Vessel information prediction method and device, equipment and storage medium

Abstract

The application provides a blood vessel information prediction method and device, equipment and a storage medium, and relates to the technical field of biomedicine. The method comprises the following steps: obtaining to-be-predicted data; inputting the to-be-predicted data into a prediction model to obtain blood vessel information of a target blood vessel; wherein the prediction model is obtained through the following steps: obtaining a training data set and a corresponding physical model, wherein the training data set comprises historical physiological signals and corresponding blood pressure waveform data or blood flow rate waveform data; determining a loss function of a to-be-trained neural network based on the physical model; and performing neural network training according to the loss function and the training data set to obtain the prediction model. The implementation of the application can predict blood vessel information by using physiological signals, improve the acquisition efficiency of blood pressure and blood flow rate related data, facilitate long-time acquisition of the data, and promote the promotion of cardiovascular disease diagnosis and prevention in daily life.

Description

Technical Field

[0001] This application relates to the field of biomedical technology, and more specifically, to a method, apparatus, device, and storage medium for predicting vascular information. Background Technology

[0002] Cardiovascular disease is one of the leading causes of death worldwide. Human hemodynamics contains a wealth of useful information, often used in the diagnosis and prevention of cardiovascular disease. Hemodynamics in human arteries is typically modeled using two variables: arterial blood pressure and blood flow rate.

[0003] Current technologies often employ invasive methods to obtain blood pressure and blood flow rate, which not only carries certain risks but also places high demands on the clinical skills of doctors and the technology of medical equipment. Moreover, these methods are limited to specific locations (such as hospitals) and require specialized equipment, making the testing process cumbersome, inefficient, and unable to obtain blood pressure and blood flow rate data over extended periods, thus hindering the advancement of cardiovascular disease diagnosis and prevention in daily life. Summary of the Invention

[0004] This application provides a method, apparatus, device, and storage medium for predicting vascular information, which can solve the problem that existing blood pressure and blood flow data can only be obtained in specific locations and cannot be obtained for extended periods. To achieve this objective, this application provides the following solutions.

[0005] According to one aspect of the embodiments of this application, a method for predicting blood vessel information is provided, the method comprising:

[0006] Acquire the data to be predicted, which includes physiological signals and the spatiotemporal coordinates of the target blood vessel. The physiological signals include at least one of photoplethysmography (PPG) signals and electrocardiogram (ECG) signals.

[0007] The data to be predicted is input into the prediction model to obtain the vascular information of the target blood vessel, including blood pressure waveform and blood flow rate waveform.

[0008] The prediction model is obtained through the following steps:

[0009] Obtain the training dataset and its corresponding physical model. The training dataset includes historical physiological signals and corresponding blood pressure waveform data or blood flow rate waveform data.

[0010] The loss function of the neural network to be trained is determined based on the physical model.

[0011] The prediction model is obtained by training a neural network based on the loss function and the training dataset.

[0012] In one possible implementation, the prediction model includes a backbone network and branch networks. The step of inputting the data to be predicted into the prediction model to obtain the vascular information of the target blood vessel includes:

[0013] The physiological signal is input into the branch network, and the spatiotemporal coordinates corresponding to the predicted location in the target blood vessel are input into the trunk network. The blood vessel information corresponding to the spatiotemporal coordinates is obtained based on the output information of the branch network and the trunk network.

[0014] In one possible implementation, the backbone network includes a fully connected neural network with multiple fully connected layers, and the branch network includes a residual neural network with multiple residual units. Obtaining the blood vessel information corresponding to the spatiotemporal coordinates based on the output information of the branch network and the backbone network includes:

[0015] Obtain the first prediction result output by the branch network and the second prediction result output by the backbone network, and obtain the blood pressure waveform and blood flow rate waveform at the location to be predicted based on the first prediction result and the second prediction result.

[0016] In one possible implementation, obtaining the training dataset and its corresponding physical model includes:

[0017] The physical model corresponding to the training dataset is determined based on the sample groups in the training dataset. The physical model includes a hemodynamic model and the initial and boundary conditions related to the hemodynamic model.

[0018] In one possible implementation, the initial conditions include conditions for periodic changes in the blood pressure or blood flow rate, wherein the period length corresponding to the periodic changes is data obtained from the historical physiological signals.

[0019] The boundary conditions include at least one of the following: Dirichlet boundary conditions, boundary conditions indicated by the Wedxel model, Neumann boundary conditions, Robin boundary conditions, and reflection boundary conditions.

[0020] In one possible implementation, determining the loss function of the neural network to be trained based on the physical model further includes:

[0021] Obtain the output of the neural network to be trained, and obtain the residual term of the hemodynamic model based on the output and the hemodynamic model;

[0022] The empirical loss function of the neural network is corrected by the residual term.

[0023] In one possible implementation, the step of training the neural network based on the loss function and the training dataset to obtain the prediction model includes:

[0024] The neural network is trained using the training dataset and iterated using a preset iterative method. The prediction model is obtained based on the iteration results. The preset iterative method includes the stochastic gradient descent algorithm.

[0025] This application provides a vascular information prediction device, the device comprising:

[0026] The data to be predicted module is used to acquire the data to be predicted, which includes physiological signals and the spatiotemporal coordinates of the target blood vessel. The physiological signals include at least one of photoplethysmography signals and electrocardiogram signals.

[0027] The vascular information prediction module is used to input the data to be predicted into the prediction model to obtain the vascular information of the target vascular vessel, including blood pressure waveform and blood flow rate waveform.

[0028] The prediction model is obtained through the following steps:

[0029] Obtain the training dataset and its corresponding physical model. The training dataset includes historical physiological signals and corresponding blood pressure waveform data or blood flow rate waveform data.

[0030] The loss function of the neural network to be trained is determined based on the physical model.

[0031] The prediction model is obtained by training a neural network based on the loss function and the training dataset.

[0032] This application provides an electronic device, including a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the steps of the method described above.

[0033] According to another aspect of the embodiments of this application, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0034] The beneficial effects of the technical solutions provided in this application are:

[0035] In the vascular information prediction method provided in this application, during offline training, historical physiological signals and corresponding blood pressure waveform data or blood flow velocity waveform data are used as the training dataset for the prediction model. The neural network is trained using this training dataset, and the loss function of the neural network is adjusted using the physical model corresponding to the training dataset to obtain a prediction model for predicting vascular information. When predicting vascular information, data to be predicted, including physiological signals and the spatiotemporal coordinates of the target blood vessel, is acquired. This data is input into the prediction model, and the blood pressure waveform and blood flow velocity waveform of the target blood vessel are obtained from the output of the prediction model. Therefore, the embodiments of this application can predict vascular information using physiological signals, improve the efficiency of acquiring blood pressure and blood flow velocity related data, and facilitate long-term acquisition of this data, thus promoting the advancement of cardiovascular disease diagnosis and prevention in daily life. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0037] Figure 1 A flowchart of the vascular information prediction method provided in the embodiments of this application;

[0038] Figure 2 A schematic diagram of an embodiment of the vascular information prediction method provided in this application;

[0039] Figure 3 This is a schematic diagram of an embodiment of the backbone network and branch network in the vascular information prediction method of this application;

[0040] Figure 4 This is a schematic diagram of the target blood vessel in the blood vessel information prediction method provided in the embodiments of this application;

[0041] Figure 5 This is a schematic diagram of the backbone network structure in the vascular information prediction method provided in the embodiments of this application;

[0042] Figure 6 This is a schematic diagram of the branch network structure in the vascular information prediction method provided in the embodiments of this application;

[0043] Figure 7 This is a schematic diagram of the structure of the residual unit in the vascular information prediction method provided in the embodiments of this application;

[0044] Figure 8 This is a flowchart of the prediction model training process in the vascular information prediction method provided in the embodiments of this application;

[0045] Figure 9 A waveform diagram of the electrocardiogram signal provided in the vascular information prediction method of this application embodiment;

[0046] Figure 10 The waveform of the photoplethysmography signal in the vascular information prediction method provided in the embodiments of this application;

[0047] Figure 11 The blood pressure waveform at the outflow end of the target blood vessel in the vascular information prediction method provided in this application embodiment;

[0048] Figure 12 A blood pressure waveform at the midpoint of the target blood vessel in the vascular information prediction method provided in this application embodiment;

[0049] Figure 13 The blood pressure waveform at the inflow end of the target blood vessel in the vascular information prediction method provided in this application embodiment;

[0050] Figure 14 The waveform of blood flow velocity at the outflow end of the target blood vessel in the vascular information prediction method provided in the embodiments of this application is shown.

[0051] Figure 15 The waveform of blood flow velocity at the midpoint of the target blood vessel in the vascular information prediction method provided in the embodiments of this application;

[0052] Figure 16 The waveform of blood flow rate at the inflow end of the target blood vessel in the vascular information prediction method provided in the embodiments of this application;

[0053] Figure 17 This is a structural diagram of the vascular information prediction device provided in the embodiments of this application;

[0054] Figure 18 This is a structural diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0055] The embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the embodiments described below with reference to the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions of the embodiments of this application.

[0056] Those skilled in the art will understand that, unless otherwise stated, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the terms “comprising” and “including” as used in embodiments of this application mean that the corresponding feature can be implemented as the presented feature, information, data, step, operation, element, and / or component, but do not exclude implementation as other features, information, data, step, operation, element, component, and / or combinations thereof supported by the art. It should be understood that when we say that an element is “connected” or “coupled” to another element, the one element can be directly connected or coupled to the other element, or it can mean that the one element and the other element establish a connection relationship through an intermediate element. Furthermore, “connected” or “coupled” as used herein can include wireless connection or wireless coupling. The term “and / or” as used herein indicates at least one of the items defined by the term; for example, “A and / or B” indicates implementation as “A,” or implementation as “A,” or implementation as “A and B.”

[0057] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0058] The technical solutions of the embodiments of the present invention and the technical effects produced by the technical solutions of the present invention will be described below through several exemplary embodiments. It should be noted that the following embodiments can be referred to, learned from, or combined with each other, and the same terms, similar features, and similar implementation steps in different embodiments will not be described again.

[0059] The vascular information prediction method, apparatus, device, and storage medium provided in this application are intended to solve at least one technical problem existing in the prior art.

[0060] This application provides a method for predicting vascular information. Devices using this method can be mobile phones, computers, servers, cloud platforms, and other devices capable of acquiring data to be predicted and loading prediction models. Figures 1-8 As shown, the vascular information prediction method includes steps S101-S102.

[0061] S101: Obtain the data to be predicted.

[0062] Optionally, the data to be predicted includes physiological signals and the spatiotemporal coordinates of the target blood vessel. The physiological signals include at least one of photoplethysmography (PPG) and electrocardiogram (ECG).

[0063] In one embodiment, the physiological signal is the physiological signal of the object whose vascular information needs to be predicted. This can be acquired using an electrocardiogram (ECG) sensor and a photoplethysmography (PPG) signal acquired using a photoelectric sensor. Furthermore, the physiological signal can be preprocessed before being input into the prediction model. For example, noise and redundant information in the physiological signal can be removed using a filter (which could be a bandpass filter).

[0064] Optionally, the spatiotemporal coordinates are a set of spatiotemporal coordinates of the location to be predicted in the target blood vessel. Each spatiotemporal coordinate includes the position coordinates of the location to be predicted in the target blood vessel and the time coordinates related to that position coordinates. There can be one or more locations to be predicted. When there are multiple locations to be predicted, the blood pressure waveform and blood flow velocity waveform at different locations in the target blood vessel are obtained through the data output by the prediction model.

[0065] S102: Input the data to be predicted into the prediction model to obtain the vascular information of the target blood vessel.

[0066] Optionally, the vascular information includes blood pressure waveforms and blood flow velocity waveforms. Specifically, the vascular information includes blood pressure waveforms and blood flow velocity waveforms corresponding to spatiotemporal coordinates.

[0067] Optionally, the target blood vessel is an artery, and the blood pressure waveform is an arterial blood pressure waveform. In other embodiments, the target blood vessel can also be a vein, and the blood pressure waveform can also be set to a venous blood pressure waveform as needed.

[0068] Optionally, after acquiring the vascular information of the target blood vessel, the vascular information can also be output to an output device, which can be a display screen, a wireless information transmission module, or other devices capable of displaying or transmitting the vascular information.

[0069] In one embodiment, the spatiotemporal coordinates are (z,t), which contain two variables: a spatial variable z∈[0,L] and a temporal variable t∈[0,T], where T is the time length corresponding to the physiological signal. For example, Figure 4As shown, the domain [0, L] of the spatial variable z is a segment of an artery, where L is the length of the artery, and the value of the variable z represents its position within this segment. For example, z = 0 represents the inflow end of the vessel; z = 0.5L represents the midpoint of the vessel; and z = L represents the outflow end of the vessel. To obtain the predicted waveforms of arterial blood pressure and blood flow velocity at a specified location in the artery, a set of spatiotemporal coordinates related to that location is obtained and input into the prediction model. This set of spatiotemporal coordinates can be: (z0, 0), (z0, Δt), (z0, 2Δt), ..., (z0, NΔt). z0 is the coordinate of that location within the artery, Δt is a very small time step, and N = T / Δt, where N is a positive integer. The values ​​of Δt and N can be set according to actual needs. Thus, the prediction model outputs the blood pressure and blood flow velocity data corresponding to each spatiotemporal coordinate, and based on these data, the predicted waveforms of arterial blood pressure and blood flow velocity at the location z = z0 ∈ [0, L] can be obtained.

[0070] Optionally, the prediction model includes a backbone network and branch networks. The data to be predicted is input into the prediction model to obtain vascular information of the target blood vessel. This includes: inputting physiological signals into the branch network and inputting the spatiotemporal coordinates corresponding to the predicted location in the target blood vessel into the backbone network; and obtaining the vascular information corresponding to the spatiotemporal coordinates based on the output information of the branch network and the backbone network. Features related to blood pressure and blood flow velocity are extracted through the backbone network and branch networks, and blood pressure waveforms and blood flow velocity waveforms are obtained based on these features.

[0071] Optionally, the prediction model adopts the structure of DeepONet, the trunk network includes a fully connected neural network with multiple fully connected layers, and the branch network includes a residual neural network (ResNet) with multiple residual units. The vascular information corresponding to the spatiotemporal coordinates is obtained based on the output information of the branch network and the trunk network, including: obtaining the first prediction result output by the branch network and the second prediction result output by the trunk network, and obtaining the blood pressure waveform and blood flow velocity waveform at the location to be predicted based on the first prediction result and the second prediction result.

[0072] In one embodiment, such as Figure 3 As shown, the input spatiotemporal coordinates y=(z,t)∈[0,L]×[0,T] The first prediction result includes two feature vectors [b 11 ,b 21 ,b 31 ...b p1 ]、[b 12 ,b 22 ,b 32 ...b p2The second prediction result includes two feature vectors [t] 11 ,t 21 ,t 31 ...t p1 ]、[t 12 ,t 22 ,t 32 ...t p2 The first preset result and the second prediction result are input into the preset formula to obtain the blood pressure P(z,t) and blood flow velocity Q(z,t) at different times. The preset formula is as follows:

[0073]

[0074] The range of z is a closed interval [0, L], and the range of t is a closed interval [0, T]. L is a pre-defined parameter representing the total length of the target blood vessel, such as 20 cm. T is the total duration of the input physiological signal, such as 4 seconds. P(z, t) is the blood pressure value at position z and time t within the blood vessel. Q(z, t) is the blood flow velocity value at position z and time t within the blood vessel. p is the dimension of the vectors output by the branch network and the backbone network; it is a hyperparameter whose specific value can be set according to actual needs.

[0075] Specifically, such as Figure 5 As shown, the backbone network can be a fully connected neural network. The input of a fully connected network is a set of spatiotemporal coordinates y = (z,t) ∈ [0,L] × [0,T], and the output is two feature vectors [t...]. 11 ,t 21 ,t 31 ...t p1 ]、[t 12 ,t 22 ,t 32 ...t p2 This fully connected neural network consists of multiple fully connected layers, each of which can be represented as a linear transformation of the spatiotemporal coordinates of the input. The output dimension of the previous fully connected layer should be equal to the input dimension of the next fully connected layer. In each fully connected layer, every node is connected to all nodes of the previous layer, used to synthesize the features extracted from the previous layer. The input coordinate dimension of the fully connected neural network is 2, and the dimension of the output feature vector is p.

[0076] like Figure 6 As shown, the branch network can be a residual neural network. Figure 6 This is the structure of a residual neural network. The input to the residual neural network is a set of physiological signals (all physiological signals have a uniform dimension), and the output of the residual neural network is two feature vectors [b]. 11 ,b 21 ,b 31...b p1 ]、[b 12 ,b 22 ,b 32 ...b p2 ].like Figure 7 As shown, each residual unit of the residual neural network includes three 1D convolution operations. After the outputs of two 1D convolution operations are added together, the input is a nonlinear activation function (which can be a hyperbolic tangent function) for calculation, thereby obtaining the output of the residual unit.

[0077] Optionally, the prediction model acquisition process of this application includes:

[0078] S201: Obtain the training dataset and its corresponding physical model.

[0079] Optionally, the training dataset includes historical physiological signals and corresponding blood pressure waveform data or blood flow velocity waveform data. The historical physiological signals originate from the object corresponding to the data to be predicted, and the corresponding blood pressure waveform data or blood flow velocity data are actual waveform data of a single or multiple locations within the target blood vessel of that object.

[0080] Optionally, to ensure the predicted results match the actual blood pressure and blood flow data, the physical model includes heamodynamic models describing the relationship between blood pressure and blood flow rate. This model allows training the prediction model to be performed using only blood pressure waveform data or blood flow rate data. Accordingly, obtaining the training dataset and its corresponding physical model includes: determining the physical model corresponding to the training dataset based on the sample groups in the training dataset. The physical model includes the heamodynamic model and its associated initial and boundary conditions. Each sample group can correspond to one model, or all samples can correspond to one model, with the parameters of the model adjusted according to each sample group.

[0081] In one embodiment, the hemodynamic model is a set of Navier-Stokes equations, which are used to adjust the loss function of the neural network. Specifically, for the i-th sample group, the corresponding equation is:

[0082] (z,t)∈Ω

[0083] Where i = 1, ..., N, N is the total number of sample groups, H i (P(z,t),Q(z,t)) is a matrix, B i(P(z,t),Q(z,t)) is a column vector, Ω is the domain of the input variable (z,t), where z is a spatial variable and t is a time variable. Specifically, H(P(z,t),Q(z,t)) is a 2x2 matrix, B i (P(z,t),Q(z,t)) is a column vector of length 2.

[0084] Optionally,

[0085]

[0086] Where, β i A 0,i K r,i , ρ i and P ext,i β represents the hyperparameters corresponding to the i-th sample group. Their specific values ​​can be set according to actual needs (different values ​​can be assigned to these hyperparameters for different training samples; the simplest method is to assign the same hyperparameter values ​​to all samples. Alternatively, these hyperparameters can be determined based on information about the sample collection subjects, such as gender and age). Specifically, β... i These are parameters related to the properties of the blood vessel itself, determined by Young's modulus and the thickness of the vessel wall; A 0,i It is the cross-sectional area of ​​the blood vessel at rest; K r,i It is a parameter determined by blood viscosity; ρ i It is blood density; P ext,i It is the pressure outside the blood vessel.

[0087] In one embodiment, the initial conditions include conditions for periodic changes in blood pressure or blood flow rate, wherein the period length corresponding to the periodic changes is data obtained from historical physiological signals. Boundary conditions include at least one of Dirichlet boundary conditions, boundary conditions indicated by the Wedxel model, Neumann boundary conditions, Robin boundary conditions, and reflection boundary conditions. These initial and boundary conditions constrain the equations of the hemodynamic model, thereby giving the equations a unique solution.

[0088] In one embodiment, the hemodynamic model is represented by partial differential equations, and the formulas characterizing the initial or boundary conditions are as follows:

[0089] l i,k (P(z,t),Q(z,t))=0,(z,t)∈Ω k k=1,…K,i=1,…,N

[0090] l i,k(P(z,t),Q(z,t)) represents the k-th condition related to the hemodynamic model for the i-th sample group. Where Ω k It is the domain of the k-th condition, and Ω k It is a subset of the domain Ω of the system of partial differential equations.

[0091] In one embodiment, for a set of training datasets A set of samples (S) i ,Y i (S) i As a physiological signal, Y i The periodic conditions for the hemodynamic model corresponding to blood pressure waveform data measured simultaneously at a certain location (where N is the number of sample groups) are as follows:

[0092] l i,k (P(z,t),Q(z,t))=P(z,t)-P(z,t+t p )=0,(z,t)∈Ω

[0093] or

[0094] l i,k (P(z,t),Q(z,t))=Q(z,t)-Q(z,t+t p )=0,(z,t)∈Ω,

[0095] Among them, t p It is a hyperparameter whose value can be determined by the input physiological signal or set according to the sample collection object corresponding to the sample group.

[0096] In another embodiment, for a set of training datasets A set of samples (S) i ,Y i The boundary conditions for the corresponding hemodynamic model can include the Windkessel model:

[0097]

[0098] (z,t)∈{L}×[0,T],

[0099] Where R i and C i As hyperparameters, other similar Wedxel models that can give the model a unique solution can also be used as boundary conditions for hemodynamic models.

[0100] In another embodiment, for a set of training datasets A set of samples (S) i ,Y iThe boundary conditions of the corresponding hemodynamic model can include Dirichlet boundary conditions:

[0101]

[0102] or

[0103]

[0104] Where b = 0 or b = L; and Y can be i The true value contained therein (which can be a value obtained through invasive methods with an accuracy higher than a predetermined threshold).

[0105] S202: Determine the loss function of the neural network to be trained based on the physical model.

[0106] Optionally, determining the loss function of the neural network to be trained based on the physical model further includes: obtaining the output of the neural network to be trained; obtaining the residual term of the hemodynamic model based on the output and the hemodynamic model; and correcting the empirical loss function of the neural network through the residual term.

[0107] Alternatively, the expected loss function of the neural network can be corrected by obtaining the residual term.

[0108] In one embodiment, the training dataset (including physiological signals S) i The neural network is input with the spatiotemporal coordinates (z,t), and the output of the neural network is represented as P. i,θ (z,t) and Q i,θ (z,t). Where P i,θ (z,t) represents the neural network's prediction of arterial blood pressure related to the i-th sample, and θ is the parameter to be learned in the neural network; Q i,θ (z,t) represents the neural network's prediction of the relevant blood flow velocity for the i-th sample group. The residual term is calculated using the obtained prediction result and the physical model described above. Specifically, the residual term used to correct the expected loss function is:

[0109]

[0110]

[0111] in and They represent the definitions at Ω and Ω respectively. k L2 norm on;

[0112] c kIt is a set of weighted coefficients, the magnitude of which can be set according to actual needs, k = 1, ..., K, l i,k This represents the initial or boundary conditions associated with the i-th sample group.

[0113] In one embodiment, the training dataset (including physiological signals S) i The neural network is input with the spatiotemporal coordinates (z,t), and the output of the neural network is represented as P. i,θ (z j ,t j ) and Q i,θ (z j ,t j ), where θ is the parameter to be learned in the neural network. Where P i,θ (z j ,t j ) represents the arterial blood pressure related to the i-th sample in the spatiotemporal coordinates (z). j ,t j Prediction results under Q; i,θ (z j ,t j ) represents the blood flow velocity associated with the i-th sample in the spatiotemporal coordinates (z). j ,t j The prediction results are then used. The residual terms are calculated based on the obtained prediction results and the aforementioned physical model. Specifically, the residual terms used to correct the empirical loss function are:

[0114]

[0115] as well as

[0116] in It is a set of first points obtained by random and uniform sampling from the domain Ω, where N0 is the total number of points in the first point set; It is a set of values ​​from the domain Ω k The second point set, N, is obtained by random uniform sampling. k This represents the total number of points in the second set of points.

[0117] S203: Train a neural network based on the loss function and training dataset to obtain a prediction model.

[0118] Optionally, the prediction model is obtained by training a neural network based on a loss function and a training dataset, including: training the neural network using the training dataset, performing iterative operations using a preset iterative method, and obtaining the prediction model based on the iterative results. The preset iterative method includes a stochastic gradient descent algorithm.

[0119] In one embodiment, the stochastic gradient descent algorithm is used to solve the neural network. After a certain number of iterations of the stochastic gradient descent algorithm (a small portion of the training dataset is used as validation data; during training, the neural network's output is validated on the validation data after each iteration to determine if the accuracy of the neural network has improved; iteration stops when the validation results determine that the accuracy will no longer improve with the number of iterations), the current neural network is used as the prediction model.

[0120] The following is combined Figures 9-16 The method for predicting vascular information in this application is described, wherein, Figures 9-16 The horizontal axis represents time, and the waveform within two seconds is displayed graphically.

[0121] In one embodiment, the input physiological signals are a set of electrocardiogram signals and photoplethysmography signals. Figure 9 , Figure 10 The graphs represent the waveforms of the electrocardiogram (ECG) signal and the photoplethysmography (PPG) signal, respectively. The vertical axis of these two graphs represents the standardized ECG and PPG values. A set of spatiotemporal coordinates is input, and the prediction model outputs predicted values ​​for blood pressure and blood flow velocity based on the input data. From these predicted values, the blood pressure waveform data and blood flow velocity data are obtained. For example, to obtain the arterial blood vessel data (this embodiment uses the attached diagram),... Figure 4 The input consists of arterial blood pressure and blood flow velocity waveforms at the midpoint of the artery shown. The input spatiotemporal coordinates of this midpoint are: (0.5L,0),(0.5L,Δt),(0.5L,2Δt),…,(0.5L,NΔt). The corresponding sequence of blood pressure data output by the prediction model is as follows:

[0122] P(0.5L,0),P(0.5L,Δt),P(0.5L,2Δt),…,P(0.5L,NΔt),

[0123] Each data point in this sequence corresponds to a blood pressure value at a specific time point, and the blood flow rate data form the following sequence:

[0124] Q(0.5L,0),Q(0.5L,Δt),Q(0.5L,2Δt),…,Q(0.5L,NΔt),

[0125] Each data point in the sequence corresponds to a blood flow rate at a given time point. These two sequences together constitute the prediction model's forecast of arterial blood pressure and blood flow rate at the midpoint of the blood vessel (z = 0.5L). Based on this prediction, the blood pressure waveform and blood flow rate waveform are obtained. Figures 11-16The results show the waveform predictions of the predictive model for arterial blood pressure at the outflow, midpoint, and inflow ends, as well as the waveform predictions of blood flow velocity at the outflow (660°), midpoint (670°), and inflow (680°). Figures 11-13 The vertical axis represents blood pressure values, with units of mmHg. Figures 14-16 The vertical axis represents blood flow velocity, with units of L / s.

[0126] Compared with existing technologies, the vascular information prediction method of this application has the following advantages:

[0127] 1. It can simultaneously predict arterial blood pressure waveform and blood flow rate waveform.

[0128] 2. It can simultaneously predict the waveforms of arterial blood pressure and blood flow velocity at different locations.

[0129] 3. A training method based on physical information was adopted, which can make the prediction results meet a specific hemodynamic model, thereby obtaining more realistic waveform predictions.

[0130] 4. A physical information-based training method is adopted. When training the neural network using this method, only the actual waveform of arterial blood pressure or blood flow velocity at a certain location is needed to solve the problem. This method greatly reduces the amount of training data required.

[0131] It should be noted that, in the optional embodiments of this application, the data involved (such as physiological signals, spatiotemporal coordinates, blood pressure waveforms, blood flow rate waveforms, etc.) requires the permission or consent of the user when the above embodiments of this application are applied to specific products or technologies. Furthermore, the collection, use, and processing of the relevant data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. In other words, if the embodiments of this application involve data related to the user, this data must be obtained with the user's authorization and consent, and in accordance with the relevant laws, regulations, and standards of the country and region.

[0132] According to one aspect of this application, this application provides a vascular information prediction device, such as... Figure 17 As shown, the vascular information prediction device 300 includes a data acquisition module 310 and a vascular information prediction module 320. The data acquisition module 310 is used to acquire data to be predicted, which includes physiological signals and the spatiotemporal coordinates of the target vascular vessel. The physiological signals include at least one of photoplethysmography (PPG) signals and electrocardiogram (ECG) signals. The vascular information prediction module 320 is used to input the data to be predicted into the prediction model to acquire vascular information of the target vascular vessel, which includes blood pressure waveforms and blood flow velocity waveforms.

[0133] The prediction model is obtained through the following steps:

[0134] Obtain the training dataset and its corresponding physical model. The training dataset includes historical physiological signals and corresponding blood pressure waveform data or blood flow rate waveform data.

[0135] The loss function of the neural network to be trained is determined based on the physical model;

[0136] A prediction model is obtained by training a neural network based on the loss function and the training dataset.

[0137] Optionally, the prediction model includes a backbone network and branch networks. The data to be predicted is input into the prediction model to obtain vascular information of the target blood vessel, including:

[0138] Physiological signals are input into the branch network, and the spatiotemporal coordinates corresponding to the predicted location in the target blood vessel are input into the trunk network. Blood vessel information corresponding to the spatiotemporal coordinates is obtained based on the output information of the branch network and the trunk network.

[0139] Optionally, the backbone network includes a fully connected neural network with multiple fully connected layers, and the branch network includes a residual neural network with multiple residual units. Based on the output information of the branch network and the backbone network, the blood vessel information corresponding to the spatiotemporal coordinates is obtained, including:

[0140] Obtain the first prediction result from the branch network and the second prediction result from the backbone network. Based on the first and second prediction results, obtain the blood pressure waveform and blood flow rate waveform at the location to be predicted.

[0141] Optionally, obtain the training dataset and its corresponding physical model, including:

[0142] The physical model corresponding to the training dataset is determined based on the sample groups in the training dataset. The physical model includes the hemodynamic model and the initial and boundary conditions related to the hemodynamic model.

[0143] Optionally, the initial conditions include the condition of periodic changes in blood pressure or blood flow rate, wherein the period length corresponding to the periodic changes is data obtained from historical physiological signals.

[0144] Boundary conditions include one or more of the following: Dirichlet boundary conditions, Wedxel model, Neumann boundary conditions, Robin boundary conditions, and reflection boundary conditions.

[0145] Optionally, determining the loss function of the neural network to be trained based on the physical model further includes:

[0146] Obtain the output of the neural network to be trained, and obtain the residual terms of the hemodynamic model based on the output and the hemodynamic model.

[0147] The empirical loss function of the neural network is corrected by modifying the residual term.

[0148] Optionally, a prediction model is obtained by training a neural network based on a loss function and a training dataset, including:

[0149] The neural network is trained using a training dataset, and iterative operations are performed using a preset iterative method. The prediction model is obtained based on the iteration results. The preset iterative method includes the stochastic gradient descent algorithm.

[0150] In one alternative embodiment, an electronic device is provided, such as Figure 18 As shown, Figure 18 The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of this application.

[0151] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), a FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0152] Bus 4002 may include a pathway for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the figure, but this does not indicate that there is only one bus or one type of bus.

[0153] The memory 4003 may be ROM (Read-Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read-Only Memory), CD-ROM (Compact Disc Read-Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium capable of carrying or storing computer programs and capable of being read by a computer, without limitation herein.

[0154] The memory 4003 stores computer programs that execute embodiments of this application, and its execution is controlled by the processor 4001. The processor 4001 executes the computer programs stored in the memory 4003 to implement the steps shown in the foregoing method embodiments.

[0155] Among them, electronic devices can be any electronic product that can interact with an object, such as personal computers, tablets, smartphones, personal digital assistants (PDAs), game consoles, interactive network television (IPTV), smart wearable devices, etc.

[0156] The electronic device may also include network devices and / or object devices. The network devices include, but are not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of hosts or network servers.

[0157] The networks in which the electronic devices are located include, but are not limited to, the Internet, wide area networks, metropolitan area networks, local area networks, and virtual private networks (VPNs).

[0158] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement the steps and corresponding content of the aforementioned method embodiments.

[0159] The terms "first," "second," "third," "fourth," "1," "2," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in a sequence other than that shown in the figures or text.

[0160] It should be understood that although arrows indicate various operation steps in the flowcharts of this application's embodiments, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of this application's embodiments, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all steps in each flowchart, based on the actual implementation scenario, may include multiple sub-steps or multiple stages. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured according to requirements, and this application's embodiments do not limit this.

[0161] The above description is only an optional implementation method for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application without departing from the technical concept of this application also fall within the protection scope of the embodiments of this application.

Claims

1. A method for predicting vascular information, characterized in that, include: Acquire the data to be predicted, which includes physiological signals and the spatiotemporal coordinates of the target blood vessel. The physiological signals include at least one of photoplethysmography (PPG) signals and electrocardiogram (ECG) signals. The data to be predicted is input into the prediction model to obtain the vascular information of the target blood vessel, including blood pressure waveform and blood flow rate waveform. The prediction model is obtained through the following steps: Obtain the training dataset and its corresponding physical model. The training dataset includes historical physiological signals and corresponding blood pressure waveform data or blood flow rate waveform data. The loss function of the neural network to be trained is determined based on the physical model. The prediction model is obtained by training a neural network based on the loss function and the training dataset. The prediction model includes a backbone network and branch networks. The step of inputting the data to be predicted into the prediction model to obtain the vascular information of the target blood vessel includes: The physiological signal is input into the branch network, and the spatiotemporal coordinates corresponding to the predicted location in the target blood vessel are input into the trunk network. The blood vessel information corresponding to the spatiotemporal coordinates is obtained based on the output information of the branch network and the trunk network.

2. The method according to claim 1, characterized in that, The backbone network includes a fully connected neural network with multiple fully connected layers, and the branch network includes a residual neural network with multiple residual units. Obtaining the blood vessel information corresponding to the spatiotemporal coordinates based on the output information of the branch network and the backbone network includes: Obtain the first prediction result output by the branch network and the second prediction result output by the backbone network, and obtain the blood pressure waveform and blood flow rate waveform at the location to be predicted based on the first prediction result and the second prediction result.

3. The method according to claim 1, characterized in that, The process of obtaining the training dataset and its corresponding physical model includes: The physical model corresponding to the training dataset is determined based on the sample groups in the training dataset. The physical model includes a hemodynamic model and the initial and boundary conditions related to the hemodynamic model.

4. The method according to claim 3, characterized in that, The initial conditions include the condition of periodic changes in blood pressure or blood flow rate, wherein the period length corresponding to the periodic change is data obtained from the historical physiological signals; The boundary conditions include at least one of the following: Dirichlet boundary conditions, boundary conditions indicated by the Wedxel model, Neumann boundary conditions, Robin boundary conditions, and reflection boundary conditions.

5. The method according to claim 3, characterized in that, The step of determining the loss function of the neural network to be trained based on the physical model further includes: Obtain the output of the neural network to be trained, and obtain the residual term of the hemodynamic model based on the output and the hemodynamic model; The empirical loss function of the neural network is corrected by the residual term.

6. The method according to claim 1, characterized in that, The step of training a neural network based on the loss function and the training dataset to obtain the prediction model includes: The neural network is trained using the training dataset and iterated using a preset iterative method. The prediction model is obtained based on the iteration results. The preset iterative method includes the stochastic gradient descent algorithm.

7. A vascular information prediction device, characterized in that, The device includes: The data to be predicted module is used to acquire the data to be predicted, which includes physiological signals and the spatiotemporal coordinates of the target blood vessel. The physiological signals include at least one of photoplethysmography signals and electrocardiogram signals. The vascular information prediction module is used to input the data to be predicted into the prediction model to obtain the vascular information of the target vascular vessel, including blood pressure waveform and blood flow rate waveform. The prediction model is obtained through the following steps: Obtain the training dataset and its corresponding physical model. The training dataset includes historical physiological signals and corresponding blood pressure waveform data or blood flow rate waveform data. The loss function of the neural network to be trained is determined based on the physical model. The prediction model is obtained by training a neural network based on the loss function and the training dataset. The prediction model includes a backbone network and branch networks. The step of inputting the data to be predicted into the prediction model to obtain the vascular information of the target blood vessel includes: The physiological signal is input into the branch network, and the spatiotemporal coordinates corresponding to the predicted location in the target blood vessel are input into the trunk network. The blood vessel information corresponding to the spatiotemporal coordinates is obtained based on the output information of the branch network and the trunk network.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-6.

Images