CN104983411A - Real-time calculation method for measurement of complexity of dynamic pulse rate variant signal - Google Patents

Abstract

The invention discloses a real-time calculation method for measurement of the complexity of a dynamic pulse rate variant signal. The real-time calculation method aims to rapidly and accurately achieve measurement calculation of the complexity of the dynamic pulse rate variant signal and is used for online monitoring and early warning of cardiovascular diseases. The real-time calculation method comprises the steps that 1, a dynamic pulse signal is detected and processed; 2, the dynamic pulse rate variant signal is preprocessed, extracted and marked as PP(i); 3, a dynamic pulse rate variant signal sampling point PP(N) is updated in a window sliding way; 4, the time sequence where a PP(1) and the PP(N) are located is reconstructed; 5, an X(1) sequence and an X(N-m+1) sequence are symbolized; 6, the storage positions of a [S1(j)] code and a [SN-m+2(j)] code are generated; 7, an entropy value is iterated and updated.

Description

A real-time calculation method for dynamic pulse rate variability signal complexity measure

technical field

The invention relates to real-time processing and analysis of pulse signal detection, in particular to a real-time calculation method for dynamic pulse rate variability signal complexity measurement based on sliding window iterative basic scale entropy. It can be used for real-time extraction of dynamic pulse rate and real-time complexity calculation.

Background technique

The human cardiovascular system is a complex dynamic system. It is very important for the diagnosis and treatment of cardiovascular diseases to detect its dynamic changes and distinguish the complexity changes in different time periods or different physiological states.

Heart rate variability signals are generated from changes in the cardiac beating cycle and contain rich physiological and pathological information about the cardiovascular system. However, heart rate variability signals are obtained from ECG signals, and the complicated connections and electrodes limit its application in portable and wearable medical devices. The pulse rate variability signal is also generated from the change of the heart beat cycle. Similar to the heart rate variability signal, it also contains a lot of physiological and pathological information about the cardiovascular system. Compared with the ECG signal, the pulse rate variability signal is obtained from the pulse signal, the acquisition process is simple, and it can be conveniently used in portable wearable medical instruments. The dynamic pulse rate variability signal is extracted from the dynamic pulse signal. Compared with offline signal analysis, online real-time analysis is very important for the monitoring and timely warning of cardiovascular diseases.

At present, the analysis of pulse rate variability mainly refers to the analysis methods of heart rate variability, and the commonly used methods are time domain, frequency domain, time-frequency domain and nonlinear analysis methods. The pulse rate variability signal is a non-stationary time-varying signal, and the nonlinear method can extract the complexity change of the signal more effectively. Some nonlinear features such as sample entropy, approximate entropy, symbol sequence entropy, basic scale entropy and other information entropy analysis methods can be used to calculate the complexity of heart rate variability signals. Among them, the basic scale entropy can effectively analyze the heart rate variability signal, and it can be used to identify diseases such as coronary heart disease, and has achieved good results. However, the calculation amount of the basic scale entropy increases exponentially with the increase of the signal length, and the calculation intermediate variables occupy a large amount of memory in the system, which brings great difficulties to the real-time calculation of the dynamic pulse rate variability signal complexity measure.

Contents of the invention

The purpose of the invention is to quickly and accurately realize the measurement calculation of the pulse rate variability signal complexity, which is used for online monitoring and early warning of cardiovascular diseases.

The present invention is a dynamic pulse rate variability signal complexity measure real-time calculation method, and its steps are:

(1) For dynamic pulse signal detection and processing, dynamic pulse signal detection and processing are completed by photoelectric pulse sensor, pulse signal detection module, Bluetooth module and smart phone module;

(2) Dynamic pulse signal preprocessing and dynamic pulse rate variability signal extraction, denoted as PP ( i ), where PP ( i ) is the ith sampling value;

(3) Update the dynamic pulse rate variability signal sampling point PP ( N ) by sliding window, where PP ( N ) is the Nth sampling value;

(4) Reconstruct the time series where PP (1) and PP ( N ) are located, denoted as X (1) and X ( N - m +1) respectively, where m is the length of the time series;

(5) Symbolize X (1) and X ( N - m +1) sequences, respectively denoted as { S ₁ ( j )} and { S _{N - m +2} ( j )}, where j is the symbolization of the corresponding sequence The value of the jth point after;

(6) Generate { S ₁ ( j )} and { S _{N - m +2} ( j )} encoded storage locations;

(7) Iteratively update the entropy value.

The advantage of the present invention is that the basic scale entropy analysis method using the sliding window iteration refers to the basic scale entropy calculation realized by using the sliding window iterative thought. Compared with the original basic scale entropy analysis method, single sampling point analysis can be realized. Under the premise of not affecting the calculation accuracy, the algorithm running speed can be improved and the system memory can be saved. The improvement method of sliding window iteration to basic scale entropy can also be used to improve other information entropy such as symbol sequence entropy.

Heart rate variability signals can reflect the activity of the human autonomic nervous system, and can also be used to evaluate the balance of sympathetic and vagus nerves. Through the sliding window iterative basic scale entropy analysis method, the dynamic pulse rate variability signal is analyzed in real time, and the basic scale entropy that can reflect the change of heart beat is obtained. In order to understand the state of the autonomic nervous system and realize the monitoring and early warning of diseases.

Description of drawings

Fig. 1 is a block diagram of the pulse signal detection and processing system of the present invention, and Fig. 2 is an algorithm principle diagram of the sliding window iterative basic scale entropy analysis of the present invention.

Detailed ways

As shown in Figure 1 and Figure 2, the present invention is a dynamic pulse rate variability signal complexity measure real-time calculation method, and its steps are:

(1) For dynamic pulse signal detection and processing, dynamic pulse signal detection and processing are completed by photoelectric pulse sensor, pulse signal detection module, Bluetooth module and smart phone module;

(2) Dynamic pulse signal preprocessing and dynamic pulse rate variability signal extraction, denoted as PP ( i ), where PP ( i ) is the ith sampling value;

(3) Update the dynamic pulse rate variability signal sampling point PP ( N ) by sliding window, where PP ( N ) is the Nth sampling value;

(4) Reconstruct the time series where PP (1) and PP ( N ) are located, denoted as X (1) and X ( N - m +1) respectively, where m is the length of the time series;

(5) Symbolize X (1) and X ( N - m +1) sequences, respectively denoted as { S ₁ ( j )} and { S _{N - m +2} ( j )}, where j is the symbolization of the corresponding sequence The value of the jth point after;

(6) Generate { S ₁ ( j )} and { S _{N - m +2} ( j )} encoded storage locations;

(7) Iteratively update the entropy value.

According to the above-mentioned real-time calculation method of dynamic pulse rate variability signal complexity measurement, the detection and processing of the dynamic pulse signal described in step (1) is completed by the photoelectric pulse sensor, pulse signal detection module, bluetooth module and smart phone module . Figure 1 shows the block diagram of the detection and processing system of the dynamic pulse signal. Realize the detection and processing of dynamic pulse signal, the signal sampling frequency is 250Hz.

According to the real-time calculation method of the dynamic pulse rate variability signal complexity measure described above, the preprocessing and extraction of the collected dynamic pulse signal described in the above step (2) is carried out according to the following steps:

(1) Real-time filtering of myoelectric interference and random noise through an integer coefficient low-pass filter with a cut-off frequency of 62.5Hz;

(2) 50Hz and its integer multiple harmonic notch filter to remove baseline drift and power frequency interference;

(3) The dynamic pulse rate variability signal is extracted by using the dynamic difference threshold method on the filtered signal. The pulse rate variability signal here is the main wave interval of the pulse signal, which is denoted as PP ( i ), where PP ( i ) is the i -th sampling value, and the unit is ms.

According to the real-time calculation method of the dynamic pulse rate variability signal complexity measure described above, the sliding window method is used to update the dynamic pulse rate variability signal sampling point PP ( N ) in the above step (3), where PP ( N ) is For the Nth sampling value, set the data buffer to buffer the extracted dynamic pulse rate variability signal, and record the length of the data buffer as N sampling points. Using the data buffer as a window, the dynamic pulse rate variability signal sampling points are updated by means of a sliding window. While updating the sampling points, the basic scale entropy is calculated iteratively. Proceed as follows:

(1) Open up space in the memory first, cache new sampling points, and record it as PP ( N +1);

(2) Eliminate the earliest cached sampling point PP (1), and move the high-order sampling point to the low-order, PP ( i )= PP ( i +1);

(3) Put the pre-cached PP ( N +1 ) in the highest position of the buffer area, that is, PP ( N ) = PP ( N +1 ).

According to the real-time calculation method of the dynamic pulse rate variability signal complexity measure described above, the time series where PP (1) and PP ( N ) are reconstructed in the above step (4), where m is the length of the time series, according to the basic The principle of scale entropy requires reconstruction of N cached sampling points to generate N - m + 1 time series of length m , and each time series represents a heart beat pattern. In order to reduce the amount of calculation and storage space in the reconstruction process, the present invention implements the calculation of the entropy value in an iterative manner. Proceed as follows:

(1) Reconstruct the time series where PP (1) is located as:

(1)

(2) Reconstruct the time series where PP ( N ) is located as:

(2).

According to the real-time calculation method of the dynamic pulse rate variability signal complexity measure described above, the symbolized X (1) and X ( N - m +1) sequences described in the above step (5), where j is the corresponding sequence symbolized For the value of the jth point, proceed as follows:

(1) Symbolize the X (1) sequence. Among them, PP (1) is symbolized by the following formula:

(3)

In the formula, μ ₁ is the mean value of the time series X (1), and α is a constant, which is used to adjust the symbolization boundary. BS ₁ is the basic scale used to determine symbolic boundaries. BS ₁ is:

(4)

(2) Symbolize the X ( N - m +1 ) sequence. The process is the same as (1), recorded as {S _{N - m +2} ( j )}, j =1, …, m .

According to the above-mentioned real-time calculation method of the dynamic pulse rate variability signal complexity measure, the above-mentioned step (6) generates { S ₁ ( j )} and { S _{N - m +2} ( j )} encoded storage locations. Each different time series represents a different cardiac beating mode, so a time series containing 4 symbols and a length of m can represent 4 ^m cardiac beating modes. Count the probability of each different mode accounting for the entire N - m +1 mode, and use it to calculate the entropy value of the entire time series. Therefore, it is necessary to update { S ₁ ( j )} and { S _{N - m +2} ( j ) } occurrence and number. The storage location of the S ₁ code is:

(5).

According to the above-mentioned real-time calculation method of the dynamic pulse rate variability signal complexity measure, the above-mentioned step (7) iteratively updates the entropy value.

The update of the entropy value is implemented in an iterative manner, so the entropy value of { S ₁ ( j )} needs to be subtracted from the last calculation result, and the entropy value of { S _{N - m +2} ( j )} needs to be added. Realize the calculation of the entropy value of all sampling points. After each data update, the number of cardiac beat patterns represented by { S ₁ ( j )} is reduced by 1, and the number of patterns represented by { S _{N - m +2} ( j )} is increased by 1. The numbers are represented by n ( h ) and n ( k ) respectively, as shown in Figure 2.

After updating the numbers of { S ₁ ( j )} and { S _{N - m +2} ( j )}, the entropy value can be updated iteratively. The entropy value before the sliding window is BS′ ( m ), the number of { S ₁ ( j )} is n ′ ( h ), the probability of occurrence is p ′ ( h ), { S _{N - m +2} ( j )} The number of { S 1 ( j )} is n ′ ( k ), the probability of occurrence is p ′ ( k ); the entropy value after the sliding window is BS ( m ), the number of { S ₁ ( j )} is n ( h ), and the probability of occurrence is p ( h ), the number of { S _{N - m +2} ( j )} is n ( k ), and the probability of occurrence is p ( k ). Initial BS ′=0, n ( h )= n ′( h )-1, n ( k )= n ′( k )+1 during iteration. According to the changing law of the number of the two modes, and the true number of the logarithm in the entropy calculation process must be greater than 0;

Proceed as follows:

(1) Record the entropy value before the sliding window as BS′ ( m ), the number of { S ₁ ( j )} is n ′ ( h ), calculate the probability of occurrence, and record it as p ′ ( h ), { S _{N - m} The number of ₊₂ ( j )} is n ′ ( k ), and the probability of occurrence is calculated, which is recorded as p ′ ( k );

(2) The entropy value after the sliding window is BS ( m ), the number of { S ₁ ( j )} is n ( h ), and the probability of occurrence is calculated, which is recorded as p ( h ), { S _{N - m +2} ( j )} is n ( k ), and the probability of occurrence is calculated as p ( k );

(3) Initial BS ′=0, n ( h )= n ′( h )-1, n ( k )= n ′( k )+1 during iteration;

(4) Iterative update. According to the change law of the number of the two modes, and the restriction that the true number of the logarithm must be greater than 0 in the entropy calculation process, there are the following four iterative update methods:

(a) n ( h )>0, n ( k )>1, which means that the modes represented by { S ₁ ( j )} and { S _{N - m +2} ( j )} exist both before and after the sliding window. Therefore, in the process of calculating the entropy value, there is no need to consider the case that the true logarithm is 0. because:

(6)

(7)

In the formula, M = ^4m . Before and after the sliding window, only the number of modes represented by { S ₁ ( j )} and { S _{N - m +2} ( j )} has changed, and other modes remain unchanged, then - p (1)log ₂ p (1 )=- p′ (1)log ₂ p′ (1),…, -p ( M )log ₂ p ( M )=- p′ ( M )log ₂ p′ ( M ). Then, subtract formula (7) from formula (6) to get:

(8)

Further simplify the formula (8):

(9)

Especially, when { S ₁ ( j )} and { S _{N - m +2} ( j )} before and after the sliding window represent the same pattern. First, the number of patterns represented by { S ₁ ( j )} minus 1, n ( h ) = n ′( h )-1; then, the number of patterns represented by { S _{N - m +2} ( j )} increases 1, n ( k ) = n '( k )+1= n ( h )+1= n '( h ). Then formula (8) is:

(10)

That is, the entropy value does not change before and after the sliding window. Because { S ₁ ( j )} and { S _{N - m +2} ( j )} represent the same mode, the number of modes before and after the sliding window does not change.

Or (b) n ( h )=0, n ( k )>1, which means that the mode represented by { S ₁ ( j )} disappears after the sliding window. Then n ′( h )=1, p ( h )=0, p ′( h )=1/( N - m +1). Then from formula (8):

(11)

Or (c) n ( h )>0, n ( k )=1, which means that the mode represented by { S _{N - m +2} ( j )} disappears after the sliding window. Then n ′( h )= n ( h )+1, n ( k )=1, n ′( k )=0, p ′( k )=0, p ( k )=1/( N - m +1 ). Then it can be obtained from formula (8):

(12)

Or (d) n ( h ) = 0, n ( k ) = 1, which means that the mode represented by { S ₁ ( j )} disappears, and the mode represented by { S _{N - m +2} ( j )} is the first time appear, so the number of total patterns remains unchanged. Then n ′( h )=1, n ′( k )=0, p ′( k )= p ( h )=0, p ′( h )= p ( k )=1/( N - m +1) . From formula (8) can get:

(13)

The entropy value of the basic scale of the dynamic pulse rate variability signal can be obtained through formula (9) - formula (13), and the calculation of the complexity measure of the dynamic pulse rate variability signal can be realized in real time through the change of the entropy value.

Claims (7)

1. dynamic pulse frequency Variability Signals Complexity Measurement real-time computing technique, is characterized in that, the steps include:

(1) to dynamic pulse signal detection and treatment, dynamic pulse signal detection and treatment is completed by photoelectric sphyg sensor, pulse signal detection module, bluetooth module and smart mobile phone module;

(2) dynamic pulse signal pretreatment and dynamic pulse frequency Variability Signals extract, and are designated as pP( i), wherein pP( i) be iindividual sampled value;

(3) the mode Regeneration dynamics pulse frequency Variability Signals sampled point of sliding window is adopted pP( n), wherein pP( n) be nindividual sampled value;

(4) reconstruct pP(1) and pP( n) time series at place, be designated as respectively x(1) and x( n- m+ 1), wherein mfor length of time series;

(5) symbolization x(1) and x( n- m+ 1) sequence, be designated as respectively s ₁( j) and s _{n- m+ 2} ( j), wherein jfor after corresponding sequence symbolization jpoint value;

(6) produce s ₁( j) and s _{n- m+ 2} ( j) memory location of encoding;

(7) iteration upgrades entropy.

2. dynamic pulse frequency Variability Signals Complexity Measurement real-time computing technique according to claim 1, is characterized in that step (2) is described and carries out pretreatment and extraction to collection dynamic pulse signal, carry out as follows:

(1) the real-time filtering myoelectricity interference of the integral coefficient LP filter by by frequency being 62.5Hz and random noise;

(2) 50Hz and integer harmonics wave trap thereof remove baseline drift and Hz noise;

(3) dynamic difference threshold method is adopted to extract dynamic pulse frequency Variability Signals to filtered signal;

Pulse frequency Variability Signals is herein the main ripple interval of pulse signal, is designated as pP( i), wherein pP( i) be iindividual sampled value.

3. dynamic pulse frequency Variability Signals Complexity Measurement real-time computing technique according to claim 1, is characterized in that the described mode Regeneration dynamics pulse frequency Variability Signals sampled point adopting sliding window of step (3) pP( n), wherein pP( n) be nindividual sampled value, carry out as follows:

(1) first at internal memory opening space, the sampled point that buffer memory is new, is designated as pP( n+ 1);

(2) sampled point of buffer memory is the earliest rejected pP(1), high-order sampled point moves to low level, pP( i)= pP( i+ 1);

(3) again by buffer memory in advance pP( n+ 1) highest order of buffer area is placed on, namely pP( n)= pP( n+ 1).

4. dynamic pulse frequency Variability Signals Complexity Measurement real-time computing technique according to claim 1, is characterized in that step (4) described reconstruct pP(1) and pP( n) time series at place, be designated as respectively x(1) and x( n- m+ 1), wherein mfor length of time series, carry out as follows:

(1) will pP(1) time series at place is reconstructed into:

(1)

(2) will pP( n) time series at place is reconstructed into:

(2)。

5. dynamic pulse frequency Variability Signals Complexity Measurement real-time computing technique according to claim 1, is characterized in that step (5) described symbolization x(1) and x( n- m+ 1) sequence, be designated as respectively s ₁( j) and s _{n- m+ 2} ( j), wherein jfor after corresponding sequence symbolization jpoint value, carries out as follows:

(1) symbolization x(1) sequence:

Wherein, following formula pair is adopted pP(1) symbolization:

(3)

In formula, μ ₁for time series x(1) average, αfor constant, be used for regulating symbolization border; bS ₁for cardinal scales, be used for determining symbolization border;

bS ₁for:

(4)

(2) symbolization x( n- m+ 1) sequence:

Process, with (1), is designated as { S _{n- m+ 2} ( j), j=1 ..., m.

6. dynamic pulse frequency Variability Signals Complexity Measurement real-time computing technique according to claim 1, it is characterized in that step (6) described generation s ₁( j) and s _{n- m+ 2} ( j) memory location of encoding, wherein s ₁code storage position is:

(5)。

7. dynamic pulse frequency Variability Signals Complexity Measurement real-time computing technique according to claim 1, is characterized in that the described iteration of step (7) upgrades entropy, carries out as follows:

(1) before the sliding window of note, entropy is bS '( m), s ₁( j) number be n' ( h), calculate probability of occurrence, be designated as p' ( h), s _{n- m+ 2} ( j) number be n' ( k), calculate probability of occurrence, be designated as p' ( k);

(2) after sliding window, entropy is bS( m), s ₁( j) number be n( h), calculate probability of occurrence, be designated as p( h), s _{n- m+ 2} ( j) number be n( k), calculate probability of occurrence, be designated as p( k);

(3) initial bS'=0, in iterative process n( h)= n' ( h)-1, n( k)= n' ( k)+1;

(4) iteration upgrades: according to the Changing Pattern of two kinds of number of modes, and in entropy computational process, the antilog of logarithm must be greater than the restriction of 0, has following four kinds of iteration update modes:

(a) n( h) >0, n( k) >1, represent s ₁( j) and s _{n- m+ 2} ( j) represented by pattern all exist in the front and back of sliding window;

Entropy iterative formula is:

(6)

Special, before and after sliding window, s ₁( j) and s _{n- m+ 2} ( j) when representing same mode; First, s ₁( j) number of representative pattern subtracts 1, n( h)= n' ( h)-1; Then, s _{n- m+ 2} ( j) representative number of modes adds 1, n( k)= n' ( k)+1= n( h)+1= n' ( h); Entropy iterative formula is:

(7)

Or (b) n( h)=0, n( k) >1, represent s ₁( j) representated by pattern disappear after sliding window; Then n' ( h)=1, p( h)=0, p' ( h)=1/ ( n- m+ 1); Entropy iterative formula is:

(8)

Or (c) n (h)>0, n (k)=1, represent s _{n- m+ 2} ( j) representated by pattern disappear after sliding window; Then n' ( h)= n( h)+1, n( k)=1, n' ( k)=0, p' ( k)=0, p( k)=1/ ( n- m+ 1); Entropy iterative formula is:

(9)

Or (d) n( h)=0, n( k)=1, represent s ₁( j) representated by pattern disappear, { s _{n- m+ 2} ( j) representated by pattern first time occur, therefore the number of assemble mode is constant; Then n' ( h)=1, n' ( k)=0, p' ( k)= p( h)=0, p' ( h)= p( k)=1/ ( n- m+ 1); Entropy iterative formula is:

(10)

Through type (6)-Shi (10) can obtain the cardinal scales entropy of dynamic pulse frequency Variability Signals, by the change of entropy, realizes the calculating of dynamic pulse frequency Variability Signals Complexity Measurement in real time.