WO2021184805A1 - 一种多数据源的血压预测方法和装置 - Google Patents

一种多数据源的血压预测方法和装置 Download PDF

Info

Publication number
WO2021184805A1
WO2021184805A1 PCT/CN2020/129646 CN2020129646W WO2021184805A1 WO 2021184805 A1 WO2021184805 A1 WO 2021184805A1 CN 2020129646 W CN2020129646 W CN 2020129646W WO 2021184805 A1 WO2021184805 A1 WO 2021184805A1
Authority
WO
WIPO (PCT)
Prior art keywords
ppg
signal
data
generate
identifier
Prior art date
Application number
PCT/CN2020/129646
Other languages
English (en)
French (fr)
Inventor
孙洪岱
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
Priority to US17/906,213 priority Critical patent/US20230137333A1/en
Priority to EP20925995.1A priority patent/EP4122382A4/en
Publication of WO2021184805A1 publication Critical patent/WO2021184805A1/zh

Links

Images

Classifications

    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/02Detecting, 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
    • A61B5/024Detecting, measuring or recording pulse rate or heart rate
    • A61B5/02416Detecting, measuring or recording pulse rate or heart rate using photoplethysmograph signals, e.g. generated by infrared radiation
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/02Detecting, 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
    • A61B5/021Measuring pressure in heart or blood vessels
    • A61B5/02108Measuring pressure in heart or blood vessels from analysis of pulse wave characteristics
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/02Detecting, 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
    • A61B5/026Measuring blood flow
    • A61B5/0261Measuring blood flow using optical means, e.g. infrared light
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235Details of waveform analysis
    • A61B5/7246Details of waveform analysis using correlation, e.g. template matching or determination of similarity
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235Details of waveform analysis
    • A61B5/725Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235Details of waveform analysis
    • A61B5/7253Details of waveform analysis characterised by using transforms
    • A61B5/7257Details of waveform analysis characterised by using transforms using Fourier transforms
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235Details of waveform analysis
    • A61B5/7253Details of waveform analysis characterised by using transforms
    • A61B5/726Details of waveform analysis characterised by using transforms using Wavelet transforms
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235Details of waveform analysis
    • A61B5/7264Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235Details of waveform analysis
    • A61B5/7264Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16HHEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61BDIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00Measuring for diagnostic purposes; Identification of persons
    • A61B5/72Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7271Specific aspects of physiological measurement analysis
    • A61B5/7275Determining 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 blood pressure prediction with multiple data sources.
  • the Photoplethysmograph (PPG) signal is a set of signals that uses a light sensor to identify and record the light intensity changes of a specific light source.
  • PPG Photoplethysmograph
  • 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. Absorption is the most; 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 light, causing the blood vessels in the diastolic cycle.
  • the absorption of light energy by the internal blood is reduced. Therefore, the blood pressure can be predicted by analyzing the PPG signal waveform reflecting the light energy absorbed by the blood in the blood vessel.
  • the PPG signal used for blood pressure prediction there are many ways to obtain the PPG signal used for blood pressure prediction, either directly through the PPG signal acquisition device, or indirectly through video shooting of the tester’s skin surface.
  • the directly acquired PPG signal because it is easily affected by factors such as sensor sensitivity, tester’s physiological state, environmental signal interference and other factors during the acquisition process, distortion and distortion often occur;
  • the PPG signal obtained after conversion the PPG signal generated by the conversion is often distorted due to factors such as the shooting environment light intensity. Using an excessively distorted PPG signal to predict blood pressure will result in a large deviation from the actual situation, or even errors.
  • the purpose of the present invention is to provide a blood pressure prediction method and device with multiple data sources in view of the shortcomings of the prior art, provide two signal filtering and shaping methods for the directly obtained PPG signal, and for the video data used to indirectly generate the PPG signal Provide video quality detection and normalized signal conversion methods, and finally generate a unified standard PPG data sequence for blood pressure prediction; further, the embodiment of the present invention provides two optional Convolutional Neural Networks (CNN) models Perform blood pressure prediction; by using the method and device provided in the embodiments of the present invention, the application's ability to preprocess multiple PPG signal data sources is improved, the application's ability to manage multiple blood pressure prediction models is improved, and the application to multiple data is improved Compatibility of source blood pressure prediction.
  • CNN Convolutional Neural Networks
  • a first aspect of the embodiments of the present invention provides a blood pressure prediction method with multiple data sources, and the method includes:
  • the data source identification is one of the three types of PPG original signal identification of the first type of photoplethysmography, the second type of PPG original signal identification and the third type of PPG video identification.
  • the original data is one of a type of PPG original signal, a type of PPG original signal and a type of PPG video data of three types of original data, and the original data corresponds to the data source identifier;
  • the CNN model identifier includes the first-type CNN identifier and the second-type CNN identifier;
  • the corresponding CNN model is selected to predict the blood pressure of the standard PPG data sequence; when the CNN model identifier is the first type of CNN identifier, a type of CNN model is selected for the standard PPG data sequence.
  • the data sequence is used for blood pressure prediction; when the CNN model identifier is the second type CNN identifier, the second type CNN model is selected to perform blood pressure prediction with wavelet transform on the standard PPG data sequence.
  • performing normalization filtering processing on the one type of PPG original signal to generate a standard PPG data sequence specifically includes:
  • data sampling is performed on the one type of PPG original signal to generate a type of PPG sampling data sequence (X 1 , X 2 ... X i ... X M) ; a class PPG said sample data sequence (X 1, X 2 ... X i ... X M) of the M include a class of PPG sampled data X i; M is an integer; the The value of i ranges from 1 to M;
  • Setting the Y i; the first process sequence (Y 1, Y 2 ... Y i ... Y M) comprises M first process data Y i; the a and b is said preset A type of filter constant; said c is the gain coefficient of the type of PPG original signal;
  • performing baseline drift removal and normalization filtering processing on the second-type PPG original signal to generate the standard PPG data sequence specifically includes:
  • the data source identifier is the second-type PPG original signal identifier
  • data sampling is performed on the second-type PPG original signal to generate a second-type PPG sampling data sequence (S 1 , S 2 ...S j ...S N );
  • the second-type PPG sampling data sequence (S 1 , S 2 ...S j ...S N ) includes the N second-class PPG sampling data S j ;
  • the N is an integer; so The value of j ranges from 1 to N;
  • Normalized filtering processing is performed on the second process sequence (T 1 , T 2 ...T j ...T N ) to generate a third process sequence (P 1 , P 2 ...P j ...P N ), specifically using the formula Set the P j ;
  • the third process sequence (P 1 , P 2 ...P j ...P N ) includes the N third process data P j ;
  • performing video quality detection and normalization signal conversion on the three types of PPG video data to generate the standard PPG data sequence specifically includes:
  • a video data frame image extraction operation is performed on the three types of PPG video data to generate three types of PPG video frame image sequences;
  • the three types of PPG video frame image sequences include multiple Three types of PPG video frame images;
  • the optical pixel threshold range performs one-dimensional green light source signal extraction processing on all the three types of PPG video frame images in the three types of PPG video frame image sequence to generate a first green light digital signal;
  • the preset band-pass filter frequency threshold range perform signal band-pass filter preprocessing on the first red light digital signal to generate a second red light digital signal, and perform signal band-pass filter pre-processing on the first green light digital signal. Processing to generate a second green light digital signal;
  • the first judgment result is an up-to-standard signal identifier, performing signal-to-noise ratio judgment processing on the second red light digital signal and the second green light digital signal to generate a second judgment result;
  • normalized PPG signal data sequence generation processing is performed on the second red light digital signal and the second green light digital signal to generate the standard PPG data sequence.
  • the one-dimensional red light source signal extraction processing is performed on all the three types of PPG video frame images of the three types of PPG video frame image sequence according to a preset red light pixel threshold range to generate a first red light digital signal;
  • Performing one-dimensional green light source signal extraction processing on all the three types of PPG video frame images of the three types of PPG video frame image sequence according to the preset green light pixel threshold range to generate the first green light digital signal which specifically includes:
  • Step 51 Initialize the first red light digital signal to be empty, initialize the first green light digital signal to be empty, initialize the value of the first index to 1, initialize the first total to the three types of PPG video frame image sequence The total number of three types of PPG video frame images included;
  • Step 52 Set a first index frame image as the three types of PPG video frame images corresponding to the first index in the three types of PPG video frame image sequence;
  • Step 53 Count all the pixels that meet the red light pixel threshold range in the first index frame image to generate a red pixel point set, and calculate the sum of the pixels included in the red pixel point set to generate a total number of red points.
  • the pixel values of all the pixels in the red pixel point set are summed to generate a sum of red pixel values, and the first index frame red light channel data is generated according to the quotient of the sum of the red pixel values divided by the total number of red points;
  • Step 54 Count all the pixels that meet the threshold range of the green light pixel in the first index frame image to generate a green pixel point set, and calculate the sum of the pixel points included in the green pixel point set to generate a total number of green points.
  • the pixel values of all the pixels in the green pixel point set are summed to generate a green pixel value sum, and the first index frame green light channel data is generated according to the quotient of the green pixel value sum divided by the total number of green points;
  • Step 55 Add 1 to the first index
  • Step 56 Determine whether the first index is greater than the first total, if the first index is less than or equal to the first total, then go to step 52, if the first index is greater than the first total, then Go to step 57;
  • Step 57 Use the first red light digital signal as a one-dimensional red light source signal extraction processing result, and transmit the first green light digital signal as a one-dimensional green light source signal extraction processing result to an upper processing flow.
  • said performing signal maximum frequency difference judgment processing on said second red light digital signal and said second green light digital signal to generate a first judgment result specifically includes:
  • Discrete Fourier transform is used for the second red light digital signal to perform digital signal time domain frequency domain conversion to generate a red light frequency domain signal
  • the second green light digital signal is used for discrete Fourier transform to perform digital signal time domain Frequency domain conversion generates green light frequency domain signal
  • the first judgment result is set as the standard-reaching signal identifier.
  • performing signal-to-noise ratio judgment processing on the second red light digital signal and the second green light digital signal to generate a second judgment result specifically includes :
  • the effective signal point whose signal frequency meets the band-stop filter frequency threshold range is changed from the effective signal point of the signal frequency to the band-stop filter frequency threshold range through multi-order Butterworth band-stop filtering.
  • the second red light digital signal is removed from the generated red light noise signal, and the effective signal point whose signal frequency meets the frequency threshold range of the band stop filter is changed from the second green light digital signal through a multi-order Butterworth band stop filter. Remove the generated green light noise signal;
  • the second judgment result is set as the standard-compliant signal identifier.
  • the normalized PPG signal data sequence generation processing is performed on the second red light digital signal and the second green light digital signal to generate the Standard PPG data sequence, including:
  • the second judgment result is the standard-compliant signal identifier
  • a green light signal set the red light data sequence of the standard PPG data sequence as the normalized red light signal, and set the green light data sequence of the standard PPG data sequence as the normalized green light signal;
  • the standard PPG data sequence includes the red light data sequence and the green light data sequence.
  • the one type of CNN model includes a multi-layer CNN network layer and a fully connected layer;
  • the CNN network layer includes a convolutional layer and a pooling layer;
  • the two-type CNN model includes a two-dimensional convolution layer, a maximum pooling layer, a batch uniformization layer, an activation layer, an addition layer, a global average pooling layer, a random discarding layer, and a fully connected layer.
  • selecting a type of CNN model to perform blood pressure prediction on the standard PPG data sequence specifically includes:
  • the CNN model identifier is the one-type CNN identifier, according to a preset one-type CNN input width threshold, perform one-type CNN model input data conversion processing on the standard PPG data sequence to generate a four-dimensional tensor of input data;
  • a preset threshold for the number of convolutional layers using the CNN network layer of the first type of CNN model to perform multi-layer convolution pooling calculation on the input data four-dimensional tensor to generate a feature data four-dimensional tensor;
  • the prediction mode identifier includes two identifiers of mean prediction and dynamic prediction;
  • the average blood pressure calculation operation is performed on the two-dimensional matrix of blood pressure regression data to generate an average blood pressure prediction data pair;
  • the average blood pressure prediction data pair includes average systolic blood pressure prediction data and Mean diastolic blood pressure prediction data;
  • the prediction mode identifier is the dynamic prediction, performing an ambulatory blood pressure data extraction operation on the two-dimensional matrix of blood pressure regression data to generate an ambulatory blood pressure prediction one-dimensional data sequence.
  • selecting the second-type CNN model to perform blood pressure prediction with wavelet transform on the standard PPG data sequence specifically includes:
  • the data source identifier is the first-type PPG original signal identifier or the second-type PPG original signal identifier
  • the CNN model identifier is the second-type CNN identifier
  • the scaling factor array includes H scaling factors
  • the movement factor array includes L movement factors
  • the H and the L are both integers
  • the standard PPG data segment is subjected to signal decomposition processing using continuous wavelet transform to generate a PPG wavelet coefficient matrix [H, L];
  • Real number matrix conversion is performed on the PPG wavelet coefficient matrix [H, L] by modulating the matrix elements, and the converted matrix is normalized by matrix element values to generate a PPG normalization matrix [H, L] ;
  • the bicubic interpolation algorithm is used to perform tensor shape reconstruction operation on the PPG time-frequency three-dimensional tensor [H,L,3] to generate the PPG convolution three-dimensional tensor [K,K ,3];
  • the K is the input width threshold of the second type CNN;
  • the PPG predicted blood pressure data pair includes PPG predicted systolic blood pressure data and PPG Predict diastolic blood pressure data.
  • the first aspect of the embodiments of the present invention provides a multi-data source blood pressure prediction method, which provides two signal filtering and shaping methods for directly obtained PPG signals, and provides video quality detection and normalization for video data used to indirectly generate PPG signals
  • the signal conversion method generates a unified standard PPG data sequence for blood pressure prediction; when entering the blood pressure prediction, different blood pressure prediction models are provided through the CNN model identifier for blood pressure prediction; the method and device provided by the embodiments of the present invention are used , Improve the application ability and compatibility of PPG signal preprocessing, and also provide different blood pressure prediction methods.
  • 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 blood pressure prediction method with multiple data sources according to Embodiment 1 of the present invention
  • FIG. 2 is a schematic diagram of PPG signals before and after filtering according to an embodiment of the present invention
  • FIG. 3 is a schematic diagram of a method for performing video quality detection on three types of PPG video data according to Embodiment 2 of the present invention.
  • FIG. 4 is a schematic diagram of the equipment structure of a blood pressure prediction device with multiple data sources according to Embodiment 3 of the present invention.
  • Fig. 1 is a schematic diagram of a blood pressure prediction method with multiple data sources provided in Embodiment 1 of the present invention. The method mainly includes the following steps:
  • Step 1 Obtain the data source identification and original data from the upper computer.
  • the data source identifier is one of the three types of PPG original signal identifiers, the second type of PPG original signal identifiers, and the three types of PPG video identifiers;
  • the original data is the first type of PPG original signal, the second type of PPG original signal and
  • the three types of PPG video data is one of the three types of original data, and the original data corresponds to the data source identifier.
  • a data source identifier is set to distinguish the type of original data acquired:
  • Data source identification Raw data A kind of PPG original signal identification A kind of PPG original signal
  • Type II PPG original signal identification Type 2 PPG original signal Three types of PPG video logo Three types of PPG video data
  • Fig. 2 is a schematic diagram of PPG signals before and after filtering provided in an embodiment of the present invention, they are generally collected from the skin surface of the tester through a PPG signal collection device, where the signal level Those with insignificant baseline drift can be classified as a type of PPG original signal, and those with very obvious horizontal baseline drift can be classified as a type of PPG original signal.
  • the three types of PPG video data it is generally video data in a conventional video format obtained by shooting a tester's skin surface with a video shooting device.
  • Step 2 Perform data preprocessing operations on the original data according to the data source identification
  • Step 21 when the data source is identified as a type of PPG original signal identifier, perform normalization filtering processing on the type of PPG original signal to generate a standard PPG data sequence;
  • step 211 when the data source is identified as a type of PPG original signal identifier, according to a preset type of signal sampling threshold, data sampling is performed on the type of PPG original signal to generate a type of PPG sampling data sequence (X 1 , X 2 ...X i ...X M );
  • a Class PPG sample data sequence (X 1, X 2 ... X i ... X M) includes M Class PPG a sampled data X-i; M is an integer; i ranges from 1 to M;
  • a type of PPG original signal that is considered to be acquired may not be digitally converted.
  • a sampling is performed before filtering;
  • Step 212 a class of PPG sample data sequence (X 1, X 2 ... X i ... X M) normalized filtering process to generate a first sequence (Y 1, Y 2 ... Y i ... Y M),
  • the first sequence of processes (Y 1, Y 2 ... Y i ... Y M) comprising M first process data Y i; a and b are a class of preset filter constant; C is the gain of a class of the original signal PPG coefficient;
  • Step 213 Set the standard PPG data sequence as the first process sequence (Y 1 , Y 2 ...Y i ...Y M ); go to step 3;
  • Step 22 When the data source is identified as the second-type PPG original signal identifier, perform baseline drift removal and normalization filtering processing on the second-type PPG original signal to generate a standard PPG data sequence;
  • step 221 when the data source is identified as the second-class PPG original signal identification, according to the preset second-class signal sampling threshold, data sampling is performed on the second-class PPG original signal to generate the second-class PPG sampling data sequence (S 1 , S 2 ...S j ...S N );
  • the second-class PPG sampling data sequence (S 1 , S 2 ...S j ...S N ) includes N second-class PPG sampling data S j ;
  • N is an integer; the value of j ranges from 1 to N;
  • the acquired original PPG signal of the second type may have not undergone digital conversion.
  • a sampling is performed before filtering;
  • Step 222 Perform de-baseline drift filtering processing on the second-type PPG sampling data sequence (S 1 , S 2 ... S j ... S N ) to generate a second process sequence (T 1 , T 2 ... T j ... T N ),
  • T j e 1 ⁇ S j +e 2 ⁇ S j-1 -e 3 ⁇ T j-1 to set T j;
  • the second process sequence (T 1 , T 2 ...T j ...T N ) includes N second process data T j ; e 1 , e 2 and e 3 are all preset high-pass filter coefficients;
  • the baseline of the entire signal is pulled to the same horizontal line as much as possible;
  • Step 223 Extract the maximum value in the second process sequence (T 1 , T 2 ... T j ... T N ) to generate the maximum reference value max, and extract the second process sequence (T 1 , T 2 ... T j ... T N ) The minimum value of generates the minimum reference value min;
  • Step 224 Perform normalization filtering on the second process sequence (T 1 , T 2 ... T j ... T N ) to generate a third process sequence (P 1 , P 2 ... P j ... P N ), specifically using the formula Set P j;
  • the third process sequence (P 1 , P 2 ...P j ...P N ) includes N pieces of third process data P j ;
  • Step 225 Set the standard PPG data sequence as the third process sequence (P 1 , P 2 ...P j ...P N ); go to step 3;
  • Step 23 When the data source is identified as a three-type PPG video identifier, perform video quality detection and normalization signal conversion on the three-type PPG video data to generate a standard PPG data sequence;
  • step 231 when the data source is identified as the three types of PPG video identifiers, performing video data frame image extraction operations on the three types of PPG video data to generate three types of PPG video frame image sequences;
  • the three types of PPG video frame image sequence includes a plurality of three types of PPG video frame images
  • the three types of PPG video data are common standard video format files.
  • Standard video processing software or methods can be used to extract image frames from the video file in seconds, such as a video with a length of 5 seconds.
  • Step 232 Perform one-dimensional red light source signal extraction processing on all three types of PPG video frame images of the three types of PPG video frame image sequence according to the preset red light pixel threshold range to generate a first red light digital signal;
  • the pixel threshold range performs one-dimensional green light source signal extraction processing on all three types of PPG video frame images in the three types of PPG video frame image sequence to generate the first green light digital signal;
  • step 2321 initialize the first red light digital signal to be empty, initialize the first green light digital signal to be empty, initialize the value of the first index to 1, initialize the first total to three types of PPG video frame image sequences.
  • Step 2322 Set the first index frame image as the three types of PPG video frame images corresponding to the first index in the three types of PPG video frame image sequence;
  • Step 2323 In the first index frame image, count all the pixels that meet the threshold range of the red light pixel to generate a red pixel point set, count the sum of the pixels included in the red pixel point set to generate the total number of red points, and for all the red pixel point sets The pixel values of the pixel points are calculated to generate the sum of the red pixel values, and the first index frame red light channel data is generated according to the quotient of the sum of the red pixel values divided by the total number of red points; the first index frame red light channel data is used as the signal point data Add a signal point to the first red light digital signal;
  • the red in the video will also produce a difference in color depth, so the total number of pixels in the pixel threshold range) is used to generate the total number of red points, and the pixel values of all the pixels that meet the red light pixel threshold range in the first frame of image are extracted and the sum is calculated.
  • the sum of red pixel values, then the red light channel data of the first index frame is equal to the sum of red pixel values/total number of red points;
  • Step 2324 Count all the pixels that meet the threshold range of the green pixel in the first index frame image to generate a green pixel set, and calculate the sum of the pixels included in the green pixel set to generate the total number of green pixels.
  • the pixel values of the pixels are summed and calculated to generate the total green pixel value, and the first index frame green channel data is generated according to the quotient of the total green pixel value divided by the total number of green points; the first index frame green channel data is used as the signal point data Add a signal point to the first green light digital signal;
  • Step 2325 add 1 to the first index
  • Step 2326 Determine whether the first index is greater than the first total, if the first index is less than or equal to the first total, go to step 2322, and if the first index is greater than the first total, go to step 233;
  • step 232 is to perform two types of optical channel data extraction operations on all three types of PPG video frame images in the three types of PPG video frame image sequence; red light channel data and green light channel data; extraction of optical channel data
  • the method is to obtain a pixel average value by calculating the weighted average of specific pixels in the frame image, and use this to represent the color channel data of the light source in the frame image; in chronological order, do the same for each frame in the video
  • two segments of one-dimensional digital signals can be obtained: the first red light digital signal and the first green light digital signal;
  • Step 233 Perform signal bandpass filter preprocessing on the first red light digital signal to generate a second red light digital signal according to the preset bandpass filter frequency threshold range, and perform signal bandpass filter preprocessing on the first green light digital signal Generating a second green light digital signal;
  • the signal filtering preprocessing is performed on the data representing the two optical channels before normalization, that is, noise reduction processing;
  • the noise reduction method used in the first embodiment is a band-pass filtering method, that is, a band-pass filter is preset.
  • the filter frequency threshold range is based on the principle of band-pass filtering to suppress signals, interference and noise below or above the frequency band; generally the band-pass filter frequency threshold range here is common 0.5 Hz to 10 Hz; in some When performing band-pass filtering on a mobile terminal, a finite-length unit impulse response ((Finite Impulse Response, FIR) filter module is used;
  • FIR Finite Impulse Response
  • Step 234 Perform signal maximum frequency difference judgment processing on the second red light digital signal and the second green light digital signal to generate a first judgment result
  • step 2341 using discrete Fourier transform for the second red light digital signal to perform digital signal time-domain frequency domain conversion to generate a red light frequency domain signal, and using discrete Fourier transform for the second green light digital signal to generate a digital signal Time domain frequency domain conversion to generate green light frequency domain signal;
  • Step 2342 extracting the highest energy frequency from the red light frequency domain signal to generate the maximum red light frequency, and extracting the highest energy frequency from the green light frequency domain signal to generate the maximum green light frequency;
  • Step 2343 Calculate the frequency difference between the maximum frequency of red light and the maximum frequency of green light to generate the maximum frequency difference between red and green;
  • Step 2344 When the maximum red-green frequency difference does not exceed the preset maximum frequency difference threshold range, the first judgment result is set as an up-to-standard signal identifier;
  • step 234 the frequency domain signals of the second red digital signal and the second green digital signal are obtained through discrete Fourier transform, and the frequency with the highest energy is obtained from the frequency domain signal (generally this frequency usually corresponds to the heart rate).
  • the basic principle is to check whether the frequencies with the highest energy of the two digital signals are the same. If the error is within the allowable range, the first judgment result is set as the standard signal identification; if the error is large, the first judgment result is set as the non-standard signal identification ;
  • Step 235 When the first judgment result is an up-to-standard signal identifier, perform signal-to-noise ratio judgment processing on the second red light digital signal and the second green light digital signal to generate a second judgment result;
  • step 2351 when the first judgment result is an up-to-standard signal identification, according to the preset band-stop filter frequency threshold range, through multi-order Butterworth band-stop filtering, the signal frequency meets the effective signal of the band-stop filter frequency threshold range The point is removed from the second red light digital signal to generate a red light noise signal, and the effective signal points whose signal frequency meets the frequency threshold range of the band stop filter are removed from the second green light digital signal to generate a green signal through multi-order Butterworth band stop filtering.
  • Optical noise signal when the first judgment result is an up-to-standard signal identification, according to the preset band-stop filter frequency threshold range, through multi-order Butterworth band-stop filtering, the signal frequency meets the effective signal of the band-stop filter frequency threshold range The point is removed from the second red light digital signal to generate a red light noise signal, and the effective signal points whose signal frequency meets the frequency threshold range of the band stop filter are removed from the second green light digital signal to generate a green signal through multi-order Butterworth band stop filtering.
  • Step 2352 Calculate the signal energy of the second red light digital signal to generate red light signal energy, calculate the signal energy of the red light noise signal to generate red light noise energy, and generate effective red light according to the difference of the red light signal energy minus the red light noise energy Signal energy, based on the ratio of the effective red light signal energy to the red light noise energy to generate the red light signal-to-noise ratio;
  • Step 2353 Calculate the signal energy of the second green light digital signal to generate green light signal energy, calculate the signal energy of the green light noise signal to generate green light noise energy, and generate effective green light according to the difference of the green light signal energy minus the green light noise energy Signal energy, based on the ratio of the effective green light signal energy to the green light noise energy to generate the green light signal-to-noise ratio;
  • Step 2354 when any one of the red light signal-to-noise ratio and the green light signal-to-noise ratio is greater than or equal to the signal-to-noise ratio threshold value, the second judgment result is set as a standard signal indicator;
  • step 235 is to perform secondary filtering processing on the red and green light.
  • This filtering is a band-stop filtering method, that is, to suppress the signals within the frequency threshold range of the band-stop filter.
  • the multi-order Bart is used.
  • Voss band-stop filtering method for example, 4th-order Butterworth band-stop filtering, 1st-order Butterworth band-stop filtering
  • through band-stop filtering noise and interference signals are retained to generate noise signals, and then effective signals and noise
  • the signal is calculated to generate the signal-to-noise ratio; finally, the signal-to-noise ratio is used to identify whether the red and green digital signals are up to standard;
  • Step 236 When the second judgment result is the standard signal identification, the normalized PPG signal data sequence generation processing is performed on the second red light digital signal and the second green light digital signal to generate a standard PPG data sequence;
  • the standard PPG data sequence is a normalized data structure, and the data of the second red light digital signal and the second green light digital signal are color channel data whose values are both greater than 1, it needs to be processed Normalization processing, there are many specific normalization methods, and this embodiment will not be further elaborated here;
  • the second judgment result is an up-to-standard signal identifier, perform signal data normalization processing on the second red light digital signal and the second green light digital signal respectively to generate a normalized red light signal and a normalized green light signal ;
  • the standard PPG data sequence includes the red light data sequence and the green light data sequence.
  • the standard PPG data sequence generated from the three types of video data is different from the first type of PPG original signal and the second type of PPG original signal due to the number of light sources: the first type of PPG original signal and the second type of PPG original signal
  • the generated standard PPG data sequence is always single-channel data; if the video is a single light source, the standard PPG data sequence extracted from the three types of video data is single-channel data, if the video is red and green light sources, then the three types of video data are extracted
  • the standard PPG data sequence is dual-channel data.
  • Step 3 Obtain the CNN model identifier of the convolutional neural network
  • the CNN model identifier includes a type of CNN identifier and a type of CNN identifier.
  • CNN has long been one of the core algorithms in the field of feature recognition; used in image recognition, it can be used in fine classification and recognition to extract discriminative features of images for learning by other classifiers ;
  • PPG signal feature extraction calculation is performed on the input one-dimensional standard PPG data sequence: after convolution and pooling of the input standard PPG data sequence, the feature data that meets the characteristics of the PPG signal is retained to For the fully connected layer to perform regression calculation;
  • the embodiment of the present invention provides two CNN models for blood pressure prediction on the standard PPG data sequence: a type of CNN model and a type of CNN model; the CNN model identifier is used to distinguish between the two , So its value includes the first type of CNN logo and the second type of CNN logo;
  • the first of these two types of CNN models is the difference in feature extraction objects: the first type of CNN model is to directly extract the standard PPG data sequence according to the signal time domain amplitude characteristics; the second type of CNN model is to convert the standard PPG data sequence into a time domain map The data sequence then performs feature extraction on the time domain graph data sequence;
  • the first type of CNN model includes a multi-layer CNN network layer and a fully connected layer;
  • the CNN network layer includes a convolutional layer and a pooling layer;
  • the second type of CNN model includes a two-dimensional convolutional layer, a maximum pooling layer, Batch normalization layer, activation layer, addition layer, global average pooling layer, random discarding layer, and fully connected layer;
  • a type of CNN model is a CNN model that has been trained through blood pressure feature extraction, and is specifically composed of multiple layers of CNN It consists of a network layer and a fully connected layer.
  • Each CNN network layer includes a convolutional layer and a pooling layer; among them, the convolutional layer of the CNN network layer is responsible for extracting blood pressure features from the input data of the CNN model, and the CNN network
  • the pooling layer of the layer is to downsample the extraction results of the convolutional layer.
  • the preset convolutional layer number threshold indicates the specific number of CNN network layers included.
  • the output result of each CNN network layer is used as the next CNN.
  • the input of the network layer; finally, the calculation results of the CNN network layer of the threshold number of convolutional layers are input to the fully connected layer of CNN for regression calculation;
  • the second-class CNN model uses a customized convolutional network structure.
  • the model includes: two-dimensional convolutional layer, maximum pooling layer, batch normalization layer, activation layer, phase Add layer, global average pooling layer, random drop layer and fully connected layer; among them, the two-dimensional convolutional layer can contain multiple sub-convolutional layers, responsible for multiple convolution calculations on the input data, and the output volume of the two-dimensional convolutional layer
  • the product result contains multiple one-dimensional tensors;
  • the maximum pooling layer takes the maximum value in each one-dimensional vector to sample the convolution result to reduce the amount of data;
  • the batch homogenization layer Data normalization is performed on the output results of the maximum pooling layer;
  • the activation layer uses a nonlinear activation function to connect the output results of the batch normalization layer with a neural network;
  • the addition layer performs a weighted addition calculation on the output results of the activation layer;
  • the global average pooling layer performs full data weighted
  • Step 4 According to the CNN model identifier, select the corresponding CNN model to predict the blood pressure of the standard PPG data sequence;
  • Step 41 when the CNN model identifier is a type of CNN identifier, select a type of CNN model to perform blood pressure prediction on the standard PPG data sequence;
  • step 411 according to a preset first-class CNN input width threshold, perform a first-class CNN model input data conversion processing on the standard PPG data sequence to generate a four-dimensional tensor of input data;
  • a type of CNN input width threshold is the maximum value of the initial input data length of a type of CNN model
  • the input data format of a type of CNN model in the embodiment of the present invention is a four-dimensional tensor format
  • Step 412 According to a preset threshold of the number of convolutional layers, a CNN network layer of a type of CNN model is used to perform multi-layer convolution pooling calculation on the four-dimensional tensor of the input data to generate a four-dimensional tensor of feature data;
  • the preprocessed input data four-dimensional tensor is input into the CNN network layer of the trained CNN model for feature extraction.
  • the generated data format is also the four-dimensional tensor of feature data in the shape of the four-dimensional tensor.
  • the CNN network layer is composed of multiple convolutional layers and pooling layers.
  • the general structure is a layer of convolution and a layer of pooling before connecting to the next convolutional layer.
  • the final number of layers is determined by the number of convolutional layer thresholds.
  • a network with 4 convolutional layers and 4 pooling layers is called a 4-layer convolutional network.
  • the convolutional layer performs convolution operations and converts the input into outputs of different dimensions. These outputs can be regarded as another input to the input. This is a way of expression, and the pooling layer is used to output the quantity, simplify the calculation and promote the network to extract more effective information;
  • Step 413 Perform a fully connected layer input data two-dimensional matrix construction operation according to the feature data four-dimensional tensor to generate a two-dimensional matrix of input data; and use a fully connected layer of a type of CNN model to perform feature data regression calculation on the two-dimensional matrix of input data to generate blood pressure Two-dimensional matrix of regression data;
  • the input and output data format of the fully connected layer of a type of CNN model is a two-dimensional matrix format, so before using the fully connected layer for regression calculation, the four-dimensional tensor output by the CNN network layer needs to be processed by tensor dimensionality reduction.
  • the shape is switched from a four-dimensional tensor to a two-dimensional matrix shape; the fully connected layer of a type of CNN model here is composed of multiple sub fully connected layers, and each node of each sub fully connected layer is connected to the previous sub fully connected layer. All nodes are connected to integrate the features extracted from the front.
  • Each sub-fully connected layer can set the number of nodes and activation function of the layer (ReLU function is used more frequently, and it can also be changed to other functions);
  • the node of the output sub-fully-connected layer is set to 2.
  • Step 414 Obtain a preset prediction mode identifier
  • the prediction mode identifier includes two identifiers of mean prediction and dynamic prediction;
  • the prediction mode identifier is a system variable.
  • this variable can be used to clarify the further prediction output content: when the prediction mode identifier is the mean prediction, the original signal is required The average blood pressure data in the data; when the prediction mode identifier is dynamic prediction, it means that the blood pressure change data sequence within the time period of the original signal is required to be output;
  • Step 415 When the prediction mode identifier is average prediction, perform an average blood pressure calculation operation on the two-dimensional matrix of blood pressure regression data to generate an average blood pressure prediction data pair;
  • the average blood pressure prediction data pair includes the average systolic blood pressure prediction data and the average diastolic blood pressure prediction data;
  • the two-dimensional matrix of blood pressure regression data can be understood as a vector sequence including multiple one-dimensional vectors [2], and the average diastolic blood pressure prediction data can be obtained by calculating the lower value of each one-dimensional vector [2] (The reason for the lower value of the statistics is because the systolic blood pressure is greater than the diastolic blood pressure, so the lower value of the two regression calculations is naturally the predicted value of the diastolic blood pressure), and the higher of each one-dimensional vector [2] Calculate the mean value to get the predicted data of the mean systolic blood pressure (the statistically higher value is because the systolic blood pressure is greater than the diastolic blood pressure, so the higher value of the two regression calculation values is naturally the predicted value of the systolic blood pressure);
  • Step 416 When the prediction mode identifier is dynamic prediction, perform an ambulatory blood pressure data extraction operation on a two-dimensional matrix of blood pressure regression data to generate an ambulatory blood pressure prediction one-dimensional data sequence;
  • the systolic and diastolic blood pressures in all one-dimensional vectors [2] in the two-dimensional matrix of blood pressure regression data are extracted to form a data sequence, through which the dynamic changes of blood pressure over a period of time can be presented;
  • Step 42 When the CNN model identifier is the second type CNN identifier, the second type CNN model is selected to perform blood pressure prediction with wavelet transform on the standard PPG data sequence.
  • the second-class CNN model is to perform feature extraction on the time-domain map data sequence.
  • the conventional time-frequency conversion method for the signal is through Fourier transform, but the Fourier transform because its time-frequency analysis window is a fixed size, so for non-stationary
  • the embodiment of the present invention adopts wavelet transform to realize the time domain frequency domain conversion
  • the wavelet transform is a kind of time-frequency analysis method that inherits the idea of Fourier transform, In principle, the local characteristics of the signal can be highlighted at the same time;
  • the commonly used method for converting time-frequency data sequence to time-domain graph data sequence is Use Red Green Blue (RGB
  • Step 42 specifically includes: Step 421, segmenting the standard PPG data sequence to generate standard PPG data segments;
  • Step 422 Obtain a preset wavelet base type, an array of expansion factors, and an array of movement factors;
  • the stretch factor array includes H stretch factors;
  • the movement factor array includes L movement factors; H and L are both integers;
  • Continuous wavelet transform provides an important means for signal locality analysis. Compared with short-time Fourier transform, continuous wavelet transform has a window adjustable property, so it has higher analysis ability for non-stationary signals;
  • the wavelet expansion and translation operation performs multi-scale refinement of the signal, which can achieve higher time resolution in the high-frequency component of the signal, and higher frequency resolution in the low-frequency component;
  • continuous wavelet transform has three core parameters: Wavelet base, expansion factor, and movement factor; among them, the wavelet basis type is a wavelet function specifically used for wavelet transformation, the expansion factor is a scale parameter that will transform itself during the wavelet transformation process, and the movement factor is the movement that will transform itself during the wavelet transformation process Time parameter
  • Step 423 according to the expansion factor of the expansion factor array, the movement factor of the movement factor array, and the wavelet base type, the standard PPG data segment is subjected to signal decomposition processing using continuous wavelet transform to generate a PPG wavelet coefficient matrix [H,L];
  • the PPG wavelet coefficient matrix [H,L] is composed of H*L wavelet coefficients, and each wavelet coefficient is a complex number that embodies the expansion factor and the movement factor;
  • Step 424 Perform real-number matrix conversion on the PPG wavelet coefficient matrix [H,L] by taking the modulus of the matrix elements, and perform the matrix element value normalization processing on the converted matrix to generate the PPG normalization matrix [H,L ];
  • the first step is to transform the complex matrix into a real matrix, and normalize all matrix elements in the real matrix to obtain the PPG normalization matrix.
  • the PPG normalization matrix [H,L] is regarded as a data sequence To understand, it is the time-frequency data sequence obtained by the standard PPG data sequence after time-frequency domain conversion;
  • Step 425 Obtain the RGB color wheel matrix, and perform PPG time-frequency tensor conversion on the PPG normalized matrix [H, L] according to the RGB color wheel matrix to generate the PPG time-frequency three-dimensional tensor [H, L, 3];
  • the RGB color mode is a color standard in the industry. It is obtained by changing the three color channels of red (Red, R), green (Green, G), and blue (Blue, B) and superimposing them with each other. For all kinds of colors, this standard includes almost all colors that human vision can perceive, and is one of the most widely used color systems; assuming that the RGB color wheel matrix includes 256 color vectors, the length of each color vector 3, including the values of the three primary colors;
  • the value range of all matrix elements in the PPG normalization matrix [H, L] is a value in the middle of 0-1.
  • the value range between 0-1 is first divided into 256 numerical segments ;
  • poll all the elements in the PPG normalization matrix [H,L] switch the element from the original value to the index of the value segment where the value is located (for example, the first segment is 0-1/256, then if an element If the value is 1/257, the element will be changed from 1/257 to 1; for example, the 256th paragraph is 255/256-1, then if the value of an element is 511/512, the element will be changed from 511/512 to 256);
  • the value of the elements in the converted PPG normalization matrix [H,L] changes from the original 0-1 to 1-256;
  • each element of the PPG normalization matrix [H,L] can find a corresponding RGB color vector from the RGB color wheel matrix (a color vector is a color vector that contains red and blue A one-dimensional vector of triseparation values [3]), extract this RGB color vector [3] from the RGB color wheel matrix and add it to the corresponding position of the PPG normalization matrix [H, L] to generate a time-frequency diagram
  • the data sequence is the PPG time-frequency three-dimensional tensor [H,L,3];
  • Step 426 Perform tensor shape reconstruction operation on the PPG time-frequency 3D tensor [H,L,3] according to the preset input width threshold of the second type CNN to generate the PPG convolution 3D tensor [K, K,3];
  • K is the input width threshold of the second type CNN
  • the size of the PPG time-frequency three-dimensional tensor [H,L,3] deviates from the input size requirements of the second type CNN model.
  • the size of the PPG time-frequency three-dimensional tensor [H,L,3] deviates
  • Use the bicubic interpolation algorithm also called bicubic interpolation for matrix data to increase the number of matrix points through interpolation calculation, usually use interpolation technology to increase the graphic data, expand the graphic size
  • intermediate numerical points in order to achieve a change in the three-dimensional
  • the effect of expanding the size of the time-frequency image by the volume shape will finally generate a PPG convolutional three-dimensional tensor that meets the requirements of the second-class CNN model [K, K, 3];
  • Step 427 Perform blood pressure prediction processing on the PPG convolutional three-dimensional tensor [K, K, 3] using the second-class CNN model to generate a PPG predicted blood pressure data pair;
  • the PPG predicted blood pressure data pair includes PPG predicted systolic blood pressure data and PPG predicted diastolic blood pressure data.
  • the second type of CNN model in the embodiment of the present invention includes: a two-dimensional convolution layer, a maximum pooling layer, a batch normalization layer, an activation layer, an addition layer, a global average pooling layer, a random discarding layer, and a fully connected layer ;
  • the two-dimensional convolutional layer can contain multiple sub-convolutional layers, responsible for multiple convolution calculations on the input data, and the convolution result (four-dimensional tensor) output by the two-dimensional convolutional layer contains multiple one-dimensional tensors;
  • maximum The pooling layer takes the maximum value in each one-dimensional vector to sample the convolution results to reduce the amount of data;
  • the batch normalization layer is to perform data normalization on the output results of the maximum pooling layer;
  • the activation layer uses a nonlinear activation function to connect the output results of the batch normalization layer with a neural network;
  • the addition layer performs a weighted addition calculation on the output results of the activation layer;
  • the global average pooling layer
  • FIG. 3 is a schematic diagram of a method for detecting video quality with three types of PPG video data provided in the second embodiment of the present invention. The method mainly includes the following steps:
  • Step 101 Obtain the data source identification and original data from the upper computer
  • the data source identifier is one of three types of PPG original signal identifiers, two types of PPG original signal identifiers, and three types of PPG video identifiers; corresponding to the data source identifier, the original data is a type of PPG original signal , One of the three types of original data, the second type of PPG original signal and the third type of PPG video data.
  • the data source identifier is set to distinguish the types of the acquired original data:
  • the first and second types of PPG original signals as shown in Fig. 2 before and after filtering provided by the embodiment of the present invention
  • the PPG signal it is generally collected from the skin surface of the tester through a PPG signal acquisition device.
  • the signal level baseline drift is not obvious can be classified as a type of PPG original signal, and the horizontal baseline drift is very obvious.
  • Class II PPG original signal Regarding the three types of PPG video data, it is generally video data in a conventional video format obtained by shooting a tester's skin surface with a video shooting device.
  • Step 102 When the data source is identified as the three types of PPG video identifiers, perform a video data frame image extraction operation on the three types of PPG video data to generate three types of PPG video frame image sequences;
  • the three types of PPG video frame image sequence includes a plurality of three types of PPG video frame images
  • the three types of PPG video data are common standard video format files.
  • Standard video processing software or methods can be used to extract image frames from the video file in seconds, such as a video with a length of 5 seconds.
  • Step 103 Perform one-dimensional red light source signal extraction processing on all three types of PPG video frame images in the three types of PPG video frame image sequence according to the preset red light pixel threshold range to generate a first red light digital signal; according to the preset green light The pixel threshold range performs one-dimensional green light source signal extraction processing on all three types of PPG video frame images in the three types of PPG video frame image sequence to generate the first green light digital signal;
  • all three types of PPG video frame images in the three types of PPG video frame image sequence are subjected to two types of light source information extraction operations; red light and green light;
  • the weighted average of pixels is calculated to obtain a pixel average value, which represents the color channel data of the light source in the frame of the image; in chronological order, the same processing is performed on each frame in the video, and two segments of one dimension can be obtained
  • Digital signal the first red light digital signal and the first green light digital signal.
  • Step 104 Perform signal bandpass filter preprocessing on the first red light digital signal to generate a second red light digital signal according to the preset bandpass filter frequency threshold range, and perform signal bandpass filter preprocessing on the first green light digital signal Generating a second green light digital signal;
  • the signal signal preprocessing is performed on the digital signals of the two light sources extracted from the video data, that is, noise reduction processing;
  • the noise reduction method used in the first embodiment is a band-pass filtering method, that is, a band-pass filter is preset.
  • the filter frequency threshold range is based on the principle of band-pass filtering to suppress signals, interference and noise below or above the frequency band; generally the band-pass filter frequency threshold range here is common 0.5 Hz to 10 Hz; in some When performing band-pass filtering on the mobile terminal, a finite-length unit impulse response ((Finite Impulse Response, FIR) filter module is used;
  • FIR Finite Impulse Response
  • Step 105 Perform a signal maximum frequency difference judgment process on the second red light digital signal and the second green light digital signal to generate a first judgment result as a substandard signal identifier;
  • the frequency domain signals of the second red digital signal and the second green digital signal are obtained through the discrete Fourier transform, and the frequency with the highest energy is obtained from the frequency domain signal (generally this frequency usually corresponds to the heart rate).
  • the basic principle here is Check whether the frequencies with the highest energy of the two digital signals are the same. If the error is within the allowable range, set the first judgment result as the up-to-standard signal identification; if the error is large, set the first judgment result as the non-compliant signal identification.
  • Step 106 The first judgment result is a non-compliant signal identification, the PPG signal processing flow is stopped, and a warning message indicating that the PPG original signal quality is not up to standard is returned to the upper computer.
  • the distance between the skin surface of the tester and the shooting device during the video shooting process is too large to cause light leakage, which makes the red light channel data extracted from the video frame image Or the green light channel data has a large deviation, which causes the frequency difference between the two to exceed the predetermined range; once the video quality has a problem, the naturally analyzed blood pressure data will not be accurate or even may be wrong, so the video data should be stopped.
  • the upper application will also mark the video data as unqualified after obtaining the warning information that the PPG original signal quality is not up to standard, and further may initiate a processing operation similar to re-shooting.
  • FIG. 4 is a schematic diagram of a device structure of a blood pressure prediction device with multiple data sources provided in 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 embodiment of the present invention provides a multi-data source blood pressure prediction method and device, which provides two signal filtering and shaping methods for directly obtained PPG signals, and provides video quality detection and normalization for video data used to indirectly generate PPG signals
  • the signal conversion method finally generates a unified standard PPG data sequence for blood pressure prediction; subsequently, the embodiment of the present invention provides two optional CNN models for blood pressure prediction; by using the method and device provided by the embodiment of the present invention, the application is improved
  • the preprocessing capability for multiple PPG signal data sources improves the application's ability to manage multiple blood pressure prediction models, and improves the compatibility of applications for blood pressure prediction from multiple data sources.
  • 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)
  • Public Health (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • Pathology (AREA)
  • General Health & Medical Sciences (AREA)
  • Animal Behavior & Ethology (AREA)
  • Biophysics (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Veterinary Medicine (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Physiology (AREA)
  • Artificial Intelligence (AREA)
  • Cardiology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Psychiatry (AREA)
  • Signal Processing (AREA)
  • Mathematical Physics (AREA)
  • Vascular Medicine (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Hematology (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Epidemiology (AREA)
  • Primary Health Care (AREA)
  • Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
  • Image Processing (AREA)

Abstract

一种多数据源的血压预测方法和装置,方法包括:获取数据源标识和原始数据(1);根据数据源标识对原始数据进行数据预处理操作(2);当数据源标识为一类PPG原始信号标识时对一类PPG原始信号进行归一化滤波处理生成标准PPG数据序列(21);当数据源标识为二类PPG原始信号标识时对二类PPG原始信号进行去基线漂移及归一化滤波处理生成标准PPG数据序列(22);当数据源标识为三类PPG视频标识时对三类PPG视频数据进行视频质量检测及归一化信号转换生成标准PPG数据序列(23);获取卷积神经网络CNN模型标识符(3);根据数据源标识和/或CNN模型标识符,选择对应的CNN模型对标准PPG数据序列进行血压预测(4)。

Description

一种多数据源的血压预测方法和装置
本申请要求于2020年3月17日提交中国专利局、申请号为202010189221.3、发明名称为“一种多数据源的血压预测方法和装置”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,特别涉及一种多数据源的血压预测方法和装置。
背景技术
光体积变化描记图法(Photoplethysmograph,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,通过对反映血管内血液吸收光能的PPG信号波形进行分析可以对血压进行预测。
在实际应用中,我们发现,用于进行血压预测的PPG信号有多种获取方式,有直接通过PPG信号采集装置获取的,也有通过对测试者的皮肤表面进行视频拍摄间接获取的。对于直接获取的PPG信号,因为在采集过程中容易受到诸如传感器灵敏度、测试者生理状态、环境信号干扰等因素的影响,所以 常常出现变形失真的情况;对于从视频中通过红绿光通道数据归一化转换后得到的PPG信号,因为拍摄环境光强等因素也时常会导致转换生成的PPG信号失真。使用过度失真的PPG信号对血压进行预测其结果都会与实际情况出现较大偏差、甚至发生错误。
发明内容
本发明的目的,就是针对现有技术的缺陷,提供一种多数据源的血压预测方法和装置,对直接获取的PPG信号提供两种信号滤波整形方法,对用于间接生成PPG信号的视频数据提供视频质量检测及归一化信号转换方法,最后生成统一的标准PPG数据序列用于血压预测;进一步的,本发明实施例提供两种可选的卷积神经网络(Convolutional Neural Networks,CNN)模型进行血压预测;通过使用本发明实施例提供的方法和装置,提高了应用对多种PPG信号数据源的预处理能力,提高了应用对多种血压预测模型的管理能力,完善了应用对多数据源血压预测的兼容性。
为实现上述目的,本发明实施例第一方面提供了一种多数据源的血压预测方法,所述方法包括:
从上位机获取数据源标识和原始数据;所述数据源标识为一类光体积变化描记图法PPG原始信号标识,二类PPG原始信号标识和三类PPG视频标识三种数据源标识中的一种;所述原始数据为一类PPG原始信号,二类PPG原始信号和三类PPG视频数据三种原始数据中的一种,所述原始数据与所述数据源标识相对应;
根据所述数据源标识对所述原始数据进行数据预处理操作;当所述数据源标识为所述一类PPG原始信号标识时对所述一类PPG原始信号进行归一化滤波处理生成标准PPG数据序列;当所述数据源标识为所述二类PPG原始信号标识时对所述二类PPG原始信号进行去基线漂移及归一化滤波处理生成所述标准PPG数据序列;当所述数据源标识为所述三类PPG视频标识时对所述 三类PPG视频数据进行视频质量检测及归一化信号转换生成所述标准PPG数据序列;
获取卷积神经网络CNN模型标识符;所述CNN模型标识符包括所述一类CNN标识和所述二类CNN标识;
根据所述CNN模型标识符,选择对应的CNN模型对所述标准PPG数据序列进行血压预测;当所述CNN模型标识符为所述一类CNN标识时,选择一类CNN模型对所述标准PPG数据序列进行血压预测;当所述CNN模型标识符为所述二类CNN标识时,选择二类CNN模型对所述标准PPG数据序列进行带小波变换的血压预测。
优选的,所述当所述数据源标识为所述一类PPG原始信号标识时对所述一类PPG原始信号进行归一化滤波处理生成标准PPG数据序列,具体包括:
当所述数据源标识为所述一类PPG原始信号标识时,根据预置的一类信号采样阈值,对所述一类PPG原始信号进行数据采样生成一类PPG采样数据序列(X 1,X 2…X i…X M);所述一类PPG采样数据序列(X 1,X 2…X i…X M)包括所述M个一类PPG采样数据X i;所述M为整数;所述i的取值范围从1到M;
对所述一类PPG采样数据序列(X 1,X 2…X i…X M)进行归一化滤波处理生成第一过程序列(Y 1,Y 2…Y i…Y M),当所述i的取值为1时设置Y i=X i,当所述i的取值大于1时使用公式
Figure PCTCN2020129646-appb-000001
对所述Y i进行设置;所述第一过程序列(Y 1,Y 2…Y i…Y M)包括所述M个第一过程数据Y i;所述a和所述b为预置的一类滤波常数;所述c为所述一类PPG原始信号的增益系数;
设置所述标准PPG数据序列为所述第一过程序列(Y 1,Y 2…Y i…Y M)。
优选的,所述当所述数据源标识为所述二类PPG原始信号标识时对所述二类PPG原始信号进行去基线漂移及归一化滤波处理生成所述标准PPG数据序列,具体包括:
当所述数据源标识为所述二类PPG原始信号标识时,根据预置的二类信号采样阈值,对所述二类PPG原始信号进行数据采样生成二类PPG采样数据 序列(S 1,S 2…S j…S N);所述二类PPG采样数据序列(S 1,S 2…S j…S N)包括所述N个二类PPG采样数据S j;所述N为整数;所述j的取值范围从1到N;
对所述二类PPG采样数据序列(S 1,S 2…S j…S N)进行去基线漂移滤波处理生成第二过程序列(T 1,T 2…T j…T N),当所述j的取值为1时设置T j=S j,当所述j的取值大于1时使用公式T j=e 1×S j+e 2×S j-1-e 3×T j-1对所述T j进行设置;所述第二过程序列(T 1,T 2…T j…T N)包括所述N个第二过程数据T j;所述e 1、所述e 2和所述e 3均为预置的高通滤波系数;
提取所述第二过程序列(T 1,T 2…T j…T N)中的最大值生成最大参考值max,提取所述第二过程序列(T 1,T 2…T j…T N)中的最小值生成最小参考值min;
对所述第二过程序列(T 1,T 2…T j…T N)进行归一化滤波处理生成第三过程序列(P 1,P 2…P j…P N),具体为使用公式
Figure PCTCN2020129646-appb-000002
对所述P j进行设置;所述第三过程序列(P 1,P 2…P j…P N)包括所述N个第三过程数据P j
设置所述标准PPG数据序列为所述第三过程序列(P 1,P 2…P j…P N)。
优选的,所述当所述数据源标识为所述三类PPG视频标识时对所述三类PPG视频数据进行视频质量检测及归一化信号转换生成所述标准PPG数据序列,具体包括:
当所述数据源标识为所述三类PPG视频标识时对所述三类PPG视频数据进行视频数据帧图像提取操作生成三类PPG视频帧图像序列;所述三类PPG视频帧图像序列包括多个三类PPG视频帧图像;
根据预置的红光像素阈值范围对所述三类PPG视频帧图像序列的所有所述三类PPG视频帧图像进行一维红光源信号提取处理生成第一红光数字信号;根据预置的绿光像素阈值范围对所述三类PPG视频帧图像序列的所有所述三类PPG视频帧图像进行一维绿光源信号提取处理生成第一绿光数字信号;
根据预置的带通滤波频率阈值范围,对所述第一红光数字信号进行信号带通滤波预处理生成第二红光数字信号,对所述第一绿光数字信号进行信号带通滤波预处理生成第二绿光数字信号;
对所述第二红光数字信号和所述第二绿光数字信号进行信号最大频差判断处理生成第一判断结果;
当所述第一判断结果为达标信号标识时对所述第二红光数字信号和所述第二绿光数字信号进行信号信噪比判断处理生成第二判断结果;
当所述第二判断结果为所述达标信号标识时对所述第二红光数字信号和所述第二绿光数字信号进行归一化PPG信号数据序列生成处理生成所述标准PPG数据序列。
进一步的,所述根据预置的红光像素阈值范围对所述三类PPG视频帧图像序列的所有所述三类PPG视频帧图像进行一维红光源信号提取处理生成第一红光数字信号;根据预置的绿光像素阈值范围对所述三类PPG视频帧图像序列的所有所述三类PPG视频帧图像进行一维绿光源信号提取处理生成第一绿光数字信号,具体包括:
步骤51,初始化所述第一红光数字信号为空,初始化所述第一绿光数字信号为空,初始化第一索引的值为1,初始化第一总数为所述三类PPG视频帧图像序列包括的三类PPG视频帧图像总数;
步骤52,设置第一索引帧图像为所述三类PPG视频帧图像序列中与所述第一索引对应的所述三类PPG视频帧图像;
步骤53,在所述第一索引帧图像中统计所有满足所述红光像素阈值范围的像素点生成红色像素点集合,统计所述红色像素点集合中包括的像素点总和生成红色点总数,对所述红色像素点集合的所有像素点的像素值进行求和计算生成红色像素值总和,根据所述红色像素值总和除以所述红色点总数的商生成第一索引帧红光通道数据;将所述第一索引帧红光通道数据作为信号点数据向所述第一红光数字信号进行信号点添加操作;
步骤54,在所述第一索引帧图像中统计所有满足所述绿光像素阈值范围的像素点生成绿色像素点集合,统计所述绿色像素点集合中包括的像素点总和生成绿色点总数,对所述绿色像素点集合的所有像素点的像素值进行求和 计算生成绿色像素值总和,根据所述绿色像素值总和除以所述绿色点总数的商生成第一索引帧绿光通道数据;将所述第一索引帧绿光通道数据作为信号点数据向所述第一绿光数字信号进行信号点添加操作;
步骤55,将所述第一索引加1;
步骤56,判断所述第一索引是否大于所述第一总数,如果所述第一索引小于或等于所述第一总数则转至步骤52,如果所述第一索引大于所述第一总数则转至步骤57;
步骤57,将所述第一红光数字信号作为一维红光源信号提取处理结果,将所述第一绿光数字信号作为一维绿光源信号提取处理结果向上位处理流程传送。
进一步的,所述对所述第二红光数字信号和所述第二绿光数字信号进行信号最大频差判断处理生成第一判断结果,具体包括:
对所述第二红光数字信号使用离散傅里叶变换进行数字信号时域频域转换生成红光频域信号,对所述第二绿光数字信号使用离散傅里叶变换进行数字信号时域频域转换生成绿光频域信号;
从所述红光频域信号中提取能量最高频率生成红光最大频率,从所述绿光频域信号中提取能量最高频率生成绿光最大频率;
计算所述红光最大频率与所述绿光最大频率的频率差生成红绿最大频差;
当所述红绿最大频差未超过预置的最大频差阈值范围时设置所述第一判断结果为所述达标信号标识。
进一步的,所述当所述第一判断结果为达标信号标识时对所述第二红光数字信号和所述第二绿光数字信号进行信号信噪比判断处理生成第二判断结果,具体包括:
当所述第一判断结果为达标信号标识时,根据预置的带阻滤波频率阈值范围,通过多阶巴特沃斯带阻滤波将信号频率满足所述带阻滤波频率阈值范围的有效信号点从所述第二红光数字信号中去除生成红光噪声信号,通过多 阶巴特沃斯带阻滤波将信号频率满足所述带阻滤波频率阈值范围的有效信号点从所述第二绿光数字信号中去除生成绿光噪声信号;
计算所述第二红光数字信号的信号能量生成红光信号能量,计算所述红光噪声信号的信号能量生成红光噪声能量,根据所述红光信号能量减去所述红光噪声能量的差生成有效红光信号能量,根据所述有效红光信号能量与所述红光噪声能量的比值生成红光信噪比;
计算所述第二绿光数字信号的信号能量生成绿光信号能量,计算所述绿光噪声信号的信号能量生成绿光噪声能量,根据所述绿光信号能量减去所述绿光噪声能量的差生成有效绿光信号能量,根据所述有效绿光信号能量与所述绿光噪声能量的比值生成绿光信噪比;
当所述红光信噪比与所述绿光信噪比中任一个大于或等于所述信噪比阈值则设置所述第二判断结果为所述达标信号标识。
进一步的,所述当所述第二判断结果为所述达标信号标识时对所述第二红光数字信号和所述第二绿光数字信号进行归一化PPG信号数据序列生成处理生成所述标准PPG数据序列,具体包括:
当所述第二判断结果为所述达标信号标识时,分别对所述第二红光数字信号和所述第二绿光数字信号进行信号数据归一化处理生成归一化红光信号和归一化绿光信号;设置所述标准PPG数据序列的红光数据序列为所述归一化红光信号,设置所述标准PPG数据序列的绿光数据序列为所述归一化绿光信号;所述标准PPG数据序列包括所述红光数据序列和所述绿光数据序列。
优选的,
所述一类CNN模型包括多层CNN网络层和全连接层;所述CNN网络层包括1层卷积层和1层池化层;
所述二类CNN模型包括二维卷积层、最大池化层、批量均一化层、激活层、相加层、全局平均池化层、随机丢弃层和全连接层。
优选的,所述当所述CNN模型标识符为所述一类CNN标识时,选择一类 CNN模型对所述标准PPG数据序列进行血压预测,具体包括:
当所述CNN模型标识符为所述一类CNN标识时,根据预置的一类CNN输入宽度阈值,对所述标准PPG数据序列进行一类CNN模型输入数据转换处理生成输入数据四维张量;
按预置的卷积层数阈值,利用所述一类CNN模型的所述CNN网络层对所述输入数据四维张量进行多层卷积池化计算生成特征数据四维张量;
根据所述特征数据四维张量进行全连接层输入数据二维矩阵构建操作生成输入数据二维矩阵;并利用所述一类CNN模型的所述全连接层对所述输入数据二维矩阵进行特征数据回归计算生成血压回归数据二维矩阵;
获取预置的预测模式标识符;所述预测模式标识符包括均值预测和动态预测两种标识符;
当所述预测模式标识符为所述均值预测时,对所述血压回归数据二维矩阵,进行均值血压计算操作生成均值血压预测数据对;所述均值血压预测数据对包括均值收缩压预测数据和均值舒张压预测数据;
当所述预测模式标识符为所述动态预测时,对所述血压回归数据二维矩阵,进行动态血压数据提取操作生成动态血压预测一维数据序列。
优选的,所述当所述CNN模型标识符为所述二类CNN标识时,选择二类CNN模型对所述标准PPG数据序列进行带小波变换的血压预测,具体包括:
当所述数据源标识为所述一类PPG原始信号标识或所述二类PPG原始信号标识,且所述CNN模型标识符为所述二类CNN标识时,对所述标准PPG数据序列进行片段划分生成标准PPG数据片段;
获取预置的小波基类型、伸缩因子数组和移动因子数组;所述伸缩因子数组包括H个伸缩因子;所述移动因子数组包括L个移动因子;所述H与所述L均为整数;
根据所述伸缩因子数组的伸缩因子、所述移动因子数组的移动因子和所述小波基类型,对所述标准PPG数据片段使用连续小波变换方式进行信号分 解处理,生成PPG小波系数矩阵[H,L];
通过对矩阵元素取模的方式对所述PPG小波系数矩阵[H,L]进行实数矩阵转换,并将转换后的矩阵进行矩阵元素值归一化处理,生成PPG归一矩阵[H,L];
获取RGB色盘矩阵,并且根据所述RGB色盘矩阵对所述PPG归一矩阵[H,L]进行PPG时频张量转换生成PPG时频三维张量[H,L,3];
根据预置的二类CNN输入宽度阈值,使用双三次插值算法对所述PPG时频三维张量[H,L,3]进行张量形状重构操作生成PPG卷积三维张量[K,K,3];所述K为所述二类CNN输入宽度阈值;
使用所述二类CNN模型对所述PPG卷积三维张量[K,K,3]进行血压预测处理,生成PPG预测血压数据对;所述PPG预测血压数据对包括PPG预测收缩压数据和PPG预测舒张压数据。
本发明实施例第一方面提供的一种多数据源的血压预测方法,对直接获取的PPG信号提供两种信号滤波整形方法,对用于间接生成PPG信号的视频数据提供视频质量检测及归一化信号转换方法,生成统一的标准PPG数据序列以用于血压预测;进入血压预测时,又通过CNN模型标识符提供不同的血压预测模型进行血压预测;通过使用本发明实施例提供的方法和装置,提高了应用对PPG信号进行预处理的能力和兼容性,同时还能提供不同的血压预测方法。
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多数据源的血压预测方法示意图;
图2为本发明实施例提供的滤波前后的PPG信号示意图;
图3为本发明实施例二提供的一种三类PPG视频数据进行视频质量检测的方法示意图;
图4为本发明实施例三提供的一种多数据源的血压预测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。
如图1为本发明实施例一提供的一种多数据源的血压预测方法示意图所示,本方法主要包括如下步骤:
步骤1,从上位机获取数据源标识和原始数据。
其中,数据源标识为一类PPG原始信号标识,二类PPG原始信号标识和三类PPG视频标识三种数据源标识中的一种;原始数据为一类PPG原始信号,二类PPG原始信号和三类PPG视频数据三种原始数据中的一种,原始数据与数据源标识相对应。
此处,为兼容多种PPG数据源的获取途径,设置数据源标识用以区分获取的原始数据的类型:
数据源标识 原始数据
一类PPG原始信号标识 一类PPG原始信号
二类PPG原始信号标识 二类PPG原始信号
三类PPG视频标识 三类PPG视频数据
表一
关于一类和二类PPG原始信号,如图2为本发明实施例提供的滤波前后的PPG信号示意图所示,一般是通过PPG信号采集设备从测试者皮肤表面采集而来的,其中,信号水平基线漂移不明显的可以归类为一类PPG原始信号,水平基线漂移非常明显的归类为二类PPG原始信号。关于三类PPG视频数据,一般是通过视频拍摄设备对测试者的皮肤表面进行拍摄获得的常规视频格式的视频数据。
步骤2,根据数据源标识对原始数据进行数据预处理操作;
具体包括:步骤21,当数据源标识为一类PPG原始信号标识时,对一类PPG原始信号进行归一化滤波处理生成标准PPG数据序列;
具体包括:步骤211,当数据源标识为一类PPG原始信号标识时,根据预置的一类信号采样阈值,对一类PPG原始信号进行数据采样生成一类PPG采样数据序列(X 1,X 2…X i…X M);
其中,一类PPG采样数据序列(X 1,X 2…X i…X M)包括M个一类PPG采样数据X i;M为整数;i的取值范围从1到M;
此处,考虑获取的一类PPG原始信号有可能是没经过数字转换的,为能够标准化处理信号,在滤波之前先进行一次采样;
步骤212,对一类PPG采样数据序列(X 1,X 2…X i…X M)进行归一化滤波处理生成第一过程序列(Y 1,Y 2…Y i…Y M),
当i的取值为1时设置Y i=X i
当i的取值大于1时使用公式
Figure PCTCN2020129646-appb-000003
对Y i进行设置;
其中,第一过程序列(Y 1,Y 2…Y i…Y M)包括M个第一过程数据Y i;a和b为预置的一类滤波常数;c为一类PPG原始信号的增益系数;
此处,是通过调整一类PPG原始信号每个数据点的相对幅值对其进行归 一化滤波,滤波后信号形状与幅值如图2为本发明实施例提供的滤波前后的PPG信号示意图所示;
步骤213,设置标准PPG数据序列为第一过程序列(Y 1,Y 2…Y i…Y M);转至步骤3;
步骤22,当数据源标识为二类PPG原始信号标识时,对二类PPG原始信号进行去基线漂移及归一化滤波处理生成标准PPG数据序列;
具体包括:步骤221,当数据源标识为二类PPG原始信号标识时,根据预置的二类信号采样阈值,对二类PPG原始信号进行数据采样生成二类PPG采样数据序列(S 1,S 2…S j…S N);
其中,二类PPG采样数据序列(S 1,S 2…S j…S N)包括N个二类PPG采样数据S j;N为整数;j的取值范围从1到N;
此处,考虑获取的二类PPG原始信号有可能是没经过数字转换的,为能够标准化处理信号,在滤波之前先进行一次采样;
步骤222,对二类PPG采样数据序列(S 1,S 2…S j…S N)进行去基线漂移滤波处理生成第二过程序列(T 1,T 2…T j…T N),
当j的取值为1时设置T j=S j
当j的取值大于1时使用公式T j=e 1×S j+e 2×S j-1-e 3×T j-1对T j进行设置;
其中,第二过程序列(T 1,T 2…T j…T N)包括N个第二过程数据T j;e 1、e 2和e 3均为预置的高通滤波系数;
此处,是通过调整二类PPG原始信号每两个相邻数据点的相对水准基线位置的调整,将整个信号的基准线最大可能地拉到同一个水平线上;
步骤223,提取第二过程序列(T 1,T 2…T j…T N)中的最大值生成最大参考值max,提取第二过程序列(T 1,T 2…T j…T N)中的最小值生成最小参考值min;
步骤224,对第二过程序列(T 1,T 2…T j…T N)进行归一化滤波处理生成第三过程序列(P 1,P 2…P j…P N),具体为使用公式
Figure PCTCN2020129646-appb-000004
对P j进行设置;
其中,第三过程序列(P 1,P 2…P j…P N)包括N个第三过程数据P j
此处,是对二类PPG原始信号完成去基线漂移之后将每个信号点的幅值与全信号最大幅差的比例,对第二过程序列(T 1,T 2…T j…T N)进行归一化滤波,滤波后信号形状与幅值如图2为本发明实施例提供的滤波前后的PPG信号示意图所示;
步骤225,设置标准PPG数据序列为第三过程序列(P 1,P 2…P j…P N);转至步骤3;
步骤23,当数据源标识为三类PPG视频标识时,对三类PPG视频数据进行视频质量检测及归一化信号转换生成标准PPG数据序列;
具体包括:步骤231,当数据源标识为三类PPG视频标识时对三类PPG视频数据进行视频数据帧图像提取操作生成三类PPG视频帧图像序列;
其中,三类PPG视频帧图像序列包括多个三类PPG视频帧图像;
此处,三类PPG视频数据是常见的标准视频格式文件,使用标准的视频处理软件或者方法就能将视频文件以秒为单位进行图像帧的提取操作,例如一段长度为5秒的视频,每秒视频包含24帧图像,那么提取出来的三类PPG视频帧图像序列就包括5*24=120个三类PPG视频帧图像向量数据(120张图像);
步骤232,根据预置的红光像素阈值范围对三类PPG视频帧图像序列的所有三类PPG视频帧图像进行一维红光源信号提取处理生成第一红光数字信号;根据预置的绿光像素阈值范围对三类PPG视频帧图像序列的所有三类PPG视频帧图像进行一维绿光源信号提取处理生成第一绿光数字信号;
具体包括:步骤2321,初始化第一红光数字信号为空,初始化第一绿光数字信号为空,初始化第一索引的值为1,初始化第一总数为三类PPG视频帧图像序列包括的三类PPG视频帧图像总数;
步骤2322,设置第一索引帧图像为三类PPG视频帧图像序列中与第一索引对应的三类PPG视频帧图像;
步骤2323,在第一索引帧图像中统计所有满足红光像素阈值范围的像素点生成红色像素点集合,统计红色像素点集合中包括的像素点总和生成红色点总数,对红色像素点集合的所有像素点的像素值进行求和计算生成红色像素值总和,根据红色像素值总和除以红色点总数的商生成第一索引帧红光通道数据;将第一索引帧红光通道数据作为信号点数据向第一红光数字信号进行信号点添加操作;
例如,提取第1帧图像中所有满足红光像素阈值范围(因为不同位置受到光源的照射因内部结构或者血管的反射和透射的程度不一样,会导致光通过率有差异,进一步会导致拍摄下来的视频中红色也会产生颜色深浅的差别,因此采用像素阈值范围)的像素点总数生成红色点总数,提取第1帧图像中所有满足红光像素阈值范围的像素点的像素值进行总和计算生成红色像素值总和,那么第一索引帧红光通道数据就等于红色像素值总和/红色点总数;
步骤2324,在第一索引帧图像中统计所有满足绿光像素阈值范围的像素点生成绿色像素点集合,统计绿色像素点集合中包括的像素点总和生成绿色点总数,对绿色像素点集合的所有像素点的像素值进行求和计算生成绿色像素值总和,根据绿色像素值总和除以绿色点总数的商生成第一索引帧绿光通道数据;将第一索引帧绿光通道数据作为信号点数据向第一绿光数字信号进行信号点添加操作;
例如,提取第1帧图像中所有满足绿光像素阈值范围(因为不同位置受到光源的照射因内部结构或者血管的反射和透射的程度不一样,会导致光通过率有差异,进一步会导致拍摄下来的视频中绿色也会产生颜色深浅的差别,因此采用像素阈值范围)的像素点总数生成绿色点总数,提取第1帧图像中所有满足绿光像素阈值范围的像素点的像素值进行总和计算生成绿色像素值总和,那么第一索引帧绿光通道数据就等于绿色像素值总和/绿色点总数;
步骤2325,将第一索引加1;
步骤2326,判断第一索引是否大于第一总数,如果第一索引小于或等于 第一总数则转至步骤2322,如果第一索引大于第一总数则转至步骤233;
此处,步骤232,是将三类PPG视频帧图像序列中的所有三类PPG视频帧图像进行两种光通道数据的提取操作;红光通道数据和绿光通道数据;对光通道数据的提取方式,就是通过对帧图像中特定像素的加权平均计算得到一个像素均值,并以此代表该光源在所在帧图像中的颜色通道数据;按时间先后顺序,对视频中的每一帧都做同样的处理,可以得到两段一维数字信号:第一红光数字信号和第一绿光数字信号;
步骤233,根据预置的带通滤波频率阈值范围,对第一红光数字信号进行信号带通滤波预处理生成第二红光数字信号,对第一绿光数字信号进行信号带通滤波预处理生成第二绿光数字信号;
此处,是对代表两种光通道数据在归一化之前进行信号滤波预处理,即降噪处理;此处,实施例一使用的降噪手段是带通滤波方式,即预置一个带通滤波频率阈值范围,基于带通滤波原理对低于或高于该频段的信号、干扰和噪声进行信号抑制处理;一般此处的带通滤波频率阈值范围常见的0.5赫兹到10赫兹;在某些移动终端上进行带通滤波处理时,使用的是有限长单位冲激响应((Finite Impulse Response,FIR)滤波模块;
步骤234,对第二红光数字信号和第二绿光数字信号进行信号最大频差判断处理生成第一判断结果;
具体包括:步骤2341,对第二红光数字信号使用离散傅里叶变换进行数字信号时域频域转换生成红光频域信号,对第二绿光数字信号使用离散傅里叶变换进行数字信号时域频域转换生成绿光频域信号;
步骤2342,从红光频域信号中提取能量最高频率生成红光最大频率,从绿光频域信号中提取能量最高频率生成绿光最大频率;
步骤2343,计算红光最大频率与绿光最大频率的频率差生成红绿最大频差;
步骤2344,当红绿最大频差未超过预置的最大频差阈值范围时设置第一 判断结果为达标信号标识;
此处,步骤234,通过离散傅里叶变换得到第二红色数字信号和第二绿色数字信号的频域信号,通过频域信号得到能量最高的频率(一般这个频率通常对应着心率),这里的基本原理是检查这两个数字信号的能量最高的频率是否一致,如果误差在允许范围之内则设置第一判断结果为达标信号标识;如果误差较大则设置第一判断结果为不达标信号标识;
步骤235,当第一判断结果为达标信号标识时对第二红光数字信号和第二绿光数字信号进行信号信噪比判断处理生成第二判断结果;
具体包括:步骤2351,当第一判断结果为达标信号标识时,根据预置的带阻滤波频率阈值范围,通过多阶巴特沃斯带阻滤波将信号频率满足带阻滤波频率阈值范围的有效信号点从第二红光数字信号中去除生成红光噪声信号,通过多阶巴特沃斯带阻滤波将信号频率满足带阻滤波频率阈值范围的有效信号点从第二绿光数字信号中去除生成绿光噪声信号;
步骤2352,计算第二红光数字信号的信号能量生成红光信号能量,计算红光噪声信号的信号能量生成红光噪声能量,根据红光信号能量减去红光噪声能量的差生成有效红光信号能量,根据有效红光信号能量与红光噪声能量的比值生成红光信噪比;
步骤2353,计算第二绿光数字信号的信号能量生成绿光信号能量,计算绿光噪声信号的信号能量生成绿光噪声能量,根据绿光信号能量减去绿光噪声能量的差生成有效绿光信号能量,根据有效绿光信号能量与绿光噪声能量的比值生成绿光信噪比;
步骤2354,当红光信噪比与绿光信噪比中任一个大于或等于信噪比阈值则设置第二判断结果为达标信号标识;
此处,步骤235,是对红绿光做二次滤波处理,本次滤波是一种带阻滤波方式,即将属于带阻滤波频率阈值范围内的信号进行抑制,具体的采用的是多阶巴特沃斯带阻滤波方式(例如,4阶巴特沃斯带阻滤波、1阶巴特沃斯带阻 滤波);通过带阻滤波将噪声及干扰信号进行保留生成噪声信号,再进一步对有效信号和噪声信号进行计算生成信噪比;最后使用信噪比再对红、绿光数字信号进行达标与否的识别操作;
步骤236,当第二判断结果为达标信号标识时对第二红光数字信号和第二绿光数字信号进行归一化PPG信号数据序列生成处理生成标准PPG数据序列;
此处,因为标准的PPG数据序列都是归一式数据结构,而第二红光数字信号和第二绿光数字信号的数据是颜色通道数据其取值都是大于1的,所以需要将其进行归一化处理,具体的归一化方法比较多,本实施例不在此处作进一步的阐述;
具体包括:当第二判断结果为达标信号标识时,分别对第二红光数字信号和第二绿光数字信号进行信号数据归一化处理生成归一化红光信号和归一化绿光信号;设置标准PPG数据序列的红光数据序列为归一化红光信号,设置标准PPG数据序列的绿光数据序列为归一化绿光信号;标准PPG数据序列包括红光数据序列和绿光数据序列。
此处,由三类视频数据生成的标准PPG数据序列因为光源数量的原因,相对于一类PPG原始信号和二类PPG原始信号而言有个区别:一类PPG原始信号和二类PPG原始信号产生的标准PPG数据序列始终是单通道数据;如果视频中是单一光源那么三类视频数据提取出的标准PPG数据序列是单通道数据,如果视频中是红绿光源那么三类视频数据提取出的标准PPG数据序列就是双通道数据。
步骤3,获取卷积神经网络CNN模型标识符;
其中,CNN模型标识符包括一类CNN标识和二类CNN标识。
此处,对CNN模型做一下简要介绍,CNN长期以来是特征识别领域的核心算法之一;应用在图像识别中,可以在精细分类识别中用于提取图像的判别特征以供其它分类器进行学习;应用血压特征识别领域中,是对输入的一维标准PPG数据序列进行PPG信号特征提取计算:在对输入的标准PPG数据序列进行 卷积和池化之后,保留符合PPG信号特性的特征数据以供全连接层进行回归计算;本发明实施例提供了两种CNN模型对标准PPG数据序列进行血压预测:一类CNN模型和二类CNN模型;CNN模型标识符就是用于对二者进行区分识别的,所以其取值包括一类CNN标识和二类CNN标识;
这两类CNN模型,首先是特征提取对象的不同:一类CNN模型是直接对标准PPG数据序列按信号时域幅值特征进行提取;二类CNN模型是将标准PPG数据序列转换为时域图数据序列然后对时域图数据序列的进行特征提取;
其次,二者内部网络结构有所不同:
1)、一类CNN模型包括多层CNN网络层和全连接层;CNN网络层包括1层卷积层和1层池化层;二类CNN模型包括二维卷积层、最大池化层、批量均一化层、激活层、相加层、全局平均池化层、随机丢弃层和全连接层;一类CNN模型是一种已经通过血压特征提取训练完成后的CNN模型,具体由多层CNN网络层和全连接层组成,每层CNN网络层包括1层卷积层和1层池化层;其中,CNN网络层的卷积层负责对CNN模型的输入数据进行血压特征提取计算,CNN网络层的池化层则是对卷积层的提取结果进行降采样,预置的卷积层数阈值表示包含的CNN网络层具体层数,每一层CNN网络层的输出结果作为下一层CNN网络层的输入;最后完成卷积层数阈值次数的CNN网络层计算结果输入到CNN的全连接层进行回归计算;
2)、二类CNN模型相较于一类CNN模型,使用了一种定制的卷积网络结构,该模型包括:二维卷积层、最大池化层、批量均一化层、激活层、相加层、全局平均池化层、随机丢弃层和全连接层;其中二维卷积层可以包含多个子卷积层,负责对输入数据进行多次卷积计算,二维卷积层输出的卷积结果(四维张量)包含多个一维张量;最大池化层在每个一维向量里取最大值的方式对卷积结果进行采样起到降低数据量的作用;批量均一化层,是对由最大池化层的输出结果进行数据均一化处理;激活层采用非线性激活函数对批量均一化层的输出结果进行神经网络连接;相加层对激活层输出结果进行加权相加计算; 全局平均池化层对相加层输出结果进行全数据加权平均计算;随机丢弃层按随机性将全局平均池化层的输出结果进行裁剪;最终使用全连接层对裁剪后的随机丢弃层输出结果进行二分类回归计算输出舒张压和收缩压的回归计算结果。
步骤4,根据CNN模型标识符,选择对应的CNN模型对标准PPG数据序列进行血压预测;
具体包括:步骤41,当CNN模型标识符为一类CNN标识时,选择一类CNN模型对标准PPG数据序列进行血压预测;
具体包括:步骤411,根据预置的一类CNN输入宽度阈值,对标准PPG数据序列进行一类CNN模型输入数据转换处理生成输入数据四维张量;
此处一类CNN输入宽度阈值,是一类CNN模型的初始输入数据长度的最大值;本发明实施例的一类CNN模型的输入数据格式是四维张量格式;
步骤412,按预置的卷积层数阈值,利用一类CNN模型的CNN网络层对输入数据四维张量进行多层卷积池化计算生成特征数据四维张量;
此处,将预处理好的输入数据四维张量输入训练好的一类CNN模型的CNN网络层中进行特征提取生成数据格式也同样为四维张量形状的特征数据四维张量,由前文可知,CNN网络层由多层卷积层和池化层组成,一般的结构是一层卷积搭配一层池化后再连接下一个卷积层,最终层数由卷积层数阈值的数量决定,例如4个卷积层搭配4个池化层的网络被称为4层卷积网络,其中卷积层进行卷积运算,将输入转换为维度不同的输出,这些输出可以看作对输入的另一种表达方式,而池化层则是用来输出数量,简化运算同时促使网络提取更加有效的信息;
步骤413,根据特征数据四维张量进行全连接层输入数据二维矩阵构建操作生成输入数据二维矩阵;并利用一类CNN模型的全连接层对输入数据二维矩阵进行特征数据回归计算生成血压回归数据二维矩阵;
此处,一类CNN模型的全连接层的输入输出数据格式为二维矩阵格式, 所以在使用全连接层进行回归计算之前需要将CNN网络层输出的四维张量进行张量降维处理,将形状从四维张量切换至二维矩阵形状;这里的一类CNN模型的全连接层由多个子全连接层组成,每一个子全连接层的每一个结点都与上一子全连接层的所有结点相连,用来把前边提取到的特征综合起来,每个子全连接层可以设置该层的结点个数以及激活函数(ReLU函数使用较多,也可以改成其他函数);将最终输出的子全连接层的节点设置为2,经过几层全连接计算则可以得到两个回归计算值:分别代表血压的收缩压和舒张压;
步骤414,获取预置的预测模式标识符;
其中,预测模式标识符包括均值预测和动态预测两种标识符;
此处,预测模式标识符是一个系统变量,对经由全连接层完成回归计算后的血压预测值,使用该变量可以明确进一步的预测输出内容:当预测模式标识符为均值预测时表示要求原始信号中的平均血压数据;当预测模式标识符为动态预测时表示要求输出原始信号时间周期内的血压变化数据序列;
步骤415,当预测模式标识符为均值预测时,对血压回归数据二维矩阵,进行均值血压计算操作生成均值血压预测数据对;
其中,均值血压预测数据对包括均值收缩压预测数据和均值舒张压预测数据;
此处,血压回归数据二维矩阵可以理解为包括多个一维向量[2]的向量序列,统计每个一维向量[2]中的较低值进行均值计算就能得到均值舒张压预测数据(之所以统计较低值,因为收缩压是大于舒张压的,那么在两个回归计算值中较低值自然就是舒张压的预测值),统计每个一维向量[2]中的较高值进行均值计算就能得到均值收缩压预测数据(之所以统计较高值,因为收缩压是大于舒张压的,那么在两个回归计算值中较高值自然就是收缩压的预测值);
步骤416,当预测模式标识符为动态预测时,对血压回归数据二维矩阵,进行动态血压数据提取操作生成动态血压预测一维数据序列;
此处,是将血压回归数据二维矩阵中所有一维向量[2]中的收缩压和舒张 压提取出来形成一个数据序列,通过该数据序列就能呈现一段时间内的血压动态变化情况;
步骤42,当CNN模型标识符为二类CNN标识时,选择二类CNN模型对标准PPG数据序列进行带小波变换的血压预测。
这里,如前文,二类CNN模型是要对时域图数据序列进行特征提取,那么在使用二类CNN模型之前,就需要将标准PPG数据序列从时域数据序列转换为时频数据序列,再将时频数据序列转换为时域图数据序列;常规的对信号常规的时频转换方式是通过傅里叶变换,但傅里叶变换因为它的时频分析窗口为固定大小,所以对于非平稳信号的PPG信号而言,会容易丢失特征数据;本发明实施例于是采用小波变换的方式来实现时域频域转换;小波变换是继承了傅里叶变换思想的时频分析方法的一种,从其原理上可以同时突出信号的局部特征;本发明实施例采用连续小波变换方式(小波变换中的一种)对PPG信号进行转换;常用的时频数据序列转换时域图数据序列的方法是使用红绿蓝(Red Green Blue,RGB)色彩模式进行转换。
步骤42具体包括:步骤421,对标准PPG数据序列进行片段划分生成标准PPG数据片段;
步骤422,获取预置的小波基类型、伸缩因子数组和移动因子数组;
其中,伸缩因子数组包括H个伸缩因子;移动因子数组包括L个移动因子;H与L均为整数;
连续小波变换提供了一种对信号进行局部性分析的重要手段,与短时傅里叶变换相比,由于连续小波变换具有窗口可调性质,其对非平稳信号有较高的分析能力;通过小波的伸缩平移运算对信号进行多尺度细化,可以在信号的高频分量达到较高的时间分辨率,在低频分量则具有较高的频率分辨率的特性;连续小波变换有三个核心参数:小波基、伸缩因子和移动因子;其中,小波基类型是具体用于小波变换的小波函数,伸缩因子是小波变换过程中会自行变换的尺度参数,移动因子就是小波变换过程中会自行变换的移动时间参 数;
步骤423,根据伸缩因子数组的伸缩因子、移动因子数组的移动因子和小波基类型,对标准PPG数据片段使用连续小波变换方式进行信号分解处理,生成PPG小波系数矩阵[H,L];
此处,PPG小波系数矩阵[H,L]由H*L个小波系数组成,每一个小波系数是一个体现了伸缩因子和移动因子的复数;
步骤424,通过对矩阵元素取模的方式对PPG小波系数矩阵[H,L]进行实数矩阵转换,并将转换后的矩阵进行矩阵元素值归一化处理,生成PPG归一矩阵[H,L];
此处,首先是将复数矩阵取模转换为实数矩阵,在将实数矩阵中的所有矩阵元素归一化,从而得到PPG归一矩阵,如果把PPG归一矩阵[H,L]当成一个数据序列来理解的话,就是标准PPG数据序列经过时域频域转换后获得的时频数据序列;
步骤425,获取RGB色盘矩阵,并且根据RGB色盘矩阵对PPG归一矩阵[H,L]进行PPG时频张量转换生成PPG时频三维张量[H,L,3];
RGB色彩模式是工业界的一种颜色标准,是通过对红(Red,R)、绿(Green,G)、蓝(Blue,B)三个颜色通道的变化以及它们相互之间的叠加来得到各式各样的颜色的,这个标准几乎包括了人类视力所能感知的所有颜色,是运用最广的颜色系统之一;假设RGB色盘矩阵包括256个颜色向量,则每个颜色向量的长度为3,分别包括三种基色的数值;
PPG归一矩阵[H,L]中所有矩阵元素的取值范围都是0-1中间的一个值,对其进行PPG时频张量时,首先将0-1之间划分为256个数值段;然后对PPG归一矩阵[H,L]中所有元素进行轮询,将元素从原有数值切换为其数值所在数值段的索引(例如第一段为0-1/256,那么如果某元素值为1/257则将此元素从1/257变为1;例如第256段为255/256-1,那么如果某元素值为511/512则将此元素从511/512变为256);最后,经过转换后的PPG归一矩阵[H,L]中 的元素取值从原来的0-1就变为1-256;
假设RGB色盘矩阵包括256个颜色,那么PPG归一矩阵[H,L]的每一个元素就能从RGB色盘矩阵中找到一个对应的RGB颜色向量(一个颜色向量是一个包含了红蓝色三分色数值的一维向量[3]),把这个RGB颜色向量[3]从RGB色盘矩阵中提取出来添加到PPG归一矩阵[H,L]的对应位置,就生成了时频图数据序列即PPG时频三维张量[H,L,3];
步骤426,根据预置的二类CNN输入宽度阈值,使用双三次插值算法对PPG时频三维张量[H,L,3]进行张量形状重构操作生成PPG卷积三维张量[K,K,3];
其中,K为二类CNN输入宽度阈值;
此处,有可能PPG时频三维张量[H,L,3]的尺寸与二类CNN模型的输入尺寸要求有偏差,在当PPG时频三维张量[H,L,3]的尺寸偏小时使用双三次插值算法(又叫双立方插值用于矩阵数据中通过插值计算增加矩阵点数量的一种方法,通常利用插值技术增加图形数据,扩大图形尺寸)中间数值点,以达到改变三维张量形状扩大时频图尺寸的效果,最终生成符合二类CNN模型要求的PPG卷积三维张量[K,K,3];
步骤427,使用二类CNN模型对PPG卷积三维张量[K,K,3]进行血压预测处理,生成PPG预测血压数据对;
其中,PPG预测血压数据对包括PPG预测收缩压数据和PPG预测舒张压数据。
此处,本发明实施例的二类CNN模型包括:二维卷积层、最大池化层、批量均一化层、激活层、相加层、全局平均池化层、随机丢弃层和全连接层;其中二维卷积层可以包含多个子卷积层,负责对输入数据进行多次卷积计算,二维卷积层输出的卷积结果(四维张量)包含多个一维张量;最大池化层在每个一维向量里取最大值的方式对卷积结果进行采样起到降低数据量的作用;批量均一化层,是对由最大池化层的输出结果进行数据均一化处理;激活层采用 非线性激活函数对批量均一化层的输出结果进行神经网络连接;相加层对激活层输出结果进行加权相加计算;全局平均池化层对相加层输出结果进行全数据加权平均计算;随机丢弃层按随机性将全局平均池化层的输出结果进行裁剪;最终使用全连接层对裁剪后的随机丢弃层输出结果进行二分类回归计算输出PPG预测收缩压数据和PPG预测舒张压数据的回归计算结果。
如图3为本发明实施例二提供的一种三类PPG视频数据进行视频质量检测的方法示意图所示,本方法主要包括如下步骤:
步骤101,从上位机获取数据源标识和原始数据;
其中,数据源标识为一类PPG原始信号标识,二类PPG原始信号标识和三类PPG视频标识三种数据源标识中的一种;与数据源标识对应的,原始数据为一类PPG原始信号,二类PPG原始信号和三类PPG视频数据三种原始数据中的一种。
此处,为兼容多种PPG数据源的获取途径,设置数据源标识用以区分获取的原始数据的类型:关于一类和二类PPG原始信号,如图2为本发明实施例提供的滤波前后的PPG信号示意图所示,一般是通过PPG信号采集设备从测试者皮肤表面采集而来的,其中,信号水平基线漂移不明显的可以归类为一类PPG原始信号,水平基线漂移非常明显的归类为二类PPG原始信号。关于三类PPG视频数据,一般是通过视频拍摄设备对测试者的皮肤表面进行拍摄获得的常规视频格式的视频数据。
步骤102,当数据源标识为三类PPG视频标识时对三类PPG视频数据进行视频数据帧图像提取操作生成三类PPG视频帧图像序列;
其中,三类PPG视频帧图像序列包括多个三类PPG视频帧图像;
此处,三类PPG视频数据是常见的标准视频格式文件,使用标准的视频处理软件或者方法就能将视频文件以秒为单位进行图像帧的提取操作,例如一段长度为5秒的视频,每秒视频包含24帧图像,那么提取出来的三类PPG视频帧图像序列就包括5*24=120个三类PPG视频帧图像向量数据(120张图 像);
步骤103,根据预置的红光像素阈值范围对三类PPG视频帧图像序列的所有三类PPG视频帧图像进行一维红光源信号提取处理生成第一红光数字信号;根据预置的绿光像素阈值范围对三类PPG视频帧图像序列的所有三类PPG视频帧图像进行一维绿光源信号提取处理生成第一绿光数字信号;
此处,是将三类PPG视频帧图像序列中的所有三类PPG视频帧图像进行两种光源信息的提取操作;红光和绿光;对光信号的提取方式,就是通过对帧图像中特定像素的加权平均计算得到一个像素均值,并以此代表该光源在所在帧图像中的颜色通道数据;按时间先后顺序,对视频中的每一帧都做同样的处理,可以得到两段一维数字信号:第一红光数字信号和第一绿光数字信号。
步骤104,根据预置的带通滤波频率阈值范围,对第一红光数字信号进行信号带通滤波预处理生成第二红光数字信号,对第一绿光数字信号进行信号带通滤波预处理生成第二绿光数字信号;
此处,是对通过视频数据提取的两种光源的数字信号进行信号信号预处理,即降噪处理;此处,实施例一使用的降噪手段是带通滤波方式,即预置一个带通滤波频率阈值范围,基于带通滤波原理对低于或高于该频段的信号、干扰和噪声进行信号抑制处理;一般此处的带通滤波频率阈值范围常见的0.5赫兹到10赫兹;在某些移动终端上进行带通滤波处理时,使用的是有限长单位冲激响应((Finite Impulse Response,FIR)滤波模块;
步骤105,对第二红光数字信号和第二绿光数字信号进行信号最大频差判断处理生成第一判断结果为不达标信号标识;
此处,通过离散傅里叶变换得到第二红色数字信号和第二绿色数字信号的频域信号,通过频域信号得到能量最高的频率(一般这个频率通常对应着心率),这里的基本原理是检查这两个数字信号的能量最高的频率是否一致,如果误差在允许范围之内则设置第一判断结果为达标信号标识;如果误差较大则设置第一判断结果为不达标信号标识。
步骤106,第一判断结果为不达标信号标识,停止PPG信号处理流程,并向上位机返回PPG原始信号质量不达标的警告信息。
此处,导致发生该类错误的原因很多,例如,在视频拍摄的过程中测试者的皮肤表面和拍摄装置之间距离过大导致漏光,从而使得从视频帧图像中提取出的红光通道数据或者绿光通道数据产生较大偏差,继而导致二者的频差超出预定范围;一旦视频质量出现问题,自然分析出的血压数据也不会准确甚至可能是错误的,所以应该停止对视频数据的继续分析,与此同时,上位应用在获取到PPG原始信号质量不达标的警告信息之后也会对视频数据进行不合格标记,进一步还可能发起类似重新拍摄的处理操作。
如图4为本发明实施例三提供的一种多数据源的血压预测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多数据源的血压预测方法和装置,对直接获取的PPG信号提供两种信号滤波整形方法,对用于间接生成PPG信号的视频数据提供视频质量检测及归一化信号转换方法,最后生成统一的标准PPG数据序列用于血压预测;随后,本发明实施例提供两种可选的CNN模型进行血压预测;通过使用本发明实施例提供的方法和装置,提高了应用对多种PPG信号数据源的预处理能力,提高了应用对多种血压预测模型的管理能力,完善了应 用对多数据源血压预测的兼容性。
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 (14)

  1. 一种多数据源的血压预测方法,其特征在于,所述方法包括:
    从上位机获取数据源标识和原始数据;所述数据源标识为一类光体积变化描记图法PPG原始信号标识,二类PPG原始信号标识和三类PPG视频标识三种数据源标识中的一种;所述原始数据为一类PPG原始信号,二类PPG原始信号和三类PPG视频数据三种原始数据中的一种,所述原始数据与所述数据源标识相对应;
    根据所述数据源标识对所述原始数据进行数据预处理操作;当所述数据源标识为所述一类PPG原始信号标识时对所述一类PPG原始信号进行归一化滤波处理生成标准PPG数据序列;当所述数据源标识为所述二类PPG原始信号标识时对所述二类PPG原始信号进行去基线漂移及归一化滤波处理生成所述标准PPG数据序列;当所述数据源标识为所述三类PPG视频标识时对所述三类PPG视频数据进行视频质量检测及归一化信号转换生成所述标准PPG数据序列;
    获取卷积神经网络CNN模型标识符;所述CNN模型标识符包括所述一类CNN标识和所述二类CNN标识;
    根据所述CNN模型标识符,选择对应的CNN模型对所述标准PPG数据序列进行血压预测;当所述CNN模型标识符为所述一类CNN标识时,选择一类CNN模型对所述标准PPG数据序列进行血压预测;当所述CNN模型标识符为所述二类CNN标识时,选择二类CNN模型对所述标准PPG数据序列进行带小波变换的血压预测。
  2. 根据权利要求1所述的多数据源的血压预测方法,其特征在于,所述当所述数据源标识为所述一类PPG原始信号标识时对所述一类PPG原始信号进行归一化滤波处理生成标准PPG数据序列,具体包括:
    当所述数据源标识为所述一类PPG原始信号标识时,根据预置的一类信号采样阈值,对所述一类PPG原始信号进行数据采样生成一类PPG采样数据 序列(X 1,X 2…X i…X M);所述一类PPG采样数据序列(X 1,X 2…X i…X M)包括所述M个一类PPG采样数据X i;所述M为整数;所述i的取值范围从1到M;
    对所述一类PPG采样数据序列(X 1,X 2…X i…X M)进行归一化滤波处理生成第一过程序列(Y 1,Y 2…Y i…Y M),当所述i的取值为1时设置Y i=X i,当所述i的取值大于1时使用公式
    Figure PCTCN2020129646-appb-100001
    对所述Y i进行设置;所述第一过程序列(Y 1,Y 2…Y i…Y M)包括所述M个第一过程数据Y i;所述a和所述b为预置的一类滤波常数;所述c为所述一类PPG原始信号的增益系数;
    设置所述标准PPG数据序列为所述第一过程序列(Y 1,Y 2…Y i…Y M)。
  3. 根据权利要求1所述的多数据源的血压预测方法,其特征在于,所述当所述数据源标识为所述二类PPG原始信号标识时对所述二类PPG原始信号进行去基线漂移及归一化滤波处理生成所述标准PPG数据序列,具体包括:
    当所述数据源标识为所述二类PPG原始信号标识时,根据预置的二类信号采样阈值,对所述二类PPG原始信号进行数据采样生成二类PPG采样数据序列(S 1,S 2…S j…S N);所述二类PPG采样数据序列(S 1,S 2…S j…S N)包括所述N个二类PPG采样数据S j;所述N为整数;所述j的取值范围从1到N;
    对所述二类PPG采样数据序列(S 1,S 2…S j…S N)进行去基线漂移滤波处理生成第二过程序列(T 1,T 2…T j…T N),当所述j的取值为1时设置T j=S j,当所述j的取值大于1时使用公式T j=e 1×S j+e 2×S j-1-e 3×T j-1对所述T j进行设置;所述第二过程序列(T 1,T 2…T j…T N)包括所述N个第二过程数据T j;所述e 1、所述e 2和所述e 3均为预置的高通滤波系数;
    提取所述第二过程序列(T 1,T 2…T j…T N)中的最大值生成最大参考值max,提取所述第二过程序列(T 1,T 2…T j…T N)中的最小值生成最小参考值min;
    对所述第二过程序列(T 1,T 2…T j…T N)进行归一化滤波处理生成第三过程序列(P 1,P 2…P j…P N),具体为使用公式
    Figure PCTCN2020129646-appb-100002
    对所述P j进行设置;所述第三过程序列(P 1,P 2…P j…P N)包括所述N个第三过程数据P j
    设置所述标准PPG数据序列为所述第三过程序列(P 1,P 2…P j…P N)。
  4. 根据权利要求1所述的多数据源的血压预测方法,其特征在于,所述当所述数据源标识为所述三类PPG视频标识时对所述三类PPG视频数据进行视频质量检测及归一化信号转换生成所述标准PPG数据序列,具体包括:
    当所述数据源标识为所述三类PPG视频标识时对所述三类PPG视频数据进行视频数据帧图像提取操作生成三类PPG视频帧图像序列;所述三类PPG视频帧图像序列包括多个三类PPG视频帧图像;
    根据预置的红光像素阈值范围对所述三类PPG视频帧图像序列的所有所述三类PPG视频帧图像进行一维红光源信号提取处理生成第一红光数字信号;根据预置的绿光像素阈值范围对所述三类PPG视频帧图像序列的所有所述三类PPG视频帧图像进行一维绿光源信号提取处理生成第一绿光数字信号;
    根据预置的带通滤波频率阈值范围,对所述第一红光数字信号进行信号带通滤波预处理生成第二红光数字信号,对所述第一绿光数字信号进行信号带通滤波预处理生成第二绿光数字信号;
    对所述第二红光数字信号和所述第二绿光数字信号进行信号最大频差判断处理生成第一判断结果;
    当所述第一判断结果为达标信号标识时对所述第二红光数字信号和所述第二绿光数字信号进行信号信噪比判断处理生成第二判断结果;
    当所述第二判断结果为所述达标信号标识时对所述第二红光数字信号和所述第二绿光数字信号进行归一化PPG信号数据序列生成处理生成所述标准PPG数据序列。
  5. 根据权利要求4所述的多数据源的血压预测方法,其特征在于,所述根据预置的红光像素阈值范围对所述三类PPG视频帧图像序列的所有所述三类PPG视频帧图像进行一维红光源信号提取处理生成第一红光数字信号;根据预置的绿光像素阈值范围对所述三类PPG视频帧图像序列的所有所述三类PPG视频帧图像进行一维绿光源信号提取处理生成第一绿光数字信号,具体包 括:
    步骤51,初始化所述第一红光数字信号为空,初始化所述第一绿光数字信号为空,初始化第一索引的值为1,初始化第一总数为所述三类PPG视频帧图像序列包括的三类PPG视频帧图像总数;
    步骤52,设置第一索引帧图像为所述三类PPG视频帧图像序列中与所述第一索引对应的所述三类PPG视频帧图像;
    步骤53,在所述第一索引帧图像中统计所有满足所述红光像素阈值范围的像素点生成红色像素点集合,统计所述红色像素点集合中包括的像素点总和生成红色点总数,对所述红色像素点集合的所有像素点的像素值进行求和计算生成红色像素值总和,根据所述红色像素值总和除以所述红色点总数的商生成第一索引帧红光通道数据;将所述第一索引帧红光通道数据作为信号点数据向所述第一红光数字信号进行信号点添加操作;
    步骤54,在所述第一索引帧图像中统计所有满足所述绿光像素阈值范围的像素点生成绿色像素点集合,统计所述绿色像素点集合中包括的像素点总和生成绿色点总数,对所述绿色像素点集合的所有像素点的像素值进行求和计算生成绿色像素值总和,根据所述绿色像素值总和除以所述绿色点总数的商生成第一索引帧绿光通道数据;将所述第一索引帧绿光通道数据作为信号点数据向所述第一绿光数字信号进行信号点添加操作;
    步骤55,将所述第一索引加1;
    步骤56,判断所述第一索引是否大于所述第一总数,如果所述第一索引小于或等于所述第一总数则转至步骤52,如果所述第一索引大于所述第一总数则转至步骤57;
    步骤57,将所述第一红光数字信号作为一维红光源信号提取处理结果,将所述第一绿光数字信号作为一维绿光源信号提取处理结果向上位处理流程传送。
  6. 根据权利要求4所述的多数据源的血压预测方法,其特征在于,所述 对所述第二红光数字信号和所述第二绿光数字信号进行信号最大频差判断处理生成第一判断结果,具体包括:
    对所述第二红光数字信号使用离散傅里叶变换进行数字信号时域频域转换生成红光频域信号,对所述第二绿光数字信号使用离散傅里叶变换进行数字信号时域频域转换生成绿光频域信号;
    从所述红光频域信号中提取能量最高频率生成红光最大频率,从所述绿光频域信号中提取能量最高频率生成绿光最大频率;
    计算所述红光最大频率与所述绿光最大频率的频率差生成红绿最大频差;
    当所述红绿最大频差未超过预置的最大频差阈值范围时设置所述第一判断结果为所述达标信号标识。
  7. 根据权利要求4所述的多数据源的血压预测方法,其特征在于,所述当所述第一判断结果为达标信号标识时对所述第二红光数字信号和所述第二绿光数字信号进行信号信噪比判断处理生成第二判断结果,具体包括:
    当所述第一判断结果为达标信号标识时,根据预置的带阻滤波频率阈值范围,通过多阶巴特沃斯带阻滤波将信号频率满足所述带阻滤波频率阈值范围的有效信号点从所述第二红光数字信号中去除生成红光噪声信号,通过多阶巴特沃斯带阻滤波将信号频率满足所述带阻滤波频率阈值范围的有效信号点从所述第二绿光数字信号中去除生成绿光噪声信号;
    计算所述第二红光数字信号的信号能量生成红光信号能量,计算所述红光噪声信号的信号能量生成红光噪声能量,根据所述红光信号能量减去所述红光噪声能量的差生成有效红光信号能量,根据所述有效红光信号能量与所述红光噪声能量的比值生成红光信噪比;
    计算所述第二绿光数字信号的信号能量生成绿光信号能量,计算所述绿光噪声信号的信号能量生成绿光噪声能量,根据所述绿光信号能量减去所述绿光噪声能量的差生成有效绿光信号能量,根据所述有效绿光信号能量与所述绿光噪声能量的比值生成绿光信噪比;
    当所述红光信噪比与所述绿光信噪比中任一个大于或等于所述信噪比阈值则设置所述第二判断结果为所述达标信号标识。
  8. 根据权利要求4所述的多数据源的血压预测方法,其特征在于,所述当所述第二判断结果为所述达标信号标识时对所述第二红光数字信号和所述第二绿光数字信号进行归一化PPG信号数据序列生成处理生成所述标准PPG数据序列,具体包括:
    当所述第二判断结果为所述达标信号标识时,分别对所述第二红光数字信号和所述第二绿光数字信号进行信号数据归一化处理生成归一化红光信号和归一化绿光信号;设置所述标准PPG数据序列的红光数据序列为所述归一化红光信号,设置所述标准PPG数据序列的绿光数据序列为所述归一化绿光信号;所述标准PPG数据序列包括所述红光数据序列和所述绿光数据序列。
  9. 根据权利要求1所述的多数据源的血压预测方法,其特征在于,
    所述一类CNN模型包括多层CNN网络层和全连接层;所述CNN网络层包括1层卷积层和1层池化层;
    所述二类CNN模型包括二维卷积层、最大池化层、批量均一化层、激活层、相加层、全局平均池化层、随机丢弃层和全连接层。
  10. 根据权利要求9所述的多数据源的血压预测方法,其特征在于,所述当所述CNN模型标识符为所述一类CNN标识时,选择一类CNN模型对所述标准PPG数据序列进行血压预测,具体包括:
    当所述CNN模型标识符为所述一类CNN标识时,根据预置的一类CNN输入宽度阈值,对所述标准PPG数据序列进行一类CNN模型输入数据转换处理生成输入数据四维张量;
    按预置的卷积层数阈值,利用所述一类CNN模型的所述CNN网络层对所述输入数据四维张量进行多层卷积池化计算生成特征数据四维张量;
    根据所述特征数据四维张量进行全连接层输入数据二维矩阵构建操作生成输入数据二维矩阵;并利用所述一类CNN模型的所述全连接层对所述输入 数据二维矩阵进行特征数据回归计算生成血压回归数据二维矩阵;
    获取预置的预测模式标识符;所述预测模式标识符包括均值预测和动态预测两种标识符;
    当所述预测模式标识符为所述均值预测时,对所述血压回归数据二维矩阵,进行均值血压计算操作生成均值血压预测数据对;所述均值血压预测数据对包括均值收缩压预测数据和均值舒张压预测数据;
    当所述预测模式标识符为所述动态预测时,对所述血压回归数据二维矩阵,进行动态血压数据提取操作生成动态血压预测一维数据序列。
  11. 根据权利要求1所述的多数据源的血压预测方法,其特征在于,所述当所述CNN模型标识符为所述二类CNN标识时,选择二类CNN模型对所述标准PPG数据序列进行带小波变换的血压预测,具体包括:
    当所述数据源标识为所述一类PPG原始信号标识或所述二类PPG原始信号标识,且所述CNN模型标识符为所述二类CNN标识时,对所述标准PPG数据序列进行片段划分生成标准PPG数据片段;
    获取预置的小波基类型、伸缩因子数组和移动因子数组;所述伸缩因子数组包括H个伸缩因子;所述移动因子数组包括L个移动因子;所述H与所述L均为整数;
    根据所述伸缩因子数组的伸缩因子、所述移动因子数组的移动因子和所述小波基类型,对所述标准PPG数据片段使用连续小波变换方式进行信号分解处理,生成PPG小波系数矩阵[H,L];
    通过对矩阵元素取模的方式对所述PPG小波系数矩阵[H,L]进行实数矩阵转换,并将转换后的矩阵进行矩阵元素值归一化处理,生成PPG归一矩阵[H,L];
    获取RGB色盘矩阵,并且根据所述RGB色盘矩阵对所述PPG归一矩阵[H,L]进行PPG时频张量转换生成PPG时频三维张量[H,L,3];
    根据预置的二类CNN输入宽度阈值,使用双三次插值算法对所述PPG时频三维张量[H,L,3]进行张量形状重构操作生成PPG卷积三维张量[K,K,3]; 所述K为所述二类CNN输入宽度阈值;
    使用所述二类CNN模型对所述PPG卷积三维张量[K,K,3]进行血压预测处理,生成PPG预测血压数据对;所述PPG预测血压数据对包括PPG预测收缩压数据和PPG预测舒张压数据。
  12. 一种设备,包括存储器和处理器,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11任一项所述的方法。
  13. 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11任一项所述的方法。
  14. 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11任一项所述的方法。
PCT/CN2020/129646 2020-03-17 2020-11-18 一种多数据源的血压预测方法和装置 WO2021184805A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US17/906,213 US20230137333A1 (en) 2020-03-17 2020-11-18 Blood pressure prediction method and device using multiple data sources
EP20925995.1A EP4122382A4 (en) 2020-03-17 2020-11-18 METHOD AND DEVICE FOR PREDICTING BLOOD PRESSURE USING A PLURALITY OF DATA SOURCES

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010189221.3A CN111358455B (zh) 2020-03-17 2020-03-17 一种多数据源的血压预测方法和装置
CN202010189221.3 2020-03-17

Publications (1)

Publication Number Publication Date
WO2021184805A1 true WO2021184805A1 (zh) 2021-09-23

Family

ID=71198671

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/129646 WO2021184805A1 (zh) 2020-03-17 2020-11-18 一种多数据源的血压预测方法和装置

Country Status (4)

Country Link
US (1) US20230137333A1 (zh)
EP (1) EP4122382A4 (zh)
CN (1) CN111358455B (zh)
WO (1) WO2021184805A1 (zh)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111248880B (zh) * 2020-02-21 2022-07-29 乐普(北京)医疗器械股份有限公司 一种基于光体积变化描记图法信号的血压预测方法和装置
CN111358455B (zh) * 2020-03-17 2022-07-29 乐普(北京)医疗器械股份有限公司 一种多数据源的血压预测方法和装置
CN112022125A (zh) * 2020-09-28 2020-12-04 无锡博智芯科技有限公司 一种基于CNN-BiGRU模型和PPG的智能血压预测方法
CN112022126A (zh) * 2020-09-28 2020-12-04 无锡博智芯科技有限公司 一种基于CNN-BiLSTM模型和PPG的智能血压预测方法
CN112336325B (zh) * 2020-10-12 2024-02-23 乐普(北京)医疗器械股份有限公司 一种融合标定光体积描计信号数据的血压预测方法和装置
CN112315437A (zh) * 2020-10-12 2021-02-05 乐普(北京)医疗器械股份有限公司 一种基于视频数据进行血压预测的系统
CN115429241A (zh) * 2020-12-31 2022-12-06 深圳北芯生命科技股份有限公司 用于识别正常压力值的基准获取方法及基准获取模块
CN112836647B (zh) * 2021-02-05 2023-07-21 合肥工业大学 一种基于ppg信号的无创式甘油三酯估计系统
CN113397516B (zh) * 2021-06-22 2022-03-25 山东科技大学 一种面向新生儿的视觉心率估计方法、装置及系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106073729A (zh) * 2016-05-31 2016-11-09 中国科学院苏州生物医学工程技术研究所 光电容积脉搏波信号的采集方法
CN106821355A (zh) * 2017-04-01 2017-06-13 泰康保险集团股份有限公司 血压预测的方法及装置
CN108498089A (zh) * 2018-05-08 2018-09-07 北京邮电大学 一种基于深度神经网络的无创连续血压测量方法
CN109833034A (zh) * 2017-11-24 2019-06-04 深圳市岩尚科技有限公司 一种脉搏波信号中提取血压数据的方法及装置
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
CN111358455A (zh) * 2020-03-17 2020-07-03 乐普(北京)医疗器械股份有限公司 一种多数据源的血压预测方法和装置

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10561321B2 (en) * 2013-12-12 2020-02-18 Alivecor, Inc. Continuous monitoring of a user's health with a mobile device
CN105877723B (zh) * 2016-06-22 2019-11-12 中国科学院苏州生物医学工程技术研究所 无创连续血压测量装置
CN106236049A (zh) * 2016-10-12 2016-12-21 南京工程学院 基于视频图像的血压测量方法
CN108478215A (zh) * 2018-01-25 2018-09-04 深圳市德力凯医疗设备股份有限公司 基于小波分析的脑电信号去噪方法、存储介质以及装置
CN110226925B (zh) * 2019-05-30 2020-12-18 华中科技大学 一种基于脉搏波的血压检测装置
CN110458197A (zh) * 2019-07-11 2019-11-15 启东市知微电子科技有限公司 基于光电容积脉搏波的身份识别方法及其系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106073729A (zh) * 2016-05-31 2016-11-09 中国科学院苏州生物医学工程技术研究所 光电容积脉搏波信号的采集方法
CN106821355A (zh) * 2017-04-01 2017-06-13 泰康保险集团股份有限公司 血压预测的方法及装置
CN109833034A (zh) * 2017-11-24 2019-06-04 深圳市岩尚科技有限公司 一种脉搏波信号中提取血压数据的方法及装置
CN108498089A (zh) * 2018-05-08 2018-09-07 北京邮电大学 一种基于深度神经网络的无创连续血压测量方法
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
CN111358455A (zh) * 2020-03-17 2020-07-03 乐普(北京)医疗器械股份有限公司 一种多数据源的血压预测方法和装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4122382A4

Also Published As

Publication number Publication date
CN111358455B (zh) 2022-07-29
EP4122382A1 (en) 2023-01-25
US20230137333A1 (en) 2023-05-04
CN111358455A (zh) 2020-07-03
EP4122382A4 (en) 2023-08-09

Similar Documents

Publication Publication Date Title
WO2021184805A1 (zh) 一种多数据源的血压预测方法和装置
CN106073743B (zh) 用于光学体积描记图信号的噪声清除的方法和系统
CN107529646B (zh) 一种基于欧拉影像放大的无接触式心率测量方法及装置
CN109685813A (zh) 一种自适应尺度信息的u型视网膜血管分割方法
CN110751636B (zh) 一种基于改进型编解码网络的眼底图像视网膜动脉硬化检测方法
CN106846293B (zh) 图像处理方法和装置
WO2021164349A1 (zh) 一种基于光体积变化描记图法信号的血压预测方法和装置
CN111291727B (zh) 一种光体积变化描记图法信号质量检测方法和装置
CN109919954B (zh) 目标对象的识别方法和装置
CN108567418A (zh) 一种基于PCANet的脉搏信号亚健康检测方法及检测系统
WO2021164350A1 (zh) 一种生成光体积变化描记图法信号的方法和装置
Dash et al. Curvelet Transform Based on Edge Preserving Filter for Retinal Blood Vessel Segmentation.
CN115034296A (zh) 一种基于域相似性脑电跨被试源域选择方法
Dikkala et al. A comprehensive analysis of morphological process dependent retinal blood vessel segmentation
Karthika et al. Retinal image analysis using contourlet transform and multistructure elements morphology by reconstruction
CN114066846A (zh) 一种基于深度学习的ctp非急性闭塞缺血评估方法和系统
CN115409764A (zh) 一种基于域自适应的多模态眼底血管分割方法及装置
CN116071565A (zh) 一种基于高光谱成像的苹果霉心病识别方法及系统
Ghaderi et al. Retinal vessel segmentation using the 2-D Morlet wavelet and neural network
CN112716469B (zh) 一种基于指尖视频的实时心率提取方法及装置
CN115017960A (zh) 一种基于时空联合mlp网络的脑电信号分类方法及应用
Lee et al. Enhancement of blood vessels in retinal imaging using the nonsubsampled contourlet transform
Das et al. A multiresolution method for non-contact heart rate estimation using facial video frames
Ahmed et al. Automated Glaucoma Detection based on LBP Representation and GLRLM Feature Extraction Method
R S FOE NET: Segmentation of Fetal in Ultrasound Images Using V-NET

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: 20925995

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2020925995

Country of ref document: EP

Effective date: 20221017