CN114569096A - Non-contact continuous blood pressure measuring method and system based on video stream - Google Patents
Non-contact continuous blood pressure measuring method and system based on video stream Download PDFInfo
- Publication number
- CN114569096A CN114569096A CN202210193095.8A CN202210193095A CN114569096A CN 114569096 A CN114569096 A CN 114569096A CN 202210193095 A CN202210193095 A CN 202210193095A CN 114569096 A CN114569096 A CN 114569096A
- Authority
- CN
- China
- Prior art keywords
- blood pressure
- pulse wave
- video
- feature
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000036772 blood pressure Effects 0.000 title claims abstract description 179
- 238000000034 method Methods 0.000 title claims abstract description 72
- 238000012545 processing Methods 0.000 claims abstract description 38
- 238000009530 blood pressure measurement Methods 0.000 claims abstract description 24
- 238000000605 extraction Methods 0.000 claims abstract description 24
- 238000012216 screening Methods 0.000 claims abstract description 18
- 230000035488 systolic blood pressure Effects 0.000 claims abstract description 14
- 238000010276 construction Methods 0.000 claims abstract description 12
- 230000035487 diastolic blood pressure Effects 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 29
- 238000012549 training Methods 0.000 claims description 27
- 238000001914 filtration Methods 0.000 claims description 20
- 238000013528 artificial neural network Methods 0.000 claims description 19
- 238000012360 testing method Methods 0.000 claims description 19
- 238000005516 engineering process Methods 0.000 claims description 18
- 238000010187 selection method Methods 0.000 claims description 18
- 239000002245 particle Substances 0.000 claims description 16
- 239000000284 extract Substances 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 13
- 238000000354 decomposition reaction Methods 0.000 claims description 13
- 230000002776 aggregation Effects 0.000 claims description 12
- 238000004220 aggregation Methods 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 12
- 238000011156 evaluation Methods 0.000 claims description 12
- 238000012880 independent component analysis Methods 0.000 claims description 11
- 238000010606 normalization Methods 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 230000000717 retained effect Effects 0.000 claims description 6
- 238000012706 support-vector machine Methods 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 5
- 239000008186 active pharmaceutical agent Substances 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000002790 cross-validation Methods 0.000 claims description 3
- 238000009432 framing Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 238000010183 spectrum analysis Methods 0.000 claims description 3
- 230000003068 static effect Effects 0.000 claims description 3
- 238000000658 coextraction Methods 0.000 claims description 2
- 230000008030 elimination Effects 0.000 claims description 2
- 238000003379 elimination reaction Methods 0.000 claims description 2
- 230000000630 rising effect Effects 0.000 claims description 2
- 230000002123 temporal effect Effects 0.000 claims description 2
- 238000012544 monitoring process Methods 0.000 abstract description 8
- 230000006872 improvement Effects 0.000 description 6
- 238000013461 design Methods 0.000 description 4
- 238000010989 Bland-Altman Methods 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 3
- 210000001367 artery Anatomy 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000001815 facial effect Effects 0.000 description 3
- 230000003278 mimic effect Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 239000008280 blood Substances 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000000926 separation method Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 1
- 230000004931 aggregating effect Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000002526 effect on cardiovascular system Effects 0.000 description 1
- 210000000744 eyelid Anatomy 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000013186 photoplethysmography Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000004393 prognosis Methods 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/021—Measuring pressure in heart or blood vessels
- A61B5/02108—Measuring pressure in heart or blood vessels from analysis of pulse wave characteristics
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/0059—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence
- A61B5/0075—Measuring for diagnostic purposes; Identification of persons using light, e.g. diagnosis by transillumination, diascopy, fluorescence by spectroscopy, i.e. measuring spectra, e.g. Raman spectroscopy, infrared absorption spectroscopy
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/021—Measuring pressure in heart or blood vessels
- A61B5/02108—Measuring pressure in heart or blood vessels from analysis of pulse wave characteristics
- A61B5/02116—Measuring pressure in heart or blood vessels from analysis of pulse wave characteristics of pulse wave amplitude
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/024—Detecting, measuring or recording pulse rate or heart rate
- A61B5/02416—Detecting, measuring or recording pulse rate or heart rate using photoplethysmograph signals, e.g. generated by infrared radiation
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7253—Details of waveform analysis characterised by using transforms
- A61B5/7257—Details of waveform analysis characterised by using transforms using Fourier transforms
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Abstract
The invention discloses a non-contact continuous blood pressure measuring method and system based on video streaming. Belongs to the field of blood pressure measurement, and the method comprises the following steps: a pulse wave signal processing and feature extraction module; a single model construction and screening module; a PPG signal feature selection module; an SVR general blood pressure model construction module; the video signal acquisition and processing module; and a video blood pressure prediction module. The video blood pressure prediction result shows that the prediction results of the systolic pressure and the diastolic pressure meet the A-grade standard of AAMI and BHS. The invention can achieve the purpose of accurately measuring the continuous blood pressure only by collecting the pulse wave signals in the video, has simple measuring mode, does not need to contact with the human body, and has wide application value and prospect for clinical and daily health monitoring.
Description
Technical Field
The invention belongs to the field of blood pressure measurement, and particularly relates to a non-contact continuous blood pressure measurement method and system based on video streaming.
Background
The Blood Pressure (BP) directly reflects the functional state of human cardiovascular and cerebrovascular vessels, and is an important basis for disease diagnosis, curative effect evaluation and prognosis analysis. Compared with the traditional blood pressure detection method, the continuous non-invasive non-contact blood pressure monitoring method has the advantages of portability, comfort, simplicity and the like.
In recent years, some studies estimate blood pressure based on pulse wave characteristics, such as document 1(Wang, y., et al. continuous blood pressure prediction using pulse waveforms and Elman neural network in 2017 IEEE 17th International Conference on Communication Technology (ICCT). 2017), a BP prediction model is constructed by extracting a few pulse wave characteristics using a pulse wave single channel signal. However, the prediction accuracy of the method needs to be improved, and a related extraction method of the pulse wave signal is not available.
Regarding the acquisition of pulse waves, the principle acquisition technology using the photoplethysmography pulse waves is well established. Research for acquiring pulse waves based on video is continuously developing. A pulse wave signal was successfully extracted from a face video signal using the principle of Independent Component Analysis (ICA) as in document 2(Xu, f., et al., Heart rate measured based on face video sequence.2015.941435).
With the development of pulse wave extraction technology and prediction algorithm, people begin to consider the research of continuous non-invasive non-contact blood pressure measurement method based on video. Facial and palmar signals were recorded via video and Blood Pressure was predicted using an artificial intelligence model as in document 3(Y.Fang, P.Huang, M.Chung and B.Wu, "A Feature Selection Method for Vision-Based Blood Pressure Measurement," 2018IEEE International Conference on Systems, Man, and Cybernetics (SMC), Miyazaki, Japan,2018, pp.2158-2163, doi: 10.1109/SMC.2018.00371). The signals required to be acquired by the users are more, only the difference between models is compared, and the prediction precision is not high.
Chinese patent No. CN108272449 discloses a non-contact blood pressure monitoring method and system. The method comprises the steps of extracting the transmission speed of pulse waves by obtaining video signal streams of a first sampling position and a second sampling position on the body of a monitored person, converting the transmission speed into a blood pressure value, and realizing dynamic monitoring of the blood pressure. The system and the method can realize non-contact dynamic continuous blood pressure monitoring, improve the comfortable experience of the user and feed back the blood pressure condition of the user in real time. The method needs to acquire the video signals of the face and the fingertip at the same time, and the complexity of signal acquisition is greatly increased.
Chinese patent No. CN110706826 discloses a non-contact real-time multi-person heart rate and blood pressure measuring method based on video images. Acquiring a video image of a human face tissue area of a detected person in a non-contact mode, and extracting a human face image according to the video image; performing image processing on skin tissues meeting preset conditions in all the face regions and generating image data; and calculating the heart rate and blood pressure value of the detected crowd according to the image data. The method only utilizes the relationship between the pulse wave conduction time and the blood pressure to establish, the calculation error is larger in the process of measuring the blood pressure in real time, and the accuracy of the blood pressure prediction result is not described.
Chinese patent No. CN106821356 discloses a cloud continuous blood pressure measuring method and system based on an Elman neural network. And taking the extracted characteristic points of the pulse wave signals as input of an Elman neural network, predicting the blood pressure value by adopting a trained Elman neural network model, and taking the obtained predicted value as a continuous blood pressure measured value. The method utilizes a neural network method to predict blood pressure, but does not describe a non-contact pulse wave signal acquisition and processing method.
Chinese patent No. CN110090010 discloses a non-contact blood pressure measuring method and system. The method comprises the steps of screening out image frames of a first region of interest and a second region of interest after obtaining a face video of a user, obtaining a heart rate estimation value by utilizing blind source separation, extracting and determining a photoplethysmogram of the region of interest according to the heart rate estimation value, and obtaining a blood pressure value by utilizing a PPG signal time-lag input time-lag and blood pressure relation equation, so that direct contact and compression on a human body can be avoided. This method does not enable real-time blood pressure monitoring.
Chinese patent No. CN105011921 discloses a method for measuring blood pressure by video analysis. The method comprises the steps of shooting an artery video of one side of a superficial artery of a human body, measuring the blood pressure of the human body before and after the video is shot through a conventional method, analyzing a video signal according to a measured high pressure value and a measured low pressure value, calculating a correlation coefficient between the blood pressure and the artery scale, and repeating the steps during actual measurement to obtain a specifically measured high pressure value and a specifically measured low pressure value. The method for measuring the blood pressure can be obtained by shooting videos and analyzing the videos, and the blood pressure measurement can be completed without contacting a human body. The method needs to measure the high pressure value and the low pressure value of the human body by using a conventional method before and after the video is shot, and the operation is complicated.
Chinese patent No. CN108523867 discloses a self-calibration PPG non-invasive blood pressure measurement method. The method comprises the steps of measuring a PPG blood pressure measurement value according to a linear relation between pulse wave conduction time and blood pressure, and calibrating the PPG blood pressure measurement value according to a blood pressure calibration model and personal characteristic information. The method calibrates the PPG blood pressure measurement value according to the personal characteristic information to improve the measurement efficiency and result, so that the result of the blood pressure measurement is less interfered. The method is a contact type blood pressure measurement method, and a bracelet is required to be worn to acquire PPG signals.
Disclosure of Invention
In order to solve the problems of the blood pressure measuring method and meet the requirements of people on blood pressure measurement under the rapid development of intelligent medical treatment, the invention provides a non-contact continuous blood pressure measuring method and system based on video stream, which realize the real-time monitoring of blood pressure information and have great application prospect in daily life and clinical monitoring. The prediction from the face video to the blood pressure is verified to be in accordance with the AAMI and BHS standards, the accuracy of obtaining the blood pressure based on the video signal is improved, and the blood pressure condition of the user is fed back in real time.
The invention is realized by adopting the following technical scheme:
a video stream based non-contact continuous blood pressure measurement system comprising:
the pulse wave signal processing and feature extraction module is used for preprocessing the pulse wave signal by adopting a discrete wavelet decomposition technology and a wavelet denoising method, analyzing the waveform feature of PPG and the expression of physiological information of a reference human body in PPG waveform, and extracting the time-frequency domain feature of a pulse wave sequence;
the single-person model building and screening module is used for building three single-person blood pressure prediction models which are respectively a single-person blood pressure prediction model based on an Elman neural network optimized by a particle swarm, a single-person blood pressure prediction model based on a deep belief network and an SVR single-person blood pressure prediction model according to the time-frequency domain characteristics of the pulse wave sequence extracted by the pulse wave signal processing and characteristic extracting module, evaluating and screening the three single-person blood pressure prediction models to obtain an optimal model which is an SVR single-person blood pressure prediction model;
the PPG signal feature selection module is used for carrying out feature sorting according to the time-frequency domain features of the pulse wave sequence extracted by the pulse wave signal processing and feature extraction module to obtain a feature weight sorting result;
the SVR general blood pressure model construction module is used for selecting the time-frequency domain characteristics of different numbers of pulse wave sequences to reconstruct an SVR single blood pressure prediction model according to the characteristic weight sorting result to obtain an SVR general blood pressure prediction model, and performing consistency analysis on the prediction result and the actual blood pressure to obtain and verify an optimal characteristic subset and the stability of the optimal characteristic subset;
the video signal acquisition and processing module is used for acquiring tested face video data and a fingertip PPG signal by using a camera, extracting an RGB (red, green and blue) sequence from an interested region by using a framing and selecting technology for the acquired face video data, filtering the RGB sequence by using a color distortion filtering algorithm, extracting a video pulse wave signal from the filtered RGB sequence by using an independent component analysis algorithm, and performing consistency estimation with the fingertip PPG signal;
the video blood pressure prediction module extracts the time-frequency domain characteristics of a video pulse wave sequence by adopting a method in the pulse wave signal processing and characteristic extraction module, extracts a video pulse wave characteristic subset according to the optimal pulse wave characteristic subset, performs consistency estimation on the video pulse wave characteristic subset and the optimal pulse wave characteristic subset of the fingertip PPG signal to obtain the optimal video pulse wave characteristic subset which is used as input, and performs blood pressure prediction by utilizing an SVR general blood pressure prediction model obtained by training.
The invention is further improved in that the pulse wave signal processing and feature extracting module comprises:
extracting time domain characteristics, namely positioning the wave peaks of the pulse wave by adopting a findpeaks function of Matlab and setting the minimum amplitude and the minimum distance of adjacent points, further realizing the positioning of the wave peaks of the pulse wave by searching the minimum value between the adjacent wave peaks, realizing the calculation of a rising area SS and a falling area DS by adopting a trapz function of Matlab, and realizing the calculation of first-order differential of the pulse wave signal by adopting a diff function to obtain the time domain characteristics of the pulse wave;
performing frequency domain characteristic extraction, performing Fourier transform on 10 pulse wave signals on the left and right sides which take the current pulse wave as the center by using a fft function of Matlab, determining the frequency and amplitude between 0.3-1.6Hz and 1.6-3Hz in the fundamental wave and the second harmonic of the pulse wave signals by using a max function of Matlab, performing 6-layer decomposition on the pulse wave signals by using a db6 wavelet as a mother wavelet to extract wavelet domain characteristics, performing Hilbert transform on IMF components obtained by decomposition to obtain Hilbert transform domain characteristics;
the co-extraction includes 78 time-frequency domain features including temporal features, amplitude features, and area features.
The further improvement of the invention is that the single-person model construction and screening module comprises:
establishing a single blood pressure prediction model of the particle swarm optimized Elman neural network, adopting a single hidden layer, adopting a particle swarm optimization algorithm to search for an optimal network weight and a threshold, adopting variation operation, and reinitializing particles with random probability after each particle update;
establishing an SVR single blood pressure prediction model, training by adopting a Libsvm library, determining a hyper-parameter in the model by adopting an element-SVR model of a radial basis function kernel and a grid parameter optimizing function SVMcgForRegress for regression, namely: a penalty coefficient C, wherein the tolerance of a kernel parameter g and a termination criterion in the kernel function belongs to the E;
establishing a single blood pressure prediction model based on a deep belief network, adopting a limited Boltzmann machine for pre-training, keeping as much characteristic information as possible when a characteristic parameter is subjected to dimension reduction, then initializing a neural network weight by utilizing the trained limited Boltzmann machine network parameter, and adjusting the weight by utilizing training data in combination with an error back-propagation algorithm;
when three single blood pressure prediction models are trained, 60% of data are used for model training and 40% of data are used as test data by adopting a three-fold cross validation method; according to the AAMI standard of blood pressure measurement, the average of the average absolute error and standard deviation of the predicted blood pressure and the actual blood pressure of the obtained 3 times of training is used as an evaluation index of the accuracy of the model, and the three modeling methods are compared and evaluated to obtain the optimal SVR single blood pressure prediction model.
A further improvement of the present invention is that the PPG signal feature selection module further comprises a feature selection method designed herein, specifically comprising:
using 65 tested data to participate in feature selection, wherein each tested data is about 3 data in different time periods, and 190 groups of data are provided, and the length of each data period is ten minutes;
selecting a filtration type feature selection method of RRelieff and neighbor component analysis and two embedded feature selection methods of recursive feature elimination of a support vector machine depending on the support vector machine and average influence value depending on a neural network, and selecting integrated features by adopting a stability aggregation technology based on ranking;
and performing stability aggregation on 190 groups of feature sequences obtained by one feature selection method to obtain the feature weight sequences of the feature selection method, and performing stability aggregation on feature weight sequences obtained by four different feature selection methods to obtain a final feature weight sequencing result.
The invention has the further improvement that the SVR general blood pressure model building module comprises:
the SVR general blood pressure model construction data is a random group of data in 65 tested 3 groups of data in the feature selection module, the test data is 213 groups of data in the other two groups of 65 tested 3 groups of data participating in feature selection and 37 tested data participating in feature selection, the feature parameter numbers are respectively valued at 3-20 and 30, 40, 50 and 60 for testing, and the random data in the 65 tested data is used for constructing the SVR general blood pressure model;
taking the average absolute error, the mean value of the standard deviation and the standard deviation of 213 groups of prediction data as the evaluation standard of the selected feature number, preliminarily screening out an optimal feature subset to obtain an optimal feature subset of the systolic pressure of 6 and an optimal feature subset of the diastolic pressure of 7, and adding 263 groups of 5 tested groups of data collected by an experimental platform to test and verify the SVR general blood pressure model and the optimal feature subset number;
and selecting 15, 25, 35, 45 and 55 tested data to participate in feature selection, and comparing the first 6 features of the systolic pressure prediction retained feature weight result and the first 7 features of the diastolic pressure prediction retained feature weight result with the optimal feature subset obtained by selecting 65 tested participated features according to the determined optimal feature subset to verify the stability of the optimal feature subset.
A further improvement of the invention is a video signal processing module comprising:
recording the dynamic characteristics of the face by using a camera, wherein the frame rate is 50fps, the distance is 1m, synchronously acquiring a fingertip PPG signal, the sampling rate is 1000Hz, and the testee keeps a static state in the acquisition process;
extracting the interesting region by adopting a frame selection technology, reading the video into a frame by using a VideoReader function of matlab, selecting the interesting region by taking a picture of an intermediate frame, and respectively calculating the gray average value of the interesting region of each frame of RGB three channels to obtain an RGB sequence;
filtering the video signal by using a color distortion filtering algorithm, performing time normalization processing on the RGB sequence, performing characteristic conversion after Fourier transformation, performing inverse Fourier transformation after performing energy calculation and weight calculation on the converted RGB sequence, and performing inverse time normalization to obtain the filtered RGB sequence;
extracting a component 1, a component 2 and a component 3 from the filtered RGB sequence by using an independent component analysis algorithm, automatically identifying the three components by using spectral analysis, selecting the component 1 with obvious PPG characteristics, and denoising the component 1 by using a Chebyshev I type IIR filter for band-pass filtering to obtain a video pulse wave signal;
the accuracy estimation is carried out on the video pulse wave signals by utilizing fingertip PPG signals, the similarity of the waveform contrast evaluation output signals and the correctness of the instantaneous heart rate contrast evaluation output signals containing physiological information are achieved.
In a further improvement of the present invention, the video blood pressure prediction module comprises:
extracting time-frequency domain characteristics of a video pulse wave sequence by adopting a method in a pulse wave signal processing and characteristic extraction module, extracting a video pulse wave characteristic subset according to the optimal characteristic subset of the pulse wave, evaluating the correctness of the video characteristic subset which can be used for blood pressure prediction by comparing the optimal characteristic subset of a fingertip PPG signal with the video pulse wave characteristic subset, and eliminating the characteristic that the average absolute percentage error is higher than 20% again to obtain the optimal video pulse wave characteristic subset as test data;
the optimal video pulse wave feature subset is 6 features including P5, F1, F2, delta T, RP5 and RP6, and based on 5 tested video PPG signal experimental data acquired by an experimental platform, 15-30 groups of data in different time periods are acquired for each test, and 100 groups of data are acquired in about 2 minutes for each group of data; and (3) taking the optimal video pulse wave feature subset of 100 groups of data as input, and performing blood pressure prediction by using the SVR general blood pressure prediction model obtained by training.
A non-contact continuous blood pressure measuring method based on video streaming comprises the following steps:
the pulse wave signal processing and feature extraction module is used for preprocessing the pulse wave signals by adopting a discrete wavelet decomposition technology and a wavelet denoising method, analyzing the waveform features of PPG and the expression of physiological information of a reference human body in PPG waveforms, and extracting the time-frequency domain features of a pulse wave sequence;
the single model building and screening module builds three single blood pressure prediction models which are respectively a single blood pressure prediction model based on an Elman neural network optimized by a particle swarm, a single blood pressure prediction model based on a deep belief network and an SVR single blood pressure prediction model according to the time-frequency domain characteristics of the pulse wave sequence extracted by the pulse wave signal processing and characteristic extraction module, and evaluates and screens the three single blood pressure prediction models to obtain an optimal model which is an SVR single blood pressure prediction model;
the PPG signal feature selection module performs feature sorting according to the time-frequency domain features of the pulse wave sequence extracted by the pulse wave signal processing and feature extraction module to obtain a feature weight sorting result;
the SVR general blood pressure model construction module selects time-frequency domain characteristics of different numbers of pulse wave sequences to reconstruct an SVR single blood pressure prediction model according to the characteristic weight sorting result to obtain an SVR general blood pressure prediction model, and performs consistency analysis on the prediction result and the actual blood pressure to obtain and verify an optimal characteristic subset and the stability thereof;
the video signal acquisition and processing module firstly acquires tested face video data and fingertip PPG signals by using a camera, extracts RGB sequences from an interested region by using a framing technology for the acquired face video data, filters the RGB sequences by using a color distortion filtering algorithm, extracts video pulse wave signals from the filtered RGB sequences by using an independent component analysis algorithm, and performs consistency estimation with the fingertip PPG signals;
the video blood pressure prediction module extracts the time-frequency domain characteristics of a video pulse wave sequence by adopting a method in the pulse wave signal processing and characteristic extraction module, extracts a video pulse wave characteristic subset according to the optimal pulse wave characteristic subset, performs consistency estimation on the video pulse wave characteristic subset and the optimal pulse wave characteristic subset of the fingertip PPG signal to obtain the optimal video pulse wave characteristic subset which is used as input, and performs blood pressure prediction by utilizing an SVR general blood pressure prediction model obtained by training.
The invention has at least the following beneficial technical effects:
1. non-contact portability
After the facial signals acquired by the video are processed, the pulse wave signals are extracted for blood pressure prediction, compared with the traditional method, the system has the characteristic of non-contact, can be applied to a mobile phone, and can acquire the video signals by using a mobile phone camera, so that convenient and rapid blood pressure measurement is realized.
2. Enabling continuous blood pressure prediction
Three single blood pressure prediction models and an SVR general blood pressure model are constructed based on continuous pulse wave and blood pressure data, continuous blood pressure value blood pressure estimation is carried out by adopting the continuous pulse wave data extracted from the video, the prediction result meets the AAMI and BHS standards, and continuous blood pressure prediction is realized.
3. The running speed is high
6/7 features are extracted from 78 pulse wave feature parameters by adopting a feature selection method of a stability aggregation technology based on ranking combined with data diversity and functional diversity to predict diastolic pressure and systolic pressure, an SVR (support vector regression) general blood pressure model is constructed, and the training speed and the prediction speed of the model are improved.
4. Strong stability
For the screened single blood pressure prediction model, the predicted result of the SVR general blood pressure model is subjected to consistency analysis with the existing blood pressure measuring method, so that the method has better consistency and the two methods can be used in place of each other; selecting the optimal feature subset screened out, performing tests with feature numbers of 3-20, 30, 40, 50 and 60, and respectively selecting 15, 25, 35, 45 and 55 tested participated feature choices to verify the stability and universality of the optimal feature subset; and for the pulse wave signals extracted from the video, carrying out consistency estimation on the pulse wave signals and fingertip PPG signals which are synchronously acquired, and determining an optimal video pulse wave feature subset for blood pressure prediction.
5. The blood pressure prediction precision is improved, and the prediction result meets the A-grade standard of AAMI and BSH.
The data diversity, the experimental data come from pulse wave, blood pressure and video signal and continuous blood pressure signal based on MIMIC of the blood pressure database collected from the experimental platform; extracting the time-frequency domain characteristics of the pulse wave sequence, providing an extraction method of 78 time-frequency domain characteristics of the pulse wave, establishing a pulse wave characteristic parameter library as complete as possible, and using the pulse wave characteristic parameter library for single-person model construction and screening; screening a blood pressure prediction model, namely establishing three single blood pressure prediction models according to the time-frequency domain characteristics of the pulse wave sequence extracted by the pulse wave signal processing and characteristic extraction module, and screening an optimal model for constructing a general blood pressure model; sorting features, namely eliminating relevant features of interference blood pressure prediction in pulse waves by adopting four feature selection methods and an integrated stability aggregation method; the method comprises the steps of video signal preprocessing, on the basis of utilizing the victory and optical characteristics of skin, improving the accuracy of extracting pulse wave signals through a signal separation technology, filtering RGB sequences by adopting a color distortion filtering algorithm, and extracting the video pulse wave signals from the filtered RGB sequences by utilizing an independent component analysis algorithm.
Drawings
Fig. 1 is a design flow chart of a non-contact non-invasive continuous blood pressure measurement method based on video streaming.
FIG. 2 is a flow chart of a design for ranking pulse wave features according to the present invention.
Fig. 3 is a flow chart of video signal acquisition and processing.
Fig. 4 is a graph showing the results of video blood pressure prediction, in which fig. 4(a) shows the results of SBP prediction of 100 sets of experimental data by the same SVR prediction model, and fig. 4(b) shows the results of DBP prediction of 100 sets of experimental data by the common SVR prediction model.
FIG. 5 is a graph of the results of the coincidence analysis of the blood pressure estimated from the video and the standard blood pressure measured experimentally, wherein FIG. 5(a) is the correlation between the predicted DBP and the actual DBP, FIG. 5(b) is a graph of the DBP prediction result Bland-Altman, FIG. 5(c) is the correlation between the predicted SBP and the actual SBP, and FIG. 5(d) is a graph of the SBP prediction result Bland-Altman.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art. It should be noted that the embodiments and features of the embodiments may be combined with each other without conflict. The present invention will be described in detail below with reference to the embodiments with reference to the attached drawings.
Fig. 1 is a design flow chart of a non-contact non-invasive continuous blood pressure measurement method based on video streaming. The system completely establishes a non-contact non-invasive continuous blood pressure measurement process based on video stream, extracts the time-frequency domain characteristics of 78 pulse wave sequences according to MIMIC database data and experimental acquisition data, is used for constructing and screening single models, evaluates and screens three single blood pressure prediction models, and establishes an optimal model as an SVR single blood pressure prediction model; carrying out feature screening on the time-frequency domain features of the extracted pulse wave sequence to determine an optimal feature subset, establishing an SVR general blood pressure model by adopting an optimal model screened by single blood pressure prediction, and further verifying the stability of the optimal feature subset and the model; according to a facial video acquired by an experiment platform, after an interested region is selected, pulse wave signals are extracted and accuracy evaluation is carried out on the pulse wave signals and fingertip PPG signals to obtain an optimal video pulse wave feature subset, and blood pressure prediction is carried out by utilizing an SVR general blood pressure prediction model obtained by training before, so that blood pressure measurement based on the video is realized.
The invention analyzes the waveform characteristics of PPG and the performance of physiological information of a reference human body in PPG waveform, extracts the characteristics of a pulse wave sequence as completely as possible, and the time-frequency domain characteristic parameters of the pulse wave used in the final research comprise 78 characteristics in total, namely time characteristics, amplitude characteristics, area characteristics, first-order differential characteristics, physiological parameter characteristics, Fourier transform domain characteristics, wavelet domain characteristics and Hilbert transform domain characteristics. Except the pulse wave shape characteristic quantity K, the other time characteristics and the amplitude characteristics are obtained according to mathematical operation after the peak and trough characteristic points of the pulse wave are obtained. And (3) positioning the pulse wave peaks by adopting a findpeaks function of Matlab and setting a minimum amplitude and a minimum distance between adjacent points, so that the pulse wave peaks are further positioned by searching for a minimum value between the adjacent wave peaks. In the area characteristics, the calculation of the ascending area SS and the descending area DS is realized by adopting a Matlab trapz function, and the other characteristics in the area characteristics can be obtained through formula mathematical operation by combining corresponding time parameters. After the area characteristic is calculated, when the pulse wave waveform characteristic quantity K is calculated, the value of Pm is the area variation of unit time, and the values of Ps and Pd are the peak value and the valley value of the pulse wave signal in the period. In the first-order differential characteristic, a diff function is adopted to realize the calculation of the first-order differential of the pulse wave signal, and the positioning of the wave crest and the wave trough point of the first-order differential signal is realized according to the findpeaks function, so that the corresponding first-order differential characteristic parameter can be calculated. The remaining physiological parameter features can be obtained by mathematical operation of the calculated features. In the feature extraction of the fourier transform domain, fourier transform is performed on 10 pulse wave signals (21 pulse wave signals in total) on the left and right sides around the current pulse wave as a center by using a fft function of Matlab. And (3) extracting the characteristics of a wavelet domain, namely performing 6-layer decomposition on the pulse wave signal by using a 'db 6' wavelet as a mother wavelet to obtain a low-frequency approximation signal Aj and a high-frequency detail signal Dj. EDM decomposition of pulse wave signals is realized by adopting a Maltab empirical mode decomposition tool box, a group of pulse wave signals are taken as an example, a residual sequence of 10 IMF components can be obtained after EDM decomposition, and IMF energy moment characteristics Ti can be obtained by normalization processing.
The single person model is constructed, 102 MIMIC databases are utilized together with 5 pieces of tested pulse wave-blood pressure pair data collected based on an experimental platform, and the length of each piece of tested pulse wave-blood pressure pair data is 5-100 minutes. Before the single-person model is constructed, preprocessing training data, marking the corresponding position of the characteristic parameter mean value which is more than 1.7 times or less than 0.3 times in the training data as 1, taking the systolic pressure as a characteristic, and marking the corresponding position of two adjacent systolic pressure changes which exceed 5 as 1; and counting the number of abnormal characteristic parameters in each group of training data, and if the abnormal data amount exceeds a threshold value 15, removing the group of data as abnormal data. And then, carrying out normalization operation on the characteristic parameters and the blood pressure, and respectively applying the same mapping rule to the characteristic parameters and the blood pressure data in the test data. The method comprises the steps of establishing a single blood pressure prediction model of an Elman neural network optimized by particle swarm, setting and inputting the Elman neural network into 78 characteristic parameters, determining the value of a hidden node between 6 and 19, and outputting the predicted systolic pressure and the predicted diastolic pressure. A single hidden layer is adopted, the optimal network weight and threshold are found by adopting a particle swarm optimization algorithm, variation operation is adopted, and particles are reinitialized according to random probability after each particle update; establishing an SVR single blood pressure prediction model, training by adopting a Libsvm library, determining a hyper-parameter in the model by adopting an element-SVR model of a radial basis function kernel and a grid parameter optimizing function SVMcgForRegress for regression, namely: a penalty coefficient C, wherein the tolerance of a kernel parameter g and a termination criterion in the kernel function belongs to the E; establishing a single blood pressure prediction model based on a deep belief network, adopting a limited Boltzmann machine for pre-training, keeping as much characteristic information as possible when a characteristic parameter is subjected to dimension reduction, then initializing a neural network weight by utilizing the trained limited Boltzmann machine network parameter, and adjusting the weight by utilizing training data in combination with an error back-propagation algorithm; when three single blood pressure prediction models are trained, 60% of data are used for model training and 40% of data are used as test data by adopting a three-fold cross validation method; according to the AAMI standard of blood pressure measurement, the average of the average absolute error and standard deviation of the predicted blood pressure and the actual blood pressure of the obtained 3 times of training is used as an evaluation index of model accuracy, and the three modeling methods are compared and evaluated to obtain the optimal SVR single blood pressure prediction model. For 102 tested samples, the SBP prediction results obtained by the SVR construction model all meet the conditions that the average absolute error is less than or equal to 3.1mmHg, the standard error difference is less than or equal to 3.9mmHg, the DBP prediction results all meet the conditions that the average absolute error is less than or equal to 1.75mmHg and the standard error difference is less than or equal to 2.7 mmHg.
Fig. 2 is a design flow chart of PPG feature selection according to the present invention. The feature sorting is carried out by using 65 tested data in total, wherein each tested data is about 3 data in different time periods, 190 groups are formed, and the length of each data is about 10 minutes. And (3) selecting 2 filtering type feature selection methods of RRelieff and neighbor component analysis, eliminating recursive features of a support vector machine depending on the support vector machine, and performing feature sorting on each tested data by 4 feature sorting methods of 2 embedded type feature selection methods of average influence values of a neural network. Completing RRelieff feature selection through a relieff function of matlab; selecting a fsrnca function in matlab to complete the analysis of the adjacent components, and learning the feature weight by using regularization; in the SVM-RFE algorithm, after a characteristic sequence is input, calculating the weight corresponding to each characteristic, sorting by taking the weight as a sorting standard, removing the characteristic with the minimum weight, updating the sorting sequence and then outputting the characteristic sequence; the MIV algorithm calculates the relative contribution rate of input variables to output by comparing the difference between data obtained after input to the neural network, namely the influence value of input change to output, and ranks according to the relative contribution rate to obtain the ranked feature sequence. And aggregating the feature sequences obtained by each method based on the diversity of the focusing data by adopting a stability aggregation method based on ranking to obtain 4 total feature sequences, and obtaining a final feature sequence, namely a feature weight sequencing result, according to the diversity of a focusing feature selection method by adopting a stability aggregation method based on ranking. The stability aggregation method based on ranking scores each feature through a feature selection technology, and sorts the features according to the scores from the 1 st feature to the nth feature. The process is repeated in each feature selection iteration to obtain a set of ranking lists, the ranking lists are combined according to the ranking of the set of ranking lists, a final ranking result is created by using the aggregated ranking, and the stability aggregation is integrated, namely whether the feature is ranked within the determined feature parameter number is measured, so that the selected feature ranking is further deepened and distinguished.
The SVR general blood pressure model construction data is a random group of data in 65 tested 3 groups of data in the feature selection module, the test data is 213 groups of data in the other two groups of 65 tested 3 groups of data participating in feature selection and 37 tested data participating in feature selection, the feature parameter numbers are respectively valued at 3-20 and 30, 40, 50 and 60 for testing, and the random data in the 65 tested data is used for constructing the SVR general blood pressure model; taking the average absolute error and the mean value and the standard deviation of 213 groups of prediction data as evaluation standards of the selected feature numbers, preliminarily screening out an optimal feature subset, wherein the optimal feature subset of systolic pressure is 6, the optimal feature subset of diastolic pressure is 7, pulse wave features for SBP prediction are F2, P5, W1, RP6, RP5 and delta T, and pulse wave features for DBP prediction are P5, F2, delta T, W2, RP6, F1 and RP 5. Adding 263 groups of data of 50 groups of 5 tested data collected by an experimental platform to test and verify an SVR general blood pressure model and the optimal feature subset number; and selecting 15, 25, 35, 45 and 55 tested data to participate in feature selection, and comparing the first 6 features of the systolic pressure prediction retained feature weight result and the first 7 features of the diastolic pressure prediction retained feature weight result with the optimal feature subset obtained by selecting 65 tested participated features according to the determined optimal feature subset to verify the stability of the optimal feature subset.
Fig. 3 is a flow chart of video signal acquisition and processing. The method comprises the steps of recording face dynamic features by using a camera, wherein the frame rate is 50fps, the distance is 1m, synchronously acquiring fingertip PPG signals, the sampling rate is 1000Hz, and a testee keeps a static state in the acquisition process. Selecting a middle area of a cheek below a lower eyelid of an eye as an interested area by adopting a frame selection technology, reading a video by using a VideoReader function of matlab to form a frame, selecting the interested area from a picture of an intermediate frame, and respectively calculating a gray average value of the interested area of each frame of RGB three channels to obtain an RGB sequence; and filtering the video signal by using a color distortion filtering algorithm, performing time normalization processing on the RGB sequence, performing characteristic conversion after Fourier transformation, performing energy calculation and weight calculation on the converted RGB sequence, performing inverse Fourier transformation, and performing inverse time normalization to obtain the filtered RGB sequence. Processing the filtered RGB sequence by using an independent component analysis algorithm, firstly carrying out zero-averaging processing on RGB signals, solving a demixing matrix by using a Newton iteration method, then separating to obtain a source signal, and extracting a component 1, a component 2 and a component 3. In an ICA decomposition result, the maximum value of the ratio of the maximum value to the mean value in a spectrogram is obtained from a component 1 with obvious PPG characteristics, three components are automatically identified by using spectral analysis, the component 1 with the obvious PPG characteristics is selected, a Chebyshev I-type IIR filter is used for band-pass filtering to denoise the component 1, the pass band of the band-pass filter is set to be 0.8Hz-3Hz, the blocking frequency is 0.5Hz and 4Hz, the maximum pass band attenuation of 1dB and the minimum stop band attenuation of 15dB, and finally 100 groups of video pulse wave signals are obtained.
Video blood pressure prediction, namely performing accuracy estimation on a video pulse wave signal by using a fingertip PPG signal which is synchronously acquired, and evaluating the accuracy of an output signal which can be used for blood pressure prediction by comparing the characteristics which are extracted by the fingertip PPG and the video PPG and used for predicting blood pressure by combining an optimal characteristic subset, wherein the result is shown in table 1, two characteristics of W1 and W2 with larger average absolute percentage are deleted, and 6 characteristics of P5, F1, F2, delta T, RP5 and RP6 are selected for prediction of systolic pressure and diastolic pressure which are applied to the video signal, so that the optimal video pulse wave characteristic subset is obtained. And taking the optimal video pulse wave feature subset as input, and adopting the 6 tested features as input training to obtain an SVR general blood pressure prediction model for blood pressure prediction.
Table 1100 mean values of MAPE of pulse wave and fingertip PPG extracted features extracted from group of test data video
FIG. 4 shows the result of video blood pressure prediction, and it can be seen from FIG. 4(a) (b) that the results of SBP/DBP prediction of 100 sets of experimental data all satisfy the MAE ≦ 5mmHg and the error standard deviation STD ≦ 8mmHg in the AAMI standard. As shown in Table 2, the prediction of SBP/DBP for the 100 sets of experimental data also meets and is far superior to the A-level criteria of BHS: the error is less than 5mmHg, 10mmHg and 15mmHg, the proportion is up to 80%, 99% and more than 1. Fig. 5 is a graph showing the consistency between the video-estimated blood pressure and the experimentally measured standard blood pressure, and it can be seen from fig. 5(a) and (c) that the systolic pressure and the diastolic pressure estimated from the video have a strong correlation with the experimentally measured standard systolic pressure and the diastolic pressure, and the linear correlation coefficient reaches 0.94. As can be seen from the Bland-Altman diagrams in fig. 5(b) and 5(d), the mean difference between the SBP estimated from the video and the standard SBP measured by experiment is 0.5mmHg, and the mean difference between the DBP estimated from the video and the standard DBP measured by experiment is 0.4mmHg, which are both closer to the dotted line representing the difference of 0, and it can be seen that the two blood pressure measurement methods have stronger consistency.
TABLE 2 comparison of video-based SBP/DBP predicted cumulative error percentage to BHS standards
Although the invention has been described in detail with respect to the general description and the specific embodiments thereof, it will be apparent to those skilled in the art that modifications and improvements can be made based on the invention. Accordingly, it is intended that all such modifications and alterations be included within the scope of this invention as defined in the appended claims.
Claims (8)
1. A video stream-based non-contact continuous blood pressure measurement system, comprising:
the pulse wave signal processing and feature extraction module is used for preprocessing the pulse wave signal by adopting a discrete wavelet decomposition technology and a wavelet denoising method, analyzing the waveform feature of PPG and the expression of physiological information of a reference human body in PPG waveform, and extracting the time-frequency domain feature of a pulse wave sequence;
the single-person model building and screening module is used for building three single-person blood pressure prediction models which are respectively a single-person blood pressure prediction model based on an Elman neural network optimized by a particle swarm, a single-person blood pressure prediction model based on a deep belief network and an SVR single-person blood pressure prediction model according to the time-frequency domain characteristics of the pulse wave sequence extracted by the pulse wave signal processing and characteristic extracting module, evaluating and screening the three single-person blood pressure prediction models to obtain an optimal model which is an SVR single-person blood pressure prediction model;
the PPG signal feature selection module is used for carrying out feature sorting according to the time-frequency domain features of the pulse wave sequence extracted by the pulse wave signal processing and feature extraction module to obtain a feature weight sorting result;
the SVR general blood pressure model construction module is used for selecting the time-frequency domain characteristics of different numbers of pulse wave sequences to reconstruct an SVR single blood pressure prediction model according to the characteristic weight sorting result to obtain an SVR general blood pressure prediction model, and performing consistency analysis on the prediction result and the actual blood pressure to obtain and verify an optimal characteristic subset and the stability of the optimal characteristic subset;
the video signal acquisition and processing module is used for acquiring tested face video data and a fingertip PPG signal by using a camera, extracting an RGB (red, green and blue) sequence from an interested region by using a framing and selecting technology for the acquired face video data, filtering the RGB sequence by using a color distortion filtering algorithm, extracting a video pulse wave signal from the filtered RGB sequence by using an independent component analysis algorithm, and performing consistency estimation with the fingertip PPG signal;
the video blood pressure prediction module extracts the time-frequency domain characteristics of a video pulse wave sequence by adopting a method in the pulse wave signal processing and characteristic extraction module, extracts a video pulse wave characteristic subset according to the optimal pulse wave characteristic subset, performs consistency estimation on the video pulse wave characteristic subset and the optimal pulse wave characteristic subset of the fingertip PPG signal to obtain the optimal video pulse wave characteristic subset which is used as input, and performs blood pressure prediction by using the SVR general blood pressure prediction model obtained by training.
2. The system of claim 1, wherein the pulse wave signal processing and feature extracting module comprises:
extracting time domain characteristics, namely positioning the wave peaks of the pulse wave by adopting a findpeaks function of Matlab and setting the minimum amplitude and the minimum distance of adjacent points, further realizing the positioning of the wave peaks of the pulse wave by searching the minimum value between the adjacent wave peaks, realizing the calculation of a rising area SS and a falling area DS by adopting a trapz function of Matlab, and realizing the calculation of first-order differential of the pulse wave signal by adopting a diff function to obtain the time domain characteristics of the pulse wave;
performing frequency domain characteristic extraction, performing Fourier transform on 10 pulse wave signals on the left and right sides which take the current pulse wave as the center by using a fft function of Matlab, determining the frequency and amplitude between 0.3-1.6Hz and 1.6-3Hz in the fundamental wave and the second harmonic of the pulse wave signals by using a max function of Matlab, performing 6-layer decomposition on the pulse wave signals by using a db6 wavelet as a mother wavelet to extract wavelet domain characteristics, performing Hilbert transform on IMF components obtained by decomposition to obtain Hilbert transform domain characteristics;
the co-extraction includes 78 time-frequency domain features including temporal features, amplitude features, and area features.
3. The system of claim 1, wherein the single-person model building and screening module comprises:
establishing a single blood pressure prediction model of the particle swarm optimized Elman neural network, adopting a single hidden layer, adopting a particle swarm optimization algorithm to search for an optimal network weight and a threshold, adopting variation operation, and reinitializing particles with random probability after each particle update;
establishing an SVR single blood pressure prediction model, training by adopting a Libsvm library, determining a hyper-parameter in the model by adopting an element-SVR model of a radial basis function kernel and a grid parameter optimizing function SVMcgForRegress for regression, namely: a penalty coefficient C, wherein the tolerance of a kernel parameter g and a termination criterion in the kernel function belongs to the E;
establishing a single blood pressure prediction model based on a deep belief network, adopting a limited Boltzmann machine for pre-training, keeping as much characteristic information as possible when a characteristic parameter is subjected to dimension reduction, then initializing a neural network weight by utilizing the trained limited Boltzmann machine network parameter, and adjusting the weight by utilizing training data in combination with an error back-propagation algorithm;
when three single blood pressure prediction models are trained, 60% of data are used for model training and 40% of data are used as test data by adopting a three-fold cross validation method; according to the AAMI standard of blood pressure measurement, the average of the average absolute error and standard deviation of the predicted blood pressure and the actual blood pressure of the obtained 3 times of training is used as an evaluation index of the accuracy of the model, and the three modeling methods are compared and evaluated to obtain the optimal SVR single blood pressure prediction model.
4. The system according to claim 1, wherein the PPG signal feature selection module further comprises a feature selection method designed herein, and specifically comprises:
using 65 tested data to participate in feature selection, wherein each tested data is about 3 data in different time periods, and 190 groups of data are provided, and the length of each data period is ten minutes;
selecting a filtration type feature selection method of RRelieff and neighbor component analysis and two embedded feature selection methods of recursive feature elimination of a support vector machine depending on the support vector machine and average influence value depending on a neural network, and selecting integrated features by adopting a stability aggregation technology based on ranking;
and performing stability aggregation on 190 groups of feature sequences obtained by one feature selection method to obtain the feature weight sequences of the feature selection method, and performing stability aggregation on feature weight sequences obtained by four different feature selection methods to obtain a final feature weight sequencing result.
5. The system of claim 1, wherein the SVR general blood pressure model building module comprises:
the SVR general blood pressure model construction data is a random group of data in 65 tested 3 groups of data in the feature selection module, the test data is 213 groups of data in the other two groups of 65 tested 3 groups of data participating in feature selection and 37 tested data participating in feature selection, the feature parameter numbers are respectively valued at 3-20 and 30, 40, 50 and 60 for testing, and the random data in the 65 tested data is used for constructing the SVR general blood pressure model;
taking the average absolute error, the mean value of the standard deviation and the standard deviation of 213 groups of prediction data as the evaluation standard of the selected feature number, preliminarily screening out an optimal feature subset to obtain an optimal feature subset of the systolic pressure of 6 and an optimal feature subset of the diastolic pressure of 7, and adding 263 groups of 5 tested groups of data collected by an experimental platform to test and verify the SVR general blood pressure model and the optimal feature subset number;
and selecting 15, 25, 35, 45 and 55 tested data to participate in feature selection, and comparing the first 6 features of the systolic pressure prediction retained feature weight result and the first 7 features of the diastolic pressure prediction retained feature weight result with the optimal feature subset obtained by selecting 65 tested participated features according to the determined optimal feature subset to verify the stability of the optimal feature subset.
6. The system of claim 1, wherein the video signal processing module comprises:
recording the dynamic characteristics of the face by using a camera, wherein the frame rate is 50fps, the distance is 1m, synchronously acquiring fingertip PPG signals, the sampling rate is 1000Hz, and the testee keeps a static state in the acquisition process;
extracting the region of interest by adopting a frame selection technology, reading a video into a frame by using a video reader function of matlab, selecting the region of interest by taking a picture of an intermediate frame, and respectively calculating a gray average value of the region of interest of each frame of RGB three channels to obtain an RGB sequence;
filtering the video signal by using a color distortion filtering algorithm, performing time normalization processing on the RGB sequence, performing characteristic conversion after Fourier transformation, performing inverse Fourier transformation after performing energy calculation and weight calculation on the converted RGB sequence, and performing inverse time normalization to obtain the filtered RGB sequence;
extracting a component 1, a component 2 and a component 3 from the filtered RGB sequence by using an independent component analysis algorithm, automatically identifying the three components by using spectral analysis, selecting the component 1 with obvious PPG characteristics, and denoising the component 1 by using a Chebyshev I type IIR filter for band-pass filtering to obtain a video pulse wave signal;
the accuracy estimation is carried out on the video pulse wave signals by utilizing fingertip PPG signals, the similarity of the waveform contrast evaluation output signals and the correctness of the instantaneous heart rate contrast evaluation output signals containing physiological information are achieved.
7. The system of claim 1, wherein the video blood pressure prediction module comprises:
extracting time-frequency domain characteristics of a video pulse wave sequence by adopting a method in a pulse wave signal processing and characteristic extraction module, extracting a video pulse wave characteristic subset according to the optimal characteristic subset of the pulse wave, evaluating the correctness of the video characteristic subset which can be used for blood pressure prediction by comparing the optimal characteristic subset of a fingertip PPG signal with the video pulse wave characteristic subset, and eliminating the characteristic that the average absolute percentage error is higher than 20% again to obtain the optimal video pulse wave characteristic subset as test data;
the optimal video pulse wave feature subset is 6 features including P5, F1, F2, delta T, RP5 and RP6, and based on 5 tested video PPG signal experimental data acquired by an experimental platform, 15-30 groups of data in different time periods are acquired for each test, and 100 groups of data are acquired in about 2 minutes for each group of data; and (3) taking the optimal video pulse wave feature subset of 100 groups of data as input, and performing blood pressure prediction by using the SVR general blood pressure prediction model obtained by training.
8. A non-contact continuous blood pressure measuring method based on video streaming is characterized by comprising the following steps:
the pulse wave signal processing and feature extraction module is used for preprocessing the pulse wave signals by adopting a discrete wavelet decomposition technology and a wavelet denoising method, analyzing the waveform features of PPG and the expression of physiological information of a reference human body in PPG waveforms, and extracting the time-frequency domain features of a pulse wave sequence;
the single model building and screening module builds three single blood pressure prediction models which are respectively a single blood pressure prediction model based on an Elman neural network optimized by a particle swarm, a single blood pressure prediction model based on a deep belief network and an SVR single blood pressure prediction model according to the time-frequency domain characteristics of the pulse wave sequence extracted by the pulse wave signal processing and characteristic extraction module, and evaluates and screens the three single blood pressure prediction models to obtain an optimal model which is an SVR single blood pressure prediction model;
the PPG signal feature selection module performs feature sorting according to the time-frequency domain features of the pulse wave sequence extracted by the pulse wave signal processing and feature extraction module to obtain a feature weight sorting result;
the SVR general blood pressure model construction module selects time-frequency domain characteristics of different numbers of pulse wave sequences to reconstruct an SVR single blood pressure prediction model according to the characteristic weight sorting result to obtain an SVR general blood pressure prediction model, and performs consistency analysis on the prediction result and the actual blood pressure to obtain and verify an optimal characteristic subset and the stability thereof;
the video signal acquisition and processing module firstly acquires tested face video data and a fingertip PPG signal by using a camera, extracts an RGB (red, green and blue) sequence from an interested region by adopting a frame selection technology for the acquired face video data, filters the RGB sequence by adopting a color distortion filtering algorithm, extracts a video pulse wave signal from the filtered RGB sequence by using an independent component analysis algorithm, and performs consistency estimation with the fingertip PPG signal;
the video blood pressure prediction module extracts the time-frequency domain characteristics of a video pulse wave sequence by adopting a method in the pulse wave signal processing and characteristic extraction module, extracts a video pulse wave characteristic subset according to the optimal pulse wave characteristic subset, performs consistency estimation on the video pulse wave characteristic subset and the optimal pulse wave characteristic subset of the fingertip PPG signal to obtain the optimal video pulse wave characteristic subset which is used as input, and performs blood pressure prediction by utilizing an SVR general blood pressure prediction model obtained by training.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210193095.8A CN114569096A (en) | 2022-02-28 | 2022-02-28 | Non-contact continuous blood pressure measuring method and system based on video stream |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210193095.8A CN114569096A (en) | 2022-02-28 | 2022-02-28 | Non-contact continuous blood pressure measuring method and system based on video stream |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114569096A true CN114569096A (en) | 2022-06-03 |
Family
ID=81777330
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210193095.8A Pending CN114569096A (en) | 2022-02-28 | 2022-02-28 | Non-contact continuous blood pressure measuring method and system based on video stream |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114569096A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115363598A (en) * | 2022-09-19 | 2022-11-22 | 深圳市太美亚电子科技有限公司 | Electrocardiosignal processing method and system |
| CN116299684A (en) * | 2023-05-17 | 2023-06-23 | 成都理工大学 | Novel microseismic classification method based on bimodal neurons in artificial neural network |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20090052442A (en) * | 2007-11-21 | 2009-05-26 | 엘지전자 주식회사 | Non-invasive blood pressure measuring method and device |
| CN103908236A (en) * | 2013-05-13 | 2014-07-09 | 天津点康科技有限公司 | Automatic blood pressure measuring system |
| CN106037694A (en) * | 2016-05-13 | 2016-10-26 | 吉林大学 | Continuous blood pressure measuring device based on pulse waves |
| US20180199870A1 (en) * | 2016-12-19 | 2018-07-19 | Nuralogix Corporation | System and method for contactless blood pressure determination |
| CN113288092A (en) * | 2021-05-06 | 2021-08-24 | 福建师范大学 | Blood pressure detection method and terminal for extracting pulse wave based on video |
| CN113303776A (en) * | 2021-05-28 | 2021-08-27 | 西安理工大学 | Non-contact blood pressure measuring method based on cyclic neural network |
-
2022
- 2022-02-28 CN CN202210193095.8A patent/CN114569096A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20090052442A (en) * | 2007-11-21 | 2009-05-26 | 엘지전자 주식회사 | Non-invasive blood pressure measuring method and device |
| CN103908236A (en) * | 2013-05-13 | 2014-07-09 | 天津点康科技有限公司 | Automatic blood pressure measuring system |
| CN106037694A (en) * | 2016-05-13 | 2016-10-26 | 吉林大学 | Continuous blood pressure measuring device based on pulse waves |
| US20180199870A1 (en) * | 2016-12-19 | 2018-07-19 | Nuralogix Corporation | System and method for contactless blood pressure determination |
| CN113288092A (en) * | 2021-05-06 | 2021-08-24 | 福建师范大学 | Blood pressure detection method and terminal for extracting pulse wave based on video |
| CN113303776A (en) * | 2021-05-28 | 2021-08-27 | 西安理工大学 | Non-contact blood pressure measuring method based on cyclic neural network |
Non-Patent Citations (2)
| Title |
|---|
| 吴海燕;季忠;李孟泽;: "基于脉搏波的无创连续血压监测模型簇研究", 仪器仪表学报, no. 07, 31 August 2020 (2020-08-31) * |
| 孙晟珂;李凯扬;: "基于IPPG非接触式血压测量方法", 信息技术, no. 10, 22 October 2020 (2020-10-22) * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115363598A (en) * | 2022-09-19 | 2022-11-22 | 深圳市太美亚电子科技有限公司 | Electrocardiosignal processing method and system |
| CN115363598B (en) * | 2022-09-19 | 2023-03-24 | 深圳市太美亚电子科技有限公司 | Electrocardiosignal processing method and system |
| CN116299684A (en) * | 2023-05-17 | 2023-06-23 | 成都理工大学 | Novel microseismic classification method based on bimodal neurons in artificial neural network |
| CN116299684B (en) * | 2023-05-17 | 2023-07-21 | 成都理工大学 | Novel microseismic classification method based on bimodal neurons in artificial neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110269600B (en) | Non-contact video heart rate detection method based on multivariate empirical mode decomposition and combined blind source separation | |
| CN110309817B (en) | Pulse wave motion artifact removing method for parameter adaptive optimization VMD | |
| CN114569096A (en) | Non-contact continuous blood pressure measuring method and system based on video stream | |
| CN112914527B (en) | Arterial blood pressure signal acquisition method based on pulse wave photoplethysmography | |
| CN113349752B (en) | Wearable device real-time heart rate monitoring method based on sensing fusion | |
| Motin et al. | PPG derived heart rate estimation during intensive physical exercise | |
| CN114781465B (en) | rPPG-based non-contact fatigue detection system and method | |
| CN114469124B (en) | Method for identifying abnormal electrocardiosignals in movement process | |
| Mishra et al. | A survey on denoising techniques of PPG signal | |
| CN115211858A (en) | Emotion recognition method and system based on deep learning and storable medium | |
| Bousefsaf et al. | iPPG 2 cPPG: reconstructing contact from imaging photoplethysmographic signals using U-Net architectures | |
| CN112001862B (en) | Non-contact type apparent heart rate detection method for eliminating motion noise of video heart impact signal | |
| Wan et al. | Combining parallel adaptive filtering and wavelet threshold denoising for photoplethysmography-based pulse rate monitoring during intensive physical exercise | |
| CN112120711B (en) | Noninvasive diabetes prediction system and method based on photoplethysmography pulse waves | |
| He et al. | Remote Photoplethysmography Heart Rate Variability Detection Using Signal to Noise Ratio Bandpass Filtering | |
| Botina-Monsalve et al. | Performance analysis of remote photoplethysmography deep filtering using long short-term memory neural network | |
| Suriani et al. | Non-contact Facial based Vital Sign Estimation using Convolutional Neural Network Approach | |
| CN116616709A (en) | Working memory load assessment method based on multichannel video in non-stationary state | |
| CN111134634A (en) | Photoelectric volume pulse wave analysis processing method based on cluster analysis | |
| Karmuse et al. | A robust rppg approach for continuous heart rate measurement based on face | |
| CN116369877A (en) | Noninvasive blood pressure estimation method based on photoelectric volume pulse wave | |
| Choe et al. | Improving video-based resting heart rate estimation: A comparison of two methods | |
| Li et al. | An improved MAMA-EMD for the automatic removal of EOG artifacts | |
| CN114387479A (en) | Non-contact heart rate measurement method and system based on face video | |
| Kuang et al. | Remote photoplethysmography signals enhancement based on generative adversarial networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |