CN116369877A - Noninvasive blood pressure estimation method based on photoelectric volume pulse wave - Google Patents
Noninvasive blood pressure estimation method based on photoelectric volume pulse wave Download PDFInfo
- Publication number
- CN116369877A CN116369877A CN202211608910.9A CN202211608910A CN116369877A CN 116369877 A CN116369877 A CN 116369877A CN 202211608910 A CN202211608910 A CN 202211608910A CN 116369877 A CN116369877 A CN 116369877A
- Authority
- CN
- China
- Prior art keywords
- blood pressure
- pressure estimation
- volume pulse
- model
- photoelectric volume
- 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 177
- 238000000034 method Methods 0.000 title claims abstract description 100
- 230000004927 fusion Effects 0.000 claims abstract description 34
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000012544 monitoring process Methods 0.000 claims abstract description 14
- 238000012216 screening Methods 0.000 claims abstract description 5
- 238000013186 photoplethysmography Methods 0.000 claims description 62
- 230000000877 morphologic effect Effects 0.000 claims description 42
- 238000004422 calculation algorithm Methods 0.000 claims description 36
- 238000010801 machine learning Methods 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 12
- 238000013528 artificial neural network Methods 0.000 claims description 11
- 238000012360 testing method Methods 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 9
- 238000012795 verification Methods 0.000 claims description 7
- 238000013135 deep learning Methods 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 6
- 230000008030 elimination Effects 0.000 claims description 5
- 238000003379 elimination reaction Methods 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 5
- 230000011218 segmentation Effects 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000012952 Resampling Methods 0.000 claims description 3
- 238000013145 classification model Methods 0.000 claims description 3
- 238000002790 cross-validation Methods 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 2
- 208000027796 Blood pressure disease Diseases 0.000 abstract description 3
- 238000001514 detection method Methods 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 9
- 238000010586 diagram Methods 0.000 description 7
- 239000013598 vector Substances 0.000 description 6
- 206010020772 Hypertension Diseases 0.000 description 5
- 230000003778 catagen phase Effects 0.000 description 4
- 238000013461 design Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 230000007774 longterm Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000001373 regressive effect Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000013401 experimental design Methods 0.000 description 3
- 208000001953 Hypotension Diseases 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 239000000654 additive Substances 0.000 description 2
- 230000000996 additive effect Effects 0.000 description 2
- 238000009530 blood pressure measurement Methods 0.000 description 2
- 210000004204 blood vessel Anatomy 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000035487 diastolic blood pressure Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 230000036543 hypotension Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000035488 systolic blood pressure Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 239000000853 adhesive Substances 0.000 description 1
- 230000001070 adhesive effect Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000002555 auscultation Methods 0.000 description 1
- 239000008280 blood Substances 0.000 description 1
- 210000004369 blood Anatomy 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000034994 death Effects 0.000 description 1
- 231100000517 death Toxicity 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013136 deep learning model Methods 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
- 238000001125 extrusion Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 208000015181 infectious disease Diseases 0.000 description 1
- 238000003780 insertion Methods 0.000 description 1
- 230000037431 insertion Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 238000005312 nonlinear dynamic Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000004393 prognosis Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
- 230000001225 therapeutic effect Effects 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
- 210000005239 tubule Anatomy 0.000 description 1
- 238000002604 ultrasonography Methods 0.000 description 1
- 230000000007 visual effect Effects 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
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7246—Details of waveform analysis using correlation, e.g. template matching or determination of similarity
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Pathology (AREA)
- Veterinary Medicine (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- General Health & Medical Sciences (AREA)
- Physiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Cardiology (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Vascular Medicine (AREA)
- Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention provides a noninvasive blood pressure estimation method based on photoelectric volume pulse waves, which comprises the following steps: collecting and monitoring photoelectric volume pulse wave signals and invasive continuous blood pressure waveform signals of a human body, initializing the signals, carrying out feature selection on initialized data, screening information parameters with discrimination significance, carrying out feature fusion, constructing a noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse waves according to the data after feature fusion, training the noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse waves according to the initialized data, collecting the photoelectric volume pulse wave signals of a user, inputting the photoelectric volume pulse wave signals into the noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse waves, and carrying out noninvasive blood pressure estimation. The noninvasive blood pressure estimation method based on the photoelectric volume pulse wave can realize noninvasive blood pressure estimation through the photoelectric volume pulse wave, improves estimation accuracy, and can provide reference for blood pressure disease detection.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, pulse wave and noninvasive blood pressure monitoring, in particular to a noninvasive blood pressure estimation method based on photoelectric volume pulse waves.
Background
Statistical reports showed that the prevalence of hypertension in the general population in the united states increased to 45.4% in 2017, with the estimated mortality associated with hypertension accounting for 19.2% of all deaths. Long-term continuous Blood Pressure (BP) monitoring is necessary to predict risk and make appropriate therapeutic interventions, which can effectively improve prognosis and reduce mortality. The gold standard for obtaining continuous blood pressure is the invasive arterial catheterization method. However, this method requires insertion of a tubule into the blood vessel to bring the pressure sensor into direct contact with the blood, which is not only painful, but also increases the risk of infection. Thus, continuous non-invasive blood pressure monitoring is more suitable for health management beyond intensive care. Current cuff-based auscultation and oscillometric techniques require inflation and occlusion of the blood vessel during testing, which is inconvenient for long-term monitoring.
In addition, some studies have proposed a sleeveless measurement method to achieve portable and comfortable blood pressure monitoring. These methods acquire pulse waveforms using various sensors, such as photoplethysmography (PPG), ultrasound, and radar. There are even studies combining multi-modal physiological signals, such as Electrocardiograph (ECG) or Ballistocardiogram (BCG), according to the pulse wave velocity (pulse wave velocity, PWV) theory. Among them, the blood pressure prediction method based on pulse transit time (pulse transient times, PTT) is the most common technique. However, most algorithms require the combination of PPG with ECG or BCG signals monitored by other sites, requiring additional sensors or adhesive electrode placement, which can prevent long-term blood pressure measurement by the user. In addition, current non-invasive blood pressure monitoring techniques have some challenges that must be addressed. First, the related art does not consider individual differences of optical signals. For example, elderly or cardiovascular patients may not have a weak pulse. These special populations often have symptoms such as hypertension. If modeled as a normal population, it may lead to incorrect blood pressure predictions.
Furthermore, current techniques and inventive methods remain limited by data set quality, experimental design, and algorithm bottlenecks. Finally, most techniques are limited to a single dataset for training and evaluation, which brings about a high risk of overfitting, considering the diversity of the population. Therefore, it is necessary to design a non-invasive blood pressure estimation method based on photoplethysmography.
Disclosure of Invention
The invention aims to provide a noninvasive blood pressure estimation method based on photoelectric volume pulse waves, which can realize noninvasive blood pressure estimation through the photoelectric volume pulse waves, improves estimation accuracy and can provide reference for blood pressure disease detection.
In order to achieve the above object, the present invention provides the following solutions:
a noninvasive blood pressure estimation method based on photoplethysmography waves comprises the following steps:
step 1: collecting and monitoring photoelectric volume pulse wave signals and invasive continuous blood pressure waveform signals of a human body;
step 2: initializing the obtained photoelectric volume pulse wave signal and the invasive continuous blood pressure waveform signal;
step 3: feature selection is carried out on the initialized data, information parameters with identification significance are screened, and feature fusion is carried out;
step 4: constructing a noninvasive blood pressure estimation double-stage model based on photoelectric volume pulse waves according to the data after feature fusion;
step 5: training a noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse wave according to the initialized data;
step 6: collecting a photoelectric volume pulse wave signal of a user, inputting the photoelectric volume pulse wave signal into a noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse wave, and performing noninvasive blood pressure estimation;
step 7: experiments prove the robustness of various morphological characteristics selected by the PMFL provided by the embodiment of the invention;
step 8: the noninvasive blood pressure estimation method based on the photoelectric volume pulse wave provided by the embodiment of the invention is quantitatively verified and evaluated.
Optionally, in step 1, a photoelectric volume pulse wave signal and an invasive continuous blood pressure waveform signal of the human body are collected and monitored, specifically:
the method comprises the steps of collecting a photoelectric volume pulse wave signal of a monitored human body, namely a PPG signal and an invasive continuous blood pressure waveform signal, wherein the invasive continuous blood pressure waveform signal is used as a reference signal.
Optionally, in step 2, initializing the obtained photoplethysmography signal and the invasive continuous blood pressure waveform signal, specifically:
resampling, filtering and denoising, signal segmentation, phase matching and normalization are carried out on the obtained photoelectric volume pulse wave signals and the invasive continuous blood pressure waveform signals, priori feature extraction is carried out on the signals after normalization, morphological features are extracted from the normalized PPG signals, and the signals after normalization and the morphological features obtained through extraction are used as training data for training a noninvasive blood pressure estimation dual-stage model based on the photoelectric volume pulse waves.
Optionally, in step 3, feature selection is performed on the initialized data, information parameters with identification significance are screened, feature fusion is performed, and the method specifically comprises the following steps:
the method comprises the steps of acquiring normalized signals and extracted morphological characteristics, screening the morphological characteristics through a PPG morphological characteristic learning algorithm (PPG morphological feature learning, PMFL), firstly sequencing and filtering a group of relatively important characteristics by using a base learner, then infiltrating by using a recursive characteristic elimination method, finding out an optimal characteristic combination, determining a final optimized characteristic set, fusing depth characteristics acquired by a noninvasive blood pressure estimation dual-stage model based on a photoelectric volume pulse wave with morphological characteristics in the final optimized characteristic set according to the final optimized characteristic set, and determining the proportion of different characteristic sets by setting characteristic weights:
F f ＝ε·F m +(1-ε)·F d (1)
wherein F is f Represents fusion features, F m And F d Respectively representing morphological features and depth features.
Optionally, in step 4, a noninvasive blood pressure estimation dual-stage model based on the photoelectric volume pulse wave is built according to the data after feature fusion, which specifically includes:
according to the normalized signals and the fusion characteristics corresponding to the signals, a non-invasive blood pressure estimation double-stage model based on a photoelectric volume pulse wave is built, wherein the non-invasive blood pressure estimation double-stage model, namely a SMART-BP model, consists of a non-cuff blood pressure group type classification model based on deep learning and a non-cuff blood pressure estimation pipeline based on automatic machine learning, namely an SEM-ResNet model and an Auto-regress model, wherein the SEM-ResNet model is used for dividing acquired user PPG signals into different blood pressure groups through a neural network, and the Auto-regress model is used for automatically building a fine-granularity Regressor for the PPG signals of each blood pressure group to perform non-invasive blood pressure estimation.
Optionally, an SEM-ResNet model is built, specifically:
and establishing an SEM-ResNet model, wherein the SEM-ResNet model comprises ResNet, a plurality of sub-networks and a neural network, the SEM-ResNet takes ResNet as a backbone network, residual connection of a fusion extrusion-excitation (SE) module is used for learning shared low-level features, the plurality of sub-networks cooperatively learn high-level specific signal features by utilizing convolution kernels with different scales, and the neural network is embedded into the extracted fusion features and is used for providing prior information.
Optionally, an Auto-regress model is built, specifically:
an Auto-regress model is built, a fine-granularity Regressor is automatically built for PPG signals of each blood pressure group obtained by an SEM-ResNet model, accurate noninvasive blood pressure estimation is obtained, a stacked integrated learning optimization algorithm is built through an AutoML pipeline, the Auto-Regressor model is defined through an H2O AutoML frame, firstly, a metalevel Regressor is trained and used for finding the optimal combination of the base level Regressor, if the prediction errors of the base learner are very low and the constructed models have significant differences, the model is judged to perform well, the AutoML pipeline enters a base regression stage, random search is carried out in different model sets, diversified base regressors can be generated, and an influencing set is generated when the Regressor is matched with a stacking method, wherein the metalevel Regressor is trained by using k-fold cross-validation predicted values of the base Regressor in the metalevel regression stage, and a test is carried out on a data set outside fold.
Optionally, in step 5, training is performed on a non-invasive blood pressure estimation dual-stage model based on the photoplethysmography pulse wave according to the initialized data, specifically:
dividing initialized data into a training set, a verification set and a test set according to the proportion of 7:1.5:1.5, training and parameter selection are carried out on the noninvasive blood pressure estimation double-stage model through the training set and the verification set, the generalization capability of the stored optimal system model is checked through the test set, and parameter updating is carried out through an Adam optimizer in the process of training a network, wherein the learning rate is 0.001, the weight attenuation is 0.999, and the momentum is 0.8.
Optionally, in step 6, collecting a photoelectric volume pulse wave signal of the user, inputting the photoelectric volume pulse wave signal into a noninvasive blood pressure estimation dual-stage model based on the photoelectric volume pulse wave, and performing noninvasive blood pressure estimation, specifically:
and collecting PPG signals of the user, and sending the signal waveforms into a noninvasive blood pressure estimation dual-stage model as input to obtain blood pressure values corresponding to the user, so as to realize noninvasive blood pressure estimation.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: the invention provides a noninvasive blood pressure estimation method based on photoelectric volume pulse waves, which comprises the steps of collecting photoelectric volume pulse wave signals and invasive continuous blood pressure waveform signals of a monitored human body, initializing the obtained photoelectric volume pulse wave signals and the invasive continuous blood pressure waveform signals, carrying out feature selection on initialized data, screening information parameters with discrimination significance, carrying out feature fusion, constructing a noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse waves according to the data after feature fusion, training the noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse waves according to the initialized data, collecting photoelectric volume pulse wave signals of a user, and inputting the photoelectric volume pulse wave signals into the noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse waves for noninvasive blood pressure estimation; according to the method, the SE module and the SEM-ResNet model are fused, so that trans-scale features can be effectively obtained from the multi-information fused PPG signal, in addition, priori knowledge of a depth network is provided by morphological features and is fused with the depth features, and therefore the model is helped to obtain more distinguishing features in the learning process, and higher classification precision is achieved; the method deploys an automatic machine learning (automated machine learning, autopl) pipeline for each BP interval to obtain an optimal BP prediction model without requiring a great deal of expert experience; the method designs a feature selection algorithm named PPG morphological feature learning (PPG morphological feature learning, PMFL), and quantifies the contribution degree of a feature subset through a visualized shape addition interpretation (SHapley Additive explanation, SHAP) value to prove the robustness of various morphological features selected by the PMFL.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of a non-invasive blood pressure estimation method based on a photoplethysmography pulse wave according to an embodiment of the present invention;
fig. 2 is a schematic diagram of experimental design flow of a noninvasive blood pressure estimation method based on photoplethysmography pulse waves according to an embodiment of the present invention;
FIG. 3 is a diagram of a statistical distribution of datasets;
FIG. 4 is a schematic diagram of a visual presentation of morphological features;
FIG. 5 is an overall network frame diagram;
FIG. 6 is a diagram showing input signals and output results;
FIG. 7 is a pseudo-code schematic of a two-stage frame algorithm;
FIG. 8 is a pseudo code schematic of a PPG morphological feature learning algorithm;
FIG. 9 is a pseudo code schematic diagram of a stacked integrated optimization algorithm.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention aims to provide a noninvasive blood pressure estimation method based on photoelectric volume pulse waves, which can realize noninvasive blood pressure estimation through the photoelectric volume pulse waves, improves estimation accuracy and can provide reference for blood pressure disease detection.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
As shown in fig. 1, the non-invasive blood pressure estimation method based on the photoplethysmography pulse wave provided by the embodiment of the present invention includes the following steps:
step 1: collecting and monitoring photoelectric volume pulse wave signals and invasive continuous blood pressure waveform signals of a human body;
step 2: initializing the obtained photoelectric volume pulse wave signal and the invasive continuous blood pressure waveform signal;
step 3: feature selection is carried out on the initialized data, information parameters with identification significance are screened, and feature fusion is carried out;
step 4: constructing a noninvasive blood pressure estimation double-stage model based on photoelectric volume pulse waves according to the data after feature fusion;
step 5: training a noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse wave according to the initialized data;
step 6: the method comprises the steps of collecting a photoelectric volume pulse wave signal of a user, inputting the photoelectric volume pulse wave signal into a noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse wave, and carrying out noninvasive blood pressure estimation.
Step 7: experiments prove the robustness of various morphological characteristics selected by the PMFL provided by the embodiment of the invention;
step 8: the noninvasive blood pressure estimation method based on the photoelectric volume pulse wave provided by the embodiment of the invention is quantitatively verified and evaluated.
In step 1, collecting and monitoring photoelectric volume pulse wave signals and invasive continuous blood pressure waveform signals of a human body, wherein the method specifically comprises the following steps:
collecting and monitoring a photoelectric volume pulse wave signal of a human body, namely a PPG signal and an invasive continuous blood pressure waveform signal, wherein the invasive continuous blood pressure waveform signal is used as a reference signal;
the method adopts a MIMIMI (multi-parameter intelligent monitoring) data set, and comprises 20000 pieces of physiological data recorded on patient monitors of medical, surgical and intensive care units in Boston Bass Israeli hospitals, wherein each record usually comprises 24-48 hours of continuous data;
the method adopts a private data set as an independent database to verify the proposed model, wherein the private data set comprises 60000 pieces of physiological data recorded on a patient monitor of an intensive care unit, and each record usually comprises 30 seconds of continuous data; predicting a blood pressure waveform based on the PPG signal can be generalized as a real-time long-sequence prediction problem. Therefore, the model needs to learn and acquire the mapping relation between two PPG and blood pressure values through a time lag window with a fixed window length. In the method of the present invention, the input data PPG sequence (i.e. PPG waveform) and corresponding blood pressure waveform for the time window t are expressed as
X t Representing an input dimension d of the PPG signal x There are N signals, each signal has s x The output is the corresponding predicted blood pressure value (i.e. the corresponding systolic and diastolic blood pressure values within the time window t), Y t Representative has s y The blood pressure sequence (i.e. the blood pressure waveform) of the sample points is calculated here, and the corresponding blood pressure values are obtained, which contain M peaks +.>
And valley->
Sparse set of->
The input signal dimension of the model is not limited to univariate cases, and the derivative of the PPG sequence, i.e. the velocity sequence of the PPG, the acceleration sequence of the PPG may be chosen. The proposed model accepts inputs and outputs, given by the following formula:
Y s ＝F(X；P；θ) (1)
where F (·) is a model function,
representing introduced priori knowledge, namely morphological characteristics of the PPG waveform screened by a characteristic selection algorithm, wherein θ represents super-parameters of the deep learning model.
In step 2, initializing the obtained photoelectric volume pulse wave signal and the invasive continuous blood pressure waveform signal, specifically:
the specific experimental design flow chart of the method is shown in fig. 2, the lengths of the PPG and blood pressure signals collected in the database are different and are interfered by outliers and baseline drift, so that the signals need to be subjected to preliminary processing, and as shown in fig. 5, the acquired photoplethysmographic pulse wave signals and invasive continuous blood pressure waveform signals are subjected to resampling, filtering and noise reduction, signal segmentation, phase matching and normalization processing, and the method specifically comprises the following steps: in addition, because the wavelet transformation is suitable for analyzing non-stationary signals, and has better time-frequency positioning in signal mutation, compression reconstruction and signal denoising, the method selects the wavelet transformation of sym4 wavelet as the basis wavelet to perform signal denoising, and performs two-stage independent decomposition on the noisy PPG signal according to a soft threshold function, then uses a time lag window with a fixed size of 8 seconds and a sliding step length of 3 seconds to divide the filtered signal, and finally adopts maximum and minimum normalization to ensure that the model can be quickly converged, and the statistical distribution of a data set is shown in figure 3 after data preliminary processing;
finally, extracting priori features of the normalized signals, providing priori information for model decision, taking the normalized signals and morphological features obtained by extraction as training data for training a noninvasive blood pressure estimation dual-stage model based on a photoelectric volume pulse wave, wherein 75 interpretable features are extracted from PPG signals respectively, including time domain, frequency domain and nonlinear features, such as time parameter signal skewness is formula (2), dimensionless index edge factor is formula (3), and area parameter K value is formula (4), wherein most of the features are shown in fig. 4;
wherein p is max ，p min And p mean Respectively representing the maximum value, the minimum value and the average value of PPG amplitude in the period;
in addition, the method additionally uses pulse rate variability (pulse rate variability, PRV) to describe the variation modes in sequence windows, such as sample entropy (SampEn) for describing the sequence disorder degree, in particular, multi-scale entropy (MSE) focuses on quantifying the nonlinear dynamic characteristics of complex systems on multiple scales, and the calculation process of MSE is detailed herein, and the PPG sequence is represented by the following steps
Setting a scale factor tau to divide the time series to form a set of constructed coarse-grained sequences
Sampenn can then be derived for coarse-grained sequences, first, a set of consecutive sequence vectors with an embedding dimension of m is constructed
G m (j)＝{g(j+k)},0≤k≤m-1,1≤j≤S x -m (6)
Second, the vector sequence G with m-dimension coarse granularity m (i) And G m (j) The distance between the two vectors is defined as the maximum value of the difference between the corresponding elements of the two vectors, and the calculation formula is:
d[g(i),g(j)]＝max(|g(i+k)-g(j+k)|),0≤k≤m-1 (7)
then, a similarity tolerance r is set, and each i (1.ltoreq.j.ltoreq.S is calculated x -m) number of values d [ g (i), g (j)]R is less than or equal to r, and the template matching number B is calculated i Ratio to total distance number
And find the average B of all i m (r) is:
increasing the vector dimension to m+1, reconstructing the m+1-dimension vector G m+1 (i) And G m+1 (j) Wherein j is more than or equal to 1 and S is more than or equal to x -m, j not equal i, calculate G m+1 (i) And G m+1 (j) The distance number between the two is smaller than the similarity tolerance value r, namely the template matching number A i Template matching number A i The ratio of the total distance is
Then calculate the average A of all i m (r) is:
sampenn of coarse-grained PPG sequences can be calculated as:
in the method of the present invention, the embedding dimension may be selected to be m=2, like the threshold r=0.15×sdnn, where SDNN represents the standard deviation of the PPG peak interval.
In step 3, feature selection is performed on the initialized data, information parameters with identification significance are screened, feature fusion is performed, and the method specifically comprises the following steps:
a large number of PPG morphological parameters are calculated from the training data obtained in the preliminary data processing stage in step 2, however, too many feature sets may contain redundant information, and still require feature selection and fusion operations to make the deep neural network learn the morphological related prior information better, so that normalized signals and extracted morphological features are obtained, the morphological features are screened by a PMFL algorithm, specifically a feature selection algorithm for PPG morphological feature learning (PPG morphological feature learning, PMFL), the PMFL algorithm is used to screen features with discriminative significance as the prior information of the deep model, the method can also quantify the contribution degree of feature subsets through visualized SHAP values to prove the robustness of various morphological features selected by PMFL, specifically, the PMFL algorithm first uses a base learner to rank and filter a set of relatively important features, then uses a recursive feature cancellation (recursive feature elimination, RFE) to rank to find an optimal feature combination, specifically, the PMFL algorithm is divided into a backward cancellation stage of baseline set generation and feature parameters. The filtering method screens the feature subset with the top k bits for estimating systolic and diastolic pressures and takes the filtered feature set as a baseline set. Then, in the backward elimination stage of the characteristic parameters, the characteristics with the lowest importance are eliminated by a REF method in sequence, the rest characteristic sets are input into a base regressive device to fit the blood pressure value, the final characteristic set quantity is considered according to the optimized regression result, the final optimized characteristic set is determined,
according to the final optimized feature set, the method considers designing a feature fusion strategy, fuses depth features acquired by an SEM-ResNet model with morphological features in the final optimized feature set, and determines the proportion of different feature sets by setting feature weights:
F f ＝ε·F m +(1-ε)·F d (11)
wherein F is f Represents fusion features, F m And F d Respectively representing morphological features and depth features.
In step 4, a noninvasive blood pressure estimation double-stage model based on photoelectric volume pulse waves is built according to the data after feature fusion, and specifically comprises the following steps:
according to the normalized signals and the fusion characteristics corresponding to the signals, a non-invasive blood pressure estimation double-stage model based on a photoplethysmogram pulse wave, namely a SMART-BP model is built, wherein the non-invasive blood pressure estimation double-stage model comprises two stages, namely a coarse-granularity classification stage (coarse-grained classification phase, CCP) and a fine-granularity regression stage (fine-grained regression phase, FRP), a non-cuff blood pressure group classification model based on deep learning is corresponding to the coarse-granularity classification stage (coarse-grained classification phase, CCP), namely an SEM-ResNet model, and a non-cuff blood pressure estimation pipeline model based on automatic machine learning is corresponding to the fine-granularity regression stage (fine-grained regression phase, FRP);
for coarse-granularity classification stage (CCP), the method uses a depth residual network (res) fused with a squeeze-and-excitation (SE) module and a multiscale kernel (SEM-res), so that trans-scale features are effectively obtained from the multi-information fused PPG signal, and in addition, priori knowledge of the depth network is provided by morphological features and fused with the depth features, thereby helping the model obtain more distinguishing features in the learning process and achieving higher classification accuracy;
correspondingly, the SEM-res net model is used to divide the acquired PPG signal of the user into different blood pressure groups, such as hypotension, normal blood pressure and hypertension, through a neural network;
in a fine-grained regression stage (fine-grained regression phase, FRP), the method deploys an automatic machine learning (automated machine learning, autoML) pipeline for each BP interval to obtain an optimal BP prediction model without requiring a great deal of expert experience;
correspondingly, the Auto-regress model is used for automatically establishing a fine-grained Regressor for the PPG signal of each blood pressure group, and carrying out noninvasive blood pressure estimation.
An SEM-ResNet model is built, and the SEM-ResNet model is specifically as follows:
establishing an SEM-ResNet model, wherein the model can automatically learn dense morphological characteristics in a PPG waveform to obtain higher classification precision than a shallow model, the method deploys multi-scale characteristic fusion learning to realize cross-scale information so as to further mine discrimination characteristics, and specifically, the SEM-ResNet model comprises a ResNet, a plurality of sub-networks and a neural network, wherein the SEM-ResNet takes the ResNet as a backbone network, residual error connection of a fusion SE module is used for learning shared low-level characteristics, the plurality of sub-networks cooperatively learn high-level specific signal characteristics by utilizing convolution kernels of different scales, and the neural network is embedded into the extracted fusion characteristics to provide prior information;
the overall framework is shown in FIG. 6, the overall structure of SEM-ResNet and schematic form of ResNet and CBR-3-1. Conv-3-1 represents one-dimensional convolution operation with kernel size of 3 and span size of 1; batch Norm is short for Batch normalization; reLU represents a rectifying linear unit activation function; maxPool and AvgPool respectively represent maximum pooling and average pooling, and share characteristic diagrams obtained by a single-scale CNN layer are sent into sub-networks with different scales to fuse input characteristics F f Output of corresponding branch
The definition is as follows:
in the method, in the process of the invention,
representing the fusion of features F f The obtained branching features, conv b And->
Representing a single-scale CNN layer and a subnetwork of a specific scale, θ b And->
The method comprises the steps of respectively carrying out network parameters, calculating Softmax loss, carrying out single-branch model training, and calculating posterior probability of each category as follows:
in the method, in the process of the invention,
is a model to assign a label y to the input PPG signal i The probability of e {0,1,.. i ∈{0,1,2}，θ k Is category y i Thus, for all observable examples, the cross entropy loss function is used as:
wherein I {. Cndot. } refers to an index function;
the invention uses SE module to learn characteristic channel, which can fuse one-dimensional convolution operationAs learned time information, the SE module can learn to use global information to selectively emphasize useful features while suppressing other features, where the SE-ResNet structure is shown in FIG. 6, for any feature map input F= { F 1 ,f 1 ,..,f m Of f, where f m ∈R t Statistics z ε R m Is generated by compressing F by its time dimension t, the c-th element of z is calculated by the extrusion operation:
after aggregation, an excitation operation consisting of two linear layers is performed to generate channel modulation weights, and the invention allows to activate a plurality of channels, so that a simple gating mechanism is adopted, namely, the Sigmoid activation mode is as follows:
s＝Excitation(z,θ)＝σ(σ(θ T ·z)) (16)
where σ represents the Sigmoid function, θ is a parameter, and the final output of the module is obtained by rescaling F with the activated s, as:
wherein,,
scale (·) is the multiplication between channels;
the objective function defining the global fusion multi-scale features extracted by the backbone is:
wherein,,
representing a fused feature map of features of the series of branches, potential complementary information across the spatio-temporal scale can be learned by training the entire SEM-ResNet model by jointly optimizing the loss of multiple branches, and in the method of the invention the final loss function is shown as follows:
wherein, gamma j Is a weight coefficient corresponding to each branch.
Building an Auto-regress model, which specifically comprises the following steps:
building an Auto-regress model, and automatically building a fine-granularity Regressor for PPG signals of each blood pressure group obtained by an SEM-ResNet model to obtain accurate noninvasive blood pressure estimation;
the machine learning model has low dependence on data and does not depend on hardware operation. The complexity of the algorithm can also be reduced to a great extent by selecting the appropriate features through domain knowledge. However, the creation, training and super-parametric optimization of models requires experienced specialists to spend a lot of time. AutoML is generally understood as a pipeline, as a readily deployed system, that can perform scientific tasks with minimal human intervention. AutoML may optimize an algorithm to minimize a specific loss function associated with a specific content domain. AutoML algorithms have been successfully implemented using techniques such as meta-learning, reinforcement learning, genetic programming, superposition set algorithms, and the like. Compared with a machine learning model manually adjusted by a data science specialist, the AutoML has good performance;
to automate pipeline selection, the method of the present invention uses an H2O AutoML framework definition model. The frame has the following advantages: 1) Its performance exceeds other AutoML frameworks; 2) It has a highly scalable, fully automated algorithm that automatically trains a large number of candidate models. The algorithms available are tested in a fixed order with expert-defined or random grid search selected hyper-parameters, and finally the best performing configurations are aggregated to form an aggregate. The method constructs a Stacking (Stacking) integrated learning optimization algorithm through an AutoML pipeline. First, a meta-level regressor is trained to find the best combination of base level regressors. If the prediction errors of the base learner are low and there are significant differences between the constructed models, the stacked model will perform well. The AutoML pipeline then enters a base regression stage (base-level regression phase) where random searches are performed in different sets of models, which can produce a diverse set of base regressors and influential sets when paired with stacking methods. In addition, the method sets a metalevel regressive device in a metalevel regressive stage, trains by using k-fold cross validation predicted values of a base regressive device, and then tests on a data set outside fold.
In step 5, training a non-invasive blood pressure estimation dual-stage model based on the photoplethysmogram pulse wave according to the initialized data, specifically:
dividing initialized data into a training set, a verification set and a test set according to the proportion of 7:1.5:1.5, training and parameter selection are carried out on a non-invasive blood pressure estimation dual-stage model through the training set and the verification set, the generalization capability of the stored optimal system model is checked through the test set, in the process of training a network, parameter updating is carried out through an Adam optimizer, data segmentation adopts a mode of inter-individual segmentation, and each group of data does not contain data of the same patient, so that information leakage is avoided, wherein the learning rate is 0.001, the weight attenuation is 0.999, and the momentum is 0.8.
In step 6, collecting a photoelectric volume pulse wave signal of a user, inputting the photoelectric volume pulse wave signal into a noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse wave, and performing noninvasive blood pressure estimation, specifically:
as shown in fig. 6, PPG signals of the user are collected, and signal waveforms are input into a non-invasive blood pressure estimation dual-stage model to obtain blood pressure values corresponding to the user, so as to realize non-invasive blood pressure estimation.
In step 7, the robustness of various morphological characteristics of PMFL selection provided by the embodiment of the invention is proved through experiments. To compare the effect of the proposed PMFL algorithm on BP prediction performance, the present patent constructs a supervised feature weighting algorithm Relief, a weak supervision feature selection algorithm (WSF) based on spectral analysis, a maximum correlation and minimum redundancy algorithm (MRMR) based on mutual information, and an unsupervised multi-cluster feature selection (MCFS). The present patent shows BP prediction errors of various feature selection methods in table 1, and calculates the robustness index of the corresponding method. The predictive performance of the feature subset selected by the PMFL algorithm is superior to other methods. In addition, the PMFL algorithm has the lowest ANHI and CD values, indicating strong consistency of feature weights and ranks. In contrast, the highest value for PCC means that the feature subset has the highest correlation and the strongest robustness.
Table 1 comparison of Performance of feature selection algorithm with reference
Note that these are dimensionless numbers; ANHI, average normalized hamming index; CD, candela distance; PCC, pearson correlation coefficient.
In step 8, experimental verification is performed on the prediction performance of the proposed noninvasive blood pressure estimation dual-stage model based on the photoplethysmogram pulse wave. To further verify the effectiveness of the proposed method, the present patent builds a model of various structures for comparison, including: 1) A direct regression single-stage AutoML algorithm; 2) The end-to-end deep learning algorithm directly sends PPG signals into a model and automatically predicts corresponding BP values, including DenseNet, transformer and SEM-ResNet models; 3) An autopl two-stage method (autopl-autopl) is used in both CCP and FRP. Table 2 shows the comparison results of the experiments, and the main quantization indexes include Mean Error (ME), mean Standard Deviation (SDE), mean absolute difference (MAE). It can be observed that the proposed SMART-BP is superior to other works in most evaluation metrics. For DBP prediction, all models meet the AAMI standard. For SBP prediction, most models pass the AAMI standard except DenseNet.
Table 2 comparison of the performance of different frameworks on private dataset (mmHg).
The invention provides a noninvasive blood pressure estimation dual-stage model, which is used for independently modeling potential physiological differences of hypotension, normal blood pressure and hypertension groups and is named SMART-BP (SeM-resnet and Auto-Regressor based on a Two-stage framework for Blood Pressure estimation). This approach can be modeled separately for each blood pressure set to obtain highly accurate blood pressure predictions. The SMART-BP is significantly different from the prior art in that it first uses a deep learning algorithm to coarsely classify the original PPG signal into three BP categories, and then builds an automatic optimization pipeline of a machine learning regressor for each interval to accurately predict the fine-grained BP values;
the input of the invention only consists of pulse waves, so that the acquisition circuit only needs to acquire PPG signals, and compared with the traditional method based on pulse wave propagation speed, the method omits the step of acquiring electrocardiosignals, and does not need excessive derivative calculation, thereby being convenient to integrate into devices such as a bracelet and the like, not needing blood pressure measurement equipment such as a cuff and the like, getting rid of the constraint of the cuff and enabling the device to be more portable.
The blood pressure estimation algorithm can realize continuous blood pressure estimation and long-term monitoring of blood pressure, can be used for measuring blood pressure in daily life, and can not bring wound and uncomfortable influence to human body during measurement.
The invention uses the deep learning and automatic machine learning pipelines as network backbone to extract a large amount of information in the PPG signal, combines morphological characteristics extracted from the PPG signal, introduces the prior information to promote the model to utilize more distinguishing characteristics in the learning process, and the input signal contains more information, so that the measured blood pressure result is more stable and higher prediction precision is achieved.
In order to improve the blood pressure prediction accuracy, the invention adopts a double-stage framework to perform high-accuracy automatic modeling aiming at specific crowds, thereby realizing the prediction accuracy meeting medical standards.
The method designs a feature selection strategy based on a recursive feature elimination method, so that a sparse subset of useful features can be obtained to improve the fine-granularity blood pressure estimation precision of an AutoML modeling stage, and in the learning and training process, the sequence features captured by the neural network can be effectively combined.
The algorithm pseudo code used in the present invention is shown in fig. 7, 8 and 9.
The invention provides a noninvasive blood pressure estimation method based on photoelectric volume pulse waves, which comprises the steps of collecting photoelectric volume pulse wave signals and invasive continuous blood pressure waveform signals of a monitored human body, initializing the obtained photoelectric volume pulse wave signals and the invasive continuous blood pressure waveform signals, carrying out feature selection on initialized data, screening information parameters with discrimination significance, carrying out feature fusion, constructing a noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse waves according to the data after feature fusion, training the noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse waves according to the initialized data, collecting photoelectric volume pulse wave signals of a user, and inputting the photoelectric volume pulse wave signals into the noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse waves for noninvasive blood pressure estimation; according to the method, the SE module and the SEM-ResNet model are fused, so that trans-scale features can be effectively obtained from the multi-information fused PPG signal, in addition, priori knowledge of a depth network is provided by morphological features and is fused with the depth features, and therefore the model is helped to obtain more distinguishing features in the learning process, and higher classification precision is achieved; the method deploys an automatic machine learning (automated machine learning, autopl) pipeline for each BP interval to obtain an optimal BP prediction model without requiring a great deal of expert experience; the method designs a feature selection algorithm named PPG morphological feature learning (PPG morphological feature learning, PMFL), and quantifies the contribution degree of a feature subset through a visualized shape addition interpretation (SHapley Additive explanation, SHAP) value to prove the robustness of various morphological features selected by the PMFL.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the methods of the present invention and the core ideas thereof; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.
Claims (9)
1. A noninvasive blood pressure estimation method based on photoplethysmography is characterized by comprising the following steps:
step 1: collecting and monitoring photoelectric volume pulse wave signals and invasive continuous blood pressure waveform signals of a human body;
step 2: initializing the obtained photoelectric volume pulse wave signal and the invasive continuous blood pressure waveform signal;
step 3: feature selection is carried out on the initialized data, information parameters with identification significance are screened, and feature fusion is carried out;
step 4: constructing a noninvasive blood pressure estimation double-stage model based on photoelectric volume pulse waves according to the data after feature fusion;
step 5: training a noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse wave according to the initialized data;
step 6: the method comprises the steps of collecting a photoelectric volume pulse wave signal of a user, inputting the photoelectric volume pulse wave signal into a noninvasive blood pressure estimation double-stage model based on the photoelectric volume pulse wave, and carrying out noninvasive blood pressure estimation.
2. The method for noninvasive blood pressure estimation based on photoplethysmogram according to claim 1, wherein in step 1, the photoplethysmogram signal and the invasive continuous blood pressure waveform signal of the monitored human body are collected, specifically:
the method comprises the steps of collecting a photoelectric volume pulse wave signal of a monitored human body, namely a PPG signal and an invasive continuous blood pressure waveform signal, wherein the invasive continuous blood pressure waveform signal is used as a reference signal.
3. The method for non-invasive blood pressure estimation based on photoplethysmography according to claim 2, wherein in step 2, the acquired photoplethysmography signal and the invasive continuous blood pressure waveform signal are initialized, specifically:
resampling, filtering and denoising, signal segmentation, phase matching and normalization are carried out on the obtained photoelectric volume pulse wave signals and the invasive continuous blood pressure waveform signals, priori feature extraction is carried out on the signals after normalization, morphological features are extracted from the normalized PPG signals, and the signals after normalization and the morphological features obtained through extraction are used as training data for training a noninvasive blood pressure estimation dual-stage model based on the photoelectric volume pulse waves.
4. The method for non-invasive blood pressure estimation based on photoplethysmogram according to claim 3, wherein in step 3, feature selection is performed on the initialized data, information parameters with identification significance are screened, and feature fusion is performed, specifically:
the method comprises the steps of obtaining normalized signals and extracted morphological characteristics, screening the morphological characteristics through a PPG morphological characteristic learning algorithm, firstly sequencing and filtering a group of relatively important characteristics by using a base learner, then iterating by using a recursive characteristic elimination method, finding out the optimal characteristic combination, determining a final optimized characteristic set, fusing depth characteristics obtained by a noninvasive blood pressure estimation dual-stage model based on photoelectric volume pulse waves with morphological characteristics in the final optimized characteristic set according to the final optimized characteristic set, and determining the proportion of different characteristic sets by setting characteristic weights:
F f ＝ε·F m +(1-ε)·F d (1)
in the method, in the process of the invention,F f represents fusion features, F m And F d Respectively representing morphological features and depth features.
5. The method for noninvasive blood pressure estimation based on photoplethysmogram according to claim 4, wherein in step 4, a noninvasive blood pressure estimation dual-stage model based on photoplethysmogram is built according to the feature fused data, specifically:
according to the normalized signals and the fusion characteristics corresponding to the signals, a non-invasive blood pressure estimation double-stage model based on a photoelectric volume pulse wave is built, wherein the non-invasive blood pressure estimation double-stage model, namely a SMART-BP model, consists of a non-cuff blood pressure group type classification model based on deep learning and a non-cuff blood pressure estimation pipeline based on automatic machine learning, namely an SEM-ResNet model and an Auto-regress model, wherein the SEM-ResNet model is used for dividing acquired user PPG signals into different blood pressure groups through a neural network, and the Auto-regress model is used for automatically building a fine-granularity Regressor for the PPG signals of each blood pressure group to perform non-invasive blood pressure estimation.
6. The non-invasive blood pressure estimation method based on photoplethysmography according to claim 5, wherein constructing an SEM-ResNet model specifically comprises:
and establishing an SEM-ResNet model, wherein the SEM-ResNet model comprises a ResNet, a plurality of sub-networks and a neural network, the SEM-ResNet takes the ResNet as a backbone network, residual connection of the fusion extrusion-excitation module is used for learning shared low-level features, the plurality of sub-networks cooperatively learn high-level specific signal features by using convolution kernels with different scales, and the neural network is embedded into the extracted fusion features and is used for providing prior information.
7. The non-invasive blood pressure estimation method based on photoplethysmography according to claim 6, wherein the Auto-regress model is built specifically as follows:
an Auto-regress model is built, a fine-granularity Regressor is automatically built for PPG signals of each blood pressure group obtained by an SEM-ResNet model, accurate noninvasive blood pressure estimation is obtained, a stacked integrated learning optimization algorithm is built through an AutoML pipeline, the Auto-Regressor model is defined through an H2OAutoML frame, firstly, a metalevel Regressor is trained and used for finding the optimal combination of the base level Regressor, if the prediction errors of the base learner are very low and the constructed models have significant differences, the model is judged to perform well, the AutoML pipeline enters a base regression stage, random search is carried out in different model sets, diversified base regressors can be generated, and an influencing set is generated when the Regressor is matched with a stacking method, wherein the metalevel Regressor is trained by using k-fold cross-validation predicted values of the base Regressor in the metalevel regression stage, and a test is carried out on a data set outside fold.
8. The method for non-invasive blood pressure estimation based on photoplethysmogram according to claim 7, wherein in step 5, the non-invasive blood pressure estimation dual-stage model based on photoplethysmogram is trained based on the initialized data, specifically:
dividing initialized data into a training set, a verification set and a test set according to the proportion of 7:1.5:1.5, training and parameter selection are carried out on the noninvasive blood pressure estimation double-stage model through the training set and the verification set, the generalization capability of the stored optimal system model is checked through the test set, and parameter updating is carried out through an Adam optimizer in the process of training a network, wherein the learning rate is 0.001, the weight attenuation is 0.999, and the momentum is 0.8.
9. The method for non-invasive blood pressure estimation based on photoplethysmogram according to claim 8, wherein in step 6, the photoplethysmogram signal of the user is collected and input into a non-invasive blood pressure estimation two-stage model based on photoplethysmogram to perform non-invasive blood pressure estimation, specifically:
and collecting PPG signals of the user, and sending the signal waveforms into a noninvasive blood pressure estimation dual-stage model as input to obtain blood pressure values corresponding to the user, so as to realize noninvasive blood pressure estimation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211608910.9A CN116369877A (en) | 2022-12-14 | 2022-12-14 | Noninvasive blood pressure estimation method based on photoelectric volume pulse wave |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211608910.9A CN116369877A (en) | 2022-12-14 | 2022-12-14 | Noninvasive blood pressure estimation method based on photoelectric volume pulse wave |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116369877A true CN116369877A (en) | 2023-07-04 |
Family
ID=86960216
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211608910.9A Pending CN116369877A (en) | 2022-12-14 | 2022-12-14 | Noninvasive blood pressure estimation method based on photoelectric volume pulse wave |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116369877A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117137465A (en) * | 2023-11-01 | 2023-12-01 | 深圳市奋达智能技术有限公司 | Blood flow dynamic parameter measurement method and related equipment thereof |
| CN117953970A (en) * | 2024-03-27 | 2024-04-30 | 山东大学 | Lung cancer polygene detection method and system based on hyperspectral image |
| CN118211182A (en) * | 2024-05-10 | 2024-06-18 | 沈阳恒德医疗器械研发有限公司 | Identity recognition system and method based on pulse wave signal multi-index fusion analysis |
-
2022
- 2022-12-14 CN CN202211608910.9A patent/CN116369877A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117137465A (en) * | 2023-11-01 | 2023-12-01 | 深圳市奋达智能技术有限公司 | Blood flow dynamic parameter measurement method and related equipment thereof |
| CN117137465B (en) * | 2023-11-01 | 2024-04-16 | 深圳市奋达智能技术有限公司 | Blood flow dynamic parameter measurement method and related equipment thereof |
| CN117953970A (en) * | 2024-03-27 | 2024-04-30 | 山东大学 | Lung cancer polygene detection method and system based on hyperspectral image |
| CN117953970B (en) * | 2024-03-27 | 2024-06-11 | 山东大学 | Lung cancer polygene detection method and system based on hyperspectral image |
| CN118211182A (en) * | 2024-05-10 | 2024-06-18 | 沈阳恒德医疗器械研发有限公司 | Identity recognition system and method based on pulse wave signal multi-index fusion analysis |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Acharya et al. | Deep neural network for respiratory sound classification in wearable devices enabled by patient specific model tuning | |
| Zhao et al. | A hybrid spatio-temporal model for detection and severity rating of Parkinson’s disease from gait data | |
| Jafarian et al. | Automating detection and localization of myocardial infarction using shallow and end-to-end deep neural networks | |
| Gupta et al. | Neural network classification of homomorphic segmented heart sounds | |
| CN116369877A (en) | Noninvasive blood pressure estimation method based on photoelectric volume pulse wave | |
| US20220093215A1 (en) | Discovering genomes to use in machine learning techniques | |
| Rath et al. | Improved heart disease detection from ECG signal using deep learning based ensemble model | |
| Chen et al. | A new deep learning framework based on blood pressure range constraint for continuous cuffless BP estimation | |
| CN117598700B (en) | Intelligent blood oxygen saturation detection system and method | |
| Singh et al. | Short unsegmented PCG classification based on ensemble classifier | |
| CN114587310A (en) | Method for realizing invasive blood pressure waveform estimation based on photoplethysmography | |
| Liu et al. | LSTM Neural Network for Beat Classification in ECG Identity Recognition. | |
| Kumar et al. | A novel CS-NET architecture based on the unification of CNN, SVM and super-resolution spectrogram to monitor and classify blood pressure using photoplethysmography | |
| Singh et al. | Short and noisy electrocardiogram classification based on deep learning | |
| Ma et al. | PPG-based continuous BP waveform estimation using polarized attention-guided conditional adversarial learning model | |
| Jiang et al. | Attention-based multi-scale features fusion for unobtrusive atrial fibrillation detection using ballistocardiogram signal | |
| Nowroozilarki et al. | Variational autoencoders for biomedical signal morphology clustering and noise detection | |
| CN111798980A (en) | Complex medical biological signal processing method and device based on deep learning network | |
| Yen et al. | Blood Pressure and Heart Rate Measurements Using Photoplethysmography with Modified LRCN. | |
| Singh et al. | Expert diagnostic system for detection of hypertension and diabetes mellitus using discrete wavelet decomposition of photoplethysmogram signal and machine learning technique | |
| Sivanjaneyulu et al. | Cnn based ppg signal quality assessment using raw ppg signal for energy-efficient ppg analysis devices in internet of medical things | |
| Shao et al. | Predicting cardiovascular and cerebrovascular events based on instantaneous high-order singular entropy and deep belief network | |
| Surapaneni et al. | Graph Signal Processing Based Classification of Noisy and Clean PPG Signals Using Machine Learning Classifiers for Intelligent Health Monitor | |
| Gong et al. | Accurate cirrhosis identification with wrist-pulse data for mobile healthcare | |
| Yang et al. | ECG-aided PPG signal quality assessment (SQA) system for heart rate estimation |
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 |