because
Sonization.Index is PWTT (pulse wave translation time), height (height), weight (body weight), age (age), gender (property
Not), T1 (pulse wave rising phase index), T2 (pulse wave decline phase index).Target variable is TF (whether taking drugs).
Second step, drug abuse Feature Engineering.Physiological characteristic index is screened using lasso logistic regressions, rejects age indexs.
3rd step, first time drug abuse grader structure.Decision tree is built by cross validation in training set, increased according to information
Beneficial principle and rear beta pruning step, reject variable height, weight.Form grader A.
4th step, first time drug abuse grader A are assessed.Grader A classifying qualities are verified on test set, find prediction
Accuracy rate is 75%, and recall rate 55%, model quality is poor, backout feature engineering.
5th step, second of Feature Engineering and drug abuse grader structure.Because single classifier prediction effect is poor, selection is random
Forest is as strong classifier.Feature Engineering is carried out by cross validation automatically in training set, and builds random forest grader, shape
Constituent class device B.
6th step, second of drug abuse grader B are assessed.Random forest classifying quality is verified on test set, finds prediction
Accuracy rate be 95%, recall rate 91%, model quality is higher, is not required to model again, retains grader B.
7th step, the determination and use of grader.The drug abuse grader B of second of structure is effectively stored and used
The grader is classified to new physiological characteristic data.
Step 6 is Remote data processing:By networks such as product utilization GPRS network, 3G network, 4G networks or WIFI networks
Connected mode accesses internet, data transfer caused by above step and storage is arrived into remote server, by remote server pair
Data are classified, and are inquired about, statistics, the function such as analysis and management.
Step 7 is that data are shown:
The display module of product realizes the result of the data, services function of providing remote server with form or figure
Mode is shown on screen.
It should be noted last that the above embodiments are merely illustrative of the technical solutions of the present invention and it is unrestricted.Although ginseng
The present invention is described in detail according to embodiment, it will be understood by those within the art that, to the technical side of the present invention
Case is modified or equivalent substitution, and without departure from the spirit and scope of technical solution of the present invention, it all should cover in the present invention
Right among.
Claims (7)
- A kind of 1. medical product design method based on pulse wave, it is characterized in that step includes:1) pulse wave is collected using pulse wave harvester;2) the Pulse wave parameters computing module of deisgn product, for calculating and extracting all kinds of characteristic parameters of pulse wave;3) Numerical model of deisgn product or classification forecast model structure module, for each category feature ginseng based on pulse wave Number structure model, by model realization to be used as target variable physiological characteristic carry out numerical prediction or classification predict, draw by The corresponding physiological status of tester.
- 2. design method according to claim 1, it is characterized in that also including step 4) designs long-range cloud server, step The model and/or data obtained in rapid 1~3) is uploaded to high in the clouds, remotely carries out data processing.
- 3. design method according to claim 1, it is characterized in that in the step 2), in the Pulse wave parameters of deisgn product Before computing module, in addition to step:The legitimacy detection module of first deisgn product, for carrying out computing detection to pulse wave signal, whether judge this pulse wave It is legal;The digital filtering module of product is redesigned, for carrying out denoising and conversion to legal pulse wave:Pulse wave is obtained to above-mentioned steps to calculate, obtain all kinds of spies of pulse wave by the Pulse wave parameters computing module of product Parameter is levied, and exports to sample database, the model construction of Numerical model or forecast model of classifying for step 3) and provides Specimen support;Output simultaneously is predicted or judged to Numerical model or classification forecast model.
- 4. design method according to claim 1, it is characterized in that in the step 3),The big data modeling of deisgn product and determination module, the characteristic parameter of the pulse wave obtained to step 2) extract, structure Numerical model or classification forecast model are built, the human body physiological characteristics of testee are precisely judged and commented using model Estimate.
- 5. design method according to claim 1, it is characterized in that in the step 4):Internet is accessed using GPRS network, 3G network, 4G networks and/or WIFI network, by step 1)~3) obtained data Transmit and store long-range cloud server, data are classified, inquire about, count, analyze and managed by server;The result for the data, services function that the display module the reception server of product provides is shown in a manner of form or figure On the screen of product.
- 6. design method according to claim 1, it is characterized in that in the step 2), the characteristic parameter of pulse wave is divided into three Class:Pulse wave time domain class parameter, pulse wave frequency domain class parameter, pulse wave statistics class parameter;The pulse wave time domain class parameter be to position i.e. time point of the characteristic point of pulse wave sample sequence in time series, The parameter that the analysis of the combined factors such as the area that amplitude and each point surround is drawn;Crest of the characteristic point here including pulse wave, Trough, flex point, dicrotic pulse point and pip;The pulse wave frequency domain class parameter be to position i.e. frequency of the characteristic point of pulse wave frequency spectrum on frequency domain sequence, amplitude, And the parameter that the combined factors analysis such as area for surrounding of each point is drawn;Here characteristic point includes maximal peak point, each harmonic wave Peak point and specific frequency;The pulse wave statistics class parameter is pulse wave time domain class parameter and pulse wave frequency domain the class ginseng to multiple pulse wave signals The parameter that number statistics construction obtains.
- 7. design method according to claim 1, it is characterized in that in the step 3), Numerical model or classification prediction Model building method is:3.1) data predictionThe Pulse wave parameters data obtained to step 2) pre-process, and step is:The first step, invalid value present in data is rejected, clean exceptional value, interpolation is carried out to missing values;Second step, data conversion, method include normalization, discretization, exponential transform or logarithmic transformation;3.2) Feature EngineeringIndex generation is carried out to Pulse wave parameters caused by data prediction, mode is:Mode one, Data Dimensionality Reduction, take the dimension-reduction algorithms such as PCA, LDA to carry out Data Dimensionality Reduction to Pulse wave parameters, obtain modeling and refer to Mark;Mode two, feature selecting, scheme have three kinds, are respectivelyThe feature selecting of 1.Filter schemes:It is used as the Pulse wave parameters of independent variable and the physiology as target variable by investigating Association between feature, so as to generate modeling index;The feature selecting of 2.Wrapper schemes:Decide whether a Pulse wave parameters being added to model by object function In, so as to generate modeling index;The feature selecting of 3.Embedded schemes:By machine learning carry out feature selecting automatic to Pulse wave parameters, so as to raw Into modeling index.Mode three, variable derive, the Pulse wave parameters as initial parameter are processed, generate significant variable;Obtain needing the index for establishing model by three of the above mode;Phase is established to the relation of modeling index and physiological characteristic again The model answered;3.3) Numerical model or classification forecast model are constructedPrediction is divided into two kinds of scenes, and the first scene is to carry out numerical prediction to physiological characteristic;Second is that physiological characteristic is carried out Classification;According to different application scenarios, Numerical model or classification forecast model are constructed;1. physiological characteristic numerical predictionWhen physiological characteristic is numeric type variable, according to different application scenarios and to modeling index and the reason of physiological characteristic relation Solution is different, including following three kinds of modeling methods:1.1 Application of Parametric Model Forecasting physiological characteristic numerical valueFor modeling index and the clear and definite scene of relation of physiological characteristic, model is built using parameter model method to realize to life The numerical prediction of feature is managed, step is as follows:The first step, the data being made up of modeling index and physiological characteristic are divided into training set and test set;Second step, on the basis of training set, using cross validation method, it is special that modeling index and physiology are fitted by homing method Relation between sign;1.2 nonparametric models predict physiological characteristic numerical valueFor modeling index and the indefinite scene of relation of physiological characteristic, model is built using Non-parameter modeling method to realize Numerical prediction to physiological characteristic, step are as follows:The first step, the data being made up of modeling index and physiological characteristic are divided into training set and test set;Second step, on the basis of training set, using cross validation method, pass through KERNEL FUNCTION METHOD, arest neighbors function method, spline function Method or wavelet function method structure nonparametric model;1.3 semi-parameter models predict physiological characteristic numerical valueClear and definite for the relation of part modeling index and physiological characteristic, the relation of part modeling index and physiological characteristic is indefinite Scene, model is built using half parameter model method to realize the numerical prediction to physiological characteristic, step is as follows:The first step, data are divided into training set and test set;Second step, on the basis of training set, using cross validation method, by by the parameter model and nonparametric of a part of index Models coupling constructs semi-parameter model;2. physiological characteristic classification predictionWhen physiological characteristic is classification type variable, physiological characteristic is classified, structure classification forecast model, forecast model of classifying Method is divided into two kinds of single classifier and integrated classifier.2.1. method one:Single classifierSingle classifier include decision tree classifier, neural network classifier, support vector machine classifier, Bayes classifier and Logistic regression grader;A. decision tree classifierIn the case where modeling index and physiological characteristic have more obvious linear relationship, using decision tree classifier to physiology Feature is classified;According to existing modeling index and its corresponding physiological characteristic combination producing physiological characteristic data, selected part physiological characteristic number According to as training set, based on training set, decision tree is constructed, step is as follows:The first step, in the case where information gain maximizes or the maximized principle of Gini coefficient instructs, selective goal;Second step, using predictive pruning or rear Pruning strategy, decision tree is trimmed, finally meets that loss function minimizes this restriction bar Part, form the decision tree that may finally classify to data;3rd step, realized by decision tree to classification type physiological characteristic exact classification;B. neural network classifierIn the case where modeling data amount is big, physiological characteristic is classified using neural network classifier;According to existing modeling index and its corresponding physiological characteristic combination producing physiological characteristic data, selected part physiological characteristic number According to as training set, based on training set, artificial neural network is constructed;According to practical application scene, the number of plies of neutral net is determined And the number of each hidden layer neuron;It is determined that after the number of plies and neuron number, adjusted based on principles such as gradient declines The parameter of neutral net, parameter is constantly updated, until the condition for meeting to terminate, finally meet that loss function minimizes this restriction Condition, so as to form the neutral net finally classified to the data being made up of modeling index and physiological characteristic;For mass data collection, the scene of high dimensional feature index, based on deep learning constructing neural network model realization to physiology Efficient, the accurate differentiation of feature;C. support vector machine classifierIt is higher in modeling index dimension, in the case that physiological characteristic data amount is less, using support vector machine classifier to physiology Feature is classified;According to existing modeling index and its corresponding physiological characteristic combination producing physiological characteristic data, selected part physiological characteristic number According to as training set, based on training set, SVMs is constructed;Suitable kernel function is chosen for different scenes, and is punished Penalty parameter, supporting vector is finally determined, to meet that loss function minimizes this qualifications, formation may finally be special to physiology Levy the SVMs of classification;D. Bayes classifierIt is smaller in the correlation of modeling index and in the case that data target dimension is higher, it is special to physiology using Bayes classifier Sign is classified;According to existing modeling index and its corresponding physiological characteristic combination producing physiological characteristic data, selected part physiological characteristic number According to as training set, based on training set, Bayes classifier is constructed;By analyzing and training collection, physiological characteristic data distribution is obtained Prior probability, posterior probability maximize criterion guidance under, build the Bayes classifier of physiological characteristic exact classification;E. logistic regression graderModel it is index related weaker, and to the required precision of classification it is not high in the case of, using logistic regression grader make For the base grader of integrated classifier, the final exact classification realized to physiological characteristic;According to existing modeling index and its corresponding physiological characteristic combination producing physiological characteristic data, selected part physiological characteristic number According to as training set, based on training set, constitutive logic returns grader;Under conditions of meeting that loss function minimizes, calculate The parameter of logistic regression grader, form the logistic regression grader that may finally classify to physiological characteristic;2.2 methods two:Integrated classifierIn the case where single classifier is bad to physiological characteristic classifying quality, the precision of judgement is improved using integrated classifier. Integrated classifier, by the combination to single grader, is formed strong classifier, finally realizes logarithm based on single grader According to exact classification;Integrated classifier includes Boosting classes or bagging classes;Boosting integrated classifiers include Adaboost, Xgboost;A.Adaboost gradersSituation for can not voluntarily carry out characteristic index selection, Feature Engineering is carried out using Adaboost graders automatically, and Modeling is completed, realizes the accurate differentiation of physiological characteristic, step is as follows:Physiological characteristic data, selected part physiological characteristic number are combined according to existing modeling index and its corresponding physiological characteristic According to as training set, based on training set, Adaboost graders are constructed;According to Adaboost algorithm, training set is constantly updated The weight of sample weights and base grader, the linear combination of base grader is ultimately formed, may finally be special to physiology so as to be formed Levy the Adaboost graders of data classification;B.Xgboost gradersSituation about classifying for the physiological characteristic for needing voluntarily to construct loss function and discriminant criterion, using Xgboost graders Physiological characteristic is precisely differentiated, step is as follows:Physiological characteristic data, selected part physiological characteristic number are combined according to existing modeling index and its corresponding physiological characteristic According to as training set, based on training set, Xgboost graders are constructed;Iterations t is set, builds t grader;Each round Grader initial value be last round of predicted value and actual value residual error;Xgboost parameter is adjusted, it is final to obtain strong classification Device, so as to realize the exact classification to physiological characteristic data;Bagging algorithms mainly include random forest scheduling algorithm;Larger for physiological characteristic data amount, characteristic index is more, nothing Method voluntarily carries out the situation of feature selecting, and using random forest grader, the step of building random forest grader is as follows:Physiological characteristic data, selected part physiological characteristic number are combined according to existing modeling index and its corresponding physiological characteristic According to as training set, based on training set, random forest grader is constructed;A tree m for setting tree, one is extracted from original training set Partial data, m new training datas of construction;Part modeling index, m new indexs of construction are extracted from original index Collection;Combine new training data and new index set are corresponding, construct base grader, most base classifiers combination at last, according to throwing The mode taken is voted for, strong classifier is formed, so as to realize the exact classification to physiological characteristic data;3.4) Numerical model or classification forecast model are assessedPredicted and required according to different physiological characteristics, build Numerical model or classification forecast model;After model is completed, The quality of evaluation model on test set;Numerical model is mainly with the error mean and error to standard deviation of predicted value and actual value As evaluation index;Classify forecast model using a series of indexs such as concrete class and the prediction accuracy rate of classification, recall rate as Evaluation index;If model quality can not meet application request, backout feature engineering step, regenerate new physiology and refer to Mark and new model;3.5) determine and use modelIt is determined that the model of generation meets application request, the model is effectively stored, and using the model to novel physiological characteristic According to being predicted.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710916372.2A CN107582037A (en) | 2017-09-30 | 2017-09-30 | Method based on pulse wave design medical product |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710916372.2A CN107582037A (en) | 2017-09-30 | 2017-09-30 | Method based on pulse wave design medical product |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107582037A true CN107582037A (en) | 2018-01-16 |
Family
ID=61052565
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710916372.2A Pending CN107582037A (en) | 2017-09-30 | 2017-09-30 | Method based on pulse wave design medical product |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107582037A (en) |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108095708A (en) * | 2018-01-19 | 2018-06-01 | 动析智能科技有限公司 | A kind of physiology monitoring and analysis method, system based on mixing sensing |
CN109106345A (en) * | 2018-06-27 | 2019-01-01 | 北京中欧美经济技术发展中心 | Pulse signal characteristic detection method and device |
CN109256207A (en) * | 2018-08-29 | 2019-01-22 | 王雁 | A method of based on XGBoost+SVM hybrid machine Learner diagnosis keratoconus case |
CN109273097A (en) * | 2018-09-07 | 2019-01-25 | 郑州大学第附属医院 | A kind of automatic generation method and device of drug indication |
CN109602410A (en) * | 2018-11-16 | 2019-04-12 | 青岛真时科技有限公司 | A kind of wearable device and its monitoring of pulse method |
CN109886923A (en) * | 2019-01-17 | 2019-06-14 | 柳州康云互联科技有限公司 | It is a kind of for internet detection in measurement detection system and method based on machine learning |
CN110146817A (en) * | 2019-05-13 | 2019-08-20 | 上海博强微电子有限公司 | The diagnostic method of lithium battery failure |
CN110705584A (en) * | 2019-08-21 | 2020-01-17 | 深圳壹账通智能科技有限公司 | Emotion recognition method, emotion recognition device, computer device and storage medium |
CN110897618A (en) * | 2019-12-12 | 2020-03-24 | 中国科学院深圳先进技术研究院 | Pulse wave conduction calculation method and device and terminal equipment |
CN110946573A (en) * | 2019-11-01 | 2020-04-03 | 东软集团股份有限公司 | Cardiac arrest detection device, detection model training device, method and equipment |
CN111000553A (en) * | 2019-12-30 | 2020-04-14 | 山东省计算中心(国家超级计算济南中心) | Intelligent classification method for electrocardiogram data based on voting ensemble learning |
CN111528825A (en) * | 2020-05-14 | 2020-08-14 | 浙江大学 | Photoelectric volume pulse wave signal optimization method |
WO2020215942A1 (en) * | 2019-04-24 | 2020-10-29 | 京东方科技集团股份有限公司 | Physiological data processing method and device, and storage medium |
CN111939031A (en) * | 2020-08-19 | 2020-11-17 | 郑州铁路职业技术学院 | Special nursing physiotherapy pad of cardiovascular internal medicine |
CN112036483A (en) * | 2020-08-31 | 2020-12-04 | 中国平安人寿保险股份有限公司 | Object prediction classification method and device based on AutoML, computer equipment and storage medium |
CN112274120A (en) * | 2020-10-28 | 2021-01-29 | 河北工业大学 | Noninvasive arteriosclerosis detection method and device based on one-way pulse wave |
CN112438707A (en) * | 2019-08-16 | 2021-03-05 | 富士通株式会社 | Detection device, method and system for vital signs |
CN118262904A (en) * | 2024-05-30 | 2024-06-28 | 南通市肿瘤医院(南通市第五人民医院) | Intelligent artery monitoring method and system based on body surface two-point pulse wave |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104665768A (en) * | 2013-10-03 | 2015-06-03 | 塔塔咨询服务有限公司 | Monitoring physiological parameters |
CN104757955A (en) * | 2015-03-25 | 2015-07-08 | 华中科技大学 | Human body blood pressure prediction method based on pulse wave |
CN105147248A (en) * | 2015-07-30 | 2015-12-16 | 华南理工大学 | Physiological information-based depressive disorder evaluation system and evaluation method thereof |
CN106419936A (en) * | 2016-09-06 | 2017-02-22 | 深圳欧德蒙科技有限公司 | Emotion classifying method and device based on pulse wave time series analysis |
CN106821356A (en) * | 2017-02-23 | 2017-06-13 | 吉林大学 | High in the clouds continuous BP measurement method and system based on Elman neutral nets |
-
2017
- 2017-09-30 CN CN201710916372.2A patent/CN107582037A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104665768A (en) * | 2013-10-03 | 2015-06-03 | 塔塔咨询服务有限公司 | Monitoring physiological parameters |
CN104757955A (en) * | 2015-03-25 | 2015-07-08 | 华中科技大学 | Human body blood pressure prediction method based on pulse wave |
CN105147248A (en) * | 2015-07-30 | 2015-12-16 | 华南理工大学 | Physiological information-based depressive disorder evaluation system and evaluation method thereof |
CN106419936A (en) * | 2016-09-06 | 2017-02-22 | 深圳欧德蒙科技有限公司 | Emotion classifying method and device based on pulse wave time series analysis |
CN106821356A (en) * | 2017-02-23 | 2017-06-13 | 吉林大学 | High in the clouds continuous BP measurement method and system based on Elman neutral nets |
Non-Patent Citations (1)
Title |
---|
吴萍: "脉搏波时、频域特征提取与识别技术研究", 《中国优秀硕士学位论文全文数据库(电子期刊) 医药卫生科技辑》 * |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019142119A1 (en) * | 2018-01-19 | 2019-07-25 | 动析智能科技有限公司 | Hybrid sensing based physiological monitoring and analysis method and system |
CN108095708A (en) * | 2018-01-19 | 2018-06-01 | 动析智能科技有限公司 | A kind of physiology monitoring and analysis method, system based on mixing sensing |
CN109106345A (en) * | 2018-06-27 | 2019-01-01 | 北京中欧美经济技术发展中心 | Pulse signal characteristic detection method and device |
CN109256207A (en) * | 2018-08-29 | 2019-01-22 | 王雁 | A method of based on XGBoost+SVM hybrid machine Learner diagnosis keratoconus case |
CN109273097A (en) * | 2018-09-07 | 2019-01-25 | 郑州大学第附属医院 | A kind of automatic generation method and device of drug indication |
CN109602410A (en) * | 2018-11-16 | 2019-04-12 | 青岛真时科技有限公司 | A kind of wearable device and its monitoring of pulse method |
CN109886923A (en) * | 2019-01-17 | 2019-06-14 | 柳州康云互联科技有限公司 | It is a kind of for internet detection in measurement detection system and method based on machine learning |
WO2020215942A1 (en) * | 2019-04-24 | 2020-10-29 | 京东方科技集团股份有限公司 | Physiological data processing method and device, and storage medium |
CN110146817A (en) * | 2019-05-13 | 2019-08-20 | 上海博强微电子有限公司 | The diagnostic method of lithium battery failure |
CN112438707A (en) * | 2019-08-16 | 2021-03-05 | 富士通株式会社 | Detection device, method and system for vital signs |
CN110705584A (en) * | 2019-08-21 | 2020-01-17 | 深圳壹账通智能科技有限公司 | Emotion recognition method, emotion recognition device, computer device and storage medium |
CN110946573A (en) * | 2019-11-01 | 2020-04-03 | 东软集团股份有限公司 | Cardiac arrest detection device, detection model training device, method and equipment |
CN110897618A (en) * | 2019-12-12 | 2020-03-24 | 中国科学院深圳先进技术研究院 | Pulse wave conduction calculation method and device and terminal equipment |
CN111000553A (en) * | 2019-12-30 | 2020-04-14 | 山东省计算中心(国家超级计算济南中心) | Intelligent classification method for electrocardiogram data based on voting ensemble learning |
CN111000553B (en) * | 2019-12-30 | 2022-09-27 | 山东省计算中心(国家超级计算济南中心) | Intelligent classification method for electrocardiogram data based on voting ensemble learning |
CN111528825A (en) * | 2020-05-14 | 2020-08-14 | 浙江大学 | Photoelectric volume pulse wave signal optimization method |
CN111939031A (en) * | 2020-08-19 | 2020-11-17 | 郑州铁路职业技术学院 | Special nursing physiotherapy pad of cardiovascular internal medicine |
CN112036483A (en) * | 2020-08-31 | 2020-12-04 | 中国平安人寿保险股份有限公司 | Object prediction classification method and device based on AutoML, computer equipment and storage medium |
CN112036483B (en) * | 2020-08-31 | 2024-03-15 | 中国平安人寿保险股份有限公司 | AutoML-based object prediction classification method, device, computer equipment and storage medium |
CN112274120A (en) * | 2020-10-28 | 2021-01-29 | 河北工业大学 | Noninvasive arteriosclerosis detection method and device based on one-way pulse wave |
CN112274120B (en) * | 2020-10-28 | 2023-07-25 | 河北工业大学 | Noninvasive arteriosclerosis detection method and device based on one-way pulse wave |
CN118262904A (en) * | 2024-05-30 | 2024-06-28 | 南通市肿瘤医院(南通市第五人民医院) | Intelligent artery monitoring method and system based on body surface two-point pulse wave |
CN118262904B (en) * | 2024-05-30 | 2024-08-30 | 南通市肿瘤医院(南通市第五人民医院) | Intelligent artery monitoring method and system based on body surface two-point pulse wave |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107582037A (en) | Method based on pulse wave design medical product | |
CN109350032B (en) | Classification method, classification system, electronic equipment and storage medium | |
Yuan et al. | Muvan: A multi-view attention network for multivariate temporal data | |
Acharjya | A hybrid scheme for heart disease diagnosis using rough set and cuckoo search technique | |
CN104398254B (en) | Electrocardiogram analyzing system, electrocardiogram analyzing equipment and electrocardiogram predication model acquisition equipment | |
CN107822622A (en) | Electrocardiographic diagnosis method and system based on depth convolutional neural networks | |
CN107249455A (en) | Gradually gesture recognition device and vegetative nervous function information acquisition device, methods and procedures | |
CN108511057A (en) | Transfusion volume model foundation and prediction technique, device, equipment and its storage medium | |
CN109864736A (en) | Processing method, device, terminal device and the medium of electrocardiosignal | |
CN111063437B (en) | Personalized chronic disease analysis system | |
CN101785670A (en) | Intelligent blurry electrocardiogram on-line analyzer system | |
CN108304887A (en) | Naive Bayesian data processing system and method based on the synthesis of minority class sample | |
CN113656558B (en) | Method and device for evaluating association rule based on machine learning | |
CN113855038B (en) | Electrocardiosignal critical value prediction method and device based on multi-model integration | |
CN108304970A (en) | The method for quick predicting and system of Apple, air conditioned storage monitoring system | |
CN107595249A (en) | pregnant female screening method based on pulse wave | |
CN110037668A (en) | The system that pulse signal time-space domain binding model judges age, health status and malignant arrhythmia identification | |
CN109065163A (en) | Tcm diagnosis service platform | |
CN110299207A (en) | For chronic disease detection in based on computer prognosis model data processing method | |
CN109410074A (en) | Intelligent core protects method and system | |
CN113456033B (en) | Physiological index characteristic value data processing method, system and computer equipment | |
Li et al. | Deep convolutional neural networks for classifying body constitution | |
CN201414795Y (en) | Fuzzy electrocardiogram intelligent online analyzer system | |
CN109147945A (en) | Chinese Medicine Diagnoses System and bracelet | |
CN108338777A (en) | A kind of pulse signal determination method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180116 |