WO2021164347A1 - 一种对血压进行预测的方法和装置 - Google Patents

一种对血压进行预测的方法和装置 Download PDF

Info

Publication number
WO2021164347A1
WO2021164347A1 PCT/CN2020/129632 CN2020129632W WO2021164347A1 WO 2021164347 A1 WO2021164347 A1 WO 2021164347A1 CN 2020129632 W CN2020129632 W CN 2020129632W WO 2021164347 A1 WO2021164347 A1 WO 2021164347A1
Authority
WO
WIPO (PCT)
Prior art keywords
time
point
signal
ppg
generate
Prior art date
Application number
PCT/CN2020/129632
Other languages
English (en)
French (fr)
Inventor
孙洪岱
李喆
曹君
Original Assignee
乐普（北京）医疗器械股份有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 乐普（北京）医疗器械股份有限公司 filed Critical 乐普（北京）医疗器械股份有限公司
Publication of WO2021164347A1 publication Critical patent/WO2021164347A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
    • A61B5/021—Measuring pressure in heart or blood vessels
    • A61B5/02108—Measuring pressure in heart or blood vessels from analysis of pulse wave characteristics
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
    • A61B5/316—Modalities, i.e. specific diagnostic methods
    • A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7271—Specific aspects of physiological measurement analysis
    • A61B5/7275—Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor

Definitions

  • the present invention relates to the technical field of electrophysiological signal processing, in particular to a method and device for predicting blood pressure.
  • the heart is the center of human blood circulation.
  • the heart produces blood pressure through regular pulsation, and then supplies blood to the whole body to complete the body's metabolism.
  • Blood pressure is one of the very important physiological signals of the human body.
  • hypertension is getting higher and higher, which seriously harms human health.
  • a large amount of epidemiological and clinical evidence shows that long-term hypertension will increase the incidence of ischemic heart disease, stroke, renal failure, aortic and Risk of damage to target organs such as peripheral arterial disease.
  • Hypertension is a chronic disease, most of which require long-term lifelong care.
  • the effectiveness of lifestyle control in patients with hypertension, the efficacy of antihypertensive drugs and the evaluation of the efficacy of interventional therapy for hypertension require long-term dynamic monitoring of blood pressure.
  • the purpose of the present invention is to provide a method and device for predicting blood pressure against the shortcomings of the prior art, and to synchronize the electrocardiogram (ECG) signal and the photoplethysmography (PPG) signal to the tester Acquisition, feature extraction of the acquired ECG signal and PPG signal, and then matching the respective feature data, and then fusion of the matched ECG signal and PPG signal to generate sample data and use the random forest algorithm model to predict and calculate the sample data. Finally, the predicted value of blood pressure is obtained.
  • ECG electrocardiogram
  • PPG photoplethysmography
  • the ECG and PPG signals can be automatically and continuously analyzed and predicted by cooperating with the acquisition sensor, thereby not only improving the comfort of the tester, but also establishing A way to automatically monitor blood pressure.
  • the first aspect of the embodiments of the present invention provides a method for predicting blood pressure, and the method includes:
  • the R point time sequence includes multiple R point times
  • the peak point time sequence includes multiple peak point times
  • the trough point time series includes multiple trough point times
  • each of the R point times is used as a time reference point, and in the PPG signal, the time reference point is extracted The first valley point time, the first peak point time and the second valley point time thereafter; according to the R point time, the first valley point time, the first peak point time and the second Generating a matching feature time group at the valley point time; and sorting all the matching feature time groups in a sequence to generate a matching feature time group sequence;
  • the matched characteristic time group sequence perform the characteristic sample data preparation operation of the random forest algorithm model to generate a random forest sample group sequence;
  • the random forest sample group sequence includes a plurality of random forest sample groups;
  • the R point time series calculate the corresponding R point instantaneous heart rate, R point trend heart rate, and R snack rate difference; and use whether the R point rate difference is less than a preset reasonable heart rate difference threshold as the abnormal sample group judgment condition pair Performing abnormal sample group deletion processing on the random forest sample group sequence;
  • the random forest algorithm model is used to perform regression prediction calculation on the random forest sample group sequence to generate a predicted blood pressure array; the predicted blood pressure array includes systolic blood pressure data and diastolic blood pressure data.
  • the synchronous collection of the electrocardiographic physiological signal and the pulse physiological signal of the tester generates the electrocardiographic signal and the pulse physiological signal; and the electrocardiographic signal and the pulse physiological signal are signaled according to a preset sampling frequency threshold.
  • the sampling process generates the ECG signal and the PPG signal of the photoplethysmography method, including:
  • the ECG signal includes a plurality of ECG signal points;
  • the ECG signal point includes signal point amplitude data and signal point time data;
  • the pulse physiological signal is signal-sampled according to the sampling frequency threshold to generate a PPG original signal, and the PPG original signal is band-pass filtered according to a preset band-pass frequency threshold range to generate the PPG signal;
  • the PPG The signal includes a plurality of PPG signal points; the PPG signal points include signal point amplitude data and signal point time data.
  • said performing an R point time feature recognition operation on the ECG signal to generate an R point time sequence specifically includes:
  • ECG signal sequentially extract the signal point time data of the ECG signal points to generate an ECG one-dimensional data vector; perform data segment division operations on the ECG one-dimensional data vector according to a preset ECG segment length threshold to generate multiple ECG one-dimensional fragment vector;
  • All the identified R point times are sorted in order to generate the R point time series.
  • the performing the pulse wave peak point time feature recognition operation and the pulse wave valley point time feature recognition operation on the PPG signal to generate the peak point time series and the bottom point time series specifically includes:
  • the peak point time sequence performing the pulse wave valley point time feature extraction operation on the PPG signal to generate the valley point time sequence.
  • performing the pulse wave peak point time feature identification operation on the PPG signal to generate the peak point time sequence specifically includes:
  • the minimum value is extracted to reference the signal point amplitude To initialize;
  • the performing the pulse wave trough point time feature extraction operation on the PPG signal according to the peak point time series to generate the trough point time series specifically includes:
  • the peak point time sequence in the PPG signal, between two adjacent peak point times, extract the signal point time data of the PPG signal point for which the signal point amplitude data is the minimum value, Generating the valley point time; adding all the valley point times extracted to the valley point time sequence in a sequential order.
  • each of the R point times is used as a time reference point, and the PPG signal is extracted from The first valley point time, the first peak point time, and the second valley point time after the time reference point; according to the R point time, the first valley point time, and the first peak point time Generating a matching feature time group with the second valley point time; and sorting all the matching feature time groups in a sequence to generate a matching feature time group sequence, which specifically includes:
  • Set the matching feature time group initialize the matching R point time of the matching feature time group to be empty, initialize the matching PPG peak time of the matching feature time group to be empty, and initialize the matching PPG start of the matching feature time group The time is empty, and the matching PPG end time of the matching feature time group is initialized as empty;
  • the bottom point time sequence is searched in the opposite direction from the end time to the start time with the first reference R point as the start time and the second reference R point as the end time, and the distance is extracted The valley point time with the closest end time generates the second valley point time, and the next valley point time closest to the second valley point time is extracted to generate the first valley point time ;
  • the peak point time sequence is searched in the opposite direction from the end time to the start time with the first reference R point as the start time and the second reference R point as the end time , Extracting the peak point time closest to the end time to generate the first peak point time;
  • Adding the matching characteristic time group to the matching characteristic time group sequence is performed on the successfully set matching characteristic time group.
  • the step of performing the feature sample data preparation operation of the random forest algorithm model according to the matching feature time group sequence to generate a random forest sample group sequence specifically includes:
  • the matching PPG start time generation start time is extracted, the matching PPG end time generation end time of the current matching feature time group is extracted; the matching R point time of the adjacent matching feature time group is extracted to generate a second R point time;
  • the first sample parameter of the random forest sample group is the quotient of 60 divided by the heartbeat time difference, where the heartbeat time difference is the absolute value of the time difference between the first R point time and the second R point time;
  • the twenty-first sample parameter of the random forest sample group is set as the absolute value of the time difference between the end time and the start time.
  • the corresponding R point instant heart rate, R point trend heart rate, and R snack rate difference are calculated according to the R point time series; and whether the R point rate difference is less than a preset reasonable heart rate difference threshold is regarded as an abnormality
  • the sample group determination condition performs abnormal sample group deletion processing on the random forest sample group sequence, which specifically includes:
  • Step 91 Initialize the instantaneous heart rate sequence to be empty; obtain the total number of R point times included in the R point time sequence to generate the total number of R points;
  • Step 92 Extract the R point time of the R point time series in sequence to generate the current R point, and extract the R point time adjacent to the current R point according to the designated adjacent point extraction direction to generate adjacent R Point; generate a first factor according to the absolute value of the time difference between the current point R and the adjacent R point; generate the instantaneous heart rate of the point R according to the inverse of the first factor; direct the instantaneous heart rate of the point R to the
  • the instantaneous heart rate sequence performs an R point instantaneous heart rate addition operation; the instantaneous heart rate sequence includes the total number of the R points and the R point instantaneous heart rate;
  • Step 93 Perform Gaussian filtering on the instantaneous heart rate sequence according to a preset filter standard deviation to generate a trend heart rate sequence;
  • the trend heart rate sequence includes the total number of R points and the R point trend heart rate;
  • Step 94 Initialize the value of the first index to 1, and initialize the value of the first total to the total number of R points;
  • Step 95 Extract the R point instantaneous heart rate corresponding to the first index from the instantaneous heart rate sequence to generate a first index instantaneous heart rate; extract the trend heart rate sequence corresponding to the first index R-point trend heart rate generates the first index trend heart rate;
  • Step 96 Generate a first index R snack rate difference according to the absolute value of the heart rate difference between the first index instant heart rate and the first index trend heart rate;
  • Step 97 when the first index R, the difference in heart rate difference is greater than the reasonable heart rate difference threshold, mark the random forest sample group corresponding to the first index as an abnormal sample group;
  • Step 98 Add 1 to the first index
  • Step 99 Determine whether the first index is greater than the first total, if the first index is greater than the first total, then go to step 100, if the first index is less than or equal to the first total, then Go to step 95;
  • Step 100 poll the random forest sample group sequence, and delete the random forest sample group marked as the abnormal sample group from the random forest sample group sequence.
  • the first aspect of the embodiments of the present invention provides a method for predicting blood pressure. Firstly, the tester is subjected to the synchronous acquisition of the ECG signal and PPG signal; secondly, the characteristic extraction is performed on the acquired ECG signal and PPG signal: acquired The R point feature of the ECG signal obtains the pulse wave peak point and the starting and ending point features of the PPG signal; then, the respective feature data is matched: the feature points of the ECG signal and the PPG signal are correlated to generate a matching feature group; then, according to the matching The ECG signal and the PPG signal are fused to generate sample data: based on each matching feature group, the related feature parameter settings are performed; finally, the full-time matching feature group sequence is used as the input of the random forest algorithm model for blood pressure regression calculation to generate Predict blood pressure array (including diastolic blood pressure and systolic blood pressure).
  • a second aspect of the embodiments of the present invention provides a device, the device including a memory and a processor, the memory is used to store a program, and the processor is used to execute the first aspect and the methods in each implementation manner of the first aspect.
  • a third aspect of the embodiments of the present invention provides a computer program product containing instructions, which when the computer program product runs on a computer, causes the computer to execute the first aspect and the methods in the implementation manners of the first aspect.
  • a fourth aspect of the embodiments of the present invention provides a computer-readable storage medium on which a computer program is stored.
  • the computer program implements the first aspect and the methods in the first aspect when executed by a processor. .
  • FIG. 1 is a schematic diagram of a method for predicting blood pressure according to Embodiment 1 of the present invention
  • FIG. 2 is a schematic diagram of a change in the reference amplitude of a signal point according to an embodiment of the present invention
  • FIG. 3 is a schematic diagram of ECG signals and PPG signals provided by an embodiment of the present invention.
  • FIG. 4 is a schematic diagram of a method for processing an abnormal random forest sample group according to Embodiment 2 of the present invention.
  • FIG. 5 is a schematic diagram of the device structure of a device for predicting blood pressure according to Embodiment 3 of the present invention.
  • the embodiment of the present invention uses the ECG signal as the heartbeat reference data and the PPG signal as the pulse reference data.
  • the ECG signal is a group of electrophysiological signals of the heart's cardiac cycle collected from the body surface using the electrocardiographic signal acquisition equipment.
  • the conventional ECG signal waveform has 5 characteristic points, which are respectively P, Q, R, S, and T points. In actual operation, except for the R point, the other four points are more likely to be interfered by the noise signal, and the probability of being mistakenly eliminated during the signal filtering and noise reduction process is also higher. Once the P point or T point of a certain heartbeat signal is mistakenly eliminated, the current heartbeat signal will not be included in the analysis signal category, and the problems of feature omission and feature deviation are prone to occur.
  • the R point signal of the strongest signal among the 5 points is used as the characteristic point of the cardiac signal, and the maximum number of valid cardiac signal data in the current ECG data can be retained, and the problem of the loss of the cardiac data in the conventional method is solved.
  • the PPG signal is a set of signals that uses the light sensor to identify and record the change in light intensity of a specific light source.
  • the blood flow per unit area in the blood vessel changes periodically, and the corresponding blood volume also changes accordingly, resulting in a periodic change trend in the PPG signal reflecting the amount of light absorbed by the blood.
  • a cardiac cycle includes two time periods: systolic and diastolic; during systole, the heart does work on the whole body, causing continuous and periodic changes in intravascular pressure and blood flow volume. When the heart is in diastole, the pressure on the blood vessels is relatively small.
  • the blood pushed out to the whole body from the last systole hits the heart valve through the circulation, which produces a certain reflection and refraction effect on the light, resulting in the diastolic cycle.
  • the absorption of light energy by blood in the blood vessels is reduced. Therefore, the time characteristics of the PPG signal waveform reflecting the light energy absorbed by the blood in the blood vessel have two time characteristics: the signal time characteristics of the systolic period and the signal time characteristics of the diastolic period; the common PPG signal waveform before the maximum peak is considered to be a typical contraction After the maximum peak value, an absolute refractory period is set as the transition period from the nominal typical systolic period to the typical diastolic period.
  • the heart beat first produces the heart beat, then causes the blood pressure change, and then affects the PPG signal fluctuation, so for the time characteristics of a heart beat, the R point time is before the PPG signal start time in the corresponding cycle.
  • the R point time of the ECG signal mentioned above can be regarded as the reference signal of motivation for a blood pressure change
  • the PPG waveform of the PPG signal (from the PPG start time, the PPG peak time, and the PPG end time) can be regarded as the result reference signal. Matching the time characteristics of the two is to select a corresponding cause waveform and result waveform for each heartbeat.
  • the actual operation is to select a PPG waveform corresponding to the R point signal in the PPG signal to complete the matching.
  • the matching principle of this embodiment is to select only between two R points (the first R point and the second R point). Point) The PPG waveform closest to the next heartbeat (second R point) is used as the matching target of the current heartbeat (first R point).
  • the embodiment of the present invention corresponds to the feature sample data preparation operation of the random forest algorithm model.
  • the characteristic sample data includes a total of 21 sample parameters in the form of a random forest sample group, which are mainly classified into six categories: heart rate parameters, time parameters, amplitude parameters, slope parameters, area parameters, area offset time parameters, specific classifications and definitions See the table below for details:
  • a random forest sample group After fusion of ECG signals and PPG signals to generate multiple random forest sample groups, in order to ensure that the proportion of abnormal sample groups in the random forest sample group is within a reasonable range, a random forest sample group needs to be screened.
  • the previous feature matching and the current sample parameter calculation are based on a principle: the R point is regarded as the normal heartbeat time point; and in the actual data collection process, the R point also has an abnormal R point.
  • the screening is to locate the abnormal R point according to the concept of the heart rate difference of the R point, and remove the sample group corresponding to the abnormal R point.
  • the heart rate difference at point R is the absolute value of the heart rate difference between the instantaneous heart rate and the trend heart rate corresponding to the R point.
  • the regression model used in the embodiment of the present invention is a random forest algorithm model.
  • the random forest algorithm model is a classifier model that contains multiple decision trees, and the output category is determined by the total number of categories output by each decision tree. To explain from an intuitive perspective, each decision tree is a classifier, so for an input sample, multiple trees will have multiple classification probabilities; integrate all the classification probabilities and specify the category with the highest probability as the final output result.
  • the random forest algorithm is used to perform regression classification calculations on multiple input eigen The predicted value of diastolic blood pressure and predicted value of systolic blood pressure.
  • Fig. 1 is a schematic diagram of a method for predicting blood pressure according to Embodiment 1 of the present invention. The method mainly includes the following steps:
  • Step 1 Synchronize the tester's electrocardiographic physiological signal and pulse physiological signal acquisition to generate the electrocardiographic signal and pulse physiological signal; perform signal sampling processing on the electrocardiographic signal and pulse physiological signal according to the preset sampling frequency threshold to generate the electrocardiogram ECG signal PPG signal with photoplethysmography method;
  • Step 11 Collect the ECG physiological signal of the tester to generate an ECG signal whose length is a fixed duration threshold. Synchronously, collect the pulse physiological signal of the tester to generate a pulse physiological signal whose length is the fixed duration threshold;
  • Step 12 Perform signal sampling on the ECG signal according to the sampling frequency threshold to generate an ECG signal; the ECG signal includes a plurality of ECG signal points;
  • ECG signal points include signal point amplitude data and signal point time data
  • Step 13 Perform signal sampling on the pulse physiological signal according to the sampling frequency threshold to generate the PPG original signal, and perform band-pass filtering on the PPG original signal according to the preset band-pass frequency threshold range to generate the PPG signal;
  • the PPG signal includes multiple PPG signal points; the PPG signal point includes signal point amplitude data and signal point time data.
  • the signals collected in these two segments must be collected synchronously, with the same length and the same subsequent sampling frequency.
  • Step 2 Perform the R point time feature recognition operation on the ECG signal to generate the R point time series
  • the R point time series includes multiple R point times
  • Step 21 Extract the signal point time data of the ECG signal points in sequence for the ECG signal to generate an ECG one-dimensional data vector; perform data segment division operations on the ECG one-dimensional data vector according to a preset ECG segment length threshold to generate multiple ECG one-dimensional fragment vector;
  • Step 23 Sort all the identified R point times in order to generate an R point time series.
  • the mean square error can be used to sequentially poll the decision-making extraction method
  • the signal can also be converted to the time domain and frequency domain to extract the maximum energy value as the R point
  • the convolutional network feature can also be used Extraction processing methods, etc.
  • the ECG signal is divided into segments and sub-segments in order to further refine the extraction module and improve the effective accuracy of extraction and recognition.
  • Step 3 Perform pulse wave peak point time feature recognition operation and pulse wave valley point time feature recognition operation on the PPG signal to generate peak point time series and valley point time series;
  • the peak point time series includes multiple peak point times;
  • the valley point time series includes multiple valley point times;
  • Step 31 by configuring the reference amplitude of the signal point and the absolute refractory period time width, perform the pulse wave peak point time feature identification operation on the PPG signal to generate the peak point time series;
  • step 311 initialize the peak point time series to be empty; set the waveform falling edge flag to 0; obtain the preset peak calibration factor; perform the full signal standard deviation calculation on the PPG signal to generate the standard deviation factor;
  • the two calculation factors here are used for the subsequent calculation of the reference amplitude of the signal point
  • Step 312 In the PPG signal, from the signal point amplitude data of the first PPG signal point to the signal point amplitude data of the specified number of PPG signal points, extract the minimum value of the signal point reference amplitude data. initialization;
  • the initial value of the reference amplitude of the signal point is calculated using the 20 PPG waveforms at the beginning of the PPG signal.
  • the reference amplitude of the signal point is a change used for continuous determination of the PPG waveform Amplitude comparison value;
  • the basic principle of the comparison is: first use the lowest amplitude value in the PPG waveform of the specified number of PPG signals as the initial value;
  • the signal point reference amplitude data is set to the waveform amplitude corresponding to the current comparison time point for each comparison. Its characteristic is that during the rising edge process, the signal point reference amplitude data It must always be smaller than the amplitude data of the current signal point;
  • a new A old +B*(P+std)/f calculates the signal point reference amplitude data, where A new is the signal point reference amplitude data after reset; A old is the signal point reference amplitude data before reset Data; B is the peak calibration factor; P is the peak point amplitude data; std is the standard deviation factor; f is the sampling frequency threshold;
  • FIG. 2 is a schematic diagram of the change of the reference amplitude of the signal point provided by the embodiment of the present invention, it can be seen that as the signal point of the PPG signal progresses, the change trend of the reference amplitude data can be seen;
  • Step 313 Perform signal point traversal on the PPG signal from the specified number plus 1 PPG signal point to the last PPG signal point to generate the current PPG signal point;
  • Step 314 When the signal point amplitude data of the current PPG signal point is greater than the signal point reference amplitude, set the signal point reference amplitude to the signal point amplitude data of the current PPG signal point, and set the waveform falling edge flag to 0;
  • Step 315 when the signal point amplitude data of the current PPG signal point is less than the signal point reference amplitude and the waveform falling edge flag is 0, set the waveform falling edge flag to 1; extract the signal point amplitude data of the last PPG signal point to generate The current peak point amplitude, extract the signal point time data of the last PPG signal point to generate the current peak point time; obtain the absolute refractory period time width and generate the absolute refractory period based on the sum of the current peak point time plus the absolute refractory period time width Period end time; set the reference amplitude of the signal point to the amplitude of the current peak point; add the peak point time to the peak point time sequence from the current peak point time;
  • Step 316 When the signal point amplitude data of the current PPG signal point is less than the signal point reference amplitude and the waveform falling edge flag is 1, if the signal point time data of the current PPG signal point is less than or equal to the end time of the absolute refractory period, then Keep the value of the reference amplitude of the signal point unchanged;
  • the signal point reference amplitude is always equal to the peak point amplitude
  • a new is the signal point reference amplitude data after reset;
  • a old is the signal point reference amplitude data before reset;
  • B is the peak calibration factor;
  • P is the current peak point amplitude;
  • std is the standard deviation factor;
  • f is the sampling frequency threshold;
  • Step 32 Perform a pulse wave trough point time feature extraction operation on the PPG signal according to the peak point time series to generate a trough point time series;
  • Step 4 According to the peak point time series, the bottom point time series and the R point time series, with each R point time as the time reference point, in the PPG signal, extract the first bottom point time after the time reference point, The first peak point time and the second bottom point time; generate matching feature time groups according to the R point time, the first bottom point time, the first peak point time and the second bottom point time; and combine all matching feature time groups Sort in order to generate a sequence of matching feature time groups;
  • Step 41 Set the matching feature time group; initialize the matching R point time of the matching feature time group to be empty, initialize the matching PPG peak time of the matching feature time group to empty, and initialize the matching PPG start time of the matching feature time group as Empty, the matching PPG end time of the initial matching feature time group is empty;
  • Step 42 Extract the time of two adjacent R points in sequence from the R point time sequence to generate the first reference R point and the second reference R point; in the PPG signal, the bottom point time sequence starts with the first reference R point Time, with the second reference point R as the end time, search in the opposite direction from the end time to the start time, extract the lowest point time closest to the end time, generate the second valley point time, and extract the time distance from the second valley point The next most recent valley point time generates the first valley point time; in the PPG signal, for the peak point time series, the first reference R point is the start time, and the second reference R point is the end time from the end time to the end time. Search in the opposite direction at the start time, extract the peak point time closest to the end time to generate the first peak point time;
  • the first reference R point is smaller than the second reference R point
  • the extracted first valley point time, first peak point time, and second valley point time are the three characteristic points of a complete PPG waveform extracted, starting point (first valley point time), peak value Point (time of the first peak point) and end point (time of the second peak point); if there are multiple PPG waveforms between two adjacent R points, select the start of the last PPG waveform between the two R points , The peak value and the end information are used as a pairing, and the paired with the PPG adopts the previous R point of the two adjacent R points, that is, the first reference R point in the text;
  • Step 43 Set the matching R point time of the matching feature time group as the first reference R point, set the matching PPG peak time of the matching feature time group as the first peak point time, and set the matching PPG start time of the matching feature time group as the first peak point time.
  • One valley point time set the matching PPG end time of the matching feature time group to the second valley point time;
  • the four point information can be understood as a three-point (start, peak, end) PPG waveform and the information of the R point with the closest front end of the time axis.
  • start, peak, end the information of the R point with the closest front end of the time axis.
  • the embodiment of the present invention selects the last PPG waveform between the two R points as the two R points.
  • Step 44 Add the matched characteristic time group to the matched characteristic time group sequence with the successfully set matching characteristic time group.
  • Step 5 Perform the feature sample data preparation operation of the random forest algorithm model according to the matched feature time group sequence to generate a random forest sample group sequence;
  • the random forest sample group sequence includes multiple random forest sample groups
  • step 51 sequentially extracting matching feature time groups of the matching feature time group sequence to generate a current matching feature time group; extracting the next matching feature time group of the current matching feature time group to generate an adjacent matching feature time group;
  • Step 52 Extract the matching R point time of the current matching feature time group to generate the first R point time, extract the matching PPG peak time of the current matching feature time group to generate the peak time, and extract the matching PPG start time of the current matching feature time group. Start time, extract the matching PPG end time generation end time of the current matching feature time group; extract the matching R point time of the adjacent matching feature time group to generate the second R point time;
  • Step 53 In the PPG signal, extract the PPG signal waveform corresponding to the current matching feature time group to generate the current PPG waveform; calculate the area enclosed by the current PPG waveform and the horizontal axis of time from the start time to the end time to generate the current PPG Wave area S;
  • Step 54 Set the first sample parameter of the random forest sample group as the quotient of 60 divided by the time difference of the heartbeat,
  • the cardiac time difference is the absolute value of the time difference between the first R point time and the second R point time
  • FIG. 3 is a schematic diagram of ECG signals and PPG signals provided by an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of ECG signals and PPG signals provided by an embodiment of the present invention.
  • Step 56 Set the third sample parameter of the random forest sample group to the absolute value of the time difference between the first R point time and the peak time;
  • FIG. 3 is a schematic diagram of ECG signals and PPG signals provided by an embodiment of the present invention.
  • Step 57 Obtain the time point at the maximum rising edge slope of the current PPG waveform to generate the maximum rising slope time; set the fourth sample parameter of the random forest sample group to the absolute value of the time difference between the first R point time and the rising maximum slope time;
  • FIG. 3 is a schematic diagram of ECG signals and PPG signals provided by an embodiment of the present invention.
  • Step 58 Obtain the signal amplitude data corresponding to the peak time in the current PPG waveform to generate the peak amplitude, and set the fifth sample parameter of the random forest sample group to the peak amplitude;
  • FIG. 3 is a schematic diagram of ECG signals and PPG signals provided by an embodiment of the present invention.
  • Step 59 Obtain the signal amplitude data corresponding to the start time in the current PPG waveform to generate the start amplitude, and set the sixth sample parameter of the random forest sample group as the ratio of the peak amplitude to the start amplitude;
  • Step 60 Set the seventh sample parameter of the random forest sample group as the absolute value of the amplitude difference between the peak amplitude and the initial amplitude;
  • FIG. 3 is a schematic diagram of ECG signals and PPG signals provided by an embodiment of the present invention.
  • Step 61 Set the eighth sample parameter of the random forest sample group as the absolute value of the slope of the line from the amplitude point corresponding to the peak time to the amplitude point corresponding to the start time in the current PPG waveform;
  • Step 62 Set the ninth sample parameter of the random forest sample group as the absolute value of the slope of the line from the amplitude point corresponding to the peak time to the amplitude point corresponding to the end time in the current PPG waveform;
  • Step 63 Set the tenth sample parameter of the random forest sample group as the area enclosed by the current PPG waveform and the horizontal axis of time from the start time to the peak time;
  • FIG. 3 is a schematic diagram of ECG signals and PPG signals provided by an embodiment of the present invention.
  • Step 64 Set the tenth sample parameter of the random forest sample group as the area enclosed by the current PPG waveform and the horizontal axis of time from the end time to the peak time;
  • FIG. 3 is a schematic diagram of ECG signals and PPG signals provided by an embodiment of the present invention.
  • Step 65 Set the twelfth sample parameter of the random forest sample group; offset the twelfth sample parameter backward from the start time, and the area enclosed by the current PPG waveform and the horizontal axis of time is 0.1*S;
  • Step 66 Set the thirteenth sample parameter of the random forest sample group; offset the thirteenth sample parameter backward from the start time, and the area enclosed by the current PPG waveform and the horizontal axis of time is 0.2*S;
  • Step 67 Set the fourteenth sample parameter of the random forest sample group; offset the fourteenth sample parameter backward from the start time, and the area enclosed by the current PPG waveform and the horizontal axis of time is 0.3*S;
  • Step 68 Set the fifteenth sample parameter of the random forest sample group; offset the fifteenth sample parameter backward from the start time, and the area enclosed by the current PPG waveform and the horizontal axis of time is 0.4*S;
  • Step 69 Set the sixteenth sample parameter of the random forest sample group; offset the sixteenth sample parameter backward from the start time, and the area enclosed by the current PPG waveform and the horizontal axis of time is 0.5*S;
  • Step 70 Set the seventeenth sample parameter of the random forest sample group; offset the seventeenth sample parameter backward from the start time, and the area enclosed by the current PPG waveform and the horizontal axis of time is 0.6*S;
  • Step 71 Set the eighteenth sample parameter of the random forest sample group; offset the eighteenth sample parameter backward from the start time, and the area enclosed by the current PPG waveform and the horizontal axis of time is 0.7*S;
  • Step 72 Set the nineteenth sample parameter of the random forest sample group; offset the nineteenth sample parameter backward from the start time, and the area enclosed by the current PPG waveform and the horizontal axis of time is 0.8*S;
  • Step 73 Set the twentieth sample parameter of the random forest sample group; offset the twentieth sample parameter backward from the start time, and the area enclosed by the current PPG waveform and the horizontal axis of time is 0.9*S;
  • Step 74 Set the twenty-first sample parameter of the random forest sample group as the absolute value of the time difference between the end time and the start time.
  • the random forest sample group sequence of the random forest model is set.
  • the random forest sample group sequence includes multiple random forest sample groups.
  • Each random forest sample group includes a total of 21 sample parameters, which are divided into six major groups.
  • Types heart rate parameters, time parameters, amplitude parameters, slope parameters, area parameters, area offset time parameters; the corresponding relationship is as follows: heart rate parameters include the first sample parameter; time parameters include the second, third, and fourth sample parameters; The amplitude parameters include the fifth, sixth, and seventh sample parameters; the slope parameters include the eighth and ninth sample parameters; the area parameters include the tenth and eleventh sample parameters; the area offset time parameters include the twelfth to the twentieth sample This parameter.
  • Step 6 According to the R point time series, calculate the corresponding R point instantaneous heart rate, R point trend heart rate and R snack rate difference; and use whether the R point rate difference is less than the preset reasonable heart rate difference threshold as the abnormal sample group judgment condition.
  • the sequence of the forest sample group is processed for the deletion of the abnormal sample group.
  • Step 7 Use the random forest algorithm model to perform regression prediction calculation on the random forest sample group sequence to generate a predicted blood pressure array
  • the predicted blood pressure array includes systolic blood pressure data and diastolic blood pressure data.
  • the random forest algorithm model is a model that has been trained through batch ECG/PPG signals and the corresponding measured blood pressure values.
  • the predicted blood pressure data sequence output by the random forest algorithm model includes two values, namely: prediction Systolic blood pressure data and predicted diastolic blood pressure data.
  • Fig. 4 is a schematic diagram of an abnormal random forest sample group processing method provided in the second embodiment of the present invention. The method mainly includes the following steps:
  • Step 401 Obtain the R point time sequence and the random forest sample sequence from the upper application.
  • the random forest sample group sequence includes multiple random forest sample groups.
  • Step 402 initialize the instantaneous heart rate sequence to be empty; obtain the total number of R point times included in the R point time sequence to generate the total number of R points;
  • Step 403 Extract the R point time of the R point time series in turn to generate the current R point, and extract the R point time adjacent to the current R point according to the specified adjacent point extraction direction to generate the adjacent R point;
  • the absolute value of the time difference between adjacent R points generates the first factor; generates the R point instantaneous heart rate according to the reciprocal of the first factor; adds the R point instantaneous heart rate to the instantaneous heart rate sequence;
  • the instantaneous heart rate sequence includes the total number of R points and the instantaneous heart rate of R points;
  • the instantaneous heart rate calculation method is the reciprocal of the two adjacent R-R intervals of the ECG, and the instantaneous heart rate of each R point is combined to generate the instantaneous heart rate sequence, which is the instantaneous heart rate sequence of the full ECG segment;
  • Step 404 Gaussian filtering is performed on the instantaneous heart rate sequence according to a preset filtering standard deviation to generate a trending heart rate sequence;
  • the trend heart rate sequence includes the total number of R points and the trend heart rate of R points;
  • Gaussian filtering is a linear smoothing filter, which is suitable for eliminating Gaussian noise and is widely used in the denoising process of image processing.
  • smoothing and denoising trend processing of the instantaneous heart rate sequence is the smoothing and denoising trend processing of the instantaneous heart rate sequence
  • Step 405 Initialize the value of the first index to 1, and initialize the value of the first total to the total number of R points;
  • Step 406 Extract the R-point instantaneous heart rate corresponding to the first index from the instantaneous heart rate sequence to generate the first indexed instantaneous heart rate; extract the R-point trend heart rate corresponding to the first index from the trend heart rate sequence to generate the first index trend heart rate;
  • a heart rate difference can be obtained by subtracting the instantaneous heart rate and trend heart rate and taking the absolute value of the result.
  • the heart rate difference will be within a reasonable error range, if it is noise or The heart rate difference of the interference signal will definitely exceed the error range;
  • Step 407 Generate the first index R snack rate difference according to the absolute value of the heart rate difference between the first index instant heart rate and the first index trend heart rate;
  • Step 408 When the first index R is greater than the reasonable heart rate difference threshold, mark the random forest sample group corresponding to the first index as an abnormal sample group;
  • the reasonable heart rate difference threshold is the error range mentioned above, the R point whose heart rate exceeds the error range is further regarded as noise, and the random forest sample group in the corresponding random forest sample group sequence is also regarded as an abnormal sample;
  • Step 409 Add 1 to the first index
  • Step 410 Determine whether the first index is greater than the first total, if the first index is greater than the first total, go to step 411, and if the first index is less than or equal to the first total, go to step 406;
  • Step 411 poll the random forest sample group sequence, and delete the random forest sample group marked as an abnormal sample group from the random forest sample group sequence.
  • the abnormal sample group in the random forest sample group sequence that is unqualified is also fully marked.
  • the forest sample group is eliminated from the random forest sample group sequence.
  • FIG. 5 is a schematic diagram of a device structure of a device for predicting blood pressure according to Embodiment 3 of the present invention.
  • the device includes a processor and a memory.
  • the memory can be connected to the processor through a bus.
  • the memory may be a non-volatile memory, such as a hard disk drive and a flash memory, and a software program and a device driver program are stored in the memory.
  • the software program can execute various functions of the foregoing method provided by the embodiments of the present invention; the device driver may be a network and interface driver.
  • the processor is used to execute a software program, and when the software program is executed, the method provided in the embodiment of the present invention can be implemented.
  • the embodiment of the present invention also provides a computer-readable storage medium.
  • a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the method provided in the embodiment of the present invention can be implemented.
  • the embodiment of the present invention also provides a computer program product containing instructions.
  • the processor is caused to execute the above method.
  • the method and device for predicting blood pressure provided by the embodiment of the present invention. Firstly, perform synchronous acquisition of the ECG signal and PPG signal of the tester; secondly, perform feature extraction on the acquired ECG signal and PPG signal: the acquired ECG The characteristic of the R point of the signal is obtained, and the pulse wave peak point and the starting and ending point characteristics of the PPG signal are obtained; then, the respective characteristic data are matched: the characteristic points of the ECG signal and the PPG signal are correlated to generate a matching characteristic group; then, according to the matched The ECG signal and PPG signal are fused to generate sample data: based on each matching feature group, the related feature parameters are set; finally, the full-time matching feature group sequence is used as the input of the random forest algorithm model for blood pressure regression calculation to generate predictions Blood pressure array (including diastolic and systolic blood pressure).
  • Blood pressure array including diastolic and systolic blood pressure
  • the ECG and PPG signals can be automatically and continuously analyzed and predicted by cooperating with the acquisition sensor, thereby not only improving the comfort of the tester, but also establishing A way to automatically monitor blood pressure.
  • the steps of the method or algorithm described in combination with the embodiments disclosed herein can be implemented by hardware, a software module executed by a processor, or a combination of the two.
  • the software module can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or all areas in the technical field. Any other known storage media.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Animal Behavior & Ethology (AREA)
  • Veterinary Medicine (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Pathology (AREA)
  • Physiology (AREA)
  • Cardiology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Psychiatry (AREA)
  • Signal Processing (AREA)
  • Vascular Medicine (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
  • Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)

Abstract

一种对血压进行预测的方法和装置，方法包括：获取同步的ECG、PPG信号(S1)；对ECG信号进行R点特征识别生成R点时间序列(S2)；对PPG信号进行峰值点、谷值点特征识别生成峰值点、谷值点时间序列(S3)；对应每个R点时间在PPG信号中提取第一谷值点、第一峰值点和第二谷值点时间，并组成匹配特征时间组(S4)；根据匹配特征时间组序列进行随机森林算法模型的特征样本数据准备，生成随机森林样本组序列(S5)；计算R点心率差并以R点心率差是否满足阈值作为异常样本组判定条件对随机森林样本组序列进行异常样本组删除处理(S6)；利用随机森林算法模型对随机森林样本组序列进行回归预测计算生成预测血压数组(S7)。

Description

一种对血压进行预测的方法和装置
本申请要求于2020年2月21日提交中国专利局、申请号为202010110286.4、发明名称为“一种对血压进行预测的方法和装置”的中国专利申请的优先权。
技术领域
本发明涉及电生理信号处理技术领域，特别涉及一种对血压进行预测的方法和装置。
背景技术
心脏是人体血液循环的中心，心脏通过有规律的搏动产生血压，进而向全身供血完成人体的新陈代谢，血压是人体非常重要的生理信号之一。当今，高血压发病率越来越高，严重危害人体健康，大量流行病学及临床证据表明，长期患有高血压病会增加患者发生缺血性心脏病、脑卒中、肾衰竭、主动脉和外周动脉疾病等靶器官损害的风险。高血压病属于慢性疾病，多数需要长期终身护理，而对高血压患者生活方式控制的成效，降压药物的药效及高血压的介入治疗的功效评估，都需要对血压进行长时间动态监测。在日常生活中，目前最常用的血压测量设备是电子血压计，在使用电子血压计的过程中，需要对被测者施加压力，这种方法操作繁琐、不能连续监测，而且容易对被测者造成不适。
发明内容
本发明的目的，就是针对现有技术的缺陷，提供一种对血压进行预测的方法和装置,对测试者进行心电图(Electrocardiogram,ECG)信号和光体积变化描记图法(Photoplethysmography,PPG)信号的同步采集，对获取的ECG信 号与PPG信号进行特征提取，再对各自的特征数据进行匹配，继而根据匹配的ECG信号和PPG信号进行融合生成样本数据并将样本数据利用随机森林算法模型进行预测计算，最终获得血压预测值。通过本发明实施例，无需对测试者进行压力测试或者干预介入式测试，通过与采集传感器配合可以自动持续对ECG和PPG信号进行分析和预测，由此既提高了测试者的舒适度又建立了一种自动监测血压的途径。
为实现上述目的，本发明实施例第一方面提供了一种对血压进行预测的方法，所述方法包括：
同步对测试者进行心电生理信号和脉搏生理信号采集生成心电信号和脉搏生理信号；并对所述心电信号和所述脉搏生理信号按预置的采样频率阈值进行信号采样处理生成心电图ECG信号和光体积变化描记图法PPG信号；
对所述ECG信号进行R点时间特征识别操作生成R点时间序列；所述R点时间序列包括多个R点时间；
对所述PPG信号进行脉搏波峰值点时间特征识别操作和脉搏波谷值点时间特征识别操作，生成峰值点时间序列和谷值点时间序列；所述峰值点时间序列包括多个峰值点时间；所述谷值点时间序列包括多个谷值点时间；
根据所述峰值点时间序列、所述谷值点时间序列和所述R点时间序列，以每个所述R点时间为时间参考点，在所述PPG信号中，提取在所述时间参考点之后的第一谷值点时间、第一峰值点时间和第二谷值点时间；根据所述R点时间、所述第一谷值点时间、所述第一峰值点时间和所述第二谷值点时间生成匹配特征时间组；并将所有所述匹配特征时间组按先后顺序进行排序生成匹配特征时间组序列；
根据所述匹配特征时间组序列，进行随机森林算法模型的特征样本数据准备操作，生成随机森林样本组序列；所述随机森林样本组序列包括多个随机森林样本组；
根据所述R点时间序列，计算对应的R点瞬时心率、R点趋势心率和R点 心率差；并以所述R点心率差是否小于预置的合理心率差阈值作为异常样本组判定条件对所述随机森林样本组序列进行异常样本组删除处理；
利用所述随机森林算法模型对所述随机森林样本组序列进行回归预测计算生成预测血压数组；所述预测血压数组包括收缩压数据和舒张压数据。
优选的，所述同步对测试者进行心电生理信号和脉搏生理信号采集生成心电信号和脉搏生理信号；并对所述心电信号和所述脉搏生理信号按预置的采样频率阈值进行信号采样处理生成心电图ECG信号和光体积变化描记图法PPG信号，具体包括：
对所述测试者进行心电生理信号采集生成一段长度为固定时长阈值的所述心电信号，同步的，对所述测试者进行脉搏生理信号采集生成一段长度为所述固定时长阈值的所述脉搏生理信号；
按所述采样频率阈值对所述心电信号进行信号采样生成所述ECG信号；所述ECG信号包括多个ECG信号点；所述ECG信号点包括信号点幅值数据和信号点时间数据；
按所述采样频率阈值对所述脉搏生理信号进行信号采样生成PPG原始信号，并根据预置的带通频率阈值范围对所述PPG原始信号进行带通滤波处理生成所述PPG信号；所述PPG信号包括多个PPG信号点；所述PPG信号点包括信号点幅值数据和信号点时间数据。
优选的，所述对所述ECG信号进行R点时间特征识别操作生成R点时间序列，具体包括：
对所述ECG信号，依次提取所述ECG信号点的信号点时间数据，生成ECG一维数据向量；按预置的ECG片段长度阈值对所述ECG一维数据向量进行数据片段划分操作生成多个ECG一维片段向量；
以所述ECG一维片段向量作为R点时间特征识别算法的输入，利用指定的R点时间特征识别算法，识别出R点在所述ECG一维片段向量内的相对时间位移信息T 1；并根据所述ECG一维片段向量的起始ECG信号点的信号点时 间数据T 2获得所述R点时间，R点时间＝T 2+T 1；
将识别出的所有所述R点时间，按先后顺序排序生成所述R点时间序列。
优选的，所述对所述PPG信号进行脉搏波峰值点时间特征识别操作和脉搏波谷值点时间特征识别操作，生成峰值点时间序列和谷值点时间序列，具体包括：
通过配置信号点参考幅值和绝对不应期时间宽度，对所述PPG信号进行脉搏波峰值点时间特征识别操作，生成所述峰值点时间序列；
根据所述峰值点时间序列，对所述PPG信号进行所述脉搏波谷值点时间特征提取操作生成所述谷值点时间序列。
进一步的，所述通过配置信号点参考幅值和绝对不应期时间宽度，对所述PPG信号进行脉搏波峰值点时间特征识别操作，生成所述峰值点时间序列，具体包括：
初始化所述峰值点时间序列为空；设置波形下降沿标志为0；获取预置的峰值校准因子；对所述PPG信号进行全信号标准偏差计算生成标准偏差因子；
在所述PPG信号中，从第1个PPG信号点的信号点幅值数据开始，到指定数目个PPG信号点的信号点幅值数据为止，提取其中的最小值对所述信号点参考幅值进行初始化；
对所述PPG信号从指定数目加1个PPG信号点开始到最后1个PPG信号点进行信号点遍历生成当前PPG信号点；
当所述当前PPG信号点的信号点幅值数据大于所述信号点参考幅值时，设置所述信号点参考幅值为所述当前PPG信号点的信号点幅值数据，设置所述波形下降沿标志为0；
当所述当前PPG信号点的信号点幅值数据小于所述信号点参考幅值且所述波形下降沿标志为0时，设置所述波形下降沿标志为1；提取上一个PPG信号点的信号点幅值数据生成当前峰值点幅值，提取上一个PPG信号点的信号点时间数据生成当前峰值点时间；获取所述绝对不应期时间宽度并根据所述 当前峰值点时间加上所述绝对不应期时间宽度的和生成绝对不应期结束时间；设置所述信号点参考幅值为所述当前峰值点幅值；将所述当前峰值点时间向所述峰值点时间序列进行峰值点时间添加操作；
当所述当前PPG信号点的信号点幅值数据小于所述信号点参考幅值且所述波形下降沿标志为1时，如果所述当前PPG信号点的信号点时间数据小于或等于所述绝对不应期结束时间，则保持所述信号点参考幅值的取值不变；
当所述当前PPG信号点的信号点幅值数据小于所述信号点参考幅值且所述波形下降沿标志为1时，如果所述当前PPG信号点的信号点时间数据大于所述绝对不应期结束时间，则根据公式A new＝A old+B*(P+std)/f对所述信号点参考幅值进行重置；所述A new为重置后的信号点参考幅值数据；所述A old为重置前的信号点参考幅值数据；所述B为所述峰值校准因子；所述P为所述当前峰值点幅值；所述std为所述标准偏差因子；所述f为所述采样频率阈值。
进一步的，所述根据所述峰值点时间序列，对所述PPG信号进行所述脉搏波谷值点时间特征提取操作生成所述谷值点时间序列，具体包括：
根据所述峰值点时间序列，在所述PPG信号中，两个相邻峰值点时间之间，提取所述信号点幅值数据为最小值的所述PPG信号点的所述信号点时间数据，生成所述谷值点时间；将提取出的所有所述谷值点时间按先后顺序对对所述谷值点时间序列进行谷值点时间添加操作。
优选的，所述根据所述峰值点时间序列、所述谷值点时间序列和所述R点时间序列，以每个所述R点时间为时间参考点，在所述PPG信号中，提取在所述时间参考点之后的第一谷值点时间、第一峰值点时间和第二谷值点时间；根据所述R点时间、所述第一谷值点时间、所述第一峰值点时间和所述第二谷值点时间生成匹配特征时间组；并将所有所述匹配特征时间组按先后顺序进行排序生成匹配特征时间组序列，具体包括：
设置所述匹配特征时间组；初始化所述匹配特征时间组的匹配R点时间为空，初始化所述匹配特征时间组的匹配PPG峰值时间为空，初始化所述匹 配特征时间组的匹配PPG起始时间为空，初始化所述匹配特征时间组的匹配PPG结束时间为空；
从所述R点时间序列依次提取两个相邻所述R点时间生成第一参考R点和第二参考R点；所述第一参考R点小于所述第二参考R点；在所述PPG信号中，对所述谷值点时间序列以所述第一参考R点为起始时间、以所述第二参考R点为结束时间从结束时间向起始时间进行反方向查找，提取距离结束时间最近的所述谷值点时间生成所述第二谷值点时间，提取与所述第二谷值点时间距离最近的下一个所述谷值点时间生成所述第一谷值点时间；在所述PPG信号中，对所述峰值点时间序列以所述第一参考R点为起始时间、以所述第二参考R点为结束时间从结束时间向起始时间进行反方向查找，提取距离结束时间最近的所述峰值点时间生成所述第一峰值点时间；
设置所述匹配特征时间组的所述匹配R点时间为所述第一参考R点，设置所述匹配特征时间组的所述匹配PPG峰值时间为所述第一峰值点时间，设置所述匹配特征时间组的所述匹配PPG起始时间为所述第一谷值点时间，设置所述匹配特征时间组的所述匹配PPG结束时间为所述第二谷值点时间；
将设置成功的所述匹配特征时间组向所述匹配特征时间组序列进行匹配特征时间组添加操作。
优选的，所述根据所述匹配特征时间组序列，进行随机森林算法模型的特征样本数据准备操作，生成随机森林样本组序列，具体包括：
依次提取所述匹配特征时间组序列的所述匹配特征时间组生成当前匹配特征时间组；提取所述当前匹配特征时间组的下一个匹配特征时间组生成相邻匹配特征时间组；
提取所述当前匹配特征时间组的所述匹配R点时间生成第一R点时间，提取所述当前匹配特征时间组的所述匹配PPG峰值时间生成峰值时间，提取所述当前匹配特征时间组的所述匹配PPG起始时间生成起始时间，提取所述当前匹配特征时间组的所述匹配PPG结束时间生成结束时间；提取所述相邻 匹配特征时间组的所述匹配R点时间生成第二R点时间；
在所述PPG信号中，提取与所述当前匹配特征时间组对应的PPG信号波形生成当前PPG波形；计算从所述起始时间到所述结束时间之间由所述当前PPG波形与时间横轴围成的面积生成当前PPG波形面积S；
设置所述随机森林样本组的第一样本参数为60除以心动时差的商，所述心动时差为所述第一R点时间与所述第二R点时间的时间差绝对值；
设置所述随机森林样本组的第二样本参数为所述第一R点时间与所述起始时间的时间差绝对值；
设置所述随机森林样本组的第三样本参数为所述第一R点时间与所述峰值时间的时间差绝对值；
获取所述当前PPG波形的上升沿斜率最大值处的时间点生成上升最大斜率时间；设置所述随机森林样本组的第四样本参数为所述第一R点时间与所述上升最大斜率时间的时间差绝对值；
获取所述当前PPG波形中与所述峰值时间对应的信号幅值数据生成峰值幅值，设置所述随机森林样本组的第五样本参数为所述峰值幅值；
获取所述当前PPG波形中与所述起始时间对应的信号幅值数据生成起始幅值，设置所述随机森林样本组的第六样本参数为所述峰值幅值与所述起始幅值的比值；
设置所述随机森林样本组的第七样本参数为所述峰值幅值与所述起始幅的幅值差绝对值；
设置所述随机森林样本组的第八样本参数为所述当前PPG波形中从所述峰值时间对应的幅值点到所述起始时间对应的幅值点之间连线的斜率绝对值；
设置所述随机森林样本组的第九样本参数为所述当前PPG波形中从所述峰值时间对应的幅值点到所述结束时间对应的幅值点之间连线的斜率绝对值；
设置所述随机森林样本组的第十样本参数为从所述起始时间到所述峰值时间之间由所述当前PPG波形与时间横轴围成的面积；
设置所述随机森林样本组的第十一样本参数为从所述结束时间到所述峰值时间之间由所述当前PPG波形与时间横轴围成的面积；
设置所述随机森林样本组的第十二样本参数；从所述起始时间起向后偏移所述第十二样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.1*S；
设置所述随机森林样本组的第十三样本参数；从所述起始时间起向后偏移所述第十三样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.2*S；
设置所述随机森林样本组的第十四样本参数；从所述起始时间起向后偏移所述第十四样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.3*S；
设置所述随机森林样本组的第十五样本参数；从所述起始时间起向后偏移所述第十五样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.4*S；
设置所述随机森林样本组的第十六样本参数；从所述起始时间起向后偏移所述第十六样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.5*S；
设置所述随机森林样本组的第十七样本参数；从所述起始时间起向后偏移所述第十七样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.6*S；
设置所述随机森林样本组的第十八样本参数；从所述起始时间起向后偏移所述第十八样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.7*S；
设置所述随机森林样本组的第十九样本参数；从所述起始时间起向后偏移所述第十九样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.8*S；
设置所述随机森林样本组的第二十样本参数；从所述起始时间起向后偏移所述第二十样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.9*S；
设置所述随机森林样本组的第二十一样本参数为所述结束时间与所述起始时间的时间差绝对值。
优选的，所述根据所述R点时间序列，计算对应的R点瞬时心率、R点趋势心率和R点心率差；并以所述R点心率差是否小于预置的合理心率差阈值作为异常样本组判定条件对所述随机森林样本组序列进行异常样本组删除处理，具体包括：
步骤91，初始化瞬时心率序列为空；获取所述R点时间序列包括的所述R点时间的总数生成R点总数；
步骤92，依次提取所述R点时间序列的所述R点时间生成当前R点，并按指定的相邻点提取方向提取与所述当前R点相邻的所述R点时间生成相邻R点；根据所述当前R点与所述相邻R点的时间差绝对值生成第一因子；根据所述第一因子的倒数生成所述R点瞬时心率；将所述R点瞬时心率向所述瞬时心率序列进行R点瞬时心率添加操作；所述瞬时心率序列包括所述R点总数个所述R点瞬时心率；
步骤93，对所述瞬时心率序列按预置的滤波标准差进行高斯滤波生成趋势心率序列；所述趋势心率序列包括所述R点总数个所述R点趋势心率；
步骤94，初始化第一索引的值为1，初始化第一总数的值为所述R点总数；
步骤95，从所述瞬时心率序列中提取与所述第一索引对应的所述R点瞬时心率生成第一索引瞬时心率；从所述趋势心率序列中提取与所述第一索引对应的所述R点趋势心率生成第一索引趋势心率；
步骤96，根据所述第一索引瞬时心率与所述第一索引趋势心率的心率差绝对值生成第一索引R点心率差；
步骤97，当所述第一索引R点心率差大于所述合理心率差阈值时，将与所述第一索引对应的所述随机森林样本组标记为异常样本组；
步骤98，将所述第一索引加1；
步骤99，判断所述第一索引是否大于所述第一总数，如果所述第一索引大于所述第一总数则转至步骤100，如果所述第一索引小于或等于所述第一总数则转至步骤95；
步骤100，轮询所述随机森林样本组序列，将标记为所述异常样本组的所述随机森林样本组从所述随机森林样本组序列中删除。
本发明实施例第一方面提供的一种对血压进行预测的方法，首先，对测试者进行心电图ECG信号和PPG信号的同步采集；其次，对获取的ECG信号与PPG信号进行特征提取：获取的ECG信号的R点特征，获取PPG信号脉搏波峰值点与起止点特征；然后，对各自的特征数据进行匹配：对ECG信号与PPG信号的特征点建立关联性生成匹配特征组；接着，根据匹配的ECG信号和PPG信号进行融合生成样本数据：基于每一个匹配特征组进行与之相关的特征参数设置；最后，将代表全时长的匹配特征组序列作为随机森林算法模型的输入进行血压回归计算生成预测血压数组(包括舒张压和收缩压)。
本发明实施例第二方面提供了一种设备，该设备包括存储器和处理器，存储器用于存储程序，处理器用于执行第一方面及第一方面的各实现方式中的方法。
本发明实施例第三方面提供了一种包含指令的计算机程序产品，当计算机程序产品在计算机上运行时，使得计算机执行第一方面及第一方面的各实现方式中的方法。
本发明实施例第四方面提供了一种计算机可读存储介质，计算机可读存储介质上存储有计算机程序，计算机程序被处理器执行时实现第一方面及第一方面的各实现方式中的方法。
附图说明
图1为本发明实施例一提供的一种对血压进行预测的方法示意图；
图2为本发明实施例提供的信号点参考幅值变化示意图；
图3为本发明实施例提供的ECG信号与PPG信号示意图；
图4为本发明实施例二提供的一种异常随机森林样本组处理方法示意图；
图5为本发明实施例三提供的一种对血压进行预测的装置的设备结构示意图。
具体实施方式
为了使本发明的目的、技术方案和优点更加清楚，下面将结合附图对本发明作进一步地详细描述，显然，所描述的实施例仅仅是本发明一部份实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其它实施例，都属于本发明保护的范围。
在通过实施例对本发明做进一步详细阐述之前，先就文中提及的一些技术做下简要说明。
我们已知脉搏是心脏射血时血液对动脉血管产生的压力变化造成的，因此脉搏、心动都与血压是具有关联特性的。本发明实施例使用ECG信号作为心动参考数据，使用PPG信号作为脉搏参考数据。
ECG信号是一组利用心电信号采集设备从体表记录采集的心脏心动周期的电生理信号。常规ECG信号波形有5个特征点，分别成为P、Q、R、S、T点。在实际操作中，除R点外，其他四点因为受噪声信号干扰的几率偏高，在信号滤波降噪过程中被误消除的几率也较高。一旦某个心搏信号的P点或T点出现误消除，当前心动信号就不会被纳入分析信号范畴，容易出现特征遗漏与特征偏差的问题。本发明实施例以5点中最强信号R点信号作为心动信号特征点，可以保留当前心电数据中最大数目的有效心动信号数据，解决了常规方法中丢失心搏数据的问题。
PPG信号是利用光感传感器对特定光源的光强识别记录光强变化的一组信号。在心脏搏动时，对血管内单位面积的血流量形成周期性变化，与之对应的血液体积也相应发生变化，从而导致反映血液吸收光量的PPG信号也呈现周期性变化趋势。一个心动周期包括两个时间期：心脏收缩期和心脏舒张期；当心脏收缩期时，心脏对去全身做功，造成血管内压力与血流体积产生连续周期性变化，此时血管内血液对光线的吸收最多；当心脏舒张期时，对血管的压力相对性较小，此时上一次心脏收缩向全身推出的血液经过循环撞击心脏瓣膜从而对光线产生一定的反射与折射效应，造成舒张周期时血管内血液对光线能量的吸收降低。因此，反映血管内血液吸收光能的PPG信号波形的时间特性就有两段时间特性：心脏收缩时期信号时间特性和心脏舒张时期信号时间特性；常见的PPG信号波形中最大峰值前认为是典型收缩期时间，而后的时间区域被认为是舒张时间，在最大峰值之后设置一个绝对不应期时间段用作标称典型收缩期到典型舒张期的过渡期。PPG原始信号(对最初采集的脉搏生理信号进行采样之后生成的信号)中，存在较多的噪声与干扰源，那么在采集之后需要对PPG信号进行一定的滤波降噪转换，将转换后的信号我们视为基本能够正常体现测试者脉搏波动周期特性的PPG信号。
因为心动是首先产生心脏搏动，然后导致血压变化，继而影响PPG信号波动，所以针对一个心动的时间特性，R点时间是在对应周期内的PPG信号起始时间之前的。
上述的ECG信号的R点时间对一次血压变化我们可以视为动因参考信号，PPG信号的PPG波形(由PPG起始时间、PPG峰值时间、PPG结束时间)我们可以视为结果参考信号。将二者的时间特性进行匹配是为了对每一次心动的选择一个对应的成因波形与结果波形，实际操作就是在PPG信号中选择与R点信号对应的一个PPG波形完成匹配。常规状态下，每两次心动之间(两个R点之间)，只有一个完整的PPG波形(一次脉动信号)；但在人们情绪激动或者运动状态下，由实际监测可知，每两次心动之间(两个R点之间)，可能有多 个PPG波形存在，在这种情况下，本实施例的匹配原则是，只选取两个R点之间(第一R点和第二R点)距离下一次心动(第二R点)最近的PPG波形作为当次心动的(第一R点)匹配对象。
对ECG信号与PPG信号完成心动数据匹配之后，就要根据他们的时序关系进行特征融合处理，对应本发明实施例就是进行随机森林算法模型的特征样本数据准备操作。特征样本数据以随机森林样本组的形式总共包括21个样本参数，主要归结为六大类：心率参数，时间参数、幅值参数、斜率参数、面积参数、面积偏移时间参数，具体分类与定义详见下表：
Figure PCTCN2020129632-appb-000001
Figure PCTCN2020129632-appb-000002
表一
对ECG信号与PPG信号完成心动数据融合生成多个随机森林样本组之后，为保证随机森林样本组中异常样本组的占比在合理范围之内，需要对随机森 林样本组做一次筛查。在之前的特征匹配和当前的样本参数计算时，都是基于一个原则：视R点为正常心动时间点；而在实际数据收集过程中，R点也存在异常R点。此处的筛查就是根据R点的心率差概念，将异常R点进行定位，并将与异常R点对应的样本组进行剔除。此处R点的心率差，是R点对应的瞬时心率和趋势心率的心率差绝对值。
在对随机森林样本组筛查完成之后，就需要使用回归分类模型对随机森林样本组进行回归分类计算。本发明实施例使用的回归模型是随机森林算法模型。随机森林算法模型是一个包含多个决策树的分类器模型，其输出的类别是由每个决策树输出的类别的总数而定。从直观角度来解释，每棵决策树都是一个分类器，那么对于一个输入样本，多棵树会有多个分类概率；将所有的分类概率进行集成并将概率最大的类别指定为最终的输出结果。当上述决策树具体性质为回归类时，随机森林算法就被用来对多个输入特征值进行回归分类计算，分出两类：舒张压和收缩压，并输出两类的回归计算值即具体的舒张压预测值与收缩压预测值。
如图1为本发明实施例一提供的一种对血压进行预测的方法示意图所示，本方法主要包括如下步骤：
步骤1，同步对测试者进行心电生理信号和脉搏生理信号采集生成心电信号和脉搏生理信号；并对心电信号和脉搏生理信号按预置的采样频率阈值进行信号采样处理生成心电图ECG信号和光体积变化描记图法PPG信号；
具体包括：步骤11，对测试者进行心电生理信号采集生成一段长度为固定时长阈值的心电信号，同步的，对测试者进行脉搏生理信号采集生成一段长度为固定时长阈值的脉搏生理信号；
步骤12，按采样频率阈值对心电信号进行信号采样生成ECG信号；ECG信号包括多个ECG信号点；
其中，ECG信号点包括信号点幅值数据和信号点时间数据；
步骤13，按采样频率阈值对脉搏生理信号进行信号采样生成PPG原始信 号，并根据预置的带通频率阈值范围对PPG原始信号进行带通滤波处理生成PPG信号；
其中，PPG信号包括多个PPG信号点；PPG信号点包括信号点幅值数据和信号点时间数据。
此处，这两段采集的信号一定必须是同步采集的，且长度一致、后续的采样频率一致。
步骤2，对ECG信号进行R点时间特征识别操作生成R点时间序列；
其中，R点时间序列包括多个R点时间；
具体包括：步骤21，对ECG信号，依次提取ECG信号点的信号点时间数据，生成ECG一维数据向量；按预置的ECG片段长度阈值对ECG一维数据向量进行数据片段划分操作生成多个ECG一维片段向量；
步骤22，以ECG一维片段向量作为R点时间特征识别算法的输入，利用指定的R点时间特征识别算法，识别出R点在ECG一维片段向量内的相对时间位移信息T 1；并根据ECG一维片段向量的起始ECG信号点的信号点时间数据T 2获得R点时间，R点时间＝T 2+T 1；
步骤23，将识别出的所有R点时间，按先后顺序排序生成R点时间序列。
此处，可是使用多种特征提取算法进行处理，可以采用均方差依次轮询决策的提取方式，也可以将信号进行时域频域转换提取最大能量值作为R点，还可以采用卷积网络特征提取处理方式等。对ECG信号进行片段和子片段划分是为了进一步细化提取模块，提高提取识别的有效精度。
步骤3，对PPG信号进行脉搏波峰值点时间特征识别操作和脉搏波谷值点时间特征识别操作，生成峰值点时间序列和谷值点时间序列；
其中，峰值点时间序列包括多个峰值点时间；谷值点时间序列包括多个谷值点时间；
具体包括：步骤31，通过配置信号点参考幅值和绝对不应期时间宽度，对PPG信号进行脉搏波峰值点时间特征识别操作，生成峰值点时间序列；
具体包括：步骤311，初始化峰值点时间序列为空；设置波形下降沿标志为0；获取预置的峰值校准因子；对PPG信号进行全信号标准偏差计算生成标准偏差因子；
此处两个计算因子是用于后续计算信号点参考幅值用的；
步骤312，在PPG信号中，从第1个PPG信号点的信号点幅值数据开始，到指定数目个PPG信号点的信号点幅值数据为止，提取其中的最小值对信号点参考幅值进行初始化；
此处，假设指定数目为20，则是利用PPG信号起始的20个PPG波形计算信号点参考幅值的初始值，信号点参考幅值是一个对PPG波形进行连续判定时使用的一个变化的幅值比对值；
比对的基本原理是：首先使用PPG信号起始指定个数的PPG波形中的最低幅值作为初始值；
其次，在单个PPG波形上升沿时，每比较一次都将信号点参考幅值数据设置为当前的进行比较的时间点对应的波形幅值，其特点是上升沿过程中，信号点参考幅值数据一定总是小于当前信号点的幅值数据的；
再者，在单个PPG波形下降沿时，需要设置两个时间段，一个是从峰值开始的一段时间称之为绝对不应期时间宽度，一个是从绝对不应期时间宽度之后到PPG单个波形结束时间之间的时间段；在绝对不应期时间宽度内，信号点参考幅值的设置是始终保持与峰值点的幅值相等；从绝对不应期时间宽度之后，需要按公式A new＝A old+B*(P+std)/f对信号点参考幅值数据进行计算，这里，A new为重置后的信号点参考幅值数据；A old为重置前的信号点参考幅值数据；B为峰值校准因子；P为峰值点幅值数据；std为标准偏差因子；f为采样频率阈值；
具体的，如图2为本发明实施例提供的信号点参考幅值变化示意图所示，可以看见随着PPG信号信号点的递进，参考幅值数据的变化趋势；
步骤313，对PPG信号从指定数目加1个PPG信号点开始到最后1个PPG 信号点进行信号点遍历生成当前PPG信号点；
步骤314，在当前PPG信号点的信号点幅值数据大于信号点参考幅值时，设置信号点参考幅值为当前PPG信号点的信号点幅值数据，设置波形下降沿标志为0；
此处，就是在波形处于上升沿时对信号点参考幅值的设置，设置其余实际波形幅值相等；
步骤315，在当前PPG信号点的信号点幅值数据小于信号点参考幅值且波形下降沿标志为0时，设置波形下降沿标志为1；提取上一个PPG信号点的信号点幅值数据生成当前峰值点幅值，提取上一个PPG信号点的信号点时间数据生成当前峰值点时间；获取绝对不应期时间宽度并根据当前峰值点时间加上绝对不应期时间宽度的和生成绝对不应期结束时间；设置信号点参考幅值为当前峰值点幅值；将当前峰值点时间向峰值点时间序列进行峰值点时间添加操作；
此处就是在波形刚跨过峰值点的时候，也就是第一个处于下降沿的R点；此时要做5件事：1、将波形下降沿标志从0切换至1表示当前波形进入下降沿；2、以上一个R点作为当前波形的峰值点，并提取对应的峰值点幅值和峰值点时间；3、计算当前波形的绝对不应期结束时间；4、在绝对不应期内，信号点参考幅值始终等于峰值点幅值；5、将峰值点时间提取出来向峰值点时间序列添加；
步骤316，在当前PPG信号点的信号点幅值数据小于信号点参考幅值且波形下降沿标志为1时，如果当前PPG信号点的信号点时间数据小于或等于绝对不应期结束时间，则保持信号点参考幅值的取值不变；
此处就是在波形处于绝对不应期时间宽度期间，信号点参考幅值始终等于峰值点幅值；
步骤317，在当前PPG信号点的信号点幅值数据小于信号点参考幅值且波形下降沿标志为1时，如果当前PPG信号点的信号点时间数据大于绝对不应 期结束时间，则根据公式A new＝A old+B*(P+std)/f对信号点参考幅值进行重置；
其中，A new为重置后的信号点参考幅值数据；A old为重置前的信号点参考幅值数据；B为峰值校准因子；P为当前峰值点幅值；std为标准偏差因子；f为采样频率阈值；
此处就是在波形跨过绝对不应期时间宽度，在这段下降沿时期信号点参考幅值是要发生变化的，具体的变化就是按照上文公式的趋势进行变化；其中，B为峰值校准因子，一般为一个负数；
步骤32，根据峰值点时间序列，对PPG信号进行脉搏波谷值点时间特征提取操作生成谷值点时间序列；
具体包括：根据峰值点时间序列，在PPG信号中，两个相邻峰值点时间之间，提取信号点幅值数据为最小值的PPG信号点的信号点时间数据，生成谷值点时间；将提取出的所有谷值点时间按先后顺序对对谷值点时间序列进行谷值点时间添加操作。
此处，默认两个峰值点间只有一个真实的波谷，又考虑到可能有噪声信号存在，所以对两者间的数据进行遍历，提取最小值作为单个PPG信号的谷底值。
步骤4，根据峰值点时间序列、谷值点时间序列和R点时间序列，以每个R点时间为时间参考点，在PPG信号中，提取在时间参考点之后的第一谷值点时间、第一峰值点时间和第二谷值点时间；根据R点时间、第一谷值点时间、第一峰值点时间和第二谷值点时间生成匹配特征时间组；并将所有匹配特征时间组按先后顺序进行排序生成匹配特征时间组序列；
具体包括：步骤41，设置匹配特征时间组；初始化匹配特征时间组的匹配R点时间为空，初始化匹配特征时间组的匹配PPG峰值时间为空，初始化匹配特征时间组的匹配PPG起始时间为空，初始化匹配特征时间组的匹配PPG结束时间为空；
步骤42，从R点时间序列依次提取两个相邻R点时间生成第一参考R点 和第二参考R点；在PPG信号中，对谷值点时间序列以第一参考R点为起始时间、以第二参考R点为结束时间从结束时间向起始时间进行反方向查找，提取距离结束时间最近的谷值点时间生成第二谷值点时间，提取与第二谷值点时间距离最近的下一个谷值点时间生成第一谷值点时间；在PPG信号中，对峰值点时间序列以第一参考R点为起始时间、以第二参考R点为结束时间从结束时间向起始时间进行反方向查找，提取距离结束时间最近的峰值点时间生成第一峰值点时间；
其中，第一参考R点小于第二参考R点；
此处，提取的第一谷值点时间、第一峰值点时间和第二谷值点时间，就是提取的一个完整PPG波形的三个特征点，起始点(第一谷值点时间)、峰值点(第一峰值点时间)和结束点(第二峰值点时间)；如果两个相邻R点之间存在多个PPG波形，则在两个R点之间选择最后一个PPG波形的起始、峰值和结束信息作为配对，与该PPG配对的采用两个相邻R点的前一个R点即文中的第一参考R点；
步骤43，设置匹配特征时间组的匹配R点时间为第一参考R点，设置匹配特征时间组的匹配PPG峰值时间为第一峰值点时间，设置匹配特征时间组的匹配PPG起始时间为第一谷值点时间，设置匹配特征时间组的匹配PPG结束时间为第二谷值点时间；
此处，四个点位信息可以理解为一个三点(起始、峰值、结束)的PPG波形和其时间轴前端距离最近的R点信息，现实情况中一般R点与R点之间只有一个PPG波形，但是如果在激烈运动时两个R点之间可能存在多个PPG信号，为了充分体现ECG信号与PPG信号的关联性，本发明实施例选择两个R点间最后一个PPG波形作为两个R点中时间居先的那个R点的匹配对象；
步骤44，将设置成功的匹配特征时间组向匹配特征时间组序列进行匹配特征时间组添加操作。
步骤5，根据匹配特征时间组序列，进行随机森林算法模型的特征样本数 据准备操作，生成随机森林样本组序列；
其中，随机森林样本组序列包括多个随机森林样本组；
具体包括：步骤51，依次提取匹配特征时间组序列的匹配特征时间组生成当前匹配特征时间组；提取当前匹配特征时间组的下一个匹配特征时间组生成相邻匹配特征时间组；
步骤52，提取当前匹配特征时间组的匹配R点时间生成第一R点时间，提取当前匹配特征时间组的匹配PPG峰值时间生成峰值时间，提取当前匹配特征时间组的匹配PPG起始时间生成起始时间，提取当前匹配特征时间组的匹配PPG结束时间生成结束时间；提取相邻匹配特征时间组的匹配R点时间生成第二R点时间；
步骤53，在PPG信号中，提取与当前匹配特征时间组对应的PPG信号波形生成当前PPG波形；计算从起始时间到结束时间之间由当前PPG波形与时间横轴围成的面积生成当前PPG波形面积S；
步骤54，设置随机森林样本组的第一样本参数为60除以心动时差的商，
其中，心动时差为第一R点时间与第二R点时间的时间差绝对值；
此处，如图3为本发明实施例提供的ECG信号与PPG信号示意图所示；
步骤55，设置随机森林样本组的第二样本参数为第一R点时间与起始时间的时间差绝对值；
此处，如图3为本发明实施例提供的ECG信号与PPG信号示意图所示；
步骤56，设置随机森林样本组的第三样本参数为第一R点时间与峰值时间的时间差绝对值；
此处，如图3为本发明实施例提供的ECG信号与PPG信号示意图所示；
步骤57，获取当前PPG波形的上升沿斜率最大值处的时间点生成上升最大斜率时间；设置随机森林样本组的第四样本参数为第一R点时间与上升最大斜率时间的时间差绝对值；
此处，如图3为本发明实施例提供的ECG信号与PPG信号示意图所示；
步骤58，获取当前PPG波形中与峰值时间对应的信号幅值数据生成峰值幅值，设置随机森林样本组的第五样本参数为峰值幅值；
此处，如图3为本发明实施例提供的ECG信号与PPG信号示意图所示；
步骤59，获取当前PPG波形中与起始时间对应的信号幅值数据生成起始幅值，设置随机森林样本组的第六样本参数为峰值幅值与起始幅值的比值；
步骤60，设置随机森林样本组的第七样本参数为峰值幅值与起始幅的幅值差绝对值；
此处，如图3为本发明实施例提供的ECG信号与PPG信号示意图所示；
步骤61，设置随机森林样本组的第八样本参数为当前PPG波形中从峰值时间对应的幅值点到起始时间对应的幅值点之间连线的斜率绝对值；
步骤62，设置随机森林样本组的第九样本参数为当前PPG波形中从峰值时间对应的幅值点到结束时间对应的幅值点之间连线的斜率绝对值；
步骤63，设置随机森林样本组的第十样本参数为从起始时间到峰值时间之间由当前PPG波形与时间横轴围成的面积；
此处，如图3为本发明实施例提供的ECG信号与PPG信号示意图所示；
步骤64，设置随机森林样本组的第十一样本参数为从结束时间到峰值时间之间由当前PPG波形与时间横轴围成的面积；
此处，如图3为本发明实施例提供的ECG信号与PPG信号示意图所示；
步骤65，设置随机森林样本组的第十二样本参数；从起始时间起向后偏移第十二样本参数止，由当前PPG波形与时间横轴围成的面积为0.1*S；
步骤66，设置随机森林样本组的第十三样本参数；从起始时间起向后偏移第十三样本参数止，由当前PPG波形与时间横轴围成的面积为0.2*S；
步骤67，设置随机森林样本组的第十四样本参数；从起始时间起向后偏移第十四样本参数止，由当前PPG波形与时间横轴围成的面积为0.3*S；
步骤68，设置随机森林样本组的第十五样本参数；从起始时间起向后偏移第十五样本参数止，由当前PPG波形与时间横轴围成的面积为0.4*S；
步骤69，设置随机森林样本组的第十六样本参数；从起始时间起向后偏移第十六样本参数止，由当前PPG波形与时间横轴围成的面积为0.5*S；
步骤70，设置随机森林样本组的第十七样本参数；从起始时间起向后偏移第十七样本参数止，由当前PPG波形与时间横轴围成的面积为0.6*S；
步骤71，设置随机森林样本组的第十八样本参数；从起始时间起向后偏移第十八样本参数止，由当前PPG波形与时间横轴围成的面积为0.7*S；
步骤72，设置随机森林样本组的第十九样本参数；从起始时间起向后偏移第十九样本参数止，由当前PPG波形与时间横轴围成的面积为0.8*S；
步骤73，设置随机森林样本组的第二十样本参数；从起始时间起向后偏移第二十样本参数止，由当前PPG波形与时间横轴围成的面积为0.9*S；
步骤74，设置随机森林样本组的第二十一样本参数为结束时间与起始时间的时间差绝对值。
此处，是对随机森林模型的随机森林样本组序列做设置，随机森林样本组序列包括多个随机森林样本组，每个随机森林样本组包括一共二十一个样本参数，一共分为六大类：心率参数，时间参数、幅值参数、斜率参数、面积参数、面积偏移时间参数；对应关系如下是：心率参数包括第一样本参数；时间参数包括第二、三、四样本参数；幅值参数包括第五、六、七样本参数；斜率参数包括第八、九样本参数；面积参数包括第十、十一样本参数；面积偏移时间参数包括第十二到第二十一样本参数。
步骤6，根据R点时间序列，计算对应的R点瞬时心率、R点趋势心率和R点心率差；并以R点心率差是否小于预置的合理心率差阈值作为异常样本组判定条件对随机森林样本组序列进行异常样本组删除处理。
步骤7，利用随机森林算法模型对随机森林样本组序列进行回归预测计算生成预测血压数组；
其中，预测血压数组包括收缩压数据和舒张压数据。
此处，随机森林算法模型是已经通过批量ECG/PPG信号与与之对应的实 测血压值完成训练的模型，通过该随机森林算法模型输出的预测血压数据序列包括两个取值，分别是：预测收缩压数据和预测舒张压数据。
如图4为本发明实施例二提供的一种异常随机森林样本组处理方法示意图所示，本方法主要包括如下步骤：
步骤401，从上位应用获取R点时间序列和随机森林样本序列。
其中，随机森林样本组序列包括多个随机森林样本组。
步骤402，初始化瞬时心率序列为空；获取R点时间序列包括的R点时间的总数生成R点总数；
步骤403，依次提取R点时间序列的R点时间生成当前R点，并按指定的相邻点提取方向提取与当前R点相邻的R点时间生成相邻R点；根据当前R点与相邻R点的时间差绝对值生成第一因子；根据第一因子的倒数生成R点瞬时心率；将R点瞬时心率向瞬时心率序列进行R点瞬时心率添加操作；
其中，瞬时心率序列包括R点总数个R点瞬时心率；
此处，瞬时心率的计算方法就是即心电图两个相邻的R-R间期的倒数，将每个R点的瞬时心率合并生成瞬时心率序列就是全ECG片段的瞬时心率序列；
步骤404，对瞬时心率序列按预置的滤波标准差进行高斯滤波生成趋势心率序列；
其中，趋势心率序列包括R点总数个R点趋势心率；
高斯滤波是一种线性平滑滤波，适用于消除高斯噪声，广泛应用于图像处理的减噪过程，此处就是对瞬时心率序列进行平滑降噪趋势处理；
步骤405，初始化第一索引的值为1，初始化第一总数的值为R点总数；
步骤406，从瞬时心率序列中提取与第一索引对应的R点瞬时心率生成第一索引瞬时心率；从趋势心率序列中提取与第一索引对应的R点趋势心率生成第一索引趋势心率；
对应每个R点，可以通过瞬时心率和趋势心率相减并对结果取绝对值得 出一个心率差，通常如果是真实的ECG信号那么这个心率差会在一个合理误差范围之内，如果是噪声或者干扰信号这个心率差一定会超过误差范围；
步骤407，根据第一索引瞬时心率与第一索引趋势心率的心率差绝对值生成第一索引R点心率差；
步骤408，当第一索引R点心率差大于合理心率差阈值时，将与第一索引对应的随机森林样本组标记为异常样本组；
此处，合理心率差阈值就是上文提及的误差范围，心率超出误差范围的R点被进一步视为噪点，对应的随机森林样本组序列中的随机森林样本组也被视为异常样本；
步骤409，将第一索引加1；
步骤410，判断第一索引是否大于第一总数，如果第一索引大于第一总数则转至步骤411，如果第一索引小于或等于第一总数则转至步骤406；
步骤411，轮询随机森林样本组序列，将标记为异常样本组的随机森林样本组从随机森林样本组序列中删除。
此处，在对完整ECG信号中的R点是否为噪点进行全检之后，也对随机森林样本组序列中不合格的异常样本组完成了全标记，这里就是最后统一将标记为异常样本的随机森林样本组从随机森林样本组序列中进行剔除处理。
如图5为本发明实施例三提供的一种对血压进行预测的装置的设备结构示意图所示，该设备包括：处理器和存储器。存储器可通过总线与处理器连接。存储器可以是非易失存储器，例如硬盘驱动器和闪存，存储器中存储有软件程序和设备驱动程序。软件程序能够执行本发明实施例提供的上述方法的各种功能；设备驱动程序可以是网络和接口驱动程序。处理器用于执行软件程序，该软件程序被执行时，能够实现本发明实施例提供的方法。
需要说明的是，本发明实施例还提供了一种计算机可读存储介质。该计算机可读存储介质上存储有计算机程序，该计算机程序被处理器执行时，能够实现本发明实施例提供的方法。
本发明实施例还提供了一种包含指令的计算机程序产品。当该计算机程序产品在计算机上运行时，使得处理器执行上述方法。
本发明实施例提供的一种对血压进行预测的方法和装置，首先，对测试者进行心电图ECG信号和PPG信号的同步采集；其次，对获取的ECG信号与PPG信号进行特征提取：获取的ECG信号的R点特征，获取PPG信号脉搏波峰值点与起止点特征；然后，对各自的特征数据进行匹配：对ECG信号与PPG信号的特征点建立关联性生成匹配特征组；接着，根据匹配的ECG信号和PPG信号进行融合生成样本数据：基于每一个匹配特征组进行与之相关的特征参数设置；最后，将代表全时长的匹配特征组序列作为随机森林算法模型的输入进行血压回归计算生成预测血压数组(包括舒张压和收缩压)。通过本发明实施例，无需对测试者进行压力测试或者干预介入式测试，通过与采集传感器配合可以自动持续对ECG和PPG信号进行分析和预测，由此既提高了测试者的舒适度又建立了一种自动监测血压的途径。
专业人员应该还可以进一步意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、计算机软件或者二者的结合来实现，为了清楚地说明硬件和软件的可互换性，在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本发明的范围。
结合本文中所公开的实施例描述的方法或算法的步骤可以用硬件、处理器执行的软件模块，或者二者的结合来实施。软件模块可以置于随机存储器(RAM)、内存、只读存储器(ROM)、电可编程ROM、电可擦除可编程ROM、寄存器、硬盘、可移动磁盘、CD-ROM、或技术领域内所公知的任意其它形式的存储介质中。
以上所述的具体实施方式，对本发明的目的、技术方案和有益效果进行 了进一步详细说明，所应理解的是，以上所述仅为本发明的具体实施方式而已，并不用于限定本发明的保护范围，凡在本发明的精神和原则之内，所做的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (12)

  1. 一种对血压进行预测的方法，其特征在于，所述方法包括：
    同步对测试者进行心电生理信号和脉搏生理信号采集生成心电信号和脉搏生理信号；并对所述心电信号和所述脉搏生理信号按预置的采样频率阈值进行信号采样处理生成心电图ECG信号和光体积变化描记图法PPG信号；
    对所述ECG信号进行R点时间特征识别操作生成R点时间序列；所述R点时间序列包括多个R点时间；
    对所述PPG信号进行脉搏波峰值点时间特征识别操作和脉搏波谷值点时间特征识别操作，生成峰值点时间序列和谷值点时间序列；所述峰值点时间序列包括多个峰值点时间；所述谷值点时间序列包括多个谷值点时间；
    根据所述峰值点时间序列、所述谷值点时间序列和所述R点时间序列，以每个所述R点时间为时间参考点，在所述PPG信号中，提取在所述时间参考点之后的第一谷值点时间、第一峰值点时间和第二谷值点时间；根据所述R点时间、所述第一谷值点时间、所述第一峰值点时间和所述第二谷值点时间生成匹配特征时间组；并将所有所述匹配特征时间组按先后顺序进行排序生成匹配特征时间组序列；
    根据所述匹配特征时间组序列，进行随机森林算法模型的特征样本数据准备操作，生成随机森林样本组序列；所述随机森林样本组序列包括多个随机森林样本组；
    根据所述R点时间序列，计算对应的R点瞬时心率、R点趋势心率和R点心率差；并以所述R点心率差是否小于预置的合理心率差阈值作为异常样本组判定条件对所述随机森林样本组序列进行异常样本组删除处理；
    利用所述随机森林算法模型对所述随机森林样本组序列进行回归预测计算生成预测血压数组；所述预测血压数组包括收缩压数据和舒张压数据。
  2. 根据权利要求1所述的对血压进行预测的方法，其特征在于，所述同步对测试者进行心电生理信号和脉搏生理信号采集生成心电信号和脉搏生理 信号；并对所述心电信号和所述脉搏生理信号按预置的采样频率阈值进行信号采样处理生成心电图ECG信号和光体积变化描记图法PPG信号，具体包括：
    对所述测试者进行心电生理信号采集生成一段长度为固定时长阈值的所述心电信号，同步的，对所述测试者进行脉搏生理信号采集生成一段长度为所述固定时长阈值的所述脉搏生理信号；
    按所述采样频率阈值对所述心电信号进行信号采样生成所述ECG信号；所述ECG信号包括多个ECG信号点；所述ECG信号点包括信号点幅值数据和信号点时间数据；
    按所述采样频率阈值对所述脉搏生理信号进行信号采样生成PPG原始信号，并根据预置的带通频率阈值范围对所述PPG原始信号进行带通滤波处理生成所述PPG信号；所述PPG信号包括多个PPG信号点；所述PPG信号点包括信号点幅值数据和信号点时间数据。
  3. 根据权利要求2所述的对血压进行预测的方法，其特征在于，所述对所述ECG信号进行R点时间特征识别操作生成R点时间序列，具体包括：
    对所述ECG信号，依次提取所述ECG信号点的信号点时间数据，生成ECG一维数据向量；按预置的ECG片段长度阈值对所述ECG一维数据向量进行数据片段划分操作生成多个ECG一维片段向量；
    以所述ECG一维片段向量作为R点时间特征识别算法的输入，利用指定的R点时间特征识别算法，识别出R点在所述ECG一维片段向量内的相对时间位移信息T 1；并根据所述ECG一维片段向量的起始ECG信号点的信号点时间数据T 2获得所述R点时间，R点时间＝T 2+T 1；
    将识别出的所有所述R点时间，按先后顺序排序生成所述R点时间序列。
  4. 根据权利要求2所述的对血压进行预测的方法，其特征在于，所述对所述PPG信号进行脉搏波峰值点时间特征识别操作和脉搏波谷值点时间特征识别操作，生成峰值点时间序列和谷值点时间序列，具体包括：
    通过配置信号点参考幅值和绝对不应期时间宽度，对所述PPG信号进行 脉搏波峰值点时间特征识别操作，生成所述峰值点时间序列；
    根据所述峰值点时间序列，对所述PPG信号进行所述脉搏波谷值点时间特征提取操作生成所述谷值点时间序列。
  5. 根据权利要求4所述的对血压进行预测的方法，其特征在于，所述通过配置信号点参考幅值和绝对不应期时间宽度，对所述PPG信号进行脉搏波峰值点时间特征识别操作，生成所述峰值点时间序列，具体包括：
    初始化所述峰值点时间序列为空；设置波形下降沿标志为0；获取预置的峰值校准因子；对所述PPG信号进行全信号标准偏差计算生成标准偏差因子；
    在所述PPG信号中，从第1个PPG信号点的信号点幅值数据开始，到指定数目个PPG信号点的信号点幅值数据为止，提取其中的最小值对所述信号点参考幅值进行初始化；
    对所述PPG信号从指定数目加1个PPG信号点开始到最后1个PPG信号点进行信号点遍历生成当前PPG信号点；
    当所述当前PPG信号点的信号点幅值数据大于所述信号点参考幅值时，设置所述信号点参考幅值为所述当前PPG信号点的信号点幅值数据，设置所述波形下降沿标志为0；
    当所述当前PPG信号点的信号点幅值数据小于所述信号点参考幅值且所述波形下降沿标志为0时，设置所述波形下降沿标志为1；提取上一个PPG信号点的信号点幅值数据生成当前峰值点幅值，提取上一个PPG信号点的信号点时间数据生成当前峰值点时间；获取所述绝对不应期时间宽度并根据所述当前峰值点时间加上所述绝对不应期时间宽度的和生成绝对不应期结束时间；设置所述信号点参考幅值为所述当前峰值点幅值；将所述当前峰值点时间向所述峰值点时间序列进行峰值点时间添加操作；
    当所述当前PPG信号点的信号点幅值数据小于所述信号点参考幅值且所述波形下降沿标志为1时，如果所述当前PPG信号点的信号点时间数据小于或等于所述绝对不应期结束时间，则保持所述信号点参考幅值的取值不变；
    当所述当前PPG信号点的信号点幅值数据小于所述信号点参考幅值且所述波形下降沿标志为1时，如果所述当前PPG信号点的信号点时间数据大于所述绝对不应期结束时间，则根据公式A new＝A old+B*(P+std)/f对所述信号点参考幅值进行重置；所述A new为重置后的信号点参考幅值数据；所述A old为重置前的信号点参考幅值数据；所述B为所述峰值校准因子；所述P为所述当前峰值点幅值；所述std为所述标准偏差因子；所述f为所述采样频率阈值。
  6. 根据权利要求4所述的对血压进行预测的方法，其特征在于，所述根据所述峰值点时间序列，对所述PPG信号进行所述脉搏波谷值点时间特征提取操作生成所述谷值点时间序列，具体包括：
    根据所述峰值点时间序列，在所述PPG信号中，两个相邻峰值点时间之间，提取所述信号点幅值数据为最小值的所述PPG信号点的所述信号点时间数据，生成所述谷值点时间；将提取出的所有所述谷值点时间按先后顺序对对所述谷值点时间序列进行谷值点时间添加操作。
  7. 根据权利要求2所述的对血压进行预测的方法，其特征在于，所述根据所述峰值点时间序列、所述谷值点时间序列和所述R点时间序列，以每个所述R点时间为时间参考点，在所述PPG信号中，提取在所述时间参考点之后的第一谷值点时间、第一峰值点时间和第二谷值点时间；根据所述R点时间、所述第一谷值点时间、所述第一峰值点时间和所述第二谷值点时间生成匹配特征时间组；并将所有所述匹配特征时间组按先后顺序进行排序生成匹配特征时间组序列，具体包括：
    设置所述匹配特征时间组；初始化所述匹配特征时间组的匹配R点时间为空，初始化所述匹配特征时间组的匹配PPG峰值时间为空，初始化所述匹配特征时间组的匹配PPG起始时间为空，初始化所述匹配特征时间组的匹配PPG结束时间为空；
    从所述R点时间序列依次提取两个相邻所述R点时间生成第一参考R点和第二参考R点；所述第一参考R点小于所述第二参考R点；在所述PPG信 号中，对所述谷值点时间序列以所述第一参考R点为起始时间、以所述第二参考R点为结束时间从结束时间向起始时间进行反方向查找，提取距离结束时间最近的所述谷值点时间生成所述第二谷值点时间，提取与所述第二谷值点时间距离最近的下一个所述谷值点时间生成所述第一谷值点时间；在所述PPG信号中，对所述峰值点时间序列以所述第一参考R点为起始时间、以所述第二参考R点为结束时间从结束时间向起始时间进行反方向查找，提取距离结束时间最近的所述峰值点时间生成所述第一峰值点时间；
    设置所述匹配特征时间组的所述匹配R点时间为所述第一参考R点，设置所述匹配特征时间组的所述匹配PPG峰值时间为所述第一峰值点时间，设置所述匹配特征时间组的所述匹配PPG起始时间为所述第一谷值点时间，设置所述匹配特征时间组的所述匹配PPG结束时间为所述第二谷值点时间；
    将设置成功的所述匹配特征时间组向所述匹配特征时间组序列进行匹配特征时间组添加操作。
  8. 根据权利要求7所述的对血压进行预测的方法，其特征在于，所述根据所述匹配特征时间组序列，进行随机森林算法模型的特征样本数据准备操作，生成随机森林样本组序列，具体包括：
    依次提取所述匹配特征时间组序列的所述匹配特征时间组生成当前匹配特征时间组；提取所述当前匹配特征时间组的下一个匹配特征时间组生成相邻匹配特征时间组；
    提取所述当前匹配特征时间组的所述匹配R点时间生成第一R点时间，提取所述当前匹配特征时间组的所述匹配PPG峰值时间生成峰值时间，提取所述当前匹配特征时间组的所述匹配PPG起始时间生成起始时间，提取所述当前匹配特征时间组的所述匹配PPG结束时间生成结束时间；提取所述相邻匹配特征时间组的所述匹配R点时间生成第二R点时间；
    在所述PPG信号中，提取与所述当前匹配特征时间组对应的PPG信号波形生成当前PPG波形；计算从所述起始时间到所述结束时间之间由所述当前 PPG波形与时间横轴围成的面积生成当前PPG波形面积S；
    设置所述随机森林样本组的第一样本参数为60除以心动时差的商，所述心动时差为所述第一R点时间与所述第二R点时间的时间差绝对值；
    设置所述随机森林样本组的第二样本参数为所述第一R点时间与所述起始时间的时间差绝对值；
    设置所述随机森林样本组的第三样本参数为所述第一R点时间与所述峰值时间的时间差绝对值；
    获取所述当前PPG波形的上升沿斜率最大值处的时间点生成上升最大斜率时间；设置所述随机森林样本组的第四样本参数为所述第一R点时间与所述上升最大斜率时间的时间差绝对值；
    获取所述当前PPG波形中与所述峰值时间对应的信号幅值数据生成峰值幅值，设置所述随机森林样本组的第五样本参数为所述峰值幅值；
    获取所述当前PPG波形中与所述起始时间对应的信号幅值数据生成起始幅值，设置所述随机森林样本组的第六样本参数为所述峰值幅值与所述起始幅值的比值；
    设置所述随机森林样本组的第七样本参数为所述峰值幅值与所述起始幅的幅值差绝对值；
    设置所述随机森林样本组的第八样本参数为所述当前PPG波形中从所述峰值时间对应的幅值点到所述起始时间对应的幅值点之间连线的斜率绝对值；
    设置所述随机森林样本组的第九样本参数为所述当前PPG波形中从所述峰值时间对应的幅值点到所述结束时间对应的幅值点之间连线的斜率绝对值；
    设置所述随机森林样本组的第十样本参数为从所述起始时间到所述峰值时间之间由所述当前PPG波形与时间横轴围成的面积；
    设置所述随机森林样本组的第十一样本参数为从所述结束时间到所述峰值时间之间由所述当前PPG波形与时间横轴围成的面积；
    设置所述随机森林样本组的第十二样本参数；从所述起始时间起向后偏 移所述第十二样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.1*S；
    设置所述随机森林样本组的第十三样本参数；从所述起始时间起向后偏移所述第十三样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.2*S；
    设置所述随机森林样本组的第十四样本参数；从所述起始时间起向后偏移所述第十四样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.3*S；
    设置所述随机森林样本组的第十五样本参数；从所述起始时间起向后偏移所述第十五样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.4*S；
    设置所述随机森林样本组的第十六样本参数；从所述起始时间起向后偏移所述第十六样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.5*S；
    设置所述随机森林样本组的第十七样本参数；从所述起始时间起向后偏移所述第十七样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.6*S；
    设置所述随机森林样本组的第十八样本参数；从所述起始时间起向后偏移所述第十八样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.7*S；
    设置所述随机森林样本组的第十九样本参数；从所述起始时间起向后偏移所述第十九样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.8*S；
    设置所述随机森林样本组的第二十样本参数；从所述起始时间起向后偏移所述第二十样本参数止，由所述当前PPG波形与时间横轴围成的面积为0.9*S；
    设置所述随机森林样本组的第二十一样本参数为所述结束时间与所述起始时间的时间差绝对值。
  9. 根据权利要求1所述的对血压进行预测的方法，其特征在于，所述根据所述R点时间序列，计算对应的R点瞬时心率、R点趋势心率和R点心率差；并以所述R点心率差是否小于预置的合理心率差阈值作为异常样本组判定条件对所述随机森林样本组序列进行异常样本组删除处理，具体包括：
    步骤91，初始化瞬时心率序列为空；获取所述R点时间序列包括的所述R点时间的总数生成R点总数；
    步骤92，依次提取所述R点时间序列的所述R点时间生成当前R点，并按指定的相邻点提取方向提取与所述当前R点相邻的所述R点时间生成相邻R点；根据所述当前R点与所述相邻R点的时间差绝对值生成第一因子；根据所述第一因子的倒数生成所述R点瞬时心率；将所述R点瞬时心率向所述瞬时心率序列进行R点瞬时心率添加操作；所述瞬时心率序列包括所述R点总数个所述R点瞬时心率；
    步骤93，对所述瞬时心率序列按预置的滤波标准差进行高斯滤波生成趋势心率序列；所述趋势心率序列包括所述R点总数个所述R点趋势心率；
    步骤94，初始化第一索引的值为1，初始化第一总数的值为所述R点总数；
    步骤95，从所述瞬时心率序列中提取与所述第一索引对应的所述R点瞬时心率生成第一索引瞬时心率；从所述趋势心率序列中提取与所述第一索引对应的所述R点趋势心率生成第一索引趋势心率；
    步骤96，根据所述第一索引瞬时心率与所述第一索引趋势心率的心率差绝对值生成第一索引R点心率差；
    步骤97，当所述第一索引R点心率差大于所述合理心率差阈值时，将与所述第一索引对应的所述随机森林样本组标记为异常样本组；
    步骤98，将所述第一索引加1；
    步骤99，判断所述第一索引是否大于所述第一总数，如果所述第一索引大于所述第一总数则转至步骤100，如果所述第一索引小于或等于所述第一总数则转至步骤95；
    步骤100，轮询所述随机森林样本组序列，将标记为所述异常样本组的所述随机森林样本组从所述随机森林样本组序列中删除。
  10. 一种设备，包存储器和处理器，其特征在于，所述存储器用于存储程序，所述处理器用于执行如权利要求1至9任一项所述的方法。
  11. 一种包含指令的计算机程序产品，当其在计算机上运行时，使得计算机执行如权利要求1至9任一项所述的方法。
  12. 一种计算机可读存储介质，包括指令，当所述指令在计算机上运行时，使所述计算机执行根据权利要求1至9任一项所述的方法。
PCT/CN2020/129632 2020-02-21 2020-11-18 一种对血压进行预测的方法和装置 WO2021164347A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010110286.4 2020-02-21
CN202010110286.4A CN111248881B (zh) 2020-02-21 2020-02-21 一种对血压进行预测的方法和装置

Publications (1)

Publication Number Publication Date
WO2021164347A1 true WO2021164347A1 (zh) 2021-08-26

Family

ID=70943175

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/129632 WO2021164347A1 (zh) 2020-02-21 2020-11-18 一种对血压进行预测的方法和装置

Country Status (2)

Country Link
CN (1) CN111248881B (zh)
WO (1) WO2021164347A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111248881B (zh) * 2020-02-21 2022-08-02 乐普（北京）医疗器械股份有限公司 一种对血压进行预测的方法和装置
CN114451873A (zh) * 2020-10-30 2022-05-10 华为技术有限公司 一种信号处理方法及设备
CN113273980A (zh) * 2021-06-07 2021-08-20 深圳麦格米特电气股份有限公司 一种获取生命体征数据的方法、光纤传感器及智能床垫

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108403101A (zh) * 2018-03-27 2018-08-17 中国科学院深圳先进技术研究院 一种模型训练方法及装置、测量血压的方法及装置
EP3366203A1 (en) * 2017-02-23 2018-08-29 Tata Consultancy Services Limited Method and system for cuffless blood pressure estimation using photoplethysmogram features and pulse transit time
CN109219391A (zh) * 2017-05-04 2019-01-15 京东方科技集团股份有限公司 用于确定对象血压的设备和方法
CN110251105A (zh) * 2019-06-12 2019-09-20 广州视源电子科技股份有限公司 一种无创血压测量方法、装置、设备及系统
CN110477890A (zh) * 2019-08-27 2019-11-22 北京麦邦光电仪器有限公司 血压计算方法及血压测量装置
CN110680307A (zh) * 2019-11-05 2020-01-14 北京航空航天大学 一种运动环境下基于脉搏波传导时间的动态血压监测方法
US20200015755A1 (en) * 2018-07-12 2020-01-16 The Chinese University Of Hong Kong Deep learning approach for long term, cuffless, and continuous arterial blood pressure estimation
CN111248881A (zh) * 2020-02-21 2020-06-09 乐普（北京）医疗器械股份有限公司 一种对血压进行预测的方法和装置

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017024457A1 (zh) * 2015-08-08 2017-02-16 深圳先进技术研究院 连续血压测量装置、测量模型建立方法和系统
WO2017089921A1 (en) * 2015-11-24 2017-06-01 Koninklijke Philips N.V. Method to quantify photoplethysmogram (ppg) signal quality
CN109864736A (zh) * 2019-03-22 2019-06-11 深圳市理邦精密仪器股份有限公司 心电信号的处理方法、装置、终端设备及介质

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3366203A1 (en) * 2017-02-23 2018-08-29 Tata Consultancy Services Limited Method and system for cuffless blood pressure estimation using photoplethysmogram features and pulse transit time
CN109219391A (zh) * 2017-05-04 2019-01-15 京东方科技集团股份有限公司 用于确定对象血压的设备和方法
CN108403101A (zh) * 2018-03-27 2018-08-17 中国科学院深圳先进技术研究院 一种模型训练方法及装置、测量血压的方法及装置
US20200015755A1 (en) * 2018-07-12 2020-01-16 The Chinese University Of Hong Kong Deep learning approach for long term, cuffless, and continuous arterial blood pressure estimation
CN110251105A (zh) * 2019-06-12 2019-09-20 广州视源电子科技股份有限公司 一种无创血压测量方法、装置、设备及系统
CN110477890A (zh) * 2019-08-27 2019-11-22 北京麦邦光电仪器有限公司 血压计算方法及血压测量装置
CN110680307A (zh) * 2019-11-05 2020-01-14 北京航空航天大学 一种运动环境下基于脉搏波传导时间的动态血压监测方法
CN111248881A (zh) * 2020-02-21 2020-06-09 乐普（北京）医疗器械股份有限公司 一种对血压进行预测的方法和装置

Also Published As

Publication number Publication date
CN111248881B (zh) 2022-08-02
CN111248881A (zh) 2020-06-09

Similar Documents

Publication Publication Date Title
WO2021164347A1 (zh) 一种对血压进行预测的方法和装置
CN109171712B (zh) 心房颤动识别方法、装置、设备及计算机可读存储介质
WO2019161609A1 (zh) 多参数监护数据分析方法和多参数监护仪
Kachuee et al. Cuff-less high-accuracy calibration-free blood pressure estimation using pulse transit time
CN109411041B (zh) 心电信息处理方法和心电工作站系统
CN106413534B (zh) 连续血压测量装置、测量模型建立方法和系统
Oweis et al. QRS detection and heart rate variability analysis: A survey
WO2021184803A1 (zh) 一种血压分类方法和装置
JP6669409B2 (ja) 血圧値を測定するための方法、機器及びコンピュータプログラム
WO2019161608A1 (zh) 多参数监护数据分析方法和多参数监护系统
CN109411042B (zh) 心电信息处理方法和心电工作站
CN107951485A (zh) 基于人工智能自学习的动态心电图分析方法和装置
CN111493850A (zh) 一种基于深度神经网络的血压测量方法及装置
CN108289615A (zh) 用于量化光电体积描记（ppg）信号质量的方法
CN109998529A (zh) 一种基于高斯函数拟合的ecg信号中p波、t波检测方法
WO2021164345A1 (zh) 一种血压预测的方法和装置
CN109288515B (zh) 基于穿戴式心电信号中早搏信号的周期性监测方法及装置
WO2021184801A1 (zh) 一种基于同步信号进行血压预测的方法和装置
RU2016146176A (ru) Способ и система неинвазивной скрининговой оценки физиологических параметров и патологий
CN111839488A (zh) 基于脉搏波的无创连续血压测量装置和方法
CN112842355B (zh) 基于深度学习目标检测的心电信号心搏检测识别方法
Cherif et al. Detection of artifacts on photoplethysmography signals using random distortion testing
EP1062596B1 (en) Apparatus and method for blood pressure pulse waveform contour analysis
Shilvya et al. Obstructive Sleep Apnea Detection from ECG Signals with Deep Learning
Goletsis et al. A multicriteria decision based approach for ischaemia detection in long duration ECGs

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20919487

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20919487

Country of ref document: EP

Kind code of ref document: A1