CN106388824A - Respiration rate extraction method and device - Google Patents
Respiration rate extraction method and device Download PDFInfo
- Publication number
- CN106388824A CN106388824A CN201610899933.8A CN201610899933A CN106388824A CN 106388824 A CN106388824 A CN 106388824A CN 201610899933 A CN201610899933 A CN 201610899933A CN 106388824 A CN106388824 A CN 106388824A
- Authority
- CN
- China
- Prior art keywords
- breathing rate
- weight factor
- breath signal
- current time
- characteristic root
- 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.)
- Granted
Links
- 230000029058 respiratory gaseous exchange Effects 0.000 title claims abstract description 163
- 238000000605 extraction Methods 0.000 title claims abstract description 21
- 238000003062 neural network model Methods 0.000 claims abstract description 27
- 239000000284 extract Substances 0.000 claims abstract description 4
- 238000000034 method Methods 0.000 claims description 58
- 239000011159 matrix material Substances 0.000 claims description 48
- 238000012549 training Methods 0.000 claims description 27
- 239000000470 constituent Substances 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000012847 principal component analysis method Methods 0.000 claims description 8
- 230000001360 synchronised effect Effects 0.000 claims description 8
- 230000001537 neural effect Effects 0.000 claims description 7
- 230000007935 neutral effect Effects 0.000 claims description 6
- 238000000513 principal component analysis Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 3
- 230000000241 respiratory effect Effects 0.000 abstract description 12
- 230000036387 respiratory rate Effects 0.000 abstract 2
- 238000005259 measurement Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 7
- 230000008859 change Effects 0.000 description 6
- 238000001514 detection method Methods 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000033001 locomotion Effects 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000000718 qrs complex Methods 0.000 description 3
- 241001123248 Arma Species 0.000 description 2
- 210000000038 chest Anatomy 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 230000028161 membrane depolarization Effects 0.000 description 2
- 210000000056 organ Anatomy 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000004202 respiratory function Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000001932 seasonal effect Effects 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 235000006508 Nelumbo nucifera Nutrition 0.000 description 1
- 240000002853 Nelumbo nucifera Species 0.000 description 1
- 235000006510 Nelumbo pentapetala Nutrition 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 210000000481 breast Anatomy 0.000 description 1
- 230000000747 cardiac effect Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 206010016256 fatigue Diseases 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000003032 molecular docking Methods 0.000 description 1
- 231100000915 pathological change Toxicity 0.000 description 1
- 230000036285 pathological change Effects 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 230000035790 physiological processes and functions Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000034225 regulation of ventricular cardiomyocyte membrane depolarization Effects 0.000 description 1
- 210000001034 respiratory center Anatomy 0.000 description 1
- 230000036391 respiratory frequency Effects 0.000 description 1
- 210000002345 respiratory system Anatomy 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 238000002604 ultrasonography Methods 0.000 description 1
- 210000001835 viscera Anatomy 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/08—Detecting, measuring or recording devices for evaluating the respiratory organs
- A61B5/0816—Measuring devices for examining respiratory frequency
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7221—Determining signal validity, reliability or quality
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Surgery (AREA)
- Public Health (AREA)
- Pathology (AREA)
- Veterinary Medicine (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Physiology (AREA)
- Artificial Intelligence (AREA)
- Signal Processing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (AREA)
- Pulmonology (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Cardiology (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a respiratory rate extraction method, which comprises the following steps: extracting the received electrocardiosignals through a pre-trained neural network model to obtain first respiratory signals, and calculating to obtain a first respiratory rate at the current moment according to the first respiratory signals; extracting the electrocardiosignals through a constructed autoregressive model to obtain second respiration signals, and calculating according to the second respiration signals to obtain a second respiration rate of the current moment; analyzing the first respiratory signal and the second respiratory signal based on a signal quality index to obtain a first weight factor corresponding to the first respiratory signal and a second weight factor corresponding to the second respiratory signal; and calculating to obtain the breathing rate of the current moment according to the first breathing rate, the first weight factor, the second breathing rate and the second weight factor. The invention also discloses a respiration rate extraction device which can conveniently and effectively extract respiration signals and calculate to obtain accurate and stable respiration rates.
Description
Technical field
The present invention relates to respiration detection field, more particularly, to a kind of breathing rate extracting method and device.
Background technology
Breathing is the important physiological process of human body, and the monitoring detection to human body respiration is also the one of modern medicine monitoring technology
Individual important component part.Patient whether the pathological changes of respiratory system itself or the pathological development of other important organs to certain journey
Degree all can affect respiratory center.The exhaustion of respiratory function is often involved in multi viscera systemic-function exhaustion, and the exhaustion of respiratory function is again
Lead to the exhaustion of other organs function, reciprocal causation.
Prior art is mainly detected using following method to respiratory movement：Impedance volumetric method：Measure breast with high-frequency constant current source
The change of portion's impedance is extracting respiration information；Sensor method：It is used temperature, pressure, humidity and pneumatic sensor to pass as nostril
Sensor；Capacitance method：Lead to capacitance to produce corresponding change when breathing；Respiratory murmur method：By picking up respiratory murmur identification of breathing；
Ultrasonic method：Produce Doppler phenomenon using ultrasound wave, detect respiratory frequency.Increase signal is not only needed to adopt using these methods
Collection part, and it is subject to the shadow noon of motion and environment, be not suitable for daily monitoring.
A large amount of clinical datas show, respiratory movement can cause Electrocardiographic change.By electrocardiogram, we can observe that
By the change of the caused ecg wave form peak-to-peak value of chest exercise and cardiac position change within the breathing cycle.This is due to breathing
In cycle, the heart electric axis rotation in description heart electric wave main propagation direction causes QRS complex form generation to change.QRS wave
Refer to the maximum wave group of amplitude in normal ECG, the overall process of reflection sequences of ventricular depolarization.Normal ventricle depolarization starts from interventricular septum
Portion, direction depolarization from left to right, therefore QRS complex first assumes a little downward q ripple.Normal chest lead QRS complex form is more permanent
Fixed.Extracting breath signal (ECG-DerivedRespiration, EDR) from electrocardiosignal is a kind of breath signal detection skill
Art, this technology does not need sensor special and hardware module detection breath signal it is only necessary to obtain electrocardio with electrocardiogram monitor
Signal, it is to avoid the constraint to human body for the above two detection method, makes the dynamic respiration detection be possibly realized.
But the existing technology extracting breath signal from electrocardiosignal, mainly adopts Waveform Method when calculating, and the method is led to
The meansigma methodss (i.e. baseline value) of interior waveform, to judge that current respiratory wave is in the trend of rising or falling, to use extreme value after a while
Method try to achieve the crest of waveform, trough.Effective crest or trough are judged according to certain threshold condition, further according to effective
The computation of Period wave period of crest or trough, thus obtain breathing rate.Although this algorithm has, comparison is directly perceived, operand is little
Advantage, but in real process obtain respiratory waveform more or less can be affected by electrocardio-activity, when base in waveform
During line drift, the baseline value of calculating cannot update quickly, waveform missing inspection can be led to cause breathing rate value low, its result has relatively
Large deviation.
Content of the invention
For the problems referred to above, it is an object of the invention to provide a kind of breathing rate extracting method and device, realize accurately may be used
The measurement of the breathing rate leaning on, and the measurement fluctuation causing due to the interference of the external world or environment or error can be mitigated.
The invention provides a kind of breathing rate extracting method, including：
By the good neural network model with regard to breath signal of training in advance, the electrocardiosignal receiving is extracted,
Obtain the first breath signal, and be calculated the first breathing rate of current time according to described first breath signal；
By the autoregression model building with regard to breath signal, described electrocardiosignal is extracted, obtain second and exhale
Inhale signal, and be calculated the second breathing rate of current time according to described second breath signal；
Based on signal quality index, described first breath signal and described second breath signal are analyzed, obtain with
Described corresponding first weight factor of first breath signal and second weight factor corresponding with described second breath signal；
According to described first breathing rate, the first weight factor, the second breathing rate and the second weight factor, it is calculated current
The breathing rate in moment.
Preferably, in the described neural network model good by training in advance, described pending electrocardiosignal is carried
Take, obtain the first breath signal, and before the first breathing rate of current time is calculated according to described first breath signal, also
Including：
Receive multi-lead electrocardiosignal, calculate phase and R peak amplitude between the RR of each electrocardiosignal of leading respectively, inputted
Sample space, wherein, the dimension in described input sample space is p, and p/2 is the number that leads of multi-lead electrocardiosignal；
Processed to according to the covariance matrix that described input sample space is formed based on Principal component analysis method, led
Component score matrix；
The target breath signal obtained with described principal component scores matrix and by impedance method synchronous acquisition is as training sample
To being trained, obtain neural network model.
Preferably, described carried out to according to the covariance matrix that described input sample space is formed based on Principal component analysis method
Process, obtain principal component scores matrix, specifically include：
Data normalization process is carried out to described input sample space；
Described input sample space after being processed according to data normalization obtains covariance matrix；
Calculate the characteristic root of described covariance matrix and characteristic vector corresponding with each characteristic root；Wherein, described feature
The quantity of root is p, and p described characteristic root is in magnitude order；
Obtain in p described characteristic root, contribution rate sum is more than the front m characteristic root of predetermined threshold；Wherein, each is special
The contribution rate levying root is equal to the value sum divided by p whole characteristic roots for the value of described characteristic root；
According to the described corresponding characteristic vector of front m characteristic root and described input sample space, obtain main constituent and obtain
Sub-matrix.
Preferably, described autoregression model is the autoregression model after moving average method optimizes.
Preferably, described according to described first breathing rate, the first weight factor, the second breathing rate and the second weight factor,
The breathing rate being calculated current time is specially：
When judging that described first weight factor is more than default reference value and described second weight factor and is less than described benchmark
During value, described first breathing rate is set to the breathing rate of current time；
When judging that described first weight factor is less than default reference value and described second weight factor and is more than described benchmark
During value, described second breathing rate is set to the breathing rate of current time；
When judging that the second weight factor described in described first weight factor meter is all higher than default reference value, according to described
First weight factor and described second weight factor are weighted to described first breathing rate and the second breathing rate suing for peace, and calculate
Breathing rate to current time.
Present invention also offers a kind of breathing rate extraction element, including:
Neutral net extraction unit, for the neural network model with regard to breath signal good by training in advance to reception
To electrocardiosignal extracted, obtain the first breath signal, and current time be calculated according to described first breath signal
The first breathing rate；
Autoregression extraction unit, for the autoregression model with regard to breath signal by building to described electrocardiosignal
Extracted, obtain the second breath signal, and be calculated the second breathing rate of current time according to described second breath signal；
Signal quality analytic unit, for based on signal quality index, exhaling to described first breath signal and described second
Inhale signal be analyzed, obtain first weight factor corresponding with described first breath signal and with described second breath signal pair
The second weight factor answered；
Breathing rate computing unit, for according to described first breathing rate, the first weight factor, the second breathing rate and the second power
Repeated factor, is calculated the breathing rate of current time.
Preferably, also include：
Input sample space acquiring unit, for receiving multi-lead electrocardiosignal, calculates each electrocardiosignal of leading respectively
RR between phase and R peak amplitude, obtain input sample space, wherein, the dimension in described input sample space is p, and p/2 is multi-lead
The number that leads of electrocardiosignal；
Principal component analysiss unit, for based on Principal component analysis method to the covariance being formed according to described input sample space
Matrix is processed, and obtains principal component scores matrix；
Neural metwork training unit, for described principal component scores matrix with the mesh that obtained by impedance method synchronous acquisition
Mark breath signal obtains neural network model for training sample to carrying out neural metwork training.
Preferably, described principal component analysiss unit specifically includes：
Standardization module, for processing to described input sample Standardization of Spatial Data；
Covariance matrix computing module, obtains association side for the described input sample space after processing according to data normalization
Difference matrix；
Feature calculation module, for calculate the characteristic root of described covariance matrix and feature corresponding with each characteristic root to
Amount；Wherein, the quantity of described characteristic root is p, and p described characteristic root is in magnitude order；
Screening module, for obtaining in p described characteristic root, contribution rate sum is more than the front m feature of predetermined threshold
Root；Wherein, the contribution rate of each characteristic root is equal to the value sum divided by p whole characteristic roots for the value of described characteristic root；
Score matrix acquisition module, for basis and the described corresponding characteristic vector of front m characteristic root and described input
Sample space, obtains principal component scores matrix.
Preferably, described autoregression model is the autoregression model after moving average method optimizes.
Preferably, described breathing rate computing unit specifically includes：
For working as, first judge module, judges that described first weight factor is more than default reference value and described second weight
When the factor is less than described reference value, described first breathing rate is set to the breathing rate of current time；
For working as, second judge module, judges that described first weight factor is less than default reference value and described second weight
When the factor is more than described reference value, described second breathing rate is set to the breathing rate of current time；
3rd judge module, the second weight factor described in based on working as and judging described first weight factor is all higher than default
During reference value, according to described first weight factor and described second weight factor, described first breathing rate and the second breathing rate are entered
Row weighted sum, is calculated the breathing rate of current time.
Breathing rate extracting method and device that the present invention provides, by using neural network model and auto-regressive time series
The mode processing cardioelectric signals that technology combines obtain the first breathing rate and the second breathing rate, and according to described first breathing rate
Corresponding first weight factor and second weight factor corresponding with described second breathing rate obtain the breathing rate of current time, phase
Ratio in the existing scheme being obtained breath signal by monotechnics from electrocardiosignal, result of calculation more accurately and reliably, and can mitigate by
The measurement fluctuation or the error that cause in the interference of extraneous or environment are such that it is able to obtain the measurement result of more accurate stable.
Brief description
Fig. 1 is a kind of flow chart of breathing rate extracting method provided in an embodiment of the present invention；
Fig. 2 is the oscillogram of original electro-cardiologic signals provided in an embodiment of the present invention；
Fig. 3 be notch filter provided in an embodiment of the present invention after pending electrocardiosignal oscillogram；
Fig. 4 is the waveform extracting the first breath signal obtaining by neural network model provided in an embodiment of the present invention
Figure；
Fig. 5 is the oscillogram extracting the second breath signal obtaining by autoregression model provided in an embodiment of the present invention.
Fig. 6 is the structural representation of breathing rate extraction element provided in an embodiment of the present invention.
Specific embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation description is it is clear that described embodiment is only a part of embodiment of the present invention, rather than whole embodiments.It is based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of not making creative work
Embodiment, broadly falls into the scope of protection of the invention.
The invention provides a kind of breathing rate extracting method, for extracting respiration information from electrocardiosignal, due to exhaling
Baseline drift in the electrocardiogram that suction effect causes, regards respiration information as the low-frequency component of electrocardiosignal, is exhaled by removing
Signal beyond suction frequency, thus obtain the respiration information of required extraction.
Referring to Fig. 1, it is a kind of breathing rate extracting method provided in an embodiment of the present invention, comprises the steps：
S1：By the good neural network model with regard to breath signal of training in advance, the electrocardiosignal receiving is carried
Take, obtain the first breath signal, and be calculated the first breathing rate of current time according to described first breath signal.
It should be noted that as shown in Figure 2 because original electrocardiosignal usually contains substantial amounts of Hz noise, need into
Row 50Hz notch filter, to filter Hz noise, referring to Fig. 3, be according to the notch filter of the embodiment of the present invention after electrocardio letter
Number oscillogram.
In embodiments of the present invention, when being extracted to described pending electrocardiosignal using neural network model, need
First pass through the neural network model that training obtains can be used for extracting breath signal.
Specifically, in embodiments of the present invention, the calculating of neural network model can be carried out by the following method：
S01, receives multi-lead electrocardiosignal, calculates phase and R peak amplitude between the RR of each electrocardiosignal of leading respectively, obtain
Input sample space, wherein, the dimension in described input sample space is p, and p/2 is the number that leads of multi-lead electrocardiosignal.
In embodiments of the present invention, input sample space X=[x1, x2 ..., xn] represents m*n dimensional vector, wherein x1 table
Show the column vector that length is m, by calculating the phase between 1 RR leading, obtain x1, calculate the 1 R peak amplitude leading, obtain x2, calculate
Phase between 2 RR leading, obtain x3；Calculate the 2 R peak amplitudes leading, obtain x4；... obtain by that analogy.
S02, is processed to according to the covariance matrix that described input sample space is formed based on Principal component analysis method, obtains
To principal component scores matrix.
In embodiments of the present invention it is contemplated that standard multi-lead electrocardiosignal is led for 12, calculate what each led respectively
After phase and R peak amplitude between RR, the eigenvalue being input to neutral net is needed to reach 24, and interrelated between respectively leading, lead
Cause input sample dimension is larger and the input item containing linear correlation, be unfavorable for calculating analysis, need for this to utilize main constituent
Analytic process carries out dimensionality reduction to it.
Specifically, step S12 may include：
S021, carries out data normalization process to described input sample space.
Specifically,
Wherein：
Wherein, Xi'jIt is the new data after standardization；Mj、SjRespectively represent initial data string arithmetic mean of instantaneous value and
Standard (inclined) is poor.
S022, the described input sample space after being processed according to data normalization obtains covariance matrix.
Wherein, covariance matrix D=XTX, that is,：
Wherein：
S023, calculates the characteristic root of described covariance matrix and characteristic vector corresponding with each characteristic root；Wherein, described
The quantity of characteristic root is p, and p described characteristic root is in magnitude order.
Wherein, DP=P λ (6)
When only considering j-th eigenvalue, there is DPj=Pjλj, that is, solve | D- λjI |=0.Solve each λ successively, and make
Its order arrangement by size, i.e. λ1≥λ2≥…,≥λp≥0；Then corresponding characteristic vector P of each eigenvalue, Jin Erte can be obtained
Levy equation solution to complete.
S024, obtains in p described characteristic root, contribution rate sum is more than the front m characteristic root of predetermined threshold；Wherein,
The contribution rate of each characteristic root is equal to the value sum divided by p whole characteristic roots for the value of described characteristic root.
First, calculate the contribution rate of single main constituent and added up, determine the number of main constituent according to contribution rate of accumulative total
M, so that it is determined that the main constituent of required selection.The computing formula of contribution rate such as formula (7) is described.Contribution rate of accumulative total i.e. first m
The accumulation of contribution rate and, such as shown in formula (8).Described threshold value Dmax is typically taken between 85%～95%.According to previous step
In characteristic root sequence understand, λ1≥λ2≥…,≥λp>=0, from front to back (being also from big to small) successively characteristic root is tired out
Plus, work as contribution rate of accumulative totalDuring more than Dmax, stop calculating, now the number of the characteristic root λ of cumulative calculation is m,
Then only need to choose front m main constituent.
S025, according to the described corresponding characteristic vector of front m characteristic root and described input sample space, obtain main one-tenth
Get sub-matrix.
Wherein, described principal component scores matrix
It should be noted that the load of main constituent in embodiments of the present invention, also can be calculated, wherein, described main constituent carries
Lotus mainly reflects principal component scores and former variable xjCorrelation degree, computing formula is：The load obtaining each main constituent is later it is possible to know each of selection
Individual main constituent corresponding primitive character respectively, if it is desired, can go back according to the dimension conversion of primitive character.
S13, the target breath signal obtained with described principal component scores matrix and by impedance method synchronous acquisition is for training
Sample, to carrying out neural metwork training, obtains neural network model.
In embodiments of the present invention, principal component scores matrix T and the target obtaining by impedance method synchronous acquisition are being obtained
Breath signal Y is it is possible to carry out the training of neutral net, specifically：
Neural network model is made up of input layer (feature), hidden layer, output layer.Wherein, every layer all corresponds to one accordingly
Function.In training, described main constituent eigenmatrix T is input to input layer, the mesh that will obtain by impedance method synchronous acquisition
Mark breath signal is input to output layer (as output layer adopts softmax function), you can use obtains the parameter of hidden layer (such as
The parameter of sigmod function).Wherein, the parameter of neural network model, including weight, biasing etc..
In embodiments of the present invention, the hidden layer of neural network model can for 1 (desirable other values, under present case, 1
Disclosure satisfy that requirement), the number of hidden nodes is K, and learning rate is μ, generally learning rate is selected between 0.02～0.2.
In order to obtain optimal network training effect, e-learning speed can be determined using trial and error procedure.
In embodiments of the present invention, the learning algorithm of neutral net can choose Levenberg-Marquart algorithm, also may be used
Using other learning algorithms, the present invention is not specifically limited.
In embodiments of the present invention, after obtaining the neural network model training, only described pending electrocardio need to be believed
Number as the input layer being input to described neural network model, you can to obtain carrying from described electrocardiosignal in output layer
Obtain the first breath signal as shown in Figure 4, hereafter, you can the first breathing is calculated according to described first breath signal
Rate.
Specifically, the ripple of the first breath signal can be found in the oscillogram of described first breath signal by seeking extremum method
Peak (or trough), referring to the point labelling in Fig. 4；
By extracting the time interval between two crests being newly generated, to obtain the cycle T of current time.
The described cycle is carried out with the first breathing rate R1 that sampling rate conversion can get current time.
For example, R1=60/T1.
S2：By the autoregression model building with regard to breath signal, described electrocardiosignal is extracted, obtain
Two breath signals, and it is calculated the second breathing rate of current time according to described second breath signal.
In embodiments of the present invention, when described pending electrocardiosignal being extracted using described autoregression model, need
First build autoregression model, its building process is as follows：
For autoregression model AR (p), it is represented by：
φ(B)yt=at(10)
Wherein, B is delay operator, Byt=yt-1；P is the exponent number of model, represents autoregression item number, ytFor seasonal effect in time series
Currency；atFor random disturbances.Meet stationarity condition.In AR model, current time
Observation ytTo be represented by the observation of p historical juncture of past and the random disturbances of a current time.
In embodiments of the present invention, for noise reduction, especially white noise, it is also with moving average method to optimize autoregression
Model is it is assumed that the exponent number of moving average method is q, then θ (B)=1- θ1-...-θqBq, moving average model MA (q) such as formula 11 institute
Show, the observation y of current timetTo be represented by the observation of q historical juncture of past and the random disturbances of a current time, yt
For seasonal effect in time series currency；atFor random disturbances.Using this model, autoregression model is optimized, then can get and certainly return
Return-moving average model ARMA (p, q), wherein, p, q are model order (p is autoregression item number, and q is moving averages item number), such as
Shown in formula 12.
yt=θ (B) at(11)
φ(B)yt=θ (B) at(12)
In embodiments of the present invention, after obtaining described auto-regressive moving-average model, you can carry out carrying of breath signal
Take.Specifically, auto-regressive moving-average model is a kind of method for extracting signal of blind source separating.First, by estimating in model
Weighting parameters, calculate ECG mixed signal (and described pending electrocardiosignal, which includes breath signal) ARMA (p,
Q) coefficient matrix of model, as the feature of breath signal；Secondly, the feature of the breath signal obtaining in conjunction with estimation, using certainly
Related separation algorithm, extracts to described pending electrocardiosignal, reaches clean ECG signal and the detached purpose of breath signal
Extraction obtains breath signal.
Referring to Fig. 5, it is the oscillogram of the second breath signal being obtained according to the autoregression model extraction that the present invention is implemented.
In embodiments of the present invention, after obtaining described second breath signal, you can calculate the second breathing rate R2, specifically
For：
Find crest (or the ripple of the second breath signal by seeking extremum method in the oscillogram of described second breath signal
Paddy), referring to the point labelling in Fig. 5；
By extracting the time interval between two crests being newly generated, to obtain cycle T 2.
Real-time second breathing rate R2 be can get according to sampling rate conversion.
S3：Based on signal quality index, described first breath signal and described second breath signal are analyzed, obtain
First weight factor corresponding with described first breath signal and second weight factor corresponding with described second breath signal.Tool
Body is：
Described first breath signal and described second breath signal are carried out with power spectrumanalysises (or peak value analysis of spectrum), analysis
Described first breath signal and the Spectral structure of described second breath signal, obtain first power corresponding with described first breath signal
Repeated factor and second weight factor corresponding with described second breath signal.
S4：According to described first breathing rate, the first weight factor, the second breathing rate and the second weight factor, it is calculated
The breathing rate of current time.
In a preferred embodiment, can be by described first breathing rate, the first weight factor, the second breathing rate and
Two weight factors are weighted averagely obtaining the breathing rate of current time, that is,：
R=μ 1*R1+ μ 2*R2 (13)
It should be noted that before being weighted averagely, needing first μ 1 and μ 2 is normalized, specifically, false
If μ is 1+ μ 2=a, then needs respectively μ 1 and μ 2 is multiplied by with normalization coefficient 1/a and be normalized the μ 1+ μ 2 it is ensured that after normalization
=1.
In another embodiment, the breathing rate being calculated current time specifically includes：
When judging that described first weight factor is more than default reference value and described second weight factor and is less than described benchmark
During value, described first breathing rate is set to the breathing rate of current time.
When described second weight factor be less than described reference value when it is believed that the signal quality of the second breath signal relatively
Described first breathing rate R1 now, is directly set to the breathing rate R of current time by difference.
When judging that described first weight factor is less than default reference value and described second weight factor and is more than described benchmark
During value, described second breathing rate is set to the breathing rate of current time.
When described first weight factor be less than described reference value when it is believed that the signal quality of the first breath signal relatively
Described first breathing rate R1 now, is directly set to the breathing rate R of current time by difference.
When judging that the second weight factor described in described first weight factor meter is all higher than default reference value, according to described
First weight factor and described second weight factor are weighted to described first breathing rate and the second breathing rate suing for peace, and calculate
Breathing rate to current time.
I.e.：R=μ 1*R1+ μ 2*R2.
If weight factor is less, illustrates that corresponding breath signal is second-rate, then directly remove and exhale with second-rate
Inhale the corresponding breathing rate of signal it is ensured that result of calculation accurately and stable.
In the embodiment of the present invention, by way of neural network model is combined with auto-regressive time series technology, process the heart
The signal of telecommunication obtains the first breathing rate and the second breathing rate, and according to first weight factor corresponding with described first breathing rate and with
Corresponding second weight factor of described second breathing rate obtains the breathing rate of current time, and result of calculation more accurately and reliably, and can
Mitigate measurement fluctuation or the error causing due to the interference of the external world or environment such that it is able to obtain the measurement of more accurate stable
Result.
Refering to Fig. 6, the embodiment of the present invention also provides a kind of breathing rate extraction element 100, including：
Neutral net extraction unit 10, for the neural network model docking with regard to breath signal good by training in advance
The electrocardiosignal receiving is extracted, and obtains the first breath signal, and when being calculated current according to described first breath signal
The first breathing rate carved.
Autoregression extraction unit 20, believes to described electrocardio for the autoregression model with regard to breath signal by building
Number extracted, obtain the second breath signal, and be calculated the second breathing of current time according to described second breath signal
Rate.
Signal quality analytic unit 30, for based on signal quality index, to described first breath signal and described second
Breath signal is analyzed, obtain first weight factor corresponding with described first breath signal and with described second breath signal
Corresponding second weight factor.
Breathing rate computing unit 40, for according to described first breathing rate, the first weight factor, the second breathing rate and second
Weight factor, is calculated the breathing rate of current time.
In the embodiment of the present invention, by using at the mode that neural network model is combined with auto-regressive time series technology
Reason electrocardiosignal obtains the first breathing rate and the second breathing rate, and according to first weight factor corresponding with described first breathing rate
With second weight factor corresponding with described second breathing rate obtains the breathing rate of current time, result of calculation more accurately and reliably,
And measurement fluctuation or the error that can mitigate the interference due to extraneous or environment and cause are such that it is able to obtain more accurate stable
Measurement result.
Preferably, described breathing rate extraction element 100 also includes：
Input sample space acquiring unit 50, for receiving multi-lead electrocardiosignal, calculates each electrocardio letter that leads respectively
Number RR between phase and R peak amplitude, obtain input sample space, wherein, the dimension in described input sample space is p, and p/2 is to lead more
The number that leads of connection electrocardiosignal；
Principal component analysiss unit 60, for based on Principal component analysis method to the association side being formed according to described input sample space
Difference matrix is processed, and obtains principal component scores matrix；
Neural metwork training unit 70, for obtained with described principal component scores matrix with by impedance method synchronous acquisition
Target breath signal obtains neural network model for training sample to carrying out neural metwork training.
Preferably, described principal component analysiss unit 60 specifically includes：
Standardization module 61, for processing to described input sample Standardization of Spatial Data；
Covariance matrix computing module 62, is assisted for the described input sample space after being processed according to data normalization
Variance matrix；
Feature calculation module 63, for calculating the characteristic root of described covariance matrix and feature corresponding with each characteristic root
Vector；Wherein, the quantity of described characteristic root is p, and p described characteristic root is in magnitude order；
Screening module 64, for obtaining in p described characteristic root, contribution rate sum is more than the front m spy of predetermined threshold
Levy root；Wherein, the contribution rate of each characteristic root is equal to the value sum divided by p whole characteristic roots for the value of described characteristic root；
Score matrix acquisition module 65, for according to the described corresponding characteristic vector of front m characteristic root and described defeated
Enter sample space, obtain principal component scores matrix.
Preferably, described autoregression model is the autoregression model after moving average method optimizes.
Preferably, described breathing rate computing unit 40 specifically includes：
For working as, first judge module 41, judges that described first weight factor is more than default reference value and described second power
When repeated factor is less than described reference value, described first breathing rate is set to the breathing rate of current time；
For working as, second judge module 42, judges that described first weight factor is less than default reference value and described second power
When repeated factor is more than described reference value, described second breathing rate is set to the breathing rate of current time；
3rd judge module 43, the second weight factor described in based on working as and judging described first weight factor is all higher than presetting
Reference value when, according to described first weight factor and described second weight factor to described first breathing rate and the second breathing rate
It is weighted suing for peace, be calculated the breathing rate of current time.
Above disclosed be only two kinds of preferred embodiments of the present invention, certainly the power of the present invention can not be limited with this
Sharp scope, one of ordinary skill in the art will appreciate that realize all or part of flow process of above-described embodiment, and according to present invention power
Profit requires made equivalent variations, still falls within the scope that invention is covered.
One of ordinary skill in the art will appreciate that realizing all or part of flow process in above-described embodiment method, it is permissible
Instruct related hardware to complete by computer program, described program can be stored in a computer read/write memory medium
In, this program is upon execution, it may include as the flow process of the embodiment of above-mentioned each method.Wherein, described storage medium can be magnetic
Dish, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random Access
Memory, RAM) etc..
Claims (10)
1. a kind of breathing rate extracting method is it is characterised in that include：
By the good neural network model with regard to breath signal of training in advance, the electrocardiosignal receiving is extracted, obtain
First breath signal, and it is calculated the first breathing rate of current time according to described first breath signal；
By the autoregression model building with regard to breath signal, described electrocardiosignal is extracted, obtain the second breathing letter
Number, and it is calculated the second breathing rate of current time according to described second breath signal；
Based on signal quality index, described first breath signal and described second breath signal are analyzed, obtain with described
Corresponding first weight factor of first breath signal and second weight factor corresponding with described second breath signal；
According to described first breathing rate, the first weight factor, the second breathing rate and the second weight factor, it is calculated current time
Breathing rate.
2. breathing rate extracting method according to claim 1 it is characterised in that described by training in advance good with regard to
The neural network model of breath signal extracts to described pending electrocardiosignal, obtains the first breath signal, and according to institute
Before stating the first breathing rate that the first breath signal is calculated current time, also include：
Receive multi-lead electrocardiosignal, calculate phase and R peak amplitude between the RR of each electrocardiosignal of leading respectively, obtain input sample
Space, wherein, the dimension in described input sample space is p, and p/2 is the number that leads of multi-lead electrocardiosignal；
Processed to according to the covariance matrix that described input sample space is formed based on Principal component analysis method, obtain main constituent
Score matrix；
The target breath signal obtained with described principal component scores matrix and by impedance method synchronous acquisition is for training sample to entering
Row training, obtains neural network model.
3. breathing rate extracting method according to claim 2 it is characterised in that
Described processed to according to the covariance matrix that described input sample space is formed based on Principal component analysis method, led
Component score matrix, specifically includes：
Data normalization process is carried out to described input sample space；
Described input sample space after being processed according to data normalization obtains covariance matrix；
Calculate the characteristic root of described covariance matrix and characteristic vector corresponding with each characteristic root；Wherein, described characteristic root
Quantity is p, and p described characteristic root is in magnitude order；
Obtain in p described characteristic root, contribution rate sum is more than the front m characteristic root of predetermined threshold；Wherein, each characteristic root
Contribution rate be equal to described characteristic root value divided by p whole characteristic roots value sum；
According to the described corresponding characteristic vector of front m characteristic root and described input sample space, obtain principal component scores square
Battle array.
4. the breathing rate extracting method according to claims 1 to 3 any one is it is characterised in that described autoregression model
It is the autoregression model after moving average method optimizes.
5. the breathing rate extracting method according to Claims 1-4 any one it is characterised in that described according to described
One breathing rate, the first weight factor, the second breathing rate and the second weight factor, the breathing rate being calculated current time is concrete
For：
When judging that described first weight factor is more than default reference value and described second weight factor and is less than described reference value,
Described first breathing rate is set to the breathing rate of current time；
When judging that described first weight factor is less than default reference value and described second weight factor and is more than described reference value,
Described second breathing rate is set to the breathing rate of current time；
When judging that the second weight factor described in described first weight factor meter is all higher than default reference value, according to described first
Weight factor and described second weight factor are weighted to described first breathing rate and the second breathing rate suing for peace, and are calculated and work as
The breathing rate in front moment.
6. a kind of breathing rate extraction element is it is characterised in that include:
Neutral net extraction unit, for the neural network model with regard to breath signal good by training in advance to receiving
Electrocardiosignal is extracted, and obtains the first breath signal, and is calculated the of current time according to described first breath signal
One breathing rate；
Autoregression extraction unit, is carried out to described electrocardiosignal for the autoregression model with regard to breath signal by building
Extract, obtain the second breath signal, and be calculated the second breathing rate of current time according to described second breath signal；
Signal quality analytic unit, for based on signal quality index, to described first breath signal and described second breathing letter
Number it is analyzed, obtain first weight factor corresponding with described first breath signal and corresponding with described second breath signal
Second weight factor；
Breathing rate computing unit, for according to described first breathing rate, the first weight factor, the second breathing rate and the second weight because
Son, is calculated the breathing rate of current time.
7. breathing rate extraction element according to claim 6 is it is characterised in that also include：
Input sample space acquiring unit, for receiving multi-lead electrocardiosignal, calculates the RR of each electrocardiosignal of leading respectively
Between phase and R peak amplitude, obtain input sample space, wherein, the dimension in described input sample space is p, and p/2 is multi-lead electrocardio
The number that leads of signal；
Principal component analysiss unit, for based on Principal component analysis method to the covariance matrix being formed according to described input sample space
Processed, obtained principal component scores matrix；
Neural metwork training unit, is exhaled for described principal component scores matrix with by the target that impedance method synchronous acquisition obtains
Inhale signal for training sample to carrying out neural metwork training, obtain neural network model.
8. breathing rate extraction element according to claim 7 is it is characterised in that described principal component analysiss unit specifically wraps
Include：
Standardization module, for processing to described input sample Standardization of Spatial Data；
Covariance matrix computing module, obtains covariance square for the described input sample space after processing according to data normalization
Battle array；
Feature calculation module, for calculating the characteristic root of described covariance matrix and characteristic vector corresponding with each characteristic root；
Wherein, the quantity of described characteristic root is p, and p described characteristic root is in magnitude order；
Screening module, for obtaining in p described characteristic root, contribution rate sum is more than the front m characteristic root of predetermined threshold；Its
In, the contribution rate of each characteristic root is equal to the value sum divided by p whole characteristic roots for the value of described characteristic root；
Score matrix acquisition module, for basis and the described corresponding characteristic vector of front m characteristic root and described input sample
Space, obtains principal component scores matrix.
9. breathing rate extraction element according to claim 6 is it is characterised in that described autoregression model is flat through sliding
Autoregression model after all method optimizes.
10. the breathing rate extraction element according to claim 6 to 9 any one is it is characterised in that described breathing rate calculates
Unit specifically includes：
For working as, first judge module, judges that described first weight factor is more than default reference value and described second weight factor
During less than described reference value, described first breathing rate is set to the breathing rate of current time；
For working as, second judge module, judges that described first weight factor is less than default reference value and described second weight factor
During more than described reference value, described second breathing rate is set to the breathing rate of current time；
3rd judge module, the second weight factor described in based on working as and judging described first weight factor is all higher than default benchmark
During value, described first breathing rate and the second breathing rate are carried out add according to described first weight factor and described second weight factor
Power summation, is calculated the breathing rate of current time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610899933.8A CN106388824B (en) | 2016-10-14 | 2016-10-14 | Respiration rate extraction method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610899933.8A CN106388824B (en) | 2016-10-14 | 2016-10-14 | Respiration rate extraction method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106388824A true CN106388824A (en) | 2017-02-15 |
| CN106388824B CN106388824B (en) | 2019-04-16 |
Family
ID=58011630
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610899933.8A Active CN106388824B (en) | 2016-10-14 | 2016-10-14 | Respiration rate extraction method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106388824B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106901694A (en) * | 2017-02-20 | 2017-06-30 | 广州视源电子科技股份有限公司 | Respiration rate extraction method and device |
| CN109009125A (en) * | 2018-06-07 | 2018-12-18 | 上海交通大学 | Driver's fine granularity monitoring of respiration method and system based on audio frequency of mobile terminal |
| CN114190951A (en) * | 2021-12-31 | 2022-03-18 | 上海联影医疗科技股份有限公司 | ECG signal processing method, system, device and medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060058590A1 (en) * | 2004-08-24 | 2006-03-16 | Shaw Geoffrey M | Method and system for assaying agitation |
| CN101991418A (en) * | 2009-08-14 | 2011-03-30 | 深圳市理邦精密仪器股份有限公司 | Method for improving respiratory rate detection accuracy |
| CN105769122A (en) * | 2016-03-16 | 2016-07-20 | 中国科学院电子学研究所 | Method for measuring sleep apnea hyponea index |
-
2016
- 2016-10-14 CN CN201610899933.8A patent/CN106388824B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060058590A1 (en) * | 2004-08-24 | 2006-03-16 | Shaw Geoffrey M | Method and system for assaying agitation |
| CN101991418A (en) * | 2009-08-14 | 2011-03-30 | 深圳市理邦精密仪器股份有限公司 | Method for improving respiratory rate detection accuracy |
| CN105769122A (en) * | 2016-03-16 | 2016-07-20 | 中国科学院电子学研究所 | Method for measuring sleep apnea hyponea index |
Non-Patent Citations (3)
| Title |
|---|
| 王跃俊: "《基于光电容积脉搏波的呼吸频率监测》", 《北京理工大学 硕士论文》 * |
| 邓宝芸: "《基于心电和脉搏波数据融合的呼吸率估计》", 《山东大学 硕士论文》 * |
| 郎晓华: "《基于ARM的4个生理参数的快速测量系统》", 《山东大学 硕士论文》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106901694A (en) * | 2017-02-20 | 2017-06-30 | 广州视源电子科技股份有限公司 | Respiration rate extraction method and device |
| WO2018149147A1 (en) * | 2017-02-20 | 2018-08-23 | 广州视源电子科技股份有限公司 | Method and apparatus for extracting respiration rate |
| CN109009125A (en) * | 2018-06-07 | 2018-12-18 | 上海交通大学 | Driver's fine granularity monitoring of respiration method and system based on audio frequency of mobile terminal |
| CN109009125B (en) * | 2018-06-07 | 2019-05-28 | 上海交通大学 | Driver's fine granularity monitoring of respiration method and system based on audio frequency of mobile terminal |
| CN114190951A (en) * | 2021-12-31 | 2022-03-18 | 上海联影医疗科技股份有限公司 | ECG signal processing method, system, device and medium |
| CN114190951B (en) * | 2021-12-31 | 2023-11-03 | 上海联影医疗科技股份有限公司 | ECG signal processing method, system, device and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106388824B (en) | 2019-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106388825B (en) | Respiration rate extraction method and device | |
| CN106108889B (en) | Electrocardiogram classification method based on deep learning algorithm | |
| CN101801263B (en) | Monitoring physiological condition and detecting abnormalities | |
| CN102245102B (en) | Method and apparatus for the analysis of ballistocardiogram signals | |
| CN104244814B (en) | The monitoring of cardiac output | |
| CN102488503B (en) | Continuous blood pressure measurer | |
| CN106073755A (en) | The implementation method that in a kind of miniature holter devices, atrial fibrillation identifies automatically | |
| CN107358196A (en) | A kind of sorting technique of heart beat type, device and electrocardiogram equipment | |
| CN106066933B (en) | The method for obtaining atrial fibrillation identification artificial neural network weight value matrix | |
| WO2009100133A1 (en) | System, method and computer program product for detection of changes in health status and risk of imminent illness | |
| EP2688468A1 (en) | Apparatus and method for measuring physiological signal quality | |
| CN101065058A (en) | Monitoring physiological activity using partial state space reconstruction | |
| CN106073784B (en) | Respiration rate extraction method and device | |
| CN108992053B (en) | Method for real-time non-binding detection of heart rate and heartbeat interval | |
| CN114699040B (en) | Method, device, equipment and storage medium for detecting sleep awakening based on physiological signals | |
| CN107257653A (en) | Processing equipment, system and method for handling the accelerometer signal being used in being monitored to the vital sign of object | |
| CN106388824B (en) | Respiration rate extraction method and device | |
| CN110037668A (en) | The system that pulse signal time-space domain binding model judges age, health status and malignant arrhythmia identification | |
| CN106901694A (en) | Respiration rate extraction method and device | |
| CN103961089A (en) | Sinus heart rate turbulence tendency detecting method based on segmented straight line fitting | |
| CN106510676A (en) | Heart rate detection method and device | |
| CN111887858A (en) | Ballistocardiogram signal heart rate estimation method based on cross-modal mapping | |
| CN113160921A (en) | Construction method and application of digital human cardiovascular system based on hemodynamics | |
| WO2009127799A1 (en) | Method and apparatus for measuring breathing rate | |
| CN109549648A (en) | A kind of intelligent breathing internal medicine lung function Training Control system and method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |