CN110084096A - Electrocardiosignal P wave extracting method based on wavelet transformation and K means Data Cluster Algorithm - Google Patents
Electrocardiosignal P wave extracting method based on wavelet transformation and K means Data Cluster Algorithm Download PDFInfo
- Publication number
- CN110084096A CN110084096A CN201910186724.2A CN201910186724A CN110084096A CN 110084096 A CN110084096 A CN 110084096A CN 201910186724 A CN201910186724 A CN 201910186724A CN 110084096 A CN110084096 A CN 110084096A
- Authority
- CN
- China
- Prior art keywords
- wave
- qrs
- class
- voltage
- electrocardiosignal
- 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
Classifications
-
- 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]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
- G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
- G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
- G06F2218/14—Classification; Matching by matching peak patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
Abstract
The invention discloses the electrocardiosignal P wave extracting methods based on wavelet transformation and K means Data Cluster Algorithm, specifically includes the following steps: S1, eliminating first with Zero-phase Digital Filter method the 50Hz interference and the baseline drift of low frequency interference of high frequency respectively, then a certain fixed sample rate will be resampled to except the electrocardiosignal after making an uproar, S2, the electrocardiosignal after S1 resampling is normalized, make its Distribution value between -0.5 and 0.5, the present invention relates to ECG's data compression technical fields.The electrocardiosignal P wave extracting method based on wavelet transformation and K means Data Cluster Algorithm, realizing well more steady under various arrhythmia conditions and under noise jamming can reliably extract P wave, can more accurately judge whether P wave is visible simultaneously, range is greatly expanded compared with the prior art, all possible P wave can be covered, without determining P wave position by artificially defined rule, therefore the more excellent robustness of algorithm performance is more preferably.
Description
Technical field
The present invention relates to ECG's data compression technical fields, specially the heart based on wavelet transformation and K means Data Cluster Algorithm
Electric signal P wave extracting method.
Background technique
Current electrocardiosignal P wave extracting method generally includes the following steps: 1, ECG signal processing: filtering out low frequency
And high frequency noise;2, pretreated electrocardiosignal is converted using wavelet transformation or other methods;3, after using conversion
Signal identification QRS starting, terminate and crest location;4, corresponding T is extracted respectively to each QRS complex using the signal after conversion
Wave and P wave.This method chooses the 5th layer of wavelet transformation detail coefficients according to Energy distribution, frequency analysis and average mutual property
(swd5) QRS wave extraction is carried out, layer 6 wavelet transformation detail coefficients (swd6) carry out the extraction of T wave and P wave, then pass through
Two search windows determine T wave point position and P wave point position in swd6 layers respectively
The shortcomings that prior art includes:
1, the prior art often determines the search range of P wave according to the feature of normal ECG, so as to cause different in analysis
Algorithm fails when Chang Xin electricity.Such as before limiting the range of P wave search in the above method as Q wave direction in 200 milliseconds, contextual quotation
As follows: " on the basis of the Q wave in Swdnj, the 4th search window that a 100/512S is arranged forward determines the 4th search
The O point of crossing of the first max min pair in window, first max min pair is P wave point position ", however,
When atrioventricular block occurs, the interval PR (i.e. the interval of P wave initial time and Q wave initial time) is likely to significantly beyond 200
Millisecond, this makes above-mentioned algorithm that can not accurately identify P wave position when atrioventricular block occurs, on the other hand, if above-mentioned
P wave search range is extended forward in method, then can be greatly increased using the probability that its algorithm detects the P wave that makes mistake again.
2, the prior art determines P wave position by artificially defined rule, and robustness is difficult to ensure, especially in signal
The position that P wave is provided when noise is larger, such as in the above method is the O point excessively of the first max min pair in search range,
But due to noise jamming such as electromyography signal, electrode perturbations, it is specified that there may be multiple max mins pair in range, and
First max min is to being difficult to ensure that corresponding is P wave.
3, it is the important evidence for judging auricular fibrillation and occurring that whether P wave is visible, and but the prior art is difficult to invisible to P wave
The case where make accurate judgement.For example the above method can find out corresponding P wave to each QRS complex.
Summary of the invention
(1) the technical issues of solving
In view of the deficiencies of the prior art, the present invention provides the electrocardiosignals based on wavelet transformation and K means Data Cluster Algorithm
P wave extracting method solves the search range that the prior art determines P wave according to the feature of normal ECG, so as to cause dividing
Algorithm fails when analysing abnormal electrocardiogram, and artificially defined rule determines P wave position, and robustness is difficult to ensure, while being difficult to P wave
The problem of sightless situation makes accurate judgement.
(2) technical solution
In order to achieve the above object, the present invention is achieved by the following technical programs: averagely poly- based on wavelet transformation and K
The electrocardiosignal P wave extracting method of class algorithm, specifically includes the following steps:
S1, the 50Hz for eliminating high frequency respectively first with Zero-phase Digital Filter method are interfered and the baseline drift of low frequency
Then interference will be resampled to a certain fixed sample rate except the electrocardiosignal after making an uproar:
S2, the electrocardiosignal after S1 resampling is normalized, makes its Distribution value between -0.5 and 0.5, is believed according to electrocardio
The spectrum signature of number QRS wave determines that interested frequency range is 11Hz to 40Hz, respectively adds in the left and right ends of electrocardiosignal
Then one piece of data does continuous wavelet change to electrocardiosignal after normalization to eliminate boundary effect in range of target frequencies
Change, calculate after Continuous Wavelet Transform Coefficients, both ends virtual sample is removed;
S3, the wavelet conversion coefficient obtained to S2, seek square of wavelet conversion coefficient on different frequency at various moments
With, to obtain each moment electrocardiosignal in the energy value of frequency domain, the small window of movement that a width is 0.3 second is then defined,
It finds electrocardio energy maximum in each small window, is a doubtful QRS wave peak at the time of corresponding to each electrocardio energy maximum,
Sample moves right the small window of the movement one by one, and thus we can obtain the doubtful QRS wave peak of a column;
S4, during S3 obtains a column doubtful QRS wave peak, duplicate doubtful QRS wave peak is removed, from doubtful QRS
Wave crest starts, and searches the initial position of the QRS wave to the left, does not find yet if be moved to the left always more than maximum search distance
Qualified initial position, then the wave is not true QRS wave, should be abandoned, while since doubtful QRS wave peak, being searched to the right
The final position of the QRS wave does not find qualified stop bit more than maximum search distance if moved right always yet
It sets, then the wave is not true QRS wave, should be abandoned;
S5, to each doubtful QRS wave, calculate separately the energy value at QRS wave wave crest, the absolute value of voltage at QRS wave peak,
The interval, the width of the QRS wave and the average voltage rate of change of the doubtful QRS wave of the QRS wave and previous doubtful QRS wave;
S6, to the ecg database labeled through electrocardio expert, handled by above step S1-S5, obtain a heart
The eigenmatrix of electric QRS wave shape, the line number of the matrix is equal to the number of all doubtful QRS waves, to each doubtful QRS wave, with expert
The QRS wave of label is compared, and judges whether it is true QRS waveform, and the eigenmatrix for then obtaining previous step inputs
In one Logic Regression Models, using gradient descent method as optimization method, setting learning rate is 0.02, and iteration is until convergence, training
In, using the method for staying 1 cross validation, i.e., training data is randomly divided into K parts, takes and wherein do training for K-1 parts, is left 1 part
It verifies, so repeatedly n times, whole are averaged to the N number of model of gained after the completion of training, and final mask are obtained, finally, with F1
Score is selecting index probability threshold value, below threshold value, it is believed that QRS wave be it is false, more than threshold value, it is believed that QRS wave is true;
S7, to any new electrocardiogram (ECG) data to be analyzed, handled, obtained using method of the step S1 into step S5
One group of doubtful QRS wave is obtained, to each doubtful QRS wave, is classified using the model that training obtains in preceding step, obtaining it is
The probability of true QRS wave, the probability threshold value determined in recycle step S6 finally determine whether each doubtful QRS wave is true
Real QRS wave is so far completed the extraction of the QRS feature of any electrocardiogram (ECG) data, to the classification results of each QRS wave detected
Do the quality information of electrocardiogram (ECG) data belonging to averagely can get;
S8, electrocardiosignal after the resampling obtained in step S1 is reprocessed, to protrude T wave part, to each
QRS complex is handled, then does wavelet transformation in selected frequency range to obtained signal, which includes T wave
Then main information seeks at various moments the wavelet conversion coefficient of acquisition the quadratic sum of wavelet conversion coefficient on different frequency,
To obtain each time-ofday signals in the energy value of frequency domain, smooth operation is carried out to the energy signal of acquisition later, specific method is
The average value of all energy values in 20 milliseconds centered on it is sought each moment, the energy at this moment is replaced with the average value
Magnitude, marking smoothed out energy signal is TES, and the search range of corresponding T wave is then determined to each QRS complex;
S9, to each QRS complex, find maximum value of the TES in corresponding T wave search range, this is T wave wave crest or wave
Paddy searches the initial position of the T wave, then since T wave wave crest or trough, search to the right since T wave wave crest or trough to the left
The final position of the T wave is sought, further the starting of each T wave, wave crest or the trough and final position that find are optimized later,
So far, the extraction of the starting of T wave, wave crest or trough, final position is completed;
S10, wavelet transformation is done in selected frequency range to electrocardiosignal after the resampling obtained in step S1, the frequency
Rate range includes the main information of P wave, then seeks wavelet transformation on different frequency at various moments to the wavelet conversion coefficient of acquisition
The quadratic sum of coefficient is denoted as energy signal PE, then to each QRS wave to obtain each time-ofday signals in the energy value of frequency domain
The search range for determining corresponding P wave is previous T wave final position between current Q wave initial position, is searched later in each P wave
Within the scope of rope, all possible P wave is searched to the left from current Q wave initial position;
S11, to each doubtful P wave, calculate various wave characters, then K is utilized according to its interval PR to all doubtful P waves
Means Data Cluster Algorithm is classified into K class, then from K class P wave, chooses each cluster that more maximum probability is true P wave and is added
Whether all P waves that each cluster in cluster set S is included are later that " semicircle " shape is divided into four according to it by cluster set S
Class judges whether it is evident as false P wave then to all P waves in class X1 respectively, if it is determined that vacation P wave is from X1
It removes;
S12, to all P waves in class X2, judge whether it is evident as false P wave respectively, if it is determined that vacation P wave is from X2
Middle removal, then to P wave remaining in class X1, find out corresponding QRS complex, if some QRS wave correspond to it is more than one
P wave, a P wave for only retaining amplitude maximum find out corresponding QRS complex then to P wave remaining in class X2, if
Some QRS wave corresponds to more than one P wave, only retains a P wave of amplitude maximum, assumes to wrap in class X1 and class X2 later
Containing at least one P wave, then doing a selection from class X1 and class X2, each P wave that the cluster of selection is included is as final
All P waves extracted, so far, the extraction of the starting of P wave, wave crest or trough, final position are completed.
Preferably, the method in the step S2 in the left and right ends of electrocardiosignal addition data is to assume that signal is s,
The first sample value of definition signal is s (1), then the virtual sample that 128 values of addition are s (1) is repeated on the left of first sample,
The last one sample value of definition signal is s (end), then it is s (end) that 128 values of addition are repeated on the right side of the last one sample
Virtual sample.
Preferably, detailed process is as follows for the search initial position QRS in the step S4:
If A1, current time voltage value are more than 1mV, it is believed that can not be currently QRS wave initial position, continue to the left
It searches;
If A2, current time voltage value are lower than 1mV, but voltage in 20 milliseconds of turning left persistently rises or falls, then also after
It is continuous to search to the left;
If A3, current time voltage value are lower than 1mV, and voltage in 20 milliseconds of turning left does not rise or fall persistently, and
Continuous 20 milliseconds of electrocardio energy value are lower than 0.001, then the moment is the initial position of QRS wave, stop searching;
If A4, current time voltage value are lower than 1mV, and voltage in 20 milliseconds of turning left does not rise or fall persistently, and
20 milliseconds of heart piezoelectric voltage overall variations are less than 0.1mV to the left, and maximum instantaneous voltage change is less than 0.05mV, then the moment be
The initial position of QRS wave stops searching.
Preferably, detailed process is as follows for search QRS final position in the step S4:
If B1, voltage persistently rises or falls within current time to the right 20 milliseconds, continue to search to the right;
If B2, voltage does not rise or fall persistently within current time to the right 20 milliseconds, and electrocardio energy value is continuous
20 milliseconds are lower than 0.0001, then the moment is the final position of QRS wave, stop searching;
If B3, voltage does not rise or fall persistently within current time to the right 20 milliseconds, and 20 milliseconds of hearts to the right
Piezoelectric voltage overall variation is less than 0.1mV, and maximum instantaneous voltage change is less than 0.05mV, then the moment is the stop bit of QRS wave
It sets, stops searching.
Preferably, it is the energy value calculating side using step S3 that the energy value at QRS wave wave crest is calculated in the step S5
Method is calculated, and the width of QRS wave is between wave initial position and final position detected by back in step S5
Interval.
Preferably, all possible P wave, specific steps are searched to the left from current Q wave initial position in the step S10
It is as follows:
If C1, occurring in Q wave initial position and not yet energy signal minimum between current time, continuation is looked into the left
It looks for, without in next step;
If C2, current time be energy signal maximum point, i.e., current time energy value be greater than adjacent two o'clock energy
At current time, is denoted as the wave crest of a possible P wave, searches adjacent two energy to the left and to the right respectively from the wave crest by value
Signal minimum point is measured, is denoted as the starting point and ending point of the doubtful P wave respectively;
C3, check whether the amplitude of the doubtful P wave found in previous step is excessive or too small, if excessive or is too small
Give up, otherwise is added into the set of all doubtful P waves.
Preferably, to each doubtful P wave in the step S11, calculating various wave characters includes: P wave initial position and Q
Time interval, P wave initial position between wave initial position and the time interval between final position, P wave maximum value and minimum
At the difference of value, P wave wave crest voltage subtract voltage at P wave original position voltage and P wave wave crest subtract it is electric at P wave final position
Pressure.
Preferably, it in the step S11 from K class P wave, chooses each cluster that more maximum probability is true P wave and is added and gather
Class set S, the specific steps are as follows:
D1, the relative standard deviation for calculating each interval PR for clustering included P wave, i.e. PR separation standard are poor between average PR
Every ratio;
D2, the opposite the smallest cluster of PR separation standard difference is chosen, set S is added;
D3, the interval PR relative standard deviation is chosen less than 20%, and the average non-significant excessive cluster in the interval PR, collection is added
Close S;
D4, the smallest the cluster in the average interval PR is chosen, set S is added;
D5, a cluster most comprising P wave number mesh is chosen, set S is added.
Preferably, in the step S11 by all P waves that each cluster in cluster set S is included according to its whether be
" semicircle " shape is divided into four classes, the specific steps are as follows:
If poor and rear half of P wave voltage difference of preceding half of P wave voltage of E1, a P wave is positive, which is divided into class
X1;
If poor and rear half of P wave voltage difference of preceding half of P wave voltage of E2, a P wave is negative, which is divided into class
X2;
If preceding half of P wave voltage difference of E3, a P wave are positive, rear half of P wave voltage difference is negative, which is divided into class
Y1;
If preceding half of P wave voltage difference of E4, a P wave are negative, just, which is divided into class to rear half of P wave voltage difference
Y2。
Preferably, assume in class X1 and class X2 to include at least one P wave in the step S12, then from class X1 and class
A selection is done in X2, the specific steps are as follows:
G1, default choice class X1;
G2, such as fruit X1 include that P wave number mesh is few, and more all QRS wave number mesh of P wave number mesh that class X2 includes are suitable, then
Select class X2;
The PR gap size for each P wave that G3, such as fruit X2 include is more consistent, and the average interval PR of all P waves is suitable
In, and one in following two condition is satisfied, then the P wave number mesh for selecting class X2, class X2 to include significantly mostly with class X1 packet
The P wave number mesh contained, and in class X2 each P wave PR gap size it is very consistent, the PR separation standard for each P wave that class X2 includes is poor
PR separation standard than the class X1 each P wave for including is poor small, and P wave number mesh is more than or close to P wave number mesh in class X1 in class X2,
And P wave morphology is more symmetrical in class X2.
(3) beneficial effect
The present invention provides the electrocardiosignal P wave extracting methods based on wavelet transformation and K means Data Cluster Algorithm.With it is existing
Technology compared to have it is following the utility model has the advantages that
(1), it is somebody's turn to do the electrocardiosignal P wave extracting method based on wavelet transformation and K means Data Cluster Algorithm, is searched by limiting P wave
Rope range is previous T wave final position between Q wave initial position, is greatly expanded compared with the prior art, and covering P wave may
All positions occurred are included in P wave position that may be present when atrioventricular block occurs, realize all by finding
Possible P wave, they are clustered, then therefrom find out a most likely true set, by thin from selected layer wavelet transformation
Starting, wave crest and the final position that all possibility P waves are extracted in section coefficient make all P waves according to the interval PR of each P wave
Classified with K means Data Cluster Algorithm, from each P wave set after cluster, excludes the P wave collection of apparent error, such as when a certain P wave
Collect PR separation standard difference much larger than other set in PR separation standard it is poor, or when the interval a certain P wave collection PR it is excessive or too small and
The interval PR in other set is in zone of reasonableness, and the rally of these P waves is excluded, to all P waves in remaining P wave set
Further Cluster Classification is done according to whether its waveform is similar to " semicircle ", finally determines a kind of most likely true P wave, very well
Realize and more steady reliable under various arrhythmia conditions and under noise jamming can extract P wave, while can be more
Accurately judge whether P wave rises as it can be seen that having reached and having limited each P wave search range as previous T wave final position to current Q wave
Between beginning position, the purpose that range is greatly expanded compared with the prior art can cover all possible P wave, be included in chamber biography
When leading retardance generation, by the method searched for generally based on clustering algorithm, without determining P wave by artificially defined rule
Position, therefore the more excellent robustness of algorithm performance is more preferably.
(2), it is somebody's turn to do the electrocardiosignal P wave extracting method based on wavelet transformation and K means Data Cluster Algorithm, by based on cluster
The method of algorithm searched for generally does not determine P wave position by artificially defined rule, can be more when noise jamming is larger
Steadily and surely.
(3), it is somebody's turn to do the electrocardiosignal P wave extracting method based on wavelet transformation and K means Data Cluster Algorithm, solves and judges P wave
Sightless problem occurs etc. in the sightless electrocardiogram of P wave, the P wave number mesh meeting that final remaining P wave collection contains in atrial fibrillation
Much smaller than the number of QRS complex, while the rule of the obvious non-genuine P wave of one group of rejecting is defined, such as width, amplitude should be located
In in particular range, as to the supplement for selecting P wave by clustering procedure, these rules are sufficiently loose.Using these rules, lead to
Crossing a small amount of P wave that clustering procedure mistake is found usually can further be removed.
Detailed description of the invention
Fig. 1 is that the present invention pre-processes original electro-cardiologic signals, wavelet transformation and extracts each doubtful QRS wave wave
The flow chart of shape feature;
Fig. 2 is that the present invention utilizes marked ecg database training for judging each doubtful QRS wave true and false logistic regression mould
The flow chart of type;
Fig. 3 is that the present invention utilizes trained model to acquire the flow chart that each doubtful QRS wave is true QRS wave probability;
Fig. 4 is the flow chart that T wave of the present invention extracts;
Fig. 5 is the flow chart that P wave of the present invention extracts.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts it is all its
His embodiment, shall fall within the protection scope of the present invention.
Fig. 1-5 is please referred to, the embodiment of the present invention provides a kind of technical solution: based on wavelet transformation and K means Data Cluster Algorithm
Electrocardiosignal P wave extracting method, specifically includes the following steps:
S1, the 50Hz for eliminating high frequency respectively first with Zero-phase Digital Filter method are interfered and the baseline drift of low frequency
Then interference will be resampled to a certain fixed sample rate except the electrocardiosignal after making an uproar, be realized using Zero-phase Digital Filter method
One ring frequency is 1Hz low-pass filter below, reuses the low-pass filter filtering original electro-cardiologic signals of acquisition, obtains
Baseline drift signal is obtained, then original signal is subtracted to the baseline drift signal of acquisition, the electrocardio after obtaining place to go baseline drift
Signal is realized that a ring frequency is 50Hz low-pass filter below using Zero-phase Digital Filter method later, is finally made
It is the electrocardiosignal that the filtering of 50Hz or less low-pass filter obtains with the ring frequency of acquisition, obtains the letter of removal Hz noise
Number;
Assume initially that former sample rate is SR, new sample rate is RSR, and re-defining the two greatest common divisor is i, defines m=
SR/i, n=RSR/i, mn=m*n, it is assumed that original signal s0 length is SR*t, then new signal s1 length is RSR*t, defined variable
Ot, when the every assignment of s1 is primary, ot increases m, defined variable it, and when s0 is often moved to right once, it increases n, and when it value is greater than mn
Zero, in this way, defining vv=s0 (1), v=s0 (2), then s1 (1)=vv, is defined vv=s0 (3), v=s0 (4), then (2) s1
=vv+ (ot%n) * (v-vv)/n;It defines vv=s0 (4), v=s0 (5), then s1 (3)=vv+ (ot%n) * (v-vv)/n, with
This analogizes, and can get the value at new signal each moment after resampling;
S2, the electrocardiosignal after S1 resampling is normalized, makes its Distribution value between -0.5 and 0.5, is believed according to electrocardio
The spectrum signature of number QRS wave determines that interested frequency range is 11Hz to 40Hz, respectively adds in the left and right ends of electrocardiosignal
Then one piece of data does continuous wavelet change to electrocardiosignal after normalization to eliminate boundary effect in range of target frequencies
Change, calculate after Continuous Wavelet Transform Coefficients, both ends virtual sample is removed;
S3, the wavelet conversion coefficient obtained to S2, seek square of wavelet conversion coefficient on different frequency at various moments
With, to obtain each moment electrocardiosignal in the energy value of frequency domain, the small window of movement that a width is 0.3 second is then defined,
It finds electrocardio energy maximum in each small window, is a doubtful QRS wave peak at the time of corresponding to each electrocardio energy maximum,
Sample moves right the small window of the movement one by one, and thus we can obtain the doubtful QRS wave peak of a column;
S4, during S3 obtains a column doubtful QRS wave peak, duplicate doubtful QRS wave peak is removed, from doubtful QRS
Wave crest starts, and searches the initial position of the QRS wave to the left, does not find yet if be moved to the left always more than maximum search distance
Qualified initial position, then the wave is not true QRS wave, should be abandoned, while since doubtful QRS wave peak, being searched to the right
The final position of the QRS wave does not find qualified stop bit more than maximum search distance if moved right always yet
It sets, then the wave is not true QRS wave, should be abandoned;
S5, to each doubtful QRS wave, calculate separately the energy value at QRS wave wave crest, the absolute value of voltage at QRS wave peak,
The interval, the width of the QRS wave and the average voltage rate of change of the doubtful QRS wave of the QRS wave and previous doubtful QRS wave,
Average voltage rate of change is defined as follows, it is assumed that signal s, length N, then average voltage rate of change are as follows: average value
{ absolute value { s [n]-s [n-1] } }, wherein n is all integers from 1 to N;
S6, to the ecg database labeled through electrocardio expert, handled by above step S1-S5, obtain a heart
The eigenmatrix of electric QRS wave shape, the line number of the matrix are equal to the number of all doubtful QRS waves, and columns is equal to 5, i.e. previous step
The number of the QRS wave shape feature of middle acquisition is compared with the QRS wave of expert's label to each doubtful QRS wave, whether judges it
For true QRS wave shape, judgment criteria is, when the QRS wave peak separation of doubtful QRS wave wave crest and expert's label 150 milliseconds with
When interior, it is believed that doubtful QRS wave be it is true, then by previous step obtain eigenmatrix input a Logic Regression Models in,
Using gradient descent method as optimization method, setting learning rate is 0.02, and iteration is until convergence, in training, using staying 1 cross validation
Method, i.e., training data is randomly divided into K parts, takes and wherein do training for K-1 parts, remaining 1 part is verified, so repeatedly n times,
It is all averaged after the completion of training to the N number of model of gained, final mask is obtained, finally, using F1 score as selecting index probability threshold
Value, below threshold value, it is believed that QRS wave be it is false, more than threshold value, it is believed that QRS wave is true;
S7, to any new electrocardiogram (ECG) data to be analyzed, handled, obtained using method of the step S1 into step S5
One group of doubtful QRS wave is obtained, to each doubtful QRS wave, is classified using the model that training obtains in preceding step, obtaining it is
The probability of true QRS wave, the probability threshold value determined in recycle step S6 finally determine whether each doubtful QRS wave is true
Real QRS wave is so far completed the extraction of the QRS feature of any electrocardiogram (ECG) data, to the classification results of each QRS wave detected
Do the quality information of electrocardiogram (ECG) data belonging to averagely can get;
S8, electrocardiosignal after the resampling obtained in step S1 is reprocessed, to protrude T wave part, to each
The specific process flow of QRS complex is as follows: in QRS starting point, 200 milliseconds of determinations a bit, are designated as x0, corresponding voltage is forward
Y0, in QRS terminating point, 100 milliseconds of determinations a bit, are designated as x1, corresponding voltage y1 backward, be origin coordinates with (x0, y0),
(x1, y1) is end coordinate, is fitted straight line f (x), electrocardiosignal is located at the voltage value between moment x0 and x1 and is used
Value on straight line f (x) replaces, and handles each QRS complex, then does in selected frequency range to obtained signal small
Wave conversion, the frequency range include the main information of T wave, then ask different at various moments to the wavelet conversion coefficient of acquisition
The quadratic sum of wavelet conversion coefficient in frequency, to obtain each time-ofday signals in the energy value of frequency domain, later to the energy of acquisition
Signal carries out smooth operation, and specific method is that all energy values are averaged in 20 milliseconds asked each moment centered on it
Value replaces the energy value at this moment with the average value, and marking smoothed out energy signal is TES, then to each QRS complex
Determine the search range of corresponding T wave, the specific steps are as follows: the search initial position of T wave is the final position of corresponding QRS wave;Root
According to the time interval (being labeled as RR) of current QRS wave and subsequent QRS wave, it is dynamically determined the search range of T wave, decision logic is such as
Under: if current QRS wave instantaneous heart rate corresponding with subsequent QRS wave time interval is lower than 100bpm, the search length of T wave is
RR*2/3;If current QRS wave instantaneous heart rate corresponding with subsequent QRS wave time interval is higher than 100bpm but is lower than 150bpm,
The search length of T wave is RR*3/4;If current QRS wave instantaneous heart rate corresponding with subsequent QRS wave time interval is higher than
The search length of 150bpm, T wave is RR*4/5;
S9, to each QRS complex, find maximum value of the TES in corresponding T wave search range, this is T wave wave crest or wave
Paddy searches the initial position of the T wave since T wave wave crest or trough to the left, and detailed process is as follows: if current time
TES value is lower than the TES value at two neighboring moment, it is meant that and the moment is TES minimum point, and forward in 40 milliseconds,
TES value is consistently lower than 0.0001, then current time is the initial position of T wave;If out of current time forward 80 milliseconds, TES
Value is consistently lower than 0.0001, then current time is the initial position of T wave, then since T wave wave crest or trough, and searching to the right should
The final position of T wave, detailed process are as follows: if the TES value at current time is lower than the TES value at two neighboring moment, meaning
The moment be TES minimum point, and backward in 40 milliseconds, TES value is consistently lower than 0.0001, then current time is T wave
Final position;If TES value is consistently lower than 0.0001 out of current time backward 80 milliseconds, then current time is the end of T wave
Stop bit is set, and is further optimized later to the starting of each T wave, wave crest or the trough and final position that find, is found out current T wave
Maximum voltage and the moment, if maximum voltage be greater than T wave starting and final position voltage, the T wave be positive polarity, most
Big voltage value is wave crest, conversely, find out minimum voltage and the moment of current T wave, if minimum voltage be less than the starting of T wave and
The voltage of final position, then the T wave is negative polarity, and the minimum voltage moment is trough;If T wave polarity is positive, from T wave
Wave crest is moved to the left, if current time voltage value is less than each average voltage in 20 milliseconds of turning left, current time is optimization
T wave initial position afterwards is moved to the left if T wave polarity is negative from T wave trough, if current time voltage value is big
In each average voltage in 20 milliseconds of turning left, then current time is the T wave initial position after optimization;If T wave polarity is positive,
It so moves right from T wave wave crest, if current time voltage value is less than each average voltage in 20 milliseconds of turning right, currently
Moment is that the T wave final position after optimization moves right, if current time if T wave polarity is negative from T wave trough
Voltage value is greater than each average voltage in 20 milliseconds of turning right, then current time be the T wave final position after optimizing, so far, T wave
The extraction of starting, wave crest or trough, final position is completed;
S10, wavelet transformation is done in selected frequency range to electrocardiosignal after the resampling obtained in step S1, the frequency
Rate range includes the main information of P wave, then seeks wavelet transformation on different frequency at various moments to the wavelet conversion coefficient of acquisition
The quadratic sum of coefficient is denoted as energy signal PE, then to each QRS wave to obtain each time-ofday signals in the energy value of frequency domain
The search range for determining corresponding P wave is previous T wave final position between current Q wave initial position, is searched later in each P wave
Within the scope of rope, all possible P wave is searched to the left from current Q wave initial position;
S11, to each doubtful P wave, calculate various wave characters, then K is utilized according to its interval PR to all doubtful P waves
Means Data Cluster Algorithm is classified into K class, then from K class P wave, chooses each cluster that more maximum probability is true P wave and is added
Whether all P waves that each cluster in cluster set S is included are later that " semicircle " shape is divided into four according to it by cluster set S
Class judges whether it is evident as false P wave then to all P waves in class X1 respectively, if it is determined that vacation P wave is from X1
Remove, judgment basis is as follows: the amplitude of P wave is excessive or too small;P wave width is excessive;The equal voltage changing rate of P popin is excessive,
It is defined as follows, it is assumed that P wave is p, width N, then average voltage rate of change are as follows: average value { absolute value { p [n]-p [n-
1] } }, wherein n is all integers from 1 to N;
S12, to all P waves in class X2, judge whether it is evident as false P wave respectively, if it is determined that vacation P wave is from X2
Middle removal, then to P wave remaining in class X1, find out corresponding QRS complex, if some QRS wave correspond to it is more than one
P wave, a P wave for only retaining amplitude maximum find out corresponding QRS complex then to P wave remaining in class X2, if
Some QRS wave corresponds to more than one P wave, only retains a P wave of amplitude maximum, assumes to wrap in class X1 and class X2 later
Containing at least one P wave, then doing a selection from class X1 and class X2, each P wave that the cluster of selection is included is as final
All P waves extracted, so far, the extraction of the starting of P wave, wave crest or trough, final position are completed.
In the present invention, the method in step S2 in the left and right ends of electrocardiosignal addition data is to assume that signal is s, fixed
The adopted first sample value of signal is s (1), then the virtual sample that 128 values of addition are s (1) is repeated on the left of first sample, fixed
The last one sample value of adopted signal is s (end), then the void that 128 values of addition are s (end) is repeated on the right side of the last one sample
Quasi- sample.
Searching the initial position QRS in the present invention, in step S4, detailed process is as follows:
If A1, current time voltage value are more than 1mV, it is believed that can not be currently QRS wave initial position, continue to the left
It searches;
If A2, current time voltage value are lower than 1mV, but voltage in 20 milliseconds of turning left persistently rises or falls, then also after
It is continuous to search to the left;
If A3, current time voltage value are lower than 1mV, and voltage in 20 milliseconds of turning left does not rise or fall persistently, and
Continuous 20 milliseconds of electrocardio energy value are lower than 0.001, then the moment is the initial position of QRS wave, stop searching;
If A4, current time voltage value are lower than 1mV, and voltage in 20 milliseconds of turning left does not rise or fall persistently, and
20 milliseconds of heart piezoelectric voltage overall variations are less than 0.1mV to the left, and maximum instantaneous voltage change is less than 0.05mV, then the moment be
The initial position of QRS wave stops searching.
Searching QRS final position in the present invention, in step S4, detailed process is as follows:
If B1, voltage persistently rises or falls within current time to the right 20 milliseconds, continue to search to the right;
If B2, voltage does not rise or fall persistently within current time to the right 20 milliseconds, and electrocardio energy value is continuous
20 milliseconds are lower than 0.0001, then the moment is the final position of QRS wave, stop searching;
If B3, voltage does not rise or fall persistently within current time to the right 20 milliseconds, and 20 milliseconds of hearts to the right
Piezoelectric voltage overall variation is less than 0.1mV, and maximum instantaneous voltage change is less than 0.05mV, then the moment is the stop bit of QRS wave
It sets, stops searching.
It is the energy value calculating method using step S3 that the energy value at QRS wave wave crest is calculated in the present invention, in step S5
It is calculated, and the width of QRS wave is between wave initial position and final position detected by back in step S5
Interval.
In the present invention, all possible P wave is searched to the left from current Q wave initial position in step S10, specific steps are such as
Under:
If C1, occurring in Q wave initial position and not yet energy signal minimum between current time, continuation is looked into the left
It looks for, without in next step;
If C2, current time be energy signal maximum point, i.e., current time energy value be greater than adjacent two o'clock energy
At current time, is denoted as the wave crest of a possible P wave, searches adjacent two energy to the left and to the right respectively from the wave crest by value
Signal minimum point is measured, is denoted as the starting point and ending point of the doubtful P wave respectively;
C3, check whether the amplitude of the doubtful P wave found in previous step is excessive or too small, if excessive or is too small
Give up, otherwise is added into the set of all doubtful P waves.
In the present invention, to each doubtful P wave in step S11, calculating various wave characters includes: P wave initial position and Q wave
Time interval, P wave initial position between initial position and the time interval between final position, P wave maxima and minima
Difference, voltage subtracts voltage at P wave original position voltage and P wave wave crest and subtracts voltage at P wave final position at P wave wave crest.
In the present invention, in step S11 from K class P wave, chooses each cluster that more maximum probability is true P wave and cluster is added
Collect S, the specific steps are as follows:
D1, the relative standard deviation for calculating each interval PR for clustering included P wave, i.e. PR separation standard are poor between average PR
Every ratio;
D2, the opposite the smallest cluster of PR separation standard difference is chosen, set S is added;
D3, the interval PR relative standard deviation is chosen less than 20%, and the average non-significant excessive cluster in the interval PR, collection is added
Close S;
D4, the smallest the cluster in the average interval PR is chosen, set S is added;
D5, a cluster most comprising P wave number mesh is chosen, set S is added.
In the present invention, in step S11 by all P waves that each cluster in cluster set S is included according to its whether be
" semicircle " shape is divided into four classes, the specific steps are as follows:
If poor and rear half of P wave voltage difference of preceding half of P wave voltage of E1, a P wave is positive, which is divided into class
X1;
If poor and rear half of P wave voltage difference of preceding half of P wave voltage of E2, a P wave is negative, which is divided into class
X2;
If preceding half of P wave voltage difference of E3, a P wave are positive, rear half of P wave voltage difference is negative, which is divided into class
Y1;
If preceding half of P wave voltage difference of E4, a P wave are negative, just, which is divided into class to rear half of P wave voltage difference
Y2。
Assume in class X1 and class X2 to include at least one P wave in the present invention, in step S12, then from class X1 and class X2
In do a selection, the specific steps are as follows:
G1, default choice class X1;
G2, such as fruit X1 include that P wave number mesh is few, and more all QRS wave number mesh of P wave number mesh that class X2 includes are suitable, then
Select class X2;
The PR gap size for each P wave that G3, such as fruit X2 include is more consistent, and the average interval PR of all P waves is suitable
In, and one in following two condition is satisfied, then the P wave number mesh for selecting class X2, class X2 to include significantly mostly with class X1 packet
The P wave number mesh contained, and in class X2 each P wave PR gap size it is very consistent, the PR separation standard for each P wave that class X2 includes is poor
PR separation standard than the class X1 each P wave for including is poor small, and P wave number mesh is more than or close to P wave number mesh in class X1 in class X2,
And class X2
Middle P wave morphology is more symmetrical.
It should be noted that, in this document, relational terms such as first and second and the like are used merely to a reality
Body or operation are distinguished with another entity or operation, without necessarily requiring or implying between these entities or operation
There are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant are intended to
Cover non-exclusive inclusion, so that the process, method, article or equipment for including a series of elements not only includes that
A little elements, but also including other elements that are not explicitly listed, or further include for this process, method, article or
The intrinsic element of equipment.
It although an embodiment of the present invention has been shown and described, for the ordinary skill in the art, can be with
A variety of variations, modification, replacement can be carried out to these embodiments without departing from the principles and spirit of the present invention by understanding
And modification, the scope of the present invention is defined by the appended.
Claims (10)
1. the electrocardiosignal P wave extracting method based on wavelet transformation and K means Data Cluster Algorithm, it is characterised in that: specifically include with
Lower step:
S1, the 50Hz interference and the baseline drift of low frequency interference for eliminating high frequency respectively first with Zero-phase Digital Filter method,
Then a certain fixed sample rate will be resampled to except the electrocardiosignal after making an uproar:
S2, the electrocardiosignal after S1 resampling is normalized, makes its Distribution value between -0.5 and 0.5, according to electrocardiosignal QRS
The spectrum signature of wave determines that interested frequency range is 11Hz to 40Hz, respectively adds a number of segment in the left and right ends of electrocardiosignal
According to then doing continuous wavelet transform in range of target frequencies to electrocardiosignal after normalization, calculate to eliminate boundary effect
After Continuous Wavelet Transform Coefficients, both ends virtual sample is removed;
S3, the wavelet conversion coefficient obtained to S2, seek the quadratic sum of wavelet conversion coefficient on different frequency at various moments, thus
Each moment electrocardiosignal is obtained in the energy value of frequency domain, the small window of movement that a width is 0.3 second is then defined, in each small window
Electrocardio energy maximum is found, is a doubtful QRS wave peak, the movement small window at the time of corresponding to each electrocardio energy maximum
Sample moves right one by one, and thus we can obtain the doubtful QRS wave peak of a column;
S4, during S3 obtains a column doubtful QRS wave peak, duplicate doubtful QRS wave peak is removed, from doubtful QRS wave peak
Start, search the initial position of the QRS wave to the left, meets item if being moved to the left do not find yet more than maximum search distance always
The initial position of part, then the wave is not true QRS wave, should be abandoned, while since doubtful QRS wave peak, searching the QRS wave to the right
Final position, if move right always more than maximum search distance do not find qualified final position yet, the wave
It is not true QRS wave, should abandons;
S5, to each doubtful QRS wave, calculate separately the energy value at QRS wave wave crest, the absolute value of voltage at QRS wave peak, the QRS
Interval, the width of the QRS wave and the average voltage rate of change of the doubtful QRS wave of wave and previous doubtful QRS wave;
S6, to the ecg database labeled through electrocardio expert, handled by above step S1-S5, obtain an electrocardio QRS
The eigenmatrix of waveform, the line number of the matrix are equal to the number of all doubtful QRS waves, to each doubtful QRS wave, with expert's label
QRS wave is compared, and judges whether it is true QRS wave shape, and the eigenmatrix that previous step obtains then is inputted one and is patrolled
In volume regression model, using gradient descent method as optimization method, setting learning rate is 0.02, and iteration in training, uses until convergence
Training data is randomly divided into K parts by the method for staying 1 cross validation, take and wherein do training for K-1 parts, and remaining 1 part is verified, such as
This repeats n times, is all averaged after the completion of training to the N number of model of gained, final mask is obtained, finally, using F1 score as index
Probability threshold value is chosen, below threshold value, it is believed that QRS wave be it is false, more than threshold value, it is believed that QRS wave is true;
S7, to any new electrocardiogram (ECG) data to be analyzed, handled using method of the step S1 into step S5, obtain one group
Doubtful QRS wave classifies to each doubtful QRS wave using the model that training obtains in preceding step, obtains it as true QRS
The probability of wave, the probability threshold value determined in recycle step S6 finally determine whether each doubtful QRS wave is true QRS wave,
So far, the extraction of the QRS feature of any electrocardiogram (ECG) data is completed, the classification results of each QRS wave detected is done and averagely may be used
The quality information of electrocardiogram (ECG) data belonging to obtaining;
S8, electrocardiosignal after the resampling obtained in step S1 is reprocessed, to protrude T wave part, to each QRS complex
It is handled, then wavelet transformation is done in selected frequency range to obtained signal, which includes the main letter of T wave
Breath, then seeks at various moments the wavelet conversion coefficient of acquisition the quadratic sum of wavelet conversion coefficient on different frequency, to obtain
Each time-ofday signals are obtained in the energy value of frequency domain, smooth operation are carried out to the energy signal of acquisition later, specific method is to each
Moment seeks the average value of all energy values in 20 milliseconds centered on it, and the energy value at this moment, mark are replaced with the average value
Remember that smoothed out energy signal is TES, the search range of corresponding T wave is then determined to each QRS complex;
S9, to each QRS complex, find maximum value of the TES in corresponding T wave search range, this is T wave wave crest or trough, from T
Wave wave crest or trough start, and search the initial position of the T wave to the left, then since T wave wave crest or trough, search the T wave to the right
Final position, further the starting of each T wave, wave crest or trough and the final position that find are optimized later, so far, T wave
The extraction of starting, wave crest or trough, final position is completed;
S10, wavelet transformation is done in selected frequency range to electrocardiosignal after the resampling obtained in step S1, the frequency range
Main information comprising P wave, then wavelet conversion coefficient on different frequency is asked at various moments to the wavelet conversion coefficient of acquisition
Quadratic sum is denoted as energy signal PE to obtain each time-ofday signals in the energy value of frequency domain, then to determining pair of each QRS wave
The search range for answering P wave is previous T wave final position between current Q wave initial position, later in each P wave search range
It is interior, all possible P wave is searched to the left from current Q wave initial position;
S11, to each doubtful P wave, calculate various wave characters, then to all doubtful P waves, it is averagely poly- using K according to its interval PR
Class algorithm is classified into K class, then from K class P wave, chooses each cluster that more maximum probability is true P wave and cluster set is added
Whether all P waves that each cluster in cluster set S is included are later that " semicircle " shape is divided into four classes according to it by S, then right
All P waves in class X1, judge whether it is evident as false P wave respectively, if it is determined that vacation P wave is removed from X1;
S12, to all P waves in class X2, judge whether it is evident as false P wave respectively, if it is determined that vacation P wave is from X2
It removes, then to P wave remaining in class X1, finds out corresponding QRS complex, if some QRS wave corresponds to more than one P wave,
A P wave for only retaining amplitude maximum finds out corresponding QRS complex, if a certain then to P wave remaining in class X2
A QRS wave corresponds to more than one P wave, only retains a P wave of amplitude maximum, is assumed in class X1 and class X2 later comprising at least
One P wave, then doing a selection from class X1 and class X2, each P wave that the cluster of selection is included as finally is extracted
All P waves, so far, the extraction of the starting of P wave, wave crest or trough, final position are completed.
2. the electrocardiosignal P wave extracting method according to claim 1 based on wavelet transformation and K means Data Cluster Algorithm,
Be characterized in that: in the step S2 electrocardiosignal left and right ends addition data method be assume signal be s, definition signal
First sample value is s (1), then the virtual sample that 128 values of addition are s (1) is repeated on the left of first sample, definition signal is most
The latter sample value is s (end), then the virtual sample that 128 values of addition are s (end) is repeated on the right side of the last one sample.
3. the electrocardiosignal P wave extracting method according to claim 1 based on wavelet transformation and K means Data Cluster Algorithm,
Be characterized in that: the initial position QRS is searched in the step S4, and detailed process is as follows:
If A1, current time voltage value are more than 1mV, it is believed that currently can not be QRS wave initial position, continuation is searched to the left;
If A2, current time voltage value are lower than 1mV, but voltage in 20 milliseconds of turning left persistently rises or falls, then also continue to
Left search;
If A3, current time voltage value are lower than 1mV, and voltage in 20 milliseconds of turning left does not rise or fall persistently, and electrocardio
Continuous 20 milliseconds of energy value are lower than 0.001, then the moment is the initial position of QRS wave, stop searching;
If A4, current time voltage value are lower than 1mV, and voltage in 20 milliseconds of turning left does not rise or fall persistently, and to the left
20 milliseconds of heart piezoelectric voltage overall variations are less than 0.1mV, and maximum instantaneous voltage change is less than 0.05mV, then the moment is QRS wave
Initial position, stop search.
4. the electrocardiosignal P wave extracting method according to claim 1 based on wavelet transformation and K means Data Cluster Algorithm,
Be characterized in that: QRS final position is searched in the step S4, and detailed process is as follows:
If B1, voltage persistently rises or falls within current time to the right 20 milliseconds, continue to search to the right;
If B2, voltage does not rise or fall persistently within current time to the right 20 milliseconds, and continuous 20 milli of electrocardio energy value
Second is lower than 0.0001, then the moment is the final position of QRS wave, stops searching;
If B3, voltage does not rise or fall persistently within current time to the right 20 milliseconds, and electrocardio electricity in 20 milliseconds to the right
Overall variation is pressed to be less than 0.1mV, maximum instantaneous voltage change is less than 0.05mV, then the moment is the final position of QRS wave, stops
It searches.
5. the electrocardiosignal P wave extracting method according to claim 1 based on wavelet transformation and K means Data Cluster Algorithm,
Be characterized in that: the energy value at QRS wave wave crest is calculated in the step S5 to be carried out using the energy value calculating method of step S3
It calculates, and the width of QRS wave is the interval between wave initial position and final position detected by back in step S5.
6. the electrocardiosignal P wave extracting method according to claim 1 based on wavelet transformation and K means Data Cluster Algorithm,
It is characterized in that: searching all possible P wave to the left from current Q wave initial position in the step S10, the specific steps are as follows:
If C1, occurring in Q wave initial position and not yet energy signal minimum between current time, continuation is searched to the left, no
It carries out in next step;
If C2, current time are energy signal maximum point, i.e., current time energy value, will greater than the energy value of adjacent two o'clock
Current time is denoted as the wave crest of a possible P wave, searches adjacent two energy signals to the left and to the right respectively from the wave crest
Minimum point is denoted as the starting point and ending point of the doubtful P wave respectively;
C3, check whether the amplitude of the doubtful P wave found in previous step is excessive or too small, gives up if excessive or is too small,
Otherwise it is added into the set of all doubtful P waves.
7. the electrocardiosignal P wave extracting method according to claim 1 based on wavelet transformation and K means Data Cluster Algorithm,
Be characterized in that: to each doubtful P wave in the step S11, calculating various wave characters includes: P wave initial position and Q wave start bit
Time interval, P wave initial position between setting and the time interval between final position, the difference of P wave maxima and minima, P
Voltage subtracts voltage at P wave original position voltage and P wave wave crest and subtracts voltage at P wave final position at wave wave crest.
8. the electrocardiosignal P wave extracting method according to claim 1 based on wavelet transformation and K means Data Cluster Algorithm,
It is characterized in that: in the step S11 from K class P wave, choosing each cluster that more maximum probability is true P wave and cluster set S is added,
Specific step is as follows:
D1, the relative standard deviation for calculating each interval PR for clustering included P wave, the i.e. ratio of PR separation standard difference and the average interval PR
Value;
D2, the opposite the smallest cluster of PR separation standard difference is chosen, set S is added;
D3, the interval PR relative standard deviation is chosen less than 20%, and the average non-significant excessive cluster in the interval PR, set S is added;
D4, the smallest the cluster in the average interval PR is chosen, set S is added;
D5, a cluster most comprising P wave number mesh is chosen, set S is added.
9. the electrocardiosignal P wave extracting method according to claim 1 based on wavelet transformation and K means Data Cluster Algorithm,
Be characterized in that: whether all P waves for being included by each cluster in cluster set S in the step S11 are " semicircle " according to it
Shape is divided into four classes, the specific steps are as follows:
If poor and rear half of P wave voltage difference of preceding half of P wave voltage of E1, a P wave is positive, which is divided into class X1;
If poor and rear half of P wave voltage difference of preceding half of P wave voltage of E2, a P wave is negative, which is divided into class X2;
If preceding half of P wave voltage difference of E3, a P wave are positive, rear half of P wave voltage difference is negative, which is divided into class Y1;
If preceding half of P wave voltage difference of E4, a P wave are negative, just, which is divided into class Y2 to rear half of P wave voltage difference.
10. the electrocardiosignal P wave extracting method according to claim 1 based on wavelet transformation and K means Data Cluster Algorithm,
It is characterized in that: is assumed in the step S12 comprising at least one P wave in class X1 and class X2, then being done from class X1 and class X2
One selection, the specific steps are as follows:
G1, default choice class X1;
G2, such as fruit X1 include that P wave number mesh is few, and more all QRS wave numbers of P wave number mesh that class X2 includes are suitable, then select
Class X2;
The PR gap size for each P wave that G3, such as fruit X2 include is more consistent, and the average interval PR of all P waves is moderate, and
One in following two condition is satisfied, then the P wave that the P wave number mesh for selecting class X2, class X2 to include significantly mostly includes with class X1
Number, and in class X2 each P wave PR gap size it is very consistent, the PR separation standard difference of each P wave that class X2 includes is wrapped than class X1
The PR separation standard difference of each P wave contained is small, and P wave number mesh is more than or close to P wave number mesh in class X1 in class X2, and P in class X2
Wave morphology is more symmetrical.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910186724.2A CN110084096A (en) | 2019-03-12 | 2019-03-12 | Electrocardiosignal P wave extracting method based on wavelet transformation and K means Data Cluster Algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910186724.2A CN110084096A (en) | 2019-03-12 | 2019-03-12 | Electrocardiosignal P wave extracting method based on wavelet transformation and K means Data Cluster Algorithm |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110084096A true CN110084096A (en) | 2019-08-02 |
Family
ID=67413156
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910186724.2A Pending CN110084096A (en) | 2019-03-12 | 2019-03-12 | Electrocardiosignal P wave extracting method based on wavelet transformation and K means Data Cluster Algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110084096A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110403601A (en) * | 2019-08-27 | 2019-11-05 | 安徽心之声医疗科技有限公司 | Electrocardiosignal QRS wave group recognition methods based on deep learning |
CN110750770A (en) * | 2019-08-18 | 2020-02-04 | 浙江好络维医疗技术有限公司 | Method for unlocking electronic equipment based on electrocardiogram |
CN111956213A (en) * | 2020-07-29 | 2020-11-20 | 鲁东大学 | QRS point detection method of electrocardiosignal |
CN112826514A (en) * | 2019-11-22 | 2021-05-25 | 华为技术有限公司 | Atrial fibrillation signal classification method, device, terminal and storage medium |
CN113749665A (en) * | 2021-08-19 | 2021-12-07 | 深圳邦健生物医疗设备股份有限公司 | Method, device, equipment and medium for capturing abnormal indexes |
CN113947112A (en) * | 2021-09-08 | 2022-01-18 | 天津大学 | Preprocessing method of time sequence data set and application thereof |
-
2019
- 2019-03-12 CN CN201910186724.2A patent/CN110084096A/en active Pending
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110750770A (en) * | 2019-08-18 | 2020-02-04 | 浙江好络维医疗技术有限公司 | Method for unlocking electronic equipment based on electrocardiogram |
CN110750770B (en) * | 2019-08-18 | 2023-10-03 | 浙江好络维医疗技术有限公司 | Electrocardiogram-based method for unlocking electronic equipment |
CN110403601A (en) * | 2019-08-27 | 2019-11-05 | 安徽心之声医疗科技有限公司 | Electrocardiosignal QRS wave group recognition methods based on deep learning |
CN112826514A (en) * | 2019-11-22 | 2021-05-25 | 华为技术有限公司 | Atrial fibrillation signal classification method, device, terminal and storage medium |
CN111956213A (en) * | 2020-07-29 | 2020-11-20 | 鲁东大学 | QRS point detection method of electrocardiosignal |
CN113749665A (en) * | 2021-08-19 | 2021-12-07 | 深圳邦健生物医疗设备股份有限公司 | Method, device, equipment and medium for capturing abnormal indexes |
CN113749665B (en) * | 2021-08-19 | 2024-02-02 | 深圳邦健生物医疗设备股份有限公司 | Method, device, equipment and medium for capturing abnormal index |
CN113947112A (en) * | 2021-09-08 | 2022-01-18 | 天津大学 | Preprocessing method of time sequence data set and application thereof |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110084096A (en) | Electrocardiosignal P wave extracting method based on wavelet transformation and K means Data Cluster Algorithm | |
CN107822622B (en) | Electrocardiogram diagnosis method and system based on deep convolutional neural network | |
CN105426842B (en) | Multiclass hand motion recognition method based on support vector machines and surface electromyogram signal | |
CN106108889B (en) | Electrocardiogram classification method based on deep learning algorithm | |
CN109948647A (en) | A kind of electrocardiogram classification method and system based on depth residual error network | |
CN105320969A (en) | A heart rate variability feature classification method based on multi-scale Renyi entropy | |
CN109381181A (en) | The end-point detecting method of electrocardiosignal signature waveform | |
CN111407243B (en) | Pulse signal pressure identification method based on deep learning | |
CN103675610B (en) | Characterization factor extracting method in shelf depreciation on-line checking | |
CN101732033A (en) | Method and device for extracting characteristic parameter in human body waveform | |
CN104523268A (en) | Electroencephalogram signal recognition fuzzy system and method with transfer learning ability | |
CN107688553A (en) | Method based on wavelet transformation and logistic regression algorithm detection ecg wave form feature | |
CN111956208B (en) | ECG signal classification method based on ultra-lightweight convolutional neural network | |
CN110399846A (en) | A kind of gesture identification method based on multichannel electromyography signal correlation | |
CN106874722A (en) | A kind of personal identification method and its device based on electrocardiosignal | |
CN111481192A (en) | Electrocardiosignal R wave detection method based on improved U-Net | |
CN113128552A (en) | Electroencephalogram emotion recognition method based on depth separable causal graph convolution network | |
CN104537243B (en) | A kind of electrocardio sorting technique based on fuzzy reasoning combination weighted similarity measures for non-treatment purpose | |
CN111603158B (en) | Fatigue driving warning method and system based on electrophysiological signal artificial intelligent analysis | |
CN107169424A (en) | Atrial fibrillation detection algorithm based on artificial neural network | |
CN115736944A (en) | Atrial fibrillation detection model MCNN-BLSTM based on short-time single lead electrocardiosignal | |
CN114692698A (en) | One-dimensional electrocardiogram data classification method based on residual error network | |
CN106203324A (en) | The quick personal identification method of electrocardiosignal based on random tree | |
CN112215490A (en) | Power load cluster analysis method based on correlation coefficient improved K-means | |
CN111261283A (en) | Electrocardiosignal deep neural network modeling method based on pyramid type convolution layer |
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 |