CN112806977B - Physiological parameter measuring method based on multi-scale fusion network - Google Patents
Physiological parameter measuring method based on multi-scale fusion network Download PDFInfo
- Publication number
- CN112806977B CN112806977B CN202110160054.4A CN202110160054A CN112806977B CN 112806977 B CN112806977 B CN 112806977B CN 202110160054 A CN202110160054 A CN 202110160054A CN 112806977 B CN112806977 B CN 112806977B
- Authority
- CN
- China
- Prior art keywords
- physiological
- physiological parameter
- matrix
- scale fusion
- dimensional
- 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.)
- Active
Links
- 230000004927 fusion Effects 0.000 title claims abstract description 45
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000005259 measurement Methods 0.000 claims abstract description 34
- 230000003068 static effect Effects 0.000 claims abstract description 11
- 230000009466 transformation Effects 0.000 claims abstract description 6
- 238000001914 filtration Methods 0.000 claims abstract description 3
- 239000011159 matrix material Substances 0.000 claims description 42
- 238000012360 testing method Methods 0.000 claims description 15
- 238000012549 training Methods 0.000 claims description 14
- 238000011176 pooling Methods 0.000 claims description 9
- 230000007246 mechanism Effects 0.000 claims description 7
- 230000036772 blood pressure Effects 0.000 claims description 6
- 101100379080 Emericella variicolor andB gene Proteins 0.000 claims description 4
- 210000000748 cardiovascular system Anatomy 0.000 claims description 4
- 238000000354 decomposition reaction Methods 0.000 claims description 4
- 230000029058 respiratory gaseous exchange Effects 0.000 claims description 4
- 206010003210 Arteriosclerosis Diseases 0.000 claims description 3
- 208000011775 arteriosclerosis disease Diseases 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012952 Resampling Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000005520 cutting process Methods 0.000 claims description 2
- 239000012634 fragment Substances 0.000 claims description 2
- 230000004217 heart function Effects 0.000 claims description 2
- 238000010606 normalization Methods 0.000 claims description 2
- 230000001575 pathological effect Effects 0.000 claims description 2
- 238000003672 processing method Methods 0.000 claims description 2
- 230000011218 segmentation Effects 0.000 claims description 2
- 101150011264 setB gene Proteins 0.000 claims description 2
- 230000005236 sound signal Effects 0.000 claims description 2
- 238000011426 transformation method Methods 0.000 claims description 2
- 238000005070 sampling Methods 0.000 abstract description 5
- 239000000284 extract Substances 0.000 abstract description 4
- 230000000295 complement effect Effects 0.000 abstract description 2
- 238000000691 measurement method Methods 0.000 abstract 1
- 238000013441 quality evaluation Methods 0.000 abstract 1
- 230000035487 diastolic blood pressure Effects 0.000 description 8
- 230000035488 systolic blood pressure Effects 0.000 description 8
- 208000024172 Cardiovascular disease Diseases 0.000 description 5
- 238000010219 correlation analysis Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 230000009467 reduction Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 230000003278 mimic effect Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000009530 blood pressure measurement Methods 0.000 description 2
- 230000002354 daily effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000003205 diastolic effect Effects 0.000 description 2
- 230000036541 health Effects 0.000 description 2
- 230000032683 aging Effects 0.000 description 1
- 230000004872 arterial blood pressure Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 239000008280 blood Substances 0.000 description 1
- 210000004369 blood Anatomy 0.000 description 1
- 230000017531 blood circulation Effects 0.000 description 1
- 230000036770 blood supply Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000003862 health status Effects 0.000 description 1
- 230000004060 metabolic process Effects 0.000 description 1
- 230000004089 microcirculation Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000011056 performance test Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000010349 pulsation 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 for evaluating the cardiovascular system, e.g. pulse, heart rate, blood pressure or blood flow
- A61B5/0205—Simultaneously evaluating both cardiovascular conditions and different types of body conditions, e.g. heart and respiratory condition
-
- 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
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- 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
- A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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 for evaluating the cardiovascular system, e.g. pulse, heart rate, blood pressure or blood flow
- A61B5/021—Measuring pressure in heart or blood vessels
-
- 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 for evaluating the cardiovascular system, e.g. pulse, heart rate, blood pressure or blood flow
- A61B5/024—Measuring pulse rate or heart rate
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/08—Measuring devices for evaluating the respiratory organs
- A61B5/0816—Measuring devices for examining respiratory frequency
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Signal Processing (AREA)
- Physiology (AREA)
- Veterinary Medicine (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Theoretical Computer Science (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Pathology (AREA)
- Heart & Thoracic Surgery (AREA)
- Psychiatry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Cardiology (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Pulmonology (AREA)
- Fuzzy Systems (AREA)
- Power Engineering (AREA)
- Measuring Pulse, Heart Rate, Blood Pressure Or Blood Flow (AREA)
Abstract
本发明提供一种基于多尺度融合网络的生理参数测量方法。所述方法包括：对生理信号进行采样并生成一维生理信号数据序列；经数据片段划分和信号质量评估，对符合信号质量要求的数据片段进行噪声滤除；然后对数据片段进行数学变换生成多维的输入数据张量；利用多尺度融合网络对输入数据提取潜在特征得到生理参数的估计值。通过识别测量模式标识符，将所有估计值的均值作为静态模式下的生理参数测量值；将所有估计值构成的一维连续数据序列作为动态模式下的生理参数连续测量值。本方法可充分提取信号中不同尺度的互补信息，实现生理参数的准确测量，其应用范围涵盖所有生理参数的测量，在心血管疾病研究和信号处理研究领域均具有一定的应用价值。
The invention provides a physiological parameter measurement method based on a multi-scale fusion network. The method includes: sampling physiological signals and generating a one-dimensional physiological signal data sequence; performing noise filtering on data segments that meet signal quality requirements through data segment division and signal quality evaluation; and then performing mathematical transformation on the data segments to generate multi-dimensional The input data tensor is used to extract the latent features of the input data by using a multi-scale fusion network to obtain the estimated values of physiological parameters. By identifying the measurement mode identifier, the mean of all estimated values is taken as the measured value of the physiological parameter in the static mode; the one-dimensional continuous data sequence formed by all the estimated values is taken as the continuous measured value of the physiological parameter in the dynamic mode. The method can fully extract the complementary information of different scales in the signal, and realize the accurate measurement of physiological parameters.
Description
Technical Field
The invention relates to a physiological parameter measuring method based on a multi-scale fusion network.
Background
The heart is the power center of human blood circulation, and supplies blood to the whole body through regular pulsation to meet the metabolism of the human body, thereby maintaining the normal life activities of the human body. It is of utmost importance to effectively measure physiological parameters related to the cardiovascular system to monitor the health status of a human body to prevent cardiovascular diseases. With the aging population and the increase of working pressure, the prevalence of cardiovascular diseases in China is increasing year by year. Cardiovascular diseases have become a leading cause of human death worldwide as reported by the world health organization in 2020. Therefore, the dynamic detection of the physiological parameters related to the cardiovascular system and the accurate evaluation of the health condition have great practical significance for the prevention and treatment of cardiovascular diseases.
Currently, a large number of medical sensors are required to be worn on commonly used physiological parameter detection equipment, such as a dynamic blood pressure monitor, an arteriosclerosis detector, a microcirculation detector and the like. Although these devices work accurately during the measurement process, the long wear can be uncomfortable for the patient, and the devices are too bulky and expensive to use in everyday life. Therefore, portability and comfort are the key of the physiological parameter measuring system facing daily life.
In recent years, the development of wearable technology and high-performance processing chips lays a hardware foundation for physiological signal processing, and the rise of deep learning makes accurate measurement of physiological parameters possible, so more and more researches are focused on automatically extracting characteristic parameters from physiological signals by using deep learning algorithms to estimate physiological parameters. The physiological signals are mostly one-dimensional time series with periodicity, and the change of the physiological signals reflects physiological information of various systems of a human body. At present, most of researches adopt a convolutional neural network with a single scale to realize automatic feature extraction, but the method ignores information which may be useful on other scales, so that the measurement accuracy cannot be further improved. Meanwhile, for the same physiological signal, the variability among individuals and the variability of waveforms of the physiological signal also cause that the convolutional neural network with a single scale cannot effectively extract features.
Disclosure of Invention
In order to overcome the defects, the invention aims to provide a brand-new physiological parameter measuring method based on a multi-scale fusion network. The method of the invention fully exerts the complementary advantages of the information of different scales by combining the information of different scales, and realizes accurate physiological parameter measurement. Meanwhile, the adopted end-to-end neural network avoids complicated feature point detection and feature engineering. The method extracts multi-scale fusion characteristics from physiological signals through a multi-scale fusion network, obtains estimated values of physiological parameters to be measured by regression of the characteristics, and outputs a static mean value or dynamic continuous measured value of the physiological parameters to be measured by corresponding operation according to a measurement mode identifier.
The invention provides a physiological parameter measuring method based on a multi-scale fusion network, which comprises the following specific steps:
(1) collecting physiological signals of biological individuals under static or dynamic conditions; preprocessing the acquired physiological signals, namely performing data segment cutting, resampling and normalization operation, and then removing the interference of baseline drift, power frequency, respiration, motion artifacts, electromyographic noise and the like by adopting a filtering or other signal processing method to obtain a one-dimensional signal segment;
(2) for the one-dimensional signal segment obtained in the step (1), the dimensionality is expanded by adopting mathematical transformation to construct more complete and rich physiological signal representation, and a three-dimensional matrix is obtainedB, C, N]Wherein:Bas the total number of data segments after the segmentation,Cfor the dimensions after the mathematical transformation to be used,Nthe length of the data segment, namely the number of sample points;
(3) for the product obtained in step (2)Three-dimensional matrix [ 2 ]B, C, N]Divided into training sets according to a certain ratio (e.g., 9:1, 8:2 or 7: 3)B 1 , C, N]And the test setB 2 , C, N]，B 1 AndB 2 the number of fragments in the training set and test set, respectively, andB 1 andB 2 the sum being equal toB；
(4) Constructing a physiological parameter measurement model, wherein the model comprises three parts: a multi-scale fusion network, a hybrid attention mechanism and a convolution network layer; according to a predetermined scale numberIDesigning a multi-scale fusion convolutional layer, and then fusing the number of convolutional layers according to the preset multi-scaleMSetting a multi-scale fusion network, and enabling the three-dimensional matrix obtained in the step (2)B, C, N]Inputting the multi-scale fusion network to obtain a first characteristic data three-dimensional matrixS(ii) a Digging out a second characteristic data three-dimensional matrix related to the physiological parameter to be measured by using a mixed attention machineS 1 (ii) a According to the set number of layers of the convolutional networkMUsing a convolutional network layer to perform three-dimensional matrix on the second characteristic dataS 1 Performing regression calculation to generate an estimated value of the physiological parameter to obtain a two-dimensional data matrixB, X]Wherein:Xis an estimated value of the physiological parameter to be measured;
(5) obtaining a measurement mode identifier, the measurement mode identifier comprising a static mode and a dynamic mode:
when the measurement mode identifier is a static mode, the two-dimensional data matrix obtained in step (4)B, X]Carrying out mean value operation to generate a mean value data matrix [1,X M]average value ofX MAs a physiological parameter measurement in static mode;
when the measurement mode identifier is in a dynamic mode, continuously extracting all estimated values of the physiological parameters to be measured obtained in the step (4) to form a one-dimensional continuous data sequence as a continuous physiological parameter measured value in the dynamic mode;
(6) the training set [ 2 ] obtained in the step (3)B 1 , C, N]Inputting the physiological parameters into the model in the step (4) for training and optimizing to obtain a physiological parameter estimation model; will test set 2B 2 , C, N]Inputting a physiological parameter estimation model for testing, and checking the accuracy of the model.
In the present invention, the physiological parameter includes, but is not limited to, one or more of heart rate, blood pressure, respiration rate, cardiac function index or arteriosclerosis index.
In the invention, the signal collected in step (1) is a physiological signal containing the physiological and pathological information of a cardiovascular system, and mainly comprises the following types: one or more of electrocardio signals, pulse wave signals, heart impact signals or heart sound signals.
In the present invention, the mathematical transformation method for dimension expansion of physiological signals in step (2) comprises the following types: any one of difference, integral, fourier transform, wavelet transform, empirical mode decomposition, or variational mode decomposition.
In the invention, the multi-scale fusion convolutional layer in the step (4) is formed by the following method:
(4.1) setting the convolution layer with the convolution kernel size of 1 to adjust the channel number of the input data according to the output channel number of the set multi-scale fusion convolution layer;
(4.2) number of scales according to settingISetting upIMulti-scale fusion convolutional layer with different convolutional kernel sizesF 1 , F 2 ,…F I This isIThe convolution kernel sizes of the multiple multi-scale fusion convolutional layers are spaced at 2 intervals.IThe input matrix is simultaneously convolved by a plurality of multi-scale fusion convolution layers to obtainIAn output matrixY 1 , Y 2 …Y I Using global pooling pairsIReducing the dimension of each output matrix to obtain a one-dimensional embedded vectorz 1 , z 2 , …z I Then using two fully-connected layers andsoftmaxweight matrix obtained by compressing and recovering information of embedded vectorW 1 , W 2 , …W I The weight matrixW 1 , W 2 , …W I And output matrixY 1 , Y 2 …Y I And correspondingly multiplying and summing to obtain the multi-scale fusion feature.
In the present invention, convolution operation is performed on an input matrix by using convolution layers, and the convolution operation includes the following types: a classical convolution or a dilated convolution.
In the invention, global pooling is used for reducing the dimension of the output matrix, and the method comprises the following types: global average pooling or global maximum pooling.
In the invention, a mixed attention mechanism is utilized in the step (4) to further mine the characteristics related to the physiological parameters to be estimated, and the method comprises the following types: a Bottleneck Attention Module (BAM) or convolution module attention mechanism (CBAM).
In the invention, the average value of all estimated values is used as the physiological parameter measured value in the static mode in the step (5); and taking the one-dimensional continuous data sequence formed by all the estimation values as the continuous measurement value of the physiological parameter in the dynamic mode.
In the invention, the training set in the step (6) is used for training the weights in the model, and the test set is used for verifying the performance of the physiological parameter measurement model on an unknown data set. According to different division modes of the data set, the physiological parameter measurement model can be divided into a calibration model and a non-calibration model, the data of the same subject appearing in the training set and the test set is the calibration model, and otherwise, the data is the non-calibration model.
The invention has the following beneficial effects:
1. the method can be used for realizing noninvasive and portable physiological parameter measurement, and is favorable for daily monitoring;
2. the invention realizes the dimension expansion by using mathematical transformation on the physiological signals, and can more comprehensively mine the potential information contained in the physiological signals. The measurement of the physiological parameter can be achieved more accurately than using only the original physiological signal;
3. the invention fully extracts the information of different scales in the original signal by utilizing the multi-scale fusion network, realizes the complementation between different scales and can more accurately realize the measurement of physiological parameters;
4. the application range of the method disclosed by the invention covers the measurement of all physiological parameters, and the method has a certain application value in the fields of cardiovascular disease research and signal processing research.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly described below. It is noted that the following drawings illustrate only certain embodiments of the invention and are therefore not to be considered limiting of its scope.
Fig. 1 is a schematic diagram of noise reduction of a pulse wave signal. The abscissa is time and the ordinate is signal amplitude. In fig. 1, (a) is a pulse wave signal segment before noise reduction, which includes a certain power frequency interference and baseline drift; (b) is a clean signal after noise reduction.
Fig. 2 shows the architecture of a multi-scale fusion network model for measuring systolic and diastolic pressures.
FIG. 3 shows the specific architecture and operation flow of the multi-scale fusion convolutional layer with the number of scales set to 3.
Fig. 4 shows a correlation analysis between the output values and the actual values of the blood pressure measurement model. (a) Correlation analysis of diastolic pressure for a non-calibrated model, (b) correlation analysis of diastolic pressure for a calibrated model; (c) the correlation analysis of the systolic pressure of a non-calibration model, and (d) the correlation analysis of the systolic pressure of a calibration model. In the coordinates of fig. 4, the horizontal axis represents the true value, and the vertical axis represents the output value of the model.
Detailed Description
The method and the application of the invention are further explained below with reference to the figures and the examples. These embodiments do not limit the invention; variations in structure, method, or function that may be apparent to those of ordinary skill in the art upon reading the foregoing description are intended to be within the scope of the present invention.
Example 1: the physiological parameter measurement model based on the multi-scale fusion network is applied to a dynamic systolic pressure and diastolic pressure measurement task, and the task is realized by adopting an MIMIC database. The MIMIC database contains ECG (electrocardiogram signal), PPG (pulse wave signal) and ABP (arterial blood pressure signal), all at a sampling rate of 125 Hz. The ECG and PPG signals are used to measure the blood pressure value, and the ABP signal is compared as the true value with the blood pressure value measured by the model. The method for measuring the blood pressure value by adopting the multi-scale fusion network model comprises the following specific steps:
(1) the pulse wave signals in the MIMIC database are observed as shown in fig. 1 (a). The pulse wave signals have serious baseline drift and contain a certain degree of power frequency interference. The pulse wave signal is first Discrete Wavelet Transform (DWT) decomposed with a db8 wavelet basis function. Then setting the wavelet coefficient corresponding to the noise frequency range to zero, and finally reconstructing according to the wavelet coefficient. Obtaining a clean pulse wave signal through the above pre-processing, as shown in fig. 1 (b);
dividing the pulse wave signals subjected to noise reduction according to the duration of 8 seconds (namely the number of sampling points is 1000) to obtain all data segments, selecting the slope as an index to evaluate the signal quality of all the data segments, discarding the current segment if the index is less than 0, and keeping the current segment if the index is more than 0;
(2) expanding dimensionality of the pulse wave data segments by adopting first-order difference and second-order difference, and splicing to obtain a three-dimensional input matrix [3, 1000], wherein 3 is input dimensionality, and 1000 is sampling point number;
(3) dividing the data of 1825 subjects in the database into a training set and a test set according to the ratio of 8:2 to form a data set of the embodiment;
(4) the overall Multi-scale Fusion Network model framework is shown in FIG. 2, where Multi-scale Fusion CNN backbone is a Multi-scale Fusion Network, where Attention is the Attention layer, DBP Network is the diastolic convolution Network, SBP Network is the systolic convolution Network, and Loss isauxIs fromLoss function, Loss, of multi-scale converged networkssbpAnd LossdbpThe loss functions from the systolic and diastolic convolution networks, respectively. The construction of the multi-scale fusion network consists of 5 steps, wherein the 5 steps respectively comprise 1, 2, 3, 3 and 3 multi-scale fusion convolutional layers, and the first multi-scale fusion convolutional layer of each step is used for changing the number of channels and reducing sampling. The first step is to increase the number of channels to 32, and the rest of the stages are increased by 2 times on this basis. Meanwhile, the signal length is decreased by 2 times at each step;
the specific structure of the multi-scale fusion convolutional layer in this embodiment is shown in fig. 3, where ConvX (X is 1, 7, 9, 11, respectively) represents a convolutional layer with a convolution kernel size of X. The number of scales of the multi-scale fusion convolutional layers is set to be 3, the convolutional layers adopt expansion convolution, and the sizes of the three convolutional layers are respectively selected to be 7, 9 and 11. After the multi-scale fusion network model, the original input matrix [ 2 ]B, 3, 1000]First, the matrix is converted into a feature matrixB, 256, 32]，BIndicating the number of samples. The mixed attention mechanism adopts a bottleneck attention module, and a systolic pressure and diastolic pressure convolution network consists of two layers of ordinary convolution layers with the channel number of 512, a global average pooling layer and convolution layers with the convolution kernel size of 1. Characteristic matrix [ alpha ]B, 256, 32]The physiological parameter matrix is finally obtained after the mixed attention module and the convolution networkB, 2]；
(5) Acquiring a measurement mode identifier, wherein the measurement mode identifier is a dynamic mode, and extracting all physiological parameter estimation values from the physiological parameter matrix finally obtained in the step (4) to form a one-dimensional continuous data sequence: systolic pressure sequence [ alpha ], [ alpha ] and [ alpha ], [ alpha ] aB S , 1]And diastolic pressure sequence [ alpha ], [ beta ] -andB D , 1]，B S the number of measurements of the systolic blood pressure is indicated,B D representing the number of measurements of diastolic pressure;
(6) the training set is used to train the weights in the model, and the test set is used to verify the performance of the physiological parameter measurement model on the unknown data set. According to different division modes of the data set, the physiological parameter measurement model can be divided into a calibration model and a non-calibration model, the data of the same subject appearing in the training set and the test set is the calibration model, and otherwise, the data is the non-calibration model. The results of the performance test of this example are shown in fig. 4. FIGS. 4 (a), (c) reflect the correlation between the measured values and the true values of the non-calibrated model (Cal-free) on the test set; fig. 4 (b), (d) reflect the correlation between the measured values and the actual values of the calibration model (Cal-based) on the test set. The result shows that the physiological parameter measurement model constructed by the embodiment can accurately measure the systolic pressure and the diastolic pressure.
Claims (8)
1. A physiological parameter measuring method based on a multi-scale fusion network is characterized by comprising the following specific steps:
(1) collecting physiological signals of a biological individual in a static mode or a dynamic mode; preprocessing the acquired physiological signals, namely performing data segment cutting, resampling and normalization operation, and then removing the interference of baseline drift, power frequency, respiration, motion artifacts and electromyographic noise by adopting a filtering or other signal processing method to obtain one-dimensional signal segments;
(2) expanding dimensionality of the one-dimensional signal segment obtained in the step (1) by adopting mathematical transformation to construct more complete and rich physiological signal representation to obtain a three-dimensional matrix [ alpha ], [ alpha ] formB, C, N]Wherein:Bas the total number of data segments after the segmentation,Cfor the dimensions after the mathematical transformation to be used,Nthe length of the data segment, namely the number of sample points;
(3) subjecting the three-dimensional matrix obtained in step (2)B, C, N]Is divided into training sets according to a certain proportionB 1 , C, N]And the test setB 2 , C, N]，B 1 AndB 2 the number of fragments in the training set and test set, respectively, andB 1 andB 2 the sum being equal toB；
(4) Constructing a physiological parameter measurement model, wherein the model comprises three parts: multi-scale fusion network, hybrid attention mechanism and convolution networkA layer; according to a predetermined scale numberIDesigning a multi-scale fusion convolutional layer, and then fusing the number of convolutional layers according to the preset multi-scaleMSetting a multi-scale fusion network, and enabling the three-dimensional matrix obtained in the step (2)B, C, N]Inputting the data into the multi-scale fusion network to obtain a first characteristic data three-dimensional matrixS(ii) a Digging out a second characteristic data three-dimensional matrix related to the physiological parameter to be measured by using a mixed attention machineS 1 (ii) a According to the set number of layers of the convolutional networkMUsing a convolutional network layer to perform three-dimensional matrix on the second characteristic dataS 1 Performing regression calculation to generate an estimated value of the physiological parameter to obtain a two-dimensional data matrixB, X]Wherein:Xis an estimated value of the physiological parameter to be measured;
(5) obtaining a measurement mode identifier, the measurement mode identifier comprising a static mode and a dynamic mode:
when the measurement mode identifier is a static mode, the two-dimensional data matrix obtained in step (4)B, X]Carrying out mean value operation to generate a mean value data matrix [1,X M]average value ofX MAs a physiological parameter measurement in static mode;
when the measurement mode identifier is in a dynamic mode, continuously extracting all estimated values of the physiological parameters to be measured obtained in the step (4) to form a one-dimensional continuous data sequence as a continuous physiological parameter measured value in the dynamic mode;
(6) the training set [ 2 ] obtained in the step (3)B 1 , C, N]Inputting the physiological parameters into the model in the step (4) for training and optimizing to obtain a physiological parameter estimation model; will test set 2B 2 , C, N]Inputting a physiological parameter estimation model for testing, and checking the accuracy of the model.
2. The method of claim 1, wherein the physiological parameter comprises one or more of heart rate, blood pressure, respiration rate, cardiac function index, and arteriosclerosis index.
3. The method according to claim 1, wherein the acquired signal in step (1) is a physiological signal containing physiological and pathological information of cardiovascular system, specifically: one or more of electrocardio signals, pulse wave signals, heart impact signals or heart sound signals.
4. The method according to claim 1, wherein the mathematical transformation method for dimension expansion of the physiological signal in step (2) is specifically: any one of difference, integral, fourier transform, wavelet transform, empirical mode decomposition, or variational mode decomposition.
5. The method of claim 1, wherein the multi-scale fusion convolutional layer in step (4) comprises the following steps:
(4.1) according to the number of output channels of the set multi-scale fusion convolutional layer, setting the multi-scale fusion convolutional layer with the convolutional kernel size of 1 to adjust the number of channels of input data;
(4.2) number of scales according to settingISetting upIMulti-scale fusion convolutional layer with different convolutional kernel sizesF 1 , F 2 ,…F I This isIThe sizes of convolution kernels of the multi-scale fusion convolution layers are spaced by 2;Ithe input matrix is simultaneously convolved by a plurality of multi-scale fusion convolution layers to obtainIAn output matrixY 1 , Y 2 …Y I Using global pooling pairsIReducing the dimension of each output matrix to obtain a one-dimensional embedded vectorz 1 , z 2 , …z I Then using two fully-connected layers andsoftmaxweight matrix obtained by compressing and recovering information of embedded vectorW 1 , W 2 , …W I The weight matrixW 1 , W 2 , …W I And the transmissionGo out matrixY 1 , Y 2 …Y I And correspondingly multiplying and summing to obtain the multi-scale fusion feature.
6. The method of claim 5, wherein the convolution operation is performed on the input matrix using convolutional layers, and wherein the convolution operation is either a classical convolution or a dilated convolution.
7. The method of claim 5, wherein the output matrix is reduced in dimension using global pooling, and wherein the method is global average pooling or global maximum pooling.
8. The method according to claim 1, wherein the step (4) further mines a second three-dimensional matrix of feature data related to the physiological parameter to be estimated by using a mixed attention mechanism, specifically: a bottleneck attention module or a convolution module attention mechanism.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110160054.4A CN112806977B (en) | 2021-02-05 | 2021-02-05 | Physiological parameter measuring method based on multi-scale fusion network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110160054.4A CN112806977B (en) | 2021-02-05 | 2021-02-05 | Physiological parameter measuring method based on multi-scale fusion network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112806977A CN112806977A (en) | 2021-05-18 |
| CN112806977B true CN112806977B (en) | 2022-03-18 |
Family
ID=75861680
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110160054.4A Active CN112806977B (en) | 2021-02-05 | 2021-02-05 | Physiological parameter measuring method based on multi-scale fusion network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112806977B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220414181A1 (en) * | 2021-06-29 | 2022-12-29 | Analog Devices International Unlimited Company | Method and device for translating between stochastic signals |
| CN113647905A (en) * | 2021-07-13 | 2021-11-16 | 广东省科学院健康医学研究所 | Training method for predicting blood pressure by deep neural network, computer device and storage medium |
| CN114041780B (en) * | 2021-12-08 | 2022-08-30 | 中国科学院计算技术研究所 | Method for monitoring respiration based on data acquired by inertial sensor |
| CN113892930B (en) * | 2021-12-10 | 2022-04-22 | 之江实验室 | A method and device for measuring facial heart rate based on multi-scale heart rate signals |
| CN116509357B (en) * | 2023-05-16 | 2024-08-02 | 长春理工大学 | Continuous blood pressure estimation method based on multi-scale convolution |
| CN118303876A (en) * | 2024-06-11 | 2024-07-09 | 长春大学 | Multi-scale feature fusion respiratory motion prediction method and system |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6389308B1 (en) * | 2000-05-30 | 2002-05-14 | Vladimir Shusterman | System and device for multi-scale analysis and representation of electrocardiographic data |
| CN109325065A (en) * | 2018-12-04 | 2019-02-12 | 浙江科技学院 | Multi-sampling rate soft sensing method based on dynamic latent variable model |
| CN110084292A (en) * | 2019-04-18 | 2019-08-02 | 江南大学 | Object detection method based on DenseNet and multi-scale feature fusion |
| CN110495878A (en) * | 2019-08-21 | 2019-11-26 | 中国科学院深圳先进技术研究院 | ECG-based disease prediction method, device and electronic equipment |
| CN110495877A (en) * | 2019-08-21 | 2019-11-26 | 中国科学院深圳先进技术研究院 | A method and device for extracting multi-scale features based on ECG |
| WO2020028382A1 (en) * | 2018-07-30 | 2020-02-06 | Memorial Sloan Kettering Cancer Center | Multi-modal, multi-resolution deep learning neural networks for segmentation, outcomes prediction and longitudinal response monitoring to immunotherapy and radiotherapy |
| CN110801218A (en) * | 2019-11-13 | 2020-02-18 | 清华大学 | Electrocardiogram data processing method, device, electronic device and computer readable medium |
| CN111449644A (en) * | 2020-03-19 | 2020-07-28 | 复旦大学 | A Bioelectric Signal Classification Method Based on Time-Frequency Transform and Data Enhancement Technology |
| CN111839488A (en) * | 2020-07-15 | 2020-10-30 | 复旦大学 | Non-invasive continuous blood pressure measurement device and method based on pulse wave |
| CN111904411A (en) * | 2020-08-25 | 2020-11-10 | 浙江工业大学 | Multi-lead heartbeat signal classification method and device based on multi-scale feature extraction |
| CN111920420A (en) * | 2020-07-28 | 2020-11-13 | 复旦大学 | Patient behavior multi-modal analysis and prediction system based on statistical learning |
| CN112102283A (en) * | 2020-09-14 | 2020-12-18 | 北京航空航天大学 | Retina fundus blood vessel segmentation method based on depth multi-scale attention convolution neural network |
| CN112101262A (en) * | 2020-09-22 | 2020-12-18 | 中国科学技术大学 | Multi-feature fusion sign language recognition method and network model |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107788970B (en) * | 2017-11-15 | 2020-08-07 | 中国航天员科研训练中心 | Mental load assessment method based on multi-physiological parameter method fusion |
| CN109805922A (en) * | 2019-01-24 | 2019-05-28 | 复旦大学 | Non-contact ECG measurement system |
| CN110141201B (en) * | 2019-05-14 | 2020-07-24 | 中国科学院深圳先进技术研究院 | Blood pressure estimation method and device |
| CN110647794B (en) * | 2019-07-12 | 2023-01-03 | 五邑大学 | Attention mechanism-based multi-scale SAR image recognition method and device |
-
2021
- 2021-02-05 CN CN202110160054.4A patent/CN112806977B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6389308B1 (en) * | 2000-05-30 | 2002-05-14 | Vladimir Shusterman | System and device for multi-scale analysis and representation of electrocardiographic data |
| WO2020028382A1 (en) * | 2018-07-30 | 2020-02-06 | Memorial Sloan Kettering Cancer Center | Multi-modal, multi-resolution deep learning neural networks for segmentation, outcomes prediction and longitudinal response monitoring to immunotherapy and radiotherapy |
| CN109325065A (en) * | 2018-12-04 | 2019-02-12 | 浙江科技学院 | Multi-sampling rate soft sensing method based on dynamic latent variable model |
| CN110084292A (en) * | 2019-04-18 | 2019-08-02 | 江南大学 | Object detection method based on DenseNet and multi-scale feature fusion |
| CN110495878A (en) * | 2019-08-21 | 2019-11-26 | 中国科学院深圳先进技术研究院 | ECG-based disease prediction method, device and electronic equipment |
| CN110495877A (en) * | 2019-08-21 | 2019-11-26 | 中国科学院深圳先进技术研究院 | A method and device for extracting multi-scale features based on ECG |
| CN110801218A (en) * | 2019-11-13 | 2020-02-18 | 清华大学 | Electrocardiogram data processing method, device, electronic device and computer readable medium |
| CN111449644A (en) * | 2020-03-19 | 2020-07-28 | 复旦大学 | A Bioelectric Signal Classification Method Based on Time-Frequency Transform and Data Enhancement Technology |
| CN111839488A (en) * | 2020-07-15 | 2020-10-30 | 复旦大学 | Non-invasive continuous blood pressure measurement device and method based on pulse wave |
| CN111920420A (en) * | 2020-07-28 | 2020-11-13 | 复旦大学 | Patient behavior multi-modal analysis and prediction system based on statistical learning |
| CN111904411A (en) * | 2020-08-25 | 2020-11-10 | 浙江工业大学 | Multi-lead heartbeat signal classification method and device based on multi-scale feature extraction |
| CN112102283A (en) * | 2020-09-14 | 2020-12-18 | 北京航空航天大学 | Retina fundus blood vessel segmentation method based on depth multi-scale attention convolution neural network |
| CN112101262A (en) * | 2020-09-22 | 2020-12-18 | 中国科学技术大学 | Multi-feature fusion sign language recognition method and network model |
Non-Patent Citations (10)
| Title |
|---|
| A novel method for continuous blood pressure estimation based on a single-channel photoplethysmogram signal;Hu, Qihan，Deng, Xintao，Yang, Cuiwei;《PHYSIOLOGICAL MEASUREMENT》;20201231;第1-12页 * |
| Arrhythmias Classification Using Short-Time Fourier Transform and GAN Based Data Augmentation;Hu, Qihan，He, Kaiyue，Yang, Cuiwei;《IEEE》;20200724;第308-311页 * |
| Deep Multi-Scale Fusion Neural Network for Multi-Class Arrhythmia Detection;R. Wang, J. Fan and Y. Li;《IEEE Journal of Biomedical and Health Informatics》;20200413;第2461-2472页 * |
| Leader-Based Multi-Scale Attention Deep Architecture for Person Re-Identification;X. Qian, Y. Fu, T. Xiang, Y. -G. Jiang and X. Xue;《IEEE Transactions on Pattern Analysis and Machine Intelligence》;20190715;第371-385页 * |
| 基于卷积注意力机制的情感分类方法;顾军华;《计算机工程与设计》;20200116;第104-108页 * |
| 基于多级特征和混合注意力机制的室内人群检测网络;沈文祥,秦品乐,曾建潮;《计算机应用》;20191015;第3496-3502页 * |
| 基于多通道时空融合网络的人体动作识别;高璇;《工业控制计算机》;20201025;第56-57+79页 * |
| 心率变异信号的多尺度分析;朱妍雯,姚景昆;《电子测试》;20190115;第53-56页 * |
| 心电标测系统中激动时刻提取方法的比较研究;王彦磊,杨翠微,余建国;《仪器仪表学报》;20150315;第661-667页 * |
| 融合注意力机制的多尺度目标检测算法;鞠默然, 罗江宁, 王仲博;《光学学报》;20200602;第132-140页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112806977A (en) | 2021-05-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112806977B (en) | Physiological parameter measuring method based on multi-scale fusion network | |
| US11476000B2 (en) | Methods and systems using mathematical analysis and machine learning to diagnose disease | |
| CN111466876B (en) | An auxiliary diagnosis system for Alzheimer's disease based on fNIRS and graph neural network | |
| CN110420019B (en) | A deep regression heart rate estimation method for shock cardiogram signals | |
| US20160183811A1 (en) | System and method of detecting abnormal movement of a physical object | |
| US11062792B2 (en) | Discovering genomes to use in machine learning techniques | |
| CN112932486A (en) | Multi-mode-based university student emotional stress detection system and method | |
| Namadurai et al. | Multifractal analysis of uterine electromyography signals for the assessment of progression of pregnancy in term conditions | |
| CN116451110A (en) | Method for constructing blood glucose prediction model based on signal energy characteristics and pulse cycle | |
| CN113509186B (en) | ECG classification system and method based on deep convolutional neural network | |
| CN115315214A (en) | Fusion Signal Processing for Maternal Uterine Activity Detection | |
| US20230075634A1 (en) | Methods and systems for engineering conduction deviation features from biophysical signals for use in characterizing physiological systems | |
| KR101744691B1 (en) | Method and Apparatus for Detecting Heartbeat using Ballistocardiogram | |
| CN115251845A (en) | Sleep monitoring method for processing brain wave signals based on TB-TF-BiGRU model | |
| CN112842355B (en) | Electrocardiosignal heart beat detection and identification method based on deep learning target detection | |
| CN112336369B (en) | Coronary heart disease risk index evaluation system of multichannel heart sound signals | |
| CN113729648A (en) | Wearable pulse-taking bracelet system based on multiple pulse sensors | |
| CN113425272B (en) | Method for analyzing atrial fibrillation through data acquired by wearable equipment | |
| CN115836847A (en) | Blood pressure prediction device and equipment | |
| CN111345815B (en) | Method, device, equipment and storage medium for detecting QRS wave in electrocardiosignal | |
| CN115605128A (en) | Pulse wave intelligent distinguishing method and system based on human body sequence parameter model | |
| US20250040894A1 (en) | Personalized chest acceleration derived prediction of cardiovascular abnormalities using deep learning | |
| CN117204835B (en) | Blood pressure measurement method and device based on PPG technology and computer storage medium | |
| CN111387939B (en) | Method, device, computer equipment and storage medium for measuring heart displacement | |
| Cammarota et al. | Trend extraction in functional data of amplitudes of R and T waves in exercise electrocardiogram |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |