WO2017096597A1

WO2017096597A1 - Method and device for processing electrocardio signals

Info

Publication number: WO2017096597A1
Application number: PCT/CN2015/097054
Authority: WO
Inventors: 张永和; 章海峰
Original assignee: 深圳市洛书和科技发展有限公司
Priority date: 2015-12-10
Filing date: 2015-12-10
Publication date: 2017-06-15

Abstract

A method and device for processing electrocardio signals. The method comprises the following steps: obtaining a sample sequence which represents electrocardio signals (S100); denoising the sample sequence (S200); geometrically compressing the denoised sample sequence, and discarding a sample which does not have geometrical characteristics (S300), wherein the sample which does not have geometrical characteristics refers to the sample which is not an extreme value as compared with the currently adjacent samples at left and right and deviates, within a predetermined range, from a straight line determined with respect to the left and right samples. The device comprises a sample sequence obtaining module (610), a denoising module (620), and a geometrical compressing module (630). By means of the method, a denoised sample sequence is compressed by means of geometrical compression, such that the calculation is simple, the processing is efficient, and the number of samples is greatly reduced while the geometrical characteristics of signals are maintained.

Description

ECG signal processing method and device

Technical field

The invention relates to the technical field of analysis and processing of human physiological data, in particular to an ECG signal processing method and a corresponding device.

Background technique

With the development of society, people pay more and more attention to health. The collection of various human physiological data is not limited to medical uses. For example, people may collect and analyze ECG signals for the purpose of examination in hospitals, or may record ECG signals in real time through wearable devices.

In any case, it is worthwhile to study how to process a large amount of ECG signal data quickly and efficiently and minimize the amount of data that needs to be saved.

Summary of the invention

According to an aspect of the present invention, an ECG signal processing method is provided, including the steps of:

Obtaining a sample sequence {(t ₁ , d ₁ ), (t ₂ , d ₂ ), ..., (t _i , d _i ), ...} representing the electrocardiographic signal, where t _i is the sampling time of the ith sample, d _i is the sampled value of the ith sample,

Denoising the sample sequence includes: calculating the extent to which the sample in the sample sequence is affected by noise, and discarding samples that are affected by noise beyond a preset limit;

Perform geometric compression on the denoised sample sequence and discard samples without geometric features. The sample that has no geometric features means that it is neither extreme nor relative to the currently adjacent left and right samples. The deviation of the straight line determined by the two samples on the left and right is within the preset range.

According to another aspect of the present invention, an ECG signal processing apparatus is provided, comprising a functional module for performing the steps of the above method.

According to the ECG signal processing method of the present invention, the sample sequence after denoising is compressed by geometric compression, which is not only simple in calculation, high in processing efficiency, but also can greatly reduce the sample under the premise of retaining the geometric characteristics of the signal. Quantity, thereby reducing the need for storage or transmission capabilities. In addition, since the method of discarding bad samples is employed in the step of denoising, the efficiency of processing can be further improved.

Specific examples in accordance with the present invention will be described in detail below with reference to the accompanying drawings.

DRAWINGS

1 is a schematic flow chart of an ECG signal processing method according to the present invention;

2 is a schematic flow chart of denoising a sample sequence in FIG. 1;

3 is a schematic flow chart of geometric compression of the sample sequence after denoising in FIG. 1;

4 is a flow chart showing the analysis of the geometrically compressed sample sequence in FIG. 1;

Figure 5 is a block diagram of an electrocardiographic signal processing apparatus in accordance with the present invention;

6 is a (a) initial sample sequence and grouping diagram, and (b) a schematic diagram of the stationarity index of each group;

Figure 7 is a schematic illustration of geometric compression of a sample sequence;

Figure 8 is a schematic diagram of QRS analysis of the compressed signal.

detailed description

An embodiment of the ECG signal processing method according to the present invention can refer to FIG. 1 and includes the following steps:

S100. Obtain a sample sequence representing an electrocardiographic signal.

The sample sequence can be expressed as {(t ₁ , d ₁ ), (t ₂ , d ₂ ), ..., (t _i , d _i ), ...}, where t _i is the sampling time of the ith sample, and d _i is The sampled value of the ith sample. The sampled value reflects the fluctuation of the ECG signal, which is proportional to the voltage, but may not have the exact physical meaning.

S200. Denoising the sample sequence.

Existing mathematical means can be used to perform the denoising operation. A preferred embodiment is to calculate the extent to which samples in the sample sequence are affected by noise and discard samples that are affected by noise beyond a preset limit. The way in which the sample is affected by noise can be selected according to the needs or characteristics of the actual application. For example, referring to FIG. 2, the following sub-steps may be included:

S210. Group the sample sequences.

The sample sequences can be grouped according to preset grouping rules, using grouping rules such that each group contains at least one complete ECG cycle.

In some embodiments, the grouping may be performed according to the number of samples, for example, every N samples are grouped according to the chronological order of sampling. In other embodiments, the grouping may be performed at time intervals, for example, the samples per Δt time are grouped according to the time of sampling.

It is more preferable to consider that grouping according to the number of samples has stability in data processing. When grouping according to the number of samples, the size of the group can be pre-configured with The body can be set according to the sampling frequency and the normal heart rate range. For example, if the sampling frequency is F Hertz and the normal heart rate is usually greater than 60 beats/min, the packet size can be set to be greater than F.

The size of the grouping affects the discarding range of inferior samples. If the grouping is too small, it may cause the normal QRS complex to be judged as noise, so the set grouping rules need to ensure that each group contains at least one complete ECG cycle; if the group is too large, it will cause the normal sample to be affected by the inferior sample. Therefore, it should not be set too large. Typically, a packet can be set to contain 1 to 5 ECG cycles.

S220. Calculate the average value of each set of samples.

Assuming that the number of samples per group is N, the average of the j-th sample can be expressed as:

A _j =(d _j1 +...+d _jN )/N

S230. Calculate the stationarity indicator of each group.

The stationarity indicator represents the accumulation of deviations from each sample in the group relative to the average of the group, thereby indicating the extent to which the samples in the group are affected by noise. The specific function relationship can be designed according to the needs of the actual application. An optional calculation method is:

First, the deviation Δd _ji of the i-th sample in the j-th group from the average value is calculated; Δd _{ji is} generally set to a monotonic function of |d _ji -A _j |, for example, Δd _ji =|d _ji -A _j |, Δd _ji =(d _ji -A _j ) ² , Δd _ji =(d _ji -A _j ) ^-2 . It can also be set to a non-monotonic function of |d _ji -A _j |, in which case the corresponding adjustment needs to be made when setting the threshold range for filtering;

Then, the U-index _j calculated stationary j-th group; the _U-j may be directly accumulated Δd _ji may be flat Δd _ji mean, for example, _{_{U j = (Δd j1 + ...}} + Δd jN) / N.

S240. Discard the group whose stationarity index does not conform to the preset rule.

The rule of discarding can be set according to the calculation method of the stationarity index. For example, if Δd _ji is a monotonically increasing function of |d _ji -A _j |, the lower the U _j , the smoother the signal of the packet, and the corresponding rule can be set to discard the packet whose U _{j is} greater than the preset threshold; if Δd _ji Is the monotonically decreasing function of |d _ji -A _j |, the higher the U _j , the smoother the signal of the packet, and the corresponding rule can be set to discard packets whose U _{j is} less than the preset threshold; if Δd _ji is |d _ji - The non-monotonic function of A _j |, such as Δd _ji =d _ji -A _j , the filter rule can be set according to the relationship between Δd _ji and stationarity.

S300. Perform geometric compression on the denoised sample sequence, and discard samples without geometric features.

The so-called sample having no geometric feature means that it is neither an extreme value nor a deviation of a straight line determined with respect to the left and right two samples in a preset range as compared with the currently adjacent left and right two samples. Referring to Figure 3, the compression process can include the following sub-steps:

S310. Sample any three adjacent samples in the sequence.

The selected samples can be expressed as (t _i-1 , d _i-1 ), (t _i , d _i ), and (t _i+1 , di+1). The selected process can be performed in the order of the sample sequence, and after the entire sequence is executed, the execution is repeated until no samples are discarded after traversing the entire sequence.

S320. Determine whether (t _i , d _i ) is the extreme value of [t _i-1 , t _i+1 ], and if yes, continue to perform step S310 of selecting samples, and if otherwise, perform the next step.

S330. Analyzing (t _{_i,} d _i) with respect deviates from a straight line by _{_{(t i-1, d i}} -1) and _{_{(t i + 1, d i}} + 1) determined exceeds a preset threshold value e, if the The step S310 of selecting a sample is continued, and if otherwise, the next step is performed.

A straight line determined by (t _i-1 , d _i-1 ) and (t _i+1 , d _i+1 ) can be expressed as:

(d _i+1 -d _i-1 )t+(t _i-1 -t _i+1 )d+(t _i+1 d _i-1 -t _i-1 d _i+1 )=0,

Sample (t _{_i,} d _i) offset with respect to two adjacent samples of the left and right straight line defined may refer to deviate in the d-axis direction,

B1=|((d _i+1 -d _i-1 )t _i +(t _i-1 -t _i+1 )d _i +(t _i+1 d _i-1 -t _i-1 d _i+1 ))/(t _i-1 -t _i+1 )|

Alternatively, it may mean the distance from (t _i , d _i ) to the straight line.

B2=|((d _i+1 -d _i-1 )t _i +(t _i-1 -t _i+1 )d _i +(t _i+1 d _i-1 -t _i-1 d _i+1 ))/√((d _i+1 -d _i-1 ) ² +(t _i-1 -t _i+1 ) ² )|

Obviously, B1 has higher computational efficiency, and can also replace B2 in the application of ECG signals. Therefore, it is more preferable to use B1 as the basis for judgment.

S340. Discard (t _i , d _i ), and continue to perform step S310 of selecting samples.

S400. Analyze the sample sequence after geometric compression.

To further identify the characteristic information of the ECG signal, the remaining sample sequence can be analyzed and identified after the data compression is completed. The sample sequence can be searched according to a preset time span, if there are samples in the following order in a time span:

First minimum, maximum, second minimum,

And the difference between the two minimum values and the maximum value meets the preset requirement, the three samples are sequentially labeled as Q wave, R wave and S wave; wherein the first minimum value refers to the time span The minimum value of the sample before the maximum value, and the second minimum value refers to the minimum value of the sample after the maximum value within the time span. Referring to FIG. 4, the analysis process may specifically include the following sub-steps:

S410. Searching the sample sequence according to the time span 2D to obtain the maximum value (t _i , d _i ) in [t _i -D, t _i +D].

D is a preset time constant, which should generally be greater than the time span of 1/2 conventional QRS processes, but less than the time span of 1/2 packet. Since the time span of the QRS complex is generally between 0.06 and 0.10 seconds, D D ∈ [0.03, 0.05] seconds can be set.

S420. Search [t _i -D, t _i ) to obtain a minimum value (t _q , d _q ).

S430. Search (t _i , t _i +D) to obtain a minimum value (t _s , d _s ).

S440. Determine whether d _i -d _q >L1 is satisfied, and d _i -d _s >L2, if yes, execute step S450, if otherwise continue to perform step S410 until the entire sample sequence is traversed.

L1 and L2 are preset parameters. Wherein, L1 can be set to be smaller than the difference between the sampling values of the conventional R wave and the Q wave, and L2 can be set to be smaller than the difference between the sampling values of the conventional R wave and the S wave. The specific values of L1 and L2 are related to the characteristics of the sampling equipment used, and can be obtained by manual analysis and statistics of the training data. For example, training data can be used to manually count the appropriate values of L1 and L2 by manually marking the position of each waveform.

S450. The samples (t _q , d _q ), (t _i , d _i ), (t _s , d _s ) are sequentially labeled as Q waves, R waves and S waves, and step S410 is continued until the entire sample sequence is traversed. Samples that are still not specifically marked after traversal are consistently labeled as "other" types.

S500. Output the analyzed sample sequence with type identification.

The sample sequence obtained after the processing is completed can be output to facilitate the storage and use of the ECG signal. The output sample sequence can be expressed as {(t ₁ , d ₁ , c ₁ ), (t ₂ , d ₂ , c ₂ ), ..., (t _i , d _i , c _i ), ...}, where c _i is The type of the i-th sample, which is selected from the group consisting of: Q wave, R wave, S wave, and others.

A person skilled in the art can understand that all or part of the steps in the implementation of the above method may be completed by a program instruction related hardware, and the program may be stored in a computer readable storage medium, and the storage medium may include: a read only memory, a random Memory, disk or disc, etc. Therefore, according to the present invention, there is also provided an electrocardiographic signal processing apparatus, which can be realized by running a corresponding software program by hardware having program execution capability, through software The operation, establishing a functional module for performing the various steps in the above method. Referring to FIG. 5, the ECG signal processing apparatus according to the present invention specifically includes:

a sample sequence obtaining module 610, configured to acquire a sample sequence {(t ₁ , d ₁ ), (t ₂ , d ₂ ), ..., (t _i , d _i ), ...} representing an electrocardiographic signal, where t _i is The sampling time of the i-th sample, d _i is the sampling value of the i-th sample,

The denoising module 620 is configured to perform denoising on the acquired sample sequence, and the denoising module is specifically configured to: calculate a degree of noise affected by the sample in the sample sequence, and discard the sample whose noise is affected by the preset limit;

The geometric compression module 630 is configured to perform geometric compression on the denoised sample sequence, discarding samples without geometric features, and the sample having no geometric features refers to, compared with the currently adjacent left and right samples, it It is neither an extreme value, and the deviation from the straight line determined by the two left and right samples is within a preset range.

Preferably, the ECG signal processing apparatus according to the present invention may further comprise:

The analysis module 640 is configured to analyze the geometrically compressed sample sequence. The analysis module is specifically configured to: search the sample sequence according to a preset time span, if there are samples in the following order in a time span:

First minimum, maximum, second minimum,

And the difference between the two minimum and maximum values meets the preset requirement, the three samples are sequentially labeled as Q wave, R wave and S wave; wherein the first minimum value is located within the time span The minimum value of the sample before the maximum value, and the second minimum value refers to the minimum value of the sample after the maximum value within the time span.

Further preferably, the electrocardiographic signal processing apparatus according to the present invention may further comprise:

The output module 650 is configured to output the analyzed sample sequence {(t ₁ , d ₁ , c ₁ ), (t ₂ , d ₂ , c ₂ ), . . . , (t _i , d _i , c _i ), . Where c _i is the type of the ith sample selected from: Q wave, R wave, S wave, others.

In order to facilitate a better understanding of the present invention, the ECG signal processing method according to the present invention will be exemplified below in conjunction with specific data.

Referring to FIG. 6, (a) is an initial sample sequence and a grouping diagram, and (b) is a schematic diagram of the stationarity index of each group. In Figure 6, the abscissa is the sample number (corresponding to the sampling time), sitting vertically Marked as a change in the voltage indicator. In Figure 6, the initial sample sequence is divided into 4 groups, wherein the stationarity indicators of the first two groups indicate that the stationaryness of the corresponding group is better, and the stationarity indicators of the latter two groups indicate that the stability of the corresponding group is poor, requiring thrown away.

Referring to FIG. 7, is a schematic diagram of geometric compression of a sample sequence after denoising (only a part is taken as an example). In Fig. 7, solid black dots indicate samples before compression, and open dots indicate samples remaining after compression. It can be seen that the remaining samples after compression retain the basic geometric characteristics of the ECG signal well. After compression, the original 1000 samples are greatly reduced to 37, the compression rate reaches 3.7%, and an average ECG cycle requires only 14 samples, which greatly saves storage space and is more conducive to data transmission, especially wireless transmission. .

Referring to Figure 8, there is shown a schematic diagram of QRS analysis of the compressed signal. In Figure 8, each set of three adjacent solid black dots represents a set of Q waves, R waves, and S waves. Samples that have been analyzed and identified can be exported for further storage and use.

The embodiments of the present invention have been described with reference to the specific embodiments of the present invention. It is understood that the above embodiments are only used to help the understanding of the present invention and are not to be construed as limiting the invention. Variations to the above-described embodiments may be made in accordance with the teachings of the present invention.

Claims

An ECG signal processing method, comprising the steps of:

Obtaining a sample sequence {(t 1 , d 1 ), (t 2 , d 2 ), ..., (t i , d i ), ...} representing the electrocardiographic signal, where t i is the sampling time of the ith sample, d i is the sampled value of the ith sample,

Denoising the sample sequence includes: calculating a degree of noise affected by the sample in the sample sequence, and discarding the sample affected by the noise exceeding a preset limit;

Perform geometric compression on the denoised sample sequence and discard samples without geometric features. The sample that has no geometric features means that it is neither extreme nor relative to the currently adjacent left and right samples. The deviation of the straight line determined by the two samples on the left and right is within the preset range.
The method of claim 1 further comprising the step of:

The geometrically compressed sample sequence is analyzed, including: searching the sample sequence according to a preset time span, if there are samples in the following order in a time span:

First minimum, maximum, second minimum,

And the difference between the two minimum and maximum values meets the preset requirement, the three samples are sequentially labeled as Q wave, R wave and S wave; wherein the first minimum value is located within the time span The minimum value of the sample before the maximum value, and the second minimum value refers to the minimum value of the sample after the maximum value within the time span.
The method of claim 2 further comprising the step of:

Outputting the analyzed sample sequence {(t 1 , d 1 , c 1 ), (t 2 , d 2 , c 2 ), ..., (t i , d i , c i ), ...}, where c i is the first The type of i samples selected from: Q wave, R wave, S wave, others.
A method according to claim 2 or 3, wherein

The analyzing the geometrically compressed sample sequence includes:

Searching the sample sequence according to the time span 2D, obtaining the maximum value (t i , d i ) in [t i -D, t i +D],

Search for [t i -D,t i ) and obtain the minimum value (t q ,d q ),

Search (t i , t i +D) to obtain the minimum value (t s ,d s ),

If d i -d q >L1, and d i -d s >L2, the samples (t q , d q ), (t i , d i ), (t s , d s ) are sequentially labeled as Q waves. R wave and S wave, where L1 and L2 are preset parameters.
The method of claim 4 wherein D ∈ [0.03, 0.05] seconds.
A method according to any of the preceding claims, wherein

The denoising the sample sequence includes:

The sample sequences are grouped according to preset grouping rules, such that each group contains at least one complete ECG cycle;

Calculate the average of each set of samples;

Calculating a stationarity indicator for each group, the stationarity indicator indicating a cumulative of deviations of each sample in the group from an average of the group;

Drop the group whose stationarity indicator does not meet the preset rules.
A method according to any of the preceding claims, wherein

In the step of performing geometric compression, the deviation of a sample (t i , d i ) relative to a line determined by the adjacent left and right samples refers to a deviation in the d-axis direction, or means (t i , d i ) the distance to the straight line.
An apparatus for processing an electrocardiogram signal, comprising:

a module for acquiring a sample sequence {(t 1 , d 1 ), (t 2 , d 2 ), ..., (t i , d i ), ...} representing an electrocardiographic signal, where t i is the ith sample Sampling time, d i is the sampled value of the ith sample,

a module for denoising the sample sequence, the module is specifically configured to: calculate a degree of noise affected by the sample in the sample sequence, and discard samples that are affected by noise exceeding a preset limit;

A module for geometrically compressing a sample sequence after denoising, discarding a sample having no geometric features, and a sample having no geometric feature means that it is neither a pole nor a sample adjacent to the currently adjacent left and right samples. The value, and the deviation of the line determined with respect to the left and right samples is within the preset range.
The device of claim 8 further comprising:

A module for analyzing a sequence of geometrically compressed samples, the module is specifically configured to: search a sample sequence according to a preset time span, if a sample appears in the following order in a time span:

First minimum, maximum, second minimum,

And the difference between the two minimum and maximum values meets the preset requirement, the three samples are sequentially labeled as Q wave, R wave and S wave; wherein the first minimum value is located within the time span The minimum value of the sample before the maximum value, the second minimum value refers to the most within the time span The minimum value of the sample after the large value.
The device of claim 9 further comprising:

a module for outputting the analyzed sample sequence {(t 1 , d 1 , c 1 ), (t 2 , d 2 , c 2 ), ..., (t i , d i , c i ), ...}, wherein c i is the type of the i-th sample selected from: Q wave, R wave, S wave, others.