WO2021164346A1 - Procédé et dispositif pour la prédiction de la tension artérielle - Google Patents
Procédé et dispositif pour la prédiction de la tension artérielle Download PDFInfo
- Publication number
- WO2021164346A1 WO2021164346A1 PCT/CN2020/129631 CN2020129631W WO2021164346A1 WO 2021164346 A1 WO2021164346 A1 WO 2021164346A1 CN 2020129631 W CN2020129631 W CN 2020129631W WO 2021164346 A1 WO2021164346 A1 WO 2021164346A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- dimensional
- cnn
- blood pressure
- dimensional tensor
- tensor
- Prior art date
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/02—Detecting, measuring or recording pulse, heart rate, blood pressure or blood flow; Combined pulse/heart-rate/blood pressure determination; Evaluating a cardiovascular condition not otherwise provided for, e.g. using combinations of techniques provided for in this group with electrocardiography or electroauscultation; Heart catheters for measuring blood pressure
- A61B5/021—Measuring pressure in heart or blood vessels
- A61B5/02108—Measuring pressure in heart or blood vessels from analysis of pulse wave characteristics
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7271—Specific aspects of physiological measurement analysis
- A61B5/7275—Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
Definitions
- the present invention relates to the technical field of electrophysiological signal processing, in particular to a method and device for predicting blood pressure.
- the heart is the center of human blood circulation.
- the heart produces blood pressure through regular pulsation, and then supplies blood to the whole body to complete the body's metabolism.
- Blood pressure is one of the human body's very important physiological signals. Blood pressure within the normal range can ensure normal blood circulation. Many factors work together to keep blood pressure normal, so that various organs and tissues of the human body can obtain sufficient blood volume to keep the human body functioning normally.
- Human blood pressure contains two important values: systolic blood pressure and diastolic blood pressure. Medically, these two quantities are used to judge whether human blood pressure is normal or not. Long-term continuous observation of these two parameters of blood pressure can help people have a clearer understanding of their own heart health. However, most of the current traditional blood pressure measurement methods use external force upward pressure detection methods such as pressure gauges, which are not only cumbersome to operate, but also easily cause discomfort to the subject, so they cannot be used multiple times to achieve the purpose of continuous monitoring. .
- the purpose of the present invention is to provide a method and device for predicting blood pressure in view of the defects of the prior art.
- PPG Photoplethysmography
- PPG Photoplethysmography
- the blood pressure feature data and the regression calculation of the blood pressure feature data are performed to predict the blood pressure data (diastolic blood pressure, systolic blood pressure) of the tester; there are two convolution schemes in the embodiment of the present invention.
- the full vector convolution method is used If the vector is long enough, the segmented vector convolution method can be used; the embodiment of the present invention not only avoids the cumbersome and uncomfortable feeling of conventional testing methods, but also produces an automatic and intelligent data analysis method, so that the application party can Conveniently carry out multiple continuous monitoring of the measured object.
- the first aspect of the embodiments of the present invention provides a method for predicting blood pressure, and the method includes:
- the wave one-dimensional segment divides multiple pulse wave one-dimensional sub-segments and obtains the total number of sub-segments;
- the CNN scheme identifier includes two identifiers of the first scheme and the second scheme;
- the blood pressure CNN model is used to perform a full vector convolution pooling calculation operation on the pulse wave one-dimensional vector, and the calculation result is calculated according to the blood pressure long- and short-term memory LSTM network model.
- the input parameter format performs tensor dimensionality reduction processing to generate the first LSTM input three-dimensional tensor;
- the blood pressure CNN model is used to perform a segmented vector convolution pooling calculation operation on the pulse wave one-dimensional vector, and the calculation result is calculated according to the blood pressure LSTM network
- the input parameter format of the model undergoes tensor dimensionality reduction processing to generate a second LSTM input three-dimensional tensor;
- the CNN scheme identifier use the LSTM network layer of the blood pressure LSTM network model to perform blood pressure long and short-term memory calculation operations on the first LSTM input three-dimensional tensor or the second LSTM input three-dimensional tensor to generate an LSTM Output a three-dimensional tensor;
- the blood pressure LSTM network model includes the LSTM network layer and a fully connected layer;
- the predicted blood pressure data is sequentially extracted from the blood pressure prediction three-dimensional tensor [X, Y, 2] to generate a blood pressure prediction Data collection.
- the pulse wave conversion and sampling processing are performed on the PPG signal data of the photoplethysmography method to generate a pulse wave one-dimensional vector; the pulse wave one-dimensional vector is divided into a plurality of pulse wave one-dimensional fragments and the total number of fragments is obtained ; Divide the one-dimensional pulse wave segment into multiple one-dimensional pulse wave sub-segments and obtain the total number of sub-segments, which specifically includes:
- the PPG signal acquisition device to collect light intensity signals from the preset light source signal on the local skin surface of the organism to generate the PPG signal data with a length of the signal acquisition time threshold; perform pulse wave data conversion operations on the PPG signal data Generate pulse wave signal data; perform characteristic data sampling on the pulse wave signal data according to the characteristic sampling frequency threshold to generate a pulse wave one-dimensional vector;
- the preset light source signal includes at least one of a red light source signal, an infrared light source signal, and a green light source signal kind;
- the pulse wave one-dimensional vector is divided into data segments according to the segment length threshold to generate a plurality of the pulse wave one-dimensional segments, and the total number of the pulse wave one-dimensional segments included in the pulse wave one-dimensional vector is used as the Total number of fragments;
- the pulse wave one-dimensional segment is divided into data sub-segments according to the sub-segment length threshold value to generate a plurality of the pulse wave one-dimensional sub-segments, and the pulse wave one-dimensional sub-segments included in the pulse wave one-dimensional segment The total is the total number of the sub-segments.
- the blood pressure CNN model is used to perform a full vector convolution pooling calculation operation on the pulse wave one-dimensional vector, and the calculation result is calculated according to the blood pressure long and short term Memorize the input parameter format of the LSTM network model and perform tensor dimensionality reduction processing to generate the first LSTM input three-dimensional tensor, which specifically includes:
- the CNN scheme identifier is the first scheme, perform a first blood pressure CNN input parameter setting operation according to the total number of fragments and the pulse wave one-dimensional vector to generate a first CNN input four-dimensional tensor;
- the first CNN output four-dimensional tensor is subjected to tensor reduction processing to generate the first LSTM input three-dimensional tensor.
- the first blood pressure CNN input parameter setting operation is performed according to the total number of segments and the pulse wave one-dimensional vector to generate a first CNN input four-dimensional Tensor, including:
- the first CNN input four-dimensional tensor is set specifically as the first CNN input four-dimensional tensor [B 1 ,1,W 1 ,1]; the first CNN input four-dimensional tensor [B 1 ,1,W 1 ,1] includes the B 1 first CNN input three-dimensional tensor [1,W 1 ,1]; said B 1 is the first CNN input four-dimensional tensor
- the fourth dimension parameter of the tensor [B 1 ,1,W 1 ,1], and the B 1 is the total number of segments;
- the W 1 is the first CNN input four-dimensional tensor [B 1 ,1, W 1 ,1] is the second dimension parameter, and the W 1 is the segment length threshold;
- the one-dimensional pulse wave segments included in the pulse wave one-dimensional vector are sequentially extracted, and the first CNN input three-dimensional tensor corresponding to the four-dimensional tensor [B 1 ,1,W 1 ,1] is input to the first CNN.
- the tensor [1,W 1 ,1] performs matrix element assignment processing.
- the blood pressure CNN model is used to perform multi-layer convolution pooling calculation on the first CNN input four-dimensional tensor to generate the first CNN output four-dimensional tensor, which specifically includes :
- Step 51 initialize the value of the first index to 1; initialize the first total to the threshold of the number of convolutional layers; initialize the first index temporary four-dimensional tensor to be the first CNN input four-dimensional tensor;
- Step 52 Use the first index layer convolution layer of the blood pressure CNN model to perform convolution calculation processing on the first index temporary four-dimensional tensor to generate a first index convolution output data four-dimensional tensor; use the blood pressure
- the first index layer pooling layer of the CNN model performs pooling calculation processing on the four-dimensional tensor of the first index convolution output data to generate the four-dimensional tensor of the first index pooled output data;
- the blood pressure CNN model includes multiple layers The convolutional layer and multiple layers of the pooling layer;
- Step 53 Set the temporary four-dimensional tensor of the first index as the four-dimensional tensor of the first index pooling output data
- Step 54 Add 1 to the first index
- Step 55 Determine whether the first index is greater than the first total, if the first index is greater than the first total, go to step 56, if the first index is less than or equal to the first total, go to Step 52;
- Step 56 Set the first CNN output four-dimensional tensor as the first index temporary four-dimensional tensor; the first CNN output four-dimensional tensor is specifically the first CNN output four-dimensional tensor [B 2 ,1,W 2 , C 2 ]; said B 2 is the fourth dimension parameter of the first CNN output four-dimensional tensor [B 2 ,1, W 2 , C 2 ], and said B 2 is the total number of segments; W 2 is the second dimension parameter of the first CNN output four-dimensional tensor [B 2 ,1, W 2 , C 2 ], and the W 2 is the preset threshold for the total number of neurons in the LSTM layer; the C 2 Output the first dimension parameter of the four-dimensional tensor [B 2 ,1, W 2 , C 2 ] for the first CNN, and the C 2 is a preset LSTM layer neuron length threshold.
- the first LSTM input three-dimensional tensor is specifically the first LSTM input three-dimensional tensor [H 3 , W 3 , C 3 ]; the H 3 is the first LSTM input three-dimensional tensor [H 3 , W 3 , C 3 ] is the third dimension parameter, and the H 3 is the total number of segments; W 3 is the second of the first LSTM input three-dimensional tensor [H 3 , W 3 , C 3 ] Dimensional parameter, and the W 3 is the W 2 ; the C 3 is the first dimensional parameter of the first LSTM input three-dimensional tensor [H 3 , W 3 , C 3 ], and the C 3 is The C 2 .
- the CNN scheme identifier is the second scheme
- the input parameter format of the blood pressure LSTM network model performs tensor dimensionality reduction processing to generate a second LSTM input three-dimensional tensor, which specifically includes:
- the CNN scheme identifier is the second scheme, generate the total number of tensors according to the product of the total number of segments multiplied by the total number of sub-segments;
- a second blood pressure CNN input parameter setting operation is performed to generate a second CNN input four-dimensional tensor group;
- the second CNN input four-dimensional tensor group includes the tensor The total number of second CNN input four-dimensional tensors;
- the blood pressure CNN model is used to perform multi-layer convolution pooling calculation on all the second CNN input four-dimensional tensors in the second CNN input four-dimensional tensor group to generate the first
- the second CNN outputs a four-dimensional tensor group;
- the second CNN outputs a four-dimensional tensor group including the total number of tensors and the second CNN outputs a four-dimensional tensor;
- the third CNN output four-dimensional tensor is subjected to tensor reduction processing to generate the second LSTM input three-dimensional tensor.
- the second blood pressure CNN input parameter setting operation is performed to generate the second CNN input four-dimensional tensor group, which specifically includes:
- the quantity group includes the total number of tensors and the second CNN input four-dimensional tensor [1,1,W 4 ,1]; the W 4 is the second CNN input four-dimensional tensor [1,1,W 4 ,1], and the W 4 is the sub-segment length threshold;
- the second CNN output four-dimensional tensor group specifically includes the total number of tensors and the second CNN output four-dimensional tensor; the second CNN output four-dimensional tensor is specifically the second CNN output four-dimensional tensor [ 1,1,1,C 5 ]; The C 5 is the first dimension parameter of the four-dimensional tensor [1,1,1,C 5] output by the second CNN.
- the performing a four-dimensional tensor merging operation on all the second CNN output four-dimensional tensors in the second CNN output four-dimensional tensor group to generate a third CNN output four-dimensional tensor specifically includes:
- the third CNN output four-dimensional tensor is set specifically as the third CNN output four-dimensional tensor [B 6 ,1,1,C 6 ]; the B 6 is the third CNN output four-dimensional tensor [B 6 ,1 ,1,C 6 ], and the B 6 is the total number of tensors; C 6 is the third CNN output four-dimensional tensor [B 6 ,1,1,C 6 ] The first dimension parameter, and the C 6 is the C 5 ;
- the second CNN output four-dimensional tensor group sequentially extract the matrix element sequence of the second CNN output four-dimensional tensor [1,1,1,C 5 ], and output the four-dimensional tensor [ B 6 ,1,1,C 6 ] performs matrix element assignment processing.
- the second LSTM input three-dimensional tensor is specifically an LSTM input three-dimensional tensor [H 7 , W 7 , C 7 ];
- the H 7 is the second LSTM input three-dimensional tensor [H 7 , W 7 , C 7 ], and the value of the H 3 is the total number of segments;
- the W 7 is the second LSTM input three-dimensional tensor [H 7 , W 7 , C 7 ] Dimension parameter, and the W 7 is the quotient of the B 6 divided by the H 3 ;
- the C 7 is the first dimension of the second LSTM input three-dimensional tensor [H 7 , W 7 , C 7 ] Parameter, and the C 7 is the C 6 .
- the LSTM network layer of the blood pressure LSTM network model is used to perform blood pressure long- and short-term memory on the first LSTM input three-dimensional tensor or the second LSTM input three-dimensional tensor Computational operations to generate LSTM output three-dimensional tensors, including:
- the CNN scheme identifier is the first scheme, use the LSTM network layer of the blood pressure LSTM network model to perform the blood pressure long and short-term memory calculation operation on the first LSTM input three-dimensional tensor to generate The LSTM outputs a three-dimensional tensor;
- the CNN scheme identifier is the second scheme
- the predicted blood pressure is sequentially extracted from the blood pressure prediction three-dimensional tensor [X, Y, 2] according to the sequence of the multiple one-dimensional pulse wave segments and the sequence of the multiple pulse wave one-dimensional sub-segments Data to generate a blood pressure prediction data set, including:
- the blood pressure prediction one-dimensional vector [2] included in the blood pressure prediction three-dimensional tensor [X, Y, 2] is sequentially extracted to generate the current one-dimensional vector [2]; the systolic blood pressure data of the blood pressure data group is set as the The sub-segment systolic blood pressure data in the current one-dimensional vector [2], and the diastolic blood pressure data of the blood pressure data group is set to the sub-segment diastolic blood pressure data in the current one-dimensional vector [2]; and the blood pressure data
- the group performs a data group addition operation to the blood pressure prediction data set;
- the blood pressure prediction three-dimensional tensor [X, Y, 2] includes X*Y the blood pressure prediction one-dimensional vectors [2];
- the vector [2] includes the sub-segment systolic blood pressure data and the sub-segment diastolic blood pressure data.
- the first aspect of the embodiments of the present invention provides a method for predicting blood pressure, which uses PPG to collect data from a tester and converts to generate pulse wave signal data, and then uses a feature extraction regression fusion composed of a blood pressure CNN model and a blood pressure LSTM network model
- the network performs feature extraction operations on the pulse wave signal data to generate blood pressure feature data and perform regression calculations on the blood pressure feature data to predict the tester’s blood pressure data (diastolic blood pressure, systolic blood pressure);
- blood pressure data diastolic blood pressure, systolic blood pressure
- there are two convolution schemes in the embodiment of the present invention If the vector length is not enough, use the full vector convolution method, if the vector is long enough, you can use the segmented vector convolution method; through the embodiment of the present invention, it not only avoids the cumbersomeness and discomfort of conventional testing methods, but also produces a kind of automatic Intelligent data analysis method, so that the application side can conveniently carry out multiple continuous monitoring of the measured
- a second aspect of the embodiments of the present invention provides a device, the device including a memory and a processor, the memory is used to store a program, and the processor is used to execute the first aspect and the methods in each implementation manner of the first aspect.
- the third aspect of the embodiments of the present invention provides a computer program product containing instructions, which when the computer program product runs on a computer, causes the computer to execute the first aspect and the methods in the first aspect.
- a fourth aspect of the embodiments of the present invention provides a computer-readable storage medium on which a computer program is stored.
- the computer program implements the first aspect and the methods in the first aspect when executed by a processor. .
- FIG. 1 is a schematic diagram of a method for predicting blood pressure according to Embodiment 1 of the present invention
- FIG. 2 is a schematic diagram of a method for predicting blood pressure according to Embodiment 2 of the present invention.
- FIG. 3 is a schematic diagram of the device structure of a device for predicting blood pressure according to Embodiment 3 of the present invention.
- the PPG signal is a set of signals that uses the light sensor to identify and record the change in light intensity of a specific light source.
- the blood flow per unit area in the blood vessel changes periodically, and the corresponding blood volume also changes accordingly, resulting in a periodic change trend in the PPG signal reflecting the amount of light absorbed by the blood.
- a cardiac cycle consists of two time periods: systolic and diastolic; during systole, the heart does work on the blood throughout the body, causing continuous and periodic changes in intravascular pressure and blood flow volume. The absorption of light is the most; during the diastole, the pressure on the blood vessels is relatively small.
- the PPG signal waveform reflecting the light energy absorbed by the blood in the blood vessel is composed of two signals: the systolic period signal and the diastolic period signal; the common PPG signal has two peaks, the first one belongs to the systolic period, and the diastolic period signal. The latter belongs to the diastolic period.
- Pulse wave signal By performing further feature recognition and regression classification processing on the pulse wave signal, we can further obtain the predicted value of the systolic and diastolic blood pressure. Specifically: First, perform effective data extraction on the pulse wave data, that is, so-called feature extraction or feature calculation; and then use the obtained feature data to use the regression data obtained by the blood pressure regression calculation as the prediction result.
- the regression data in the blood pressure regression calculation also corresponds to two: diastolic blood pressure data and systolic blood pressure data, where the systolic blood pressure data is greater than the diastolic blood pressure data.
- CNN has long been one of the core algorithms in the field of feature recognition. Used in image recognition, it can be used in fine classification and recognition to extract the discriminative features of the image for learning by other classifiers.
- the pulse wave feature extraction calculation is performed on the input one-dimensional pulse wave data: after the input original pulse wave data is convolved and pooled, the feature data that conforms to the pulse wave characteristics is retained.
- the blood pressure CNN model mentioned in the article is a CNN model that has been trained through blood pressure feature extraction. It is specifically composed of a convolutional layer and a pooling layer. The convolutional layer is responsible for performing blood pressure feature extraction calculations on the input data of the CNN model.
- the pooling layer is to down-sample the extraction results of the convolutional layer; the blood pressure CNN model in this article is divided into multiple CNN network layers, and each CNN network layer includes a convolutional layer and a pooling layer.
- the input data and output data format of the blood pressure CNN model are both in the form of a 4-dimensional tensor: [B, H, W, C]. After each layer of convolutional layer or pooling layer, the value of some dimensional parameters of the output data will change, that is, the total data length of the tensor will be shortened.
- B is the fourth-dimensional parameter (pulse The total number of fragments of the one-dimensional data sequence) will not change
- H and W are the third and two-dimensional parameters in the four-dimensional, and the changes of the two are related to the size of the convolution kernel of each convolutional layer and the setting of the sliding step length It is also related to the pooling window size and sliding step length of the pooling layer
- C is the first-dimensional parameter in the four-dimensional, and its change is related to the selected output space dimension in the convolutional layer (the number of convolution kernels) )related.
- the embodiment of the present invention provides two convolution schemes for CNN feature calculation, respectively: scheme one, the pulse wave data is sent to the blood pressure CNN model as a whole to perform overall convolution and pooling calculations and output feature tensors; Solution 2: Divide the pulse wave into segments, divide each segment into multiple sub-segments, send all the sub-segments to the blood pressure CNN model according to the sequence of the segments to perform convolution pooling calculations and output the total number of segments * total number of sub-segments features Tensor, and then merge all the output feature tensors to generate the feature tensor of the overall pulse wave.
- scheme one the pulse wave data is sent to the blood pressure CNN model as a whole to perform overall convolution and pooling calculations and output feature tensors
- Solution 2 Divide the pulse wave into segments, divide each segment into multiple sub-segments, send all the sub-segments to the blood pressure CNN model according to the sequence of the segments to perform convolution pooling calculations and output the total number of segments * total number of
- the LSTM network is suitable for processing and predicting important events with very long intervals and delays in the time series. It can selectively remember or forget the information of the previous step in the time series. By using the time series feature data to train the LSTM network, the regression classification can be achieved. Model effect.
- the embodiment of the present invention uses the blood pressure LSTM network model that has been trained through batch pulse wave signal feature data and corresponding batch actual measured blood pressure data to perform further regression classification processing for the feature data output by the blood pressure CNN model according to the time sequence of the data, and finally Output the predicted blood pressure result.
- the blood pressure LSTM network model in the embodiment of the present invention includes a multi-layer LSTM network and a fully connected layer.
- Each layer of LSTM network is composed of multiple processor units, which are connected in sequence, and can transmit information in a specified direction to make the network have memory function. There are also parameter settings in the processor unit that can control the size of the output. In the time sequence, what is passed from front to back is called forward, and what is passed from back to front is called reverse.
- the first layer of the LSTM network of the blood pressure LSTM network model used in this embodiment is a two-way LSTM network, and the subsequent layers are all one-way LSTM networks.
- the data format output by the last processor unit of the last layer of the LSTM network is in the form of a three-dimensional tensor. This three-dimensional tensor is input into the fully connected layer for blood pressure regression calculation to finally obtain the required data for prediction, including diastolic blood pressure data and systolic blood pressure data .
- Fig. 1 is a schematic diagram of a method for predicting blood pressure provided in Embodiment 1 of the present invention. The method mainly includes the following steps:
- Step 1 Perform pulse wave conversion and sampling processing on the PPG signal data of the photoplethysmography method to generate a one-dimensional pulse wave vector; divide the one-dimensional pulse wave vector into multiple one-dimensional pulse wave segments and obtain the total number of segments; divide the pulse wave The one-dimensional segment is divided into multiple one-dimensional sub-segments of pulse wave and the total number of sub-segments is obtained;
- Step 11 call the PPG signal acquisition device, collect the light intensity signal of the preset light source signal on the local skin surface of the organism, and generate a PPG signal data whose length is the signal acquisition time threshold; perform pulse wave data on the PPG signal data The conversion operation generates pulse wave signal data; the characteristic data of the pulse wave signal data is sampled according to the characteristic sampling frequency threshold to generate a one-dimensional pulse wave vector; the preset light source signal includes at least one of the red light source signal, the infrared light source signal and the green light source signal;
- the pulse wave one-dimensional vector is specifically the pulse wave one-dimensional vector [A];
- the pulse wave one-dimensional vector [A] specifically becomes the pulse wave one-dimensional vector [1250], which is a one-dimensional vector including 1250 independent pulse wave data;
- Step 12 Divide the pulse wave one-dimensional vector into data segments according to the segment length threshold to generate multiple pulse wave one-dimensional segments, and use the total number of pulse wave one-dimensional segments included in the pulse wave one-dimensional vector as the total number of segments;
- Step 13 Divide the pulse wave one-dimensional segment into data sub-segments according to the sub-segment length threshold to generate multiple pulse wave one-dimensional sub-segments, and use the total number of pulse wave one-dimensional sub-segments included in the pulse wave one-dimensional segment as the total number of sub-segments .
- the blood pressure CNN will be used to perform feature calculations on the data in the pulse wave one-dimensional vector
- the pulse wave one-dimensional vector is segmented according to the maximum data length input by the blood pressure CNN, where ,
- the fragment length threshold is the maximum data length input by the blood pressure CNN.
- Step 2 Obtain the value of the CNN scheme identifier of the convolutional neural network as the first scheme
- the CNN scheme identifier includes two identifiers: the first scheme and the second scheme.
- the CNN scheme identifier is used to distinguish the two processing methods of CNN: when the CNN scheme identifier is the first scheme, the pulse wave one-dimensional vector is input as a whole into the CNN for full vector convolution pooling Calculate, and use the output result as the input of the next long and short-term memory calculation.
- Step 3 When the CNN scheme identifier is the first scheme, use the blood pressure CNN model to perform a full vector convolution pooling calculation operation on the pulse wave one-dimensional vector, and use the calculation result according to the input parameter format of the blood pressure long and short-term memory LSTM network model Perform tensor dimensionality reduction processing to generate the first LSTM input three-dimensional tensor;
- Step 31 When the CNN scheme identifier is the first scheme, perform the first blood pressure CNN input parameter setting operation according to the total number of fragments and the pulse wave one-dimensional vector to generate the first CNN input four-dimensional tensor;
- step 311 when the CNN scheme identifier is the first scheme, set the first CNN input four-dimensional tensor specifically to the first CNN input four-dimensional tensor [B 1 ,1,W 1 ,1];
- the first CNN input four-dimensional tensor [B 1 ,1,W 1 ,1] includes B 1 first CNN input three-dimensional tensor [1,W 1 ,1]; B 1 is the first CNN input four-dimensional tensor The fourth dimension parameter of [B 1 ,1,W 1 ,1], and B 1 is the total number of segments; W 1 is the second dimension parameter of the first CNN input four-dimensional tensor [B 1 ,1,W 1 ,1] , And W 1 is the fragment length threshold;
- the obtained one-dimensional pulse wave segment is converted into 4-dimensional data.
- the four dimensions [B, H, W, C] represent the number of segments (batch), the height of the segment data (height), and the width of the segment data. (width) and the number of channels of fragment data (channel).
- the height, width, and channel number of the segment correspond to the height, width, and color channel of the image, respectively.
- the number of segments B should be set to the total number of segments, the height H should be set to 1, the number of color channels C should be set to 1, and the width W should be the segment length threshold previously set, for example:
- the fragment length threshold is 250, and the first CNN input four-dimensional tensor obtained after converting the pulse wave one-dimensional vector [1250] is [5,1,250,1];
- Step 312 sequentially extracts a pulse wave of the pulse wave dimensional vector fragment including a one-dimensional, four-input of the first CNN tensor [B 1, 1, W 1 , 1] corresponding to the first input three-dimensional CNN tensor [1, W 1 ,1] performs matrix element assignment processing;
- the first CNN input four-dimensional tensor is [ 5,1,250,1], that is, the first CNN input four-dimensional tensor includes five first CNN input three-dimensional tensors [1,250,1] i (the value of i is 1 to 5); then,
- the first CNN input three-dimensional tensor [1,250,1] 1 ⁇ D 1 , whilD 250 ⁇ ,
- the first CNN input three-dimensional tensor [1,250,1] 2 ⁇ D 251 , whilD 500 ⁇ ,
- the first CNN input three-dimensional tensor [1,250,1] 3 ⁇ D 501 , whilD 750 ⁇ ,
- the first CNN input three-dimensional tensor [1,250,1] 4 ⁇ D 751 , whilD 1000 ⁇ ,
- the first CNN input three-dimensional tensor [1,250,1] 5 ⁇ D 1001 , whilD 1250 ⁇ ;
- Step 32 According to the preset threshold of the number of convolutional layers, use the blood pressure CNN model to perform multi-layer convolution pooling calculation on the first CNN input four-dimensional tensor to generate the first CNN output four-dimensional tensor;
- step 321 initializing the value of the first index to 1; initializing the first total to the threshold of the number of convolutional layers; initializing the temporary four-dimensional tensor of the first index as the first CNN input four-dimensional tensor;
- Step 322 Use the first index layer convolution layer of the blood pressure CNN model to perform convolution calculation processing on the first index temporary four-dimensional tensor to generate the first index convolution output data four-dimensional tensor; use the first index of the blood pressure CNN model The layer pooling layer performs pooling calculation processing on the four-dimensional tensor of the first index convolution output data to generate the four-dimensional tensor of the first index pooling output data;
- the blood pressure CNN model includes a multi-layer convolutional layer and a multi-layer pooling layer;
- Step 323 Set the temporary four-dimensional tensor of the first index as the four-dimensional tensor of the first index pooling output data
- Step 324 add 1 to the first index
- Step 325 Determine whether the first index is greater than the first total, if the first index is greater than the first total, go to step 326, if the first index is less than or equal to the first total, go to step 322;
- Step 326 Set the first CNN output four-dimensional tensor as the first index temporary four-dimensional tensor
- the first CNN output four-dimensional tensor is specifically the first CNN output four-dimensional tensor [B 2 ,1,W 2 ,C 2 ]; B 2 is the first CNN output four-dimensional tensor [B 2 ,1,W 2 , C 2 ] the fourth dimension parameter, and B 2 is the total number of segments; W 2 is the second dimension parameter of the first CNN output four-dimensional tensor [B 2 ,1,W 2 ,C 2 ], and W 2 is the preset The threshold of the total number of neurons in the LSTM layer; C 2 is the first dimension parameter of the first CNN output four-dimensional tensor [B 2 ,1,W 2 ,C 2 ], and C 2 is the preset LSTM layer neuron length threshold;
- step 32 is to perform multi-layer convolution pooling calculation on the first input data four-dimensional tensor on the blood pressure CNN to generate the first CNN output four-dimensional tensor;
- the blood pressure CNN model consists of convolutional layers and pooling Layer composition, the general structure is one layer of convolution and one layer of pooling before connecting to the next convolutional layer.
- the final number of layers of the network is determined by the number of convolutional layers, that is, the threshold for the number of convolutional layers; assuming the threshold for the number of convolutional layers If it is 4, then 4 convolution + pooling calculations need to be completed here; the output result of each layer calculation will be used as the input of the next layer calculation.
- the first CNN output four-dimensional tensor is specifically the first CNN output four-dimensional tensor [B 2 ,1,W 2 ,C 2 ]; B 2 is the first CNN output four-dimensional tensor [B 2 ,1,W 2 ,C 2 ], and the value of B 2 is the total number of segments;
- W 2 is the second dimensional parameter of the first CNN output four-dimensional tensor [B 2 ,1,W 2 ,C 2 ], and W 2 Is the preset threshold for the total number of neurons in the LSTM layer;
- C 2 is the first dimension parameter of the first CNN output four-dimensional tensor [B 2 ,1,W 2 ,C 2 ], and C 2 is the preset LSTM layer neurons Length threshold; assuming that the total number of fragments is 5, the threshold of the total number of neurons in the LSTM layer is 5, and the threshold of the length of neurons in the LSTM layer is 64, then the first CNN output four-dimensional tensor is [5,1,5,64].
- Step 33 According to the input parameter format of the blood pressure long and short-term memory LSTM network model, perform tensor reduction processing on the first CNN output four-dimensional tensor to generate the first LSTM input three-dimensional tensor;
- the first LSTM input three-dimensional tensor is specifically the first LSTM input three-dimensional tensor [H 3 ,W 3 ,C 3 ]; H 3 is the first LSTM input three-dimensional tensor [H 3 ,W 3 ,C 3 ]
- the third dimension parameter, and H 3 is the total number of fragments;
- W 3 is the second dimension parameter of the first LSTM input three-dimensional tensor [H 3 , W 3 , C 3 ], and W 3 is W 2 ;
- C 3 is the first The LSTM inputs the first dimension parameter of the three-dimensional tensor [H 3 , W 3 , C 3 ], and C 3 is C 2 .
- the LSTM input three-dimensional tensor is [5,5,64].
- the actual three-dimensional parameters after dimensionality reduction should be equal to the four-dimensional parameters before dimensionality reduction, and the two-dimensional parameters after dimensionality reduction It should be equal to the product of the three-dimensional parameter before the dimensionality reduction multiplied by the two-dimensional parameter. Because the third-dimensional parameter of the first CNN output four-dimensional tensor is actually 1, the two-dimensional parameter does not change after the dimensionality reduction.
- Step 4 Use the LSTM network layer of the blood pressure LSTM network model to perform blood pressure long and short-term memory calculation operations on the first LSTM input three-dimensional tensor to generate the LSTM output three-dimensional tensor;
- the blood pressure LSTM network model includes the LSTM network layer and the fully connected layer;
- Step 41 When the CNN scheme identifier is the first scheme, use the LSTM network layer of the blood pressure LSTM network model to perform blood pressure long and short-term memory calculation operations on the first LSTM input three-dimensional tensor to generate the LSTM output three-dimensional tensor;
- Step 42 When the CNN scheme identifier is the second scheme, use the LSTM network layer of the blood pressure LSTM network model to perform blood pressure long and short-term memory calculation operations on the second LSTM input three-dimensional tensor to generate the LSTM output three-dimensional tensor.
- Step 5 Use the fully connected layer to perform blood pressure regression calculation on the LSTM output three-dimensional tensor to generate the blood pressure prediction three-dimensional tensor [X, Y, 2];
- Each one-dimensional tensor [2] includes two blood pressure prediction data: the reference value of diastolic blood pressure and the reference value of systolic blood pressure; the three-dimensional tensor of blood pressure prediction should be [5,5,2], that is, it includes 25 pairs of blood pressure predictions value.
- Step 6 according to the sequence of multiple one-dimensional pulse wave fragments and multiple one-dimensional sub-segment sequences of pulse wave, sequentially extract the predicted blood pressure data from the blood pressure prediction three-dimensional tensor [X, Y, 2] to generate a blood pressure prediction data set;
- step 61 initialize the blood pressure prediction data set to be empty; set the blood pressure data group; initialize the diastolic blood pressure data of the blood pressure data group to be empty; initialize the systolic blood pressure data of the blood pressure data group to be empty;
- Step 62 sequentially extract the blood pressure prediction one-dimensional vector [2] included in the blood pressure prediction three-dimensional tensor [X, Y, 2] to generate the current one-dimensional vector [2]; set the systolic blood pressure data of the blood pressure data group to the current one-dimensional vector [ 2] sub-segment systolic blood pressure data, set the diastolic blood pressure data of the blood pressure data group to the sub-segment diastolic blood pressure data in the current one-dimensional vector [2]; add the blood pressure data group to the blood pressure prediction data set for data group addition operation; blood pressure
- the predicted three-dimensional tensor [X, Y, 2] includes X*Y blood pressure prediction one-dimensional vectors [2]; the blood pressure prediction one-dimensional vector [2] includes sub-segment systolic blood pressure data and sub-segment diastolic blood pressure data.
- step 6 is to extract the 25 pairs of blood pressure prediction values from [5,5,2] in the blood pressure prediction three-dimensional tensor one by one, and add them to the blood pressure prediction data set in the form of 25 blood pressure data groups.
- the blood pressure prediction data set here is the calculation result of using the fusion network of CNN and LSTM network to predict the blood pressure of PPG signal data.
- FIG. 2 is a schematic diagram of a method for predicting blood pressure according to Embodiment 2 of the present invention. The method mainly includes the following steps:
- Step 101 Perform pulse wave conversion and sampling processing on the PPG signal data of the photoplethysmography method to generate a pulse wave one-dimensional vector; divide the pulse wave one-dimensional vector into multiple pulse wave one-dimensional fragments and obtain the total number of fragments; The one-dimensional segment is divided into multiple one-dimensional sub-segments of pulse wave and the total number of sub-segments is obtained;
- Step 1011 call the PPG signal acquisition device, collect the light intensity signal of the preset light source signal on the local skin surface of the organism, and generate a PPG signal data whose length is the signal acquisition time threshold; perform pulse wave data on the PPG signal data The conversion operation generates pulse wave signal data; the characteristic data of the pulse wave signal data is sampled according to the characteristic sampling frequency threshold to generate a one-dimensional pulse wave vector; the preset light source signal includes at least one of the red light source signal, the infrared light source signal and the green light source signal;
- the pulse wave one-dimensional vector is specifically the pulse wave one-dimensional vector [A];
- the pulse wave one-dimensional vector [A] specifically becomes the pulse wave one-dimensional vector [1250], which is a one-dimensional vector including 1250 independent pulse wave data;
- Step 1012 divide the pulse wave one-dimensional vector into data segments according to the segment length threshold to generate multiple pulse wave one-dimensional segments, and use the total number of pulse wave one-dimensional segments included in the pulse wave one-dimensional vector as the total number of segments;
- Step 1013 divide the pulse wave one-dimensional segment into data sub-segments according to the sub-segment length threshold to generate multiple pulse wave one-dimensional sub-segments, and use the total number of pulse wave one-dimensional sub-segments included in the pulse wave one-dimensional segment as the total number of sub-segments .
- the blood pressure CNN will be used to perform feature calculations on the data in the pulse wave one-dimensional vector
- the pulse wave one-dimensional vector is segmented according to the maximum data length input by the blood pressure CNN, where ,
- the fragment length threshold is the maximum data length input by the blood pressure CNN.
- Step 102 Obtain the value of the CNN scheme identifier of the convolutional neural network as the second scheme
- the CNN scheme identifier includes two identifiers: the first scheme and the second scheme.
- the CNN scheme identifier is used to distinguish between the two processing methods of CNN: when the CNN scheme identifier is the second scheme, the one-dimensional pulse wave vector is divided into multiple segments and sub-segments, and each A sub-segment uses the segmented vector convolution pooling calculation in the blood pressure CNN model, and the output result is used as the input of the next long and short-term memory calculation.
- Step 103 When the CNN scheme identifier is the second scheme, use the blood pressure CNN model to perform a segmented vector convolution pooling calculation operation on the pulse wave one-dimensional vector, and perform the calculation of the calculation result according to the input parameter format of the blood pressure LSTM network model. Dimension reduction processing generates a second LSTM input three-dimensional tensor;
- step 1031 when the CNN scheme identifier is the second scheme, generate the total number of tensors according to the product of the total number of segments multiplied by the total number of sub-segments;
- Step 1032 Perform a second blood pressure CNN input parameter setting operation based on the total number of tensors and the one-dimensional pulse wave vector to generate a second CNN input four-dimensional tensor group;
- the second CNN input four-dimensional tensor group includes the total number of tensors and the second CNN input four-dimensional tensor;
- step 10321 sort all pulse wave one-dimensional sub-segments of the pulse wave one-dimensional vector to generate a full sequence of sub-segments;
- the full sequence of sub-segments includes the total number of tensors and one-dimensional sub-segments of pulse wave;
- Step 10322 set the second CNN input four-dimensional tensor group; set the second CNN input four-dimensional tensor specifically as the second CNN input four-dimensional tensor [1,1,W 4 ,1];
- the second CNN input four-dimensional tensor group includes the total number of tensors.
- the second CNN input four-dimensional tensor group includes a total of 25 second CNN input four-dimensional tensors;
- Step 10323 sequentially extract the pulse wave one-dimensional sub-segments in the full sequence of sub-segments, and input the corresponding second CNN in the four-dimensional tensor group to the second CNN to input the four-dimensional tensor [1,1,W 4 ,1] for matrix elements Assignment processing;
- Step 1033 use the blood pressure CNN model to perform multi-layer convolution pooling calculation on all the second CNN input four-dimensional tensors in the second CNN input four-dimensional tensor group to generate the second CNN output Four-dimensional tensor group;
- the second CNN output four-dimensional tensor group includes the total number of tensors; the second CNN output four-dimensional tensor group; the second CNN output four-dimensional tensor group specifically includes the total number of tensors; the second CNN outputs a four-dimensional tensor; the second CNN outputs a four-dimensional tensor
- the tensor is specifically the second CNN output four-dimensional tensor [1,1,1,C 5 ]; C 5 is the first dimensional parameter of the second CNN output four-dimensional tensor [1,1,1,C 5 ];
- the 25 second CNN input four-dimensional tensors in the second CNN input four-dimensional tensor group are respectively subjected to convolution calculations to generate the second CNN output four-dimensional tensor group, and the second CNN output four-dimensional tensor group and the input Correspondingly, it also includes 25 second CNN output four-dimensional tensors [1,1,1,C 5 ]; assuming that the length of the LSTM neural unit is 64, the second CNN output four-dimensional tensor is [1,1,1,64 ];
- Step 1034 Perform a four-dimensional tensor merging operation on all the second CNN output four-dimensional tensors in the second CNN output four-dimensional tensor group to generate a third CNN output four-dimensional tensor;
- step 10341 setting the third CNN to output a four-dimensional tensor, specifically the third CNN to output a four-dimensional tensor [B 6 ,1,1,C 6 ];
- B 6 is the fourth dimension parameter of the third CNN output four-dimensional tensor [B 6 ,1,1,C 6 ], and B 6 is the total number of tensors;
- C 6 is the third CNN output four-dimensional tensor [B 6 ,1,1,C 6 ] is the first dimension parameter, and C 6 is C 5 ;
- Step 10342 in the second CNN output four-dimensional tensor group, sequentially extract the matrix element sequence of the second CNN output four-dimensional tensor [1,1,1,C 5 ], and output the four-dimensional tensor [B 6 , 1,1,C 6 ] Perform matrix element assignment processing;
- the input of the LSTM network model is to be performed later, it is necessary to merge multiple second CNN output four-dimensional tensors of the second CNN output four-dimensional tensor group to generate a four-dimensional tensor;
- the second CNN outputs a four-dimensional tensor
- the group includes 25 second CNN output four-dimensional tensors [1,1,1,64], then the third CNN output four-dimensional tensor after the merge is completed is [25,1,1,64]; the corresponding second CNN output
- the specific data sequence of each tensor in the four-dimensional tensor group also needs to output the four-dimensional tensor to the third CNN for sequential addition;
- Step 1035 according to the input parameter format of the blood pressure LSTM network model, perform tensor reduction processing on the third CNN output four-dimensional tensor to generate a second LSTM input three-dimensional tensor;
- the second LSTM input three-dimensional tensor is specifically the LSTM input three-dimensional tensor [H 7 , W 7 , C 7 ]; H 7 is the third LSTM input three-dimensional tensor [H 7 , W 7 , C 7 ] Dimension parameter, and the value of H 3 is the total number of fragments; W 7 is the second dimension parameter of the second LSTM input three-dimensional tensor [H 7 , W 7 , C 7 ], and W 7 is the quotient of B 6 divided by H 3 ; C 7 is the first dimension parameter of the second LSTM input three-dimensional tensor [H 7 , W 7 , C 7 ], and C 7 is C 6 .
- the input parameter structure of the LSTM network layer is in the form of a three-dimensional tensor, so this step needs to perform tensor dimensionality reduction processing on the third CNN output four-dimensional tensor generated by the merge.
- the principle of this dimensionality reduction is to reduce the dimensionality
- the third dimension parameter of the three-dimensional tensor is set to the total number of fragments, the product of the third and second dimension parameters of the tensor after dimensionality reduction is equal to the fourth dimension parameter before dimensionality reduction, and the value of the first dimension parameter remains unchanged after dimensionality reduction;
- the third CNN output four-dimensional tensor is [25,1,1,64]
- the second LSTM input three-dimensional tensor after dimensionality reduction is [5,5,64]; it can be seen that the actual tensor data value sequence No increase or decrease occurs, only the shape of the tensor is transformed.
- Step 104 Use the LSTM network layer of the blood pressure LSTM network model to perform blood pressure long and short-term memory calculation operations on the second LSTM input three-dimensional tensor to generate the LSTM output three-dimensional tensor;
- the blood pressure LSTM network model includes the LSTM network layer and the fully connected layer;
- Step 1041 When the CNN scheme identifier is the first scheme, use the LSTM network layer of the blood pressure LSTM network model to perform blood pressure long and short-term memory calculation operations on the first LSTM input three-dimensional tensor to generate the LSTM output three-dimensional tensor;
- Step 1042 When the CNN scheme identifier is the second scheme, use the LSTM network layer of the blood pressure LSTM network model to perform blood pressure long and short-term memory calculation operations on the second LSTM input three-dimensional tensor to generate the LSTM output three-dimensional tensor.
- Step 105 Use the fully connected layer to perform a blood pressure regression calculation operation on the LSTM output three-dimensional tensor to generate a blood pressure prediction three-dimensional tensor [X, Y, 2];
- each one-dimensional tensor [2] includes two blood pressure prediction data: the reference value of diastolic blood pressure and the reference value of systolic blood pressure; Predicted blood pressure.
- Step 106 according to the sequence of multiple one-dimensional pulse wave fragments and multiple one-dimensional sub-segment sequences of pulse wave, sequentially extract the predicted blood pressure data from the blood pressure prediction three-dimensional tensor [X, Y, 2], and generate a blood pressure prediction data set;
- step 1061 initialize the blood pressure prediction data set to be empty; set the blood pressure data group; initialize the diastolic blood pressure data of the blood pressure data group to be empty; initialize the systolic blood pressure data of the blood pressure data group to be empty;
- Step 1062 successively extract the blood pressure prediction one-dimensional vector [2] included in the blood pressure prediction three-dimensional tensor [X, Y, 2] to generate the current one-dimensional vector [2]; set the systolic blood pressure data of the blood pressure data group to the current one-dimensional vector [ 2] sub-segment systolic blood pressure data, set the diastolic blood pressure data of the blood pressure data group to the sub-segment diastolic blood pressure data in the current one-dimensional vector [2]; add the blood pressure data group to the blood pressure prediction data set for data group addition operation; blood pressure
- the predicted three-dimensional tensor [X, Y, 2] includes X*Y blood pressure prediction one-dimensional vectors [2]; the blood pressure prediction one-dimensional vector [2] includes sub-segment systolic blood pressure data and sub-segment diastolic blood pressure data.
- the blood pressure prediction data set here is the calculation result of using the fusion network of CNN and LSTM network to predict the blood pressure of PPG signal data.
- FIG. 3 is a schematic diagram of a device structure of a device for predicting blood pressure according to Embodiment 3 of the present invention.
- the device includes a processor and a memory.
- the memory can be connected to the processor through a bus.
- the memory may be a non-volatile memory, such as a hard disk drive and a flash memory, and a software program and a device driver program are stored in the memory.
- the software program can execute various functions of the foregoing method provided by the embodiments of the present invention; the device driver may be a network and interface driver.
- the processor is used to execute a software program, and when the software program is executed, the method provided in the embodiment of the present invention can be implemented.
- the embodiment of the present invention also provides a computer-readable storage medium.
- a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the method provided in the embodiment of the present invention can be implemented.
- the embodiment of the present invention also provides a computer program product containing instructions.
- the processor is caused to execute the above method.
- the method and device for predicting blood pressure provided by the embodiment of the present invention first convert the collected PPG signal data into pulse wave signal data, and then use a fusion network composed of blood pressure CNN and LSTM network to perform feature extraction operations on the pulse wave signal data Generate blood pressure characteristic data and perform regression calculation on the blood pressure characteristic data to predict the blood pressure data (diastolic blood pressure, systolic blood pressure) of the tester.
- a fusion network composed of blood pressure CNN and LSTM network to perform feature extraction operations on the pulse wave signal data
- the steps of the method or algorithm described in combination with the embodiments disclosed herein can be implemented by hardware, a software module executed by a processor, or a combination of the two.
- the software module can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or all areas in the technical field. Any other known storage media.
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Public Health (AREA)
- Animal Behavior & Ethology (AREA)
- Veterinary Medicine (AREA)
- Physiology (AREA)
- Biophysics (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (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)
Abstract
L'invention concerne un procédé et un dispositif pour la prédiction de la tension artérielle. Le procédé comprend les étapes consistant à : convertir un signal PPG pour générer un vecteur d'onde de pouls unidimensionnelle ; diviser le vecteur d'onde de pouls unidimensionnelle en une pluralité de segments d'onde de pouls unidimensionnelle et obtenir le nombre total de segments ; diviser les segments d'onde de pouls unidimensionnelle en une pluralité de sous-segments d'onde de pouls unidimensionnelle et obtenir le nombre total de sous-segments (1, 101) ; obtenir un identifiant de schéma de réseau de neurones convolutifs (2, 102) ; lorsque l'identifiant du schéma du réseau de neurones convolutifs correspond à un premier schéma, effectuer un calcul de regroupement convolutif de vecteurs complets sur le vecteur d'onde de pouls unidimensionnelle et réduire la taille du résultat pour générer un premier tenseur tridimensionnel de mémoire à court terme longue (LSTM) d'entrée (3) ; lorsque l'identifiant du schéma du réseau de neurones convolutifs correspond à un second schéma, effectuer un calcul de regroupement convolutif de vecteurs segmentés sur le vecteur d'onde de pouls unidimensionnelle et réduire la taille du résultat pour générer un second tenseur tridimensionnel LSTM d'entrée (103) ; effectuer un calcul LSTM de tension artérielle sur le premier ou le second tenseur tridimensionnel LSTM d'entrée pour générer un tenseur tridimensionnel LSTM de sortie (4, 104) ; effectuer un calcul de régression pour la tension artérielle sur le tenseur tridimensionnel LSTM de sortie pour générer un tenseur tridimensionnel de prédiction de tension artérielle [X, Y, 2] (5, 105) ; et générer un ensemble de données de prédiction de tension artérielle (6, 106).
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010110287.9 | 2020-02-21 | ||
CN202010110287.9A CN111248882B (zh) | 2020-02-21 | 2020-02-21 | 一种预测血压的方法和装置 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2021164346A1 true WO2021164346A1 (fr) | 2021-08-26 |
Family
ID=70941767
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2020/129631 WO2021164346A1 (fr) | 2020-02-21 | 2020-11-18 | Procédé et dispositif pour la prédiction de la tension artérielle |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN111248882B (fr) |
WO (1) | WO2021164346A1 (fr) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114557691A (zh) * | 2022-04-29 | 2022-05-31 | 广东工业大学 | 基于多波长的ppg信号的无创血脂检测方法及系统 |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111248882B (zh) * | 2020-02-21 | 2022-07-29 | 乐普(北京)医疗器械股份有限公司 | 一种预测血压的方法和装置 |
CN111728602A (zh) * | 2020-08-21 | 2020-10-02 | 之江实验室 | 基于ppg的无接触血压测量装置 |
CN112022125A (zh) * | 2020-09-28 | 2020-12-04 | 无锡博智芯科技有限公司 | 一种基于CNN-BiGRU模型和PPG的智能血压预测方法 |
CN112022126A (zh) * | 2020-09-28 | 2020-12-04 | 无锡博智芯科技有限公司 | 一种基于CNN-BiLSTM模型和PPG的智能血压预测方法 |
CN116982950A (zh) * | 2023-06-26 | 2023-11-03 | 深圳先进技术研究院 | 一种无袖带血压监测系统 |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106782602A (zh) * | 2016-12-01 | 2017-05-31 | 南京邮电大学 | 基于长短时间记忆网络和卷积神经网络的语音情感识别方法 |
CN107961007A (zh) * | 2018-01-05 | 2018-04-27 | 重庆邮电大学 | 一种结合卷积神经网络和长短时记忆网络的脑电识别方法 |
US20190104951A1 (en) * | 2013-12-12 | 2019-04-11 | Alivecor, Inc. | Continuous monitoring of a user's health with a mobile device |
CN109886347A (zh) * | 2019-02-28 | 2019-06-14 | 泉州师范学院 | 基于rbf和lstm模型的多因素网络的血压预测方法 |
CN109965862A (zh) * | 2019-04-16 | 2019-07-05 | 重庆大学 | 一种无袖带式长时连续血压无创监测方法 |
CN110755059A (zh) * | 2019-10-11 | 2020-02-07 | 中国科学院深圳先进技术研究院 | 一种血压波形监测方法及装置 |
CN111248882A (zh) * | 2020-02-21 | 2020-06-09 | 乐普(北京)医疗器械股份有限公司 | 一种预测血压的方法和装置 |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106037694B (zh) * | 2016-05-13 | 2019-11-05 | 吉林大学 | 一种基于脉搏波的连续血压测量装置 |
CN108926334A (zh) * | 2017-05-26 | 2018-12-04 | 深圳市玉成创新科技有限公司 | 基于脉搏波的血压获取方法及其系统和装置 |
CN108498089B (zh) * | 2018-05-08 | 2022-03-25 | 北京邮电大学 | 一种基于深度神经网络的无创连续血压测量方法 |
KR102098561B1 (ko) * | 2018-07-10 | 2020-04-08 | 재단법인 아산사회복지재단 | 동맥압 파형을 이용한 심박출량 획득 방법 및 그 프로그램 |
CN110680278B (zh) * | 2019-09-10 | 2022-07-19 | 广州视源电子科技股份有限公司 | 基于卷积神经网络的心电信号识别装置 |
-
2020
- 2020-02-21 CN CN202010110287.9A patent/CN111248882B/zh active Active
- 2020-11-18 WO PCT/CN2020/129631 patent/WO2021164346A1/fr active Application Filing
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190104951A1 (en) * | 2013-12-12 | 2019-04-11 | Alivecor, Inc. | Continuous monitoring of a user's health with a mobile device |
CN106782602A (zh) * | 2016-12-01 | 2017-05-31 | 南京邮电大学 | 基于长短时间记忆网络和卷积神经网络的语音情感识别方法 |
CN107961007A (zh) * | 2018-01-05 | 2018-04-27 | 重庆邮电大学 | 一种结合卷积神经网络和长短时记忆网络的脑电识别方法 |
CN109886347A (zh) * | 2019-02-28 | 2019-06-14 | 泉州师范学院 | 基于rbf和lstm模型的多因素网络的血压预测方法 |
CN109965862A (zh) * | 2019-04-16 | 2019-07-05 | 重庆大学 | 一种无袖带式长时连续血压无创监测方法 |
CN110755059A (zh) * | 2019-10-11 | 2020-02-07 | 中国科学院深圳先进技术研究院 | 一种血压波形监测方法及装置 |
CN111248882A (zh) * | 2020-02-21 | 2020-06-09 | 乐普(北京)医疗器械股份有限公司 | 一种预测血压的方法和装置 |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114557691A (zh) * | 2022-04-29 | 2022-05-31 | 广东工业大学 | 基于多波长的ppg信号的无创血脂检测方法及系统 |
CN114557691B (zh) * | 2022-04-29 | 2022-08-02 | 广东工业大学 | 基于多波长的ppg信号的无创血脂检测方法及系统 |
Also Published As
Publication number | Publication date |
---|---|
CN111248882A (zh) | 2020-06-09 |
CN111248882B (zh) | 2022-07-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2021164346A1 (fr) | Procédé et dispositif pour la prédiction de la tension artérielle | |
WO2021164345A1 (fr) | Procédé et dispositif de prédiction de pression artérielle | |
CN110619322A (zh) | 一种基于多流态卷积循环神经网络的多导联心电异常信号识别方法及系统 | |
CN111657926B (zh) | 一种基于多导联信息融合的心律失常分类方法 | |
CN111772619B (zh) | 一种基于深度学习的心搏识别方法、终端设备及存储介质 | |
WO2021184801A1 (fr) | Procédé et appareil de prédiction de pression artérielle sur la base d'un signal de synchronisation | |
CN111134662B (zh) | 一种基于迁移学习和置信度选择的心电异常信号识别方法及装置 | |
CN111297349A (zh) | 一种基于机器学习的心律分类系统 | |
CN111626114B (zh) | 基于卷积神经网络的心电信号心律失常分类系统 | |
CN111759345B (zh) | 基于卷积神经网络的心脏瓣膜异常分析方法、系统和装置 | |
KR20210085867A (ko) | 사용자의 혈압을 추정하기 위한 장치 및 방법 | |
WO2021184802A1 (fr) | Procédé et appareil de prédiction de classification de pression artérielle | |
CN111291727B (zh) | 一种光体积变化描记图法信号质量检测方法和装置 | |
WO2021164347A1 (fr) | Procédé et appareil de prédiction de la tension artérielle | |
CN111358451B (zh) | 一种血压预测方法和装置 | |
CN112070067A (zh) | 一种光体积描计信号的散点图分类方法和装置 | |
CN111528832A (zh) | 一种心律失常分类方法及其有效性验证方法 | |
CN117137451A (zh) | 基于远程脉搏波信号的非接触式应激检测方法和系统 | |
CN116269426A (zh) | 一种十二导联ecg辅助的心脏疾病多模态融合筛查方法 | |
CN116451129A (zh) | 一种脉象分类识别方法及系统 | |
Sangeetha et al. | A CNN based similarity learning for cardiac arrhythmia prediction | |
CN117752326B (zh) | 一种基于bcg信号的递归图和波形分布的心脏状态判断方法 | |
CN116584908A (zh) | 一种基于领域自适应的无创连续血压测量方法 | |
KR20230102956A (ko) | 광용적맥파의 품질 평가 방법 및 장치 | |
Soumiaa et al. | The Use of the Multi-Scale Discrete Wavelet Transform and Deep Neural Networks on ECGs for the Diagnosis of 8 Cardio-Vascular Diseases |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20919572 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 20919572 Country of ref document: EP Kind code of ref document: A1 |