Hypertemporal-spatial resolution blood flow velocity field and pressure field analysis method and system

Through deep learning and physical information neural networks, combined with ultrasonic super-resolution imaging technology, the problem of low blood flow velocity field resolution in ultrasonic imaging is solved, and the ultra-time and space-resolution blood flow velocity field and pressure field in tumor microvascular and vascular networks is realized. The rapid calculation provides high-precision tools for early tumor diagnosis.

CN118981985AActive Publication Date: 2024-11-19XI AN JIAOTONG UNIV

Patent Information

Application Number
CN202411066013.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-11-19
Estimated Expiration
2044-08-05

AI Technical Summary

Technical Problem

At this stage, ultrasound imaging technology has low imaging resolution in early tumor diagnosis, and it is impossible to quickly calculate the blood flow velocity field and pressure field intravascular ultra-time and space resolution, especially in microvascular and vascular networks, resulting in unclear vascular morphology and hemodynamics. Difficult to learn parameters.

Method used

Using deep learning ideas combined with physical information neural networks, the auxiliary data set is constructed and iteratively trained by extracting the geometric features of blood vessel boundaries and spatial coordinate points under time sequence in the blood flow velocity diagram. Function, realizes the calculation of the blood flow velocity field and pressure field with super-time and space resolution.

Benefits of technology

It significantly improves the resolution and timeliness of the blood flow velocity field, can calculate the pressure field based on the coordinates of any spatial position in the blood vessel, breaks through the resolution limit of ultrasound imaging, and provides a high-precision monitoring tool for the development of tumor microvascular and peripheral blood vessels.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118981985A_ABST
    Figure CN118981985A_ABST
Patent Text Reader

Abstract

The invention discloses a hyper-temporal-spatial-resolution blood flow velocity field and pressure field analysis method and a hyper-temporal-spatial-resolution blood flow velocity field and pressure field analysis system, and belongs to the technical field of ultrasonic medical image processing. Extracting geometric features of a blood vessel boundary and a spatial coordinate point set under a time sequence in a blood vessel to form a coordinate data set, performing data assimilation on a high-resolution instantaneous velocity vector and a hemodynamic simulation value, and constructing an auxiliary data set; constructing a physical information neural network, taking the coordinate data set and the auxiliary data set as training data sets, carrying out iterative training on the physical information neural network, and obtaining an optimal model of the physical information neural network for estimating the super-resolution blood flow velocity field and the pressure field after iterative optimization; predicting an instantaneous super-resolution blood flow velocity field and a pressure field according to the optimal model of the physical information neural network without other constraint conditions; according to the method, the problem of low time and space resolution of velocity field estimation in the ultrasonic super-resolution imaging process is solved, and rapid calculation of the hyper-temporal spatial resolution blood flow velocity field and the pressure field is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of ultrasonic medical image processing, and in particular relates to a method and system for analyzing blood flow velocity field and pressure field with ultra-temporal and spatial resolution. Background Art

[0002] Malignant tumors have become one of the main causes of death in humans, seriously threatening human health and life safety. At present, tumor diagnosis and treatment are facing multiple tests and challenges from early screening and diagnosis to various treatment methods, as well as efficacy evaluation and recurrence monitoring. Therefore, tumor diagnosis and treatment have become the core of clinical medical research. Imaging is currently an important means of early detection of tumors. Clinical imaging methods such as CT, MRI, and conventional ultrasound mainly rely on morphological and conventional perfusion information for tumor diagnosis, and lack the capture of micro-blood flow information that is sensitive to changes in early tumor microvascular dynamics. In addition, one of the malignant natures of tumors is the induction of new blood vessels. Tumor angiogenesis is the growth of new capillaries in existing blood vessels, that is, the process of microvascularization and blood circulation in tumors, which is an important condition for tumor occurrence, growth and metastasis. For example, in liver tumors, the vascular network in the tumor usually changes from large blood vessels to abnormal microvessels. Therefore, studying the hemodynamic parameters of tumor microvessels and blood vessels around tumors is of great significance for the early detection of tumor-related diseases.

[0003] Ultrasound imaging is a non-invasive diagnostic tool that can achieve non-destructive real-time imaging, and is simple to operate and inexpensive. However, due to the diffraction limit of ultrasound, its imaging resolution is low, which limits its application in early diagnosis of tumors. Ultrasound super-resolution imaging detects, locates and tracks microbubbles in ultrasonic blood flow signal data, and after time accumulation, superimposes to form a super-resolution image of the blood vessel, breaking through the diffraction limit and realizing the reconstruction of micron-level vascular structure. At the same time, super-resolution imaging can obtain the motion displacement of microbubbles through tracking algorithms, and further obtain the blood flow velocity in the blood vessel. However, in the process of super-resolution imaging, the number of instantaneous microbubbles in the blood vessel that can be located is limited, and the final super-resolution velocity map needs to be formed by superimposing thousands of single-frame velocity maps. The velocity obtained by the super-resolution velocity map tends to the average velocity of the blood vessel. Therefore, it is still a huge challenge to obtain the super-resolution velocity distribution in the blood vessel while ensuring the time resolution. In addition, in the process of super-resolution imaging, only the blood flow velocity value in the blood vessel can be obtained, and other hemodynamic parameters such as pressure distribution can not be calculated. In summary, at present, there is no method to quickly calculate the super-temporal and spatial resolution blood flow velocity field and pressure field of the target blood vessels and vascular networks based on the geometric characteristics of super-resolution imaging and the sparse spatial position coordinates within the blood vessels. The difficulty lies in the unclear morphology of blood vessels and vascular networks caused by the low resolution of ultrasound imaging, the limited number of microbubble distribution in the target blood vessels and the limitations of existing motion tracking algorithms. Summary of the invention

[0004] In order to solve the problem of using contrast-enhanced ultrasonic blood flow data to quickly calculate the ultra-temporal and spatial resolution blood flow velocity field and pressure field in blood vessels, especially for microvessels and vascular networks, the present invention provides a method and system for analyzing blood flow velocity field and pressure field with ultra-temporal and spatial resolution. Based on ultrasonic super-resolution imaging and from the perspective of hemodynamics, combined with the idea of ​​deep learning, the method solves the current problems of low resolution, poor real-time performance, and lack of physical correctness in calculating the blood flow velocity field in blood vessels. The corresponding pressure field can be calculated based on the spatial position coordinates of any sample in the blood vessel, which significantly improves the timeliness and temporal and spatial resolution of the calculation.

[0005] The present invention is achieved through the following technical solutions:

[0006] A method for analyzing blood flow velocity field and pressure field with ultra-temporal and spatial resolution, comprising the following steps:

[0007] Step 1: According to the ultrasonic super-resolution vascular structure map and blood flow velocity map of the target object, the geometric features of the vascular boundary and the set of spatial coordinate points in the time sequence in the blood vessel are extracted to form a coordinate data set;

[0008] A high-resolution instantaneous velocity vector is obtained according to the blood flow velocity map, and the high-resolution instantaneous velocity vector is data assimilated with the hemodynamic simulation value to construct an auxiliary data set;

[0009] Step 2: construct a physical information neural network, and construct the loss function term corresponding to the physical residual network and data-driven according to the mean square error of the physical residual network, the mean square error of the intravascular velocity vector and its predicted value, and construct the boundary condition term of the loss function by extracting the given velocity of the boundary position point and the mean square error of its predicted value according to the geometric characteristics of the blood vessel boundary;

[0010] Step 3: Use the coordinate data set and the auxiliary data set as training data sets, iteratively train the physical information neural network, and obtain the optimal model of the physical information neural network for estimating the super-resolution blood flow velocity field and pressure field after iterative optimization. Predict the instantaneous super-resolution blood flow velocity field and pressure field based on the optimal model of the physical information neural network.

[0011] Preferably, in step 1, contrast-enhanced ultrasonic blood flow data of the target area blood vessels or blood vessel networks are extracted, and super-resolution imaging is performed on the ultrasonic blood flow data to obtain a super-resolution blood vessel structure map and a blood flow velocity map.

[0012] Preferably, the method for acquiring the ultrasonic super-resolution vascular structure image and blood flow velocity image is as follows:

[0013] The tissue signal in the contrast-enhanced ultrasound blood flow data of the target object is filtered to obtain the blood flow signal, and the super-resolution vascular structure map and blood flow velocity map are obtained by using a multi-scale statistical feature positioning algorithm and a tracking algorithm that minimizes the total matching distance between adjacent frames.

[0014] Preferably, the method for constructing the auxiliary data set in step 1 is as follows:

[0015] The super-resolution blood flow velocity vector within the set time set is stacked according to the blood flow velocity map to construct a high-resolution instantaneous velocity vector, which is then fused with the hemodynamic simulation through data assimilation to obtain an auxiliary data set.

[0016] Preferably, the velocity vectors obtained by ultrasound super-resolution tracking are preprocessed by standardized median filtering and boundary value solver smoothing interpolation, and the preprocessed super-resolution velocity vectors within a set time set are stacked, and an auxiliary data set is constructed through data assimilation with hemodynamic simulation.

[0017] Preferably, in step 3, the Navier-Stokes equations are encoded in the physical information neural network as a physical residual network as follows:

[0018]

[0019] e3=e1+e2

[0020] Among them, e1, e2 and e3 represent the residual network of the physical information network, is the dynamic viscosity of blood flow; u, v are dimensionless velocity vectors, p is dimensionless pressure, t is dimensionless time, is the gradient operator, is the Laplace operator U representing the two-dimensional velocity vector.

[0021] Preferably, the training method of the physical information neural network is as follows:

[0022] Input the training data set into the physical information neural network model to obtain the error function of the output value and the input value. Use the optimization algorithm to minimize the loss function and update the network model parameters. After multiple iterations of optimization, the optimal model is obtained.

[0023] The chain rule is applied in physical information neural networks to solve differential functions using automatic differentiation, so that the physical control equations participate in the updating and optimization of the network model.

[0024] Preferably, during the training process of the physical information neural network, an optimizer is used to minimize the loss function, and the optimizer is an Adam optimizer and an L-BFGS optimizer.

[0025] Preferably, the loss function of the physical information neural network is as follows:

[0026] L=L D +L E +L B

[0027]

[0028] Among them, L D , L E and L B They are the loss function terms corresponding to data-driven, physical residual network and super-resolution blood vessel boundary features, and β are the weights of the corresponding items, N represents the total number of data points, i represents the number of residual networks, and p represents the prediction result. Indicates the velocity value corresponding to the data point on the nth boundary.

[0029] A system for analyzing a blood flow velocity field and pressure field with ultra-temporal and spatial resolution, comprising:

[0030] A feature extraction module is used to extract the geometric features of the blood vessel boundary and the set of spatial coordinate points in the blood vessel in time sequence to form a coordinate data set based on the ultrasonic super-resolution blood vessel structure map and blood flow velocity map of the target object;

[0031] A high-resolution instantaneous velocity vector is obtained according to the blood flow velocity map, and the high-resolution instantaneous velocity vector is data assimilated with the hemodynamic simulation value to construct an auxiliary data set;

[0032] A network module is used to construct a physical information neural network, construct a loss function term corresponding to the physical residual network and data driving according to the mean square error of the physical residual network, the velocity vector and its predicted value, and construct a boundary condition term of the loss function by extracting the given velocity of the boundary position point and the mean square error of its predicted value according to the geometric characteristics of the blood vessel boundary;

[0033] The prediction module is used to use the coordinate data set and the auxiliary data set as training data sets to iteratively train the physical information neural network. After iterative optimization, the optimal model of the physical information neural network for estimating the super-resolution blood flow velocity field and pressure field is obtained. Based on the optimal model, the instantaneous super-resolution blood flow velocity field and pressure field are predicted according to the set of spatial coordinate points in the intravascular time series.

[0034] Compared with the prior art, the present invention has the following beneficial technical effects:

[0035] The present invention provides a fast super-temporal and spatial resolution blood flow velocity field and pressure field calculation and analysis method. First, the geometric features of the blood vessel boundary and the time-series spatial coordinate point set in the blood vessel are extracted according to the ultrasonic super-resolution result to form a coordinate data set. At the same time, the pre-processed velocity vector obtained by ultrasonic super-resolution tracking and the hemodynamic simulation are used to construct an auxiliary data set through data assimilation. The training data set composed of the auxiliary data set and the coordinate data set is used to optimize the physical information neural network to obtain the optimal model for calculating the super-temporal and spatial resolution blood flow velocity field and pressure field, and the time-series spatial coordinate point set in the blood vessel is input into the above-mentioned optimal model without other constraints, so that the super-temporal and spatial resolution blood flow velocity field and pressure field can be quickly obtained.

[0036] This method is based on a physical information neural network model. It uses the sparse super-resolution blood flow velocity vectors in a stacked short-time set and the geometric features of the super-resolution blood vessel boundary to achieve the calculation of the super-temporal and spatial resolution blood flow velocity field and pressure field. It breaks through the obstacle that the existing super-resolution imaging is limited by microbubble concentration and tracking algorithm and cannot quickly and accurately estimate the blood flow velocity field. It also predicts the corresponding pressure field while calculating the velocity field, ensuring the physical correctness of the hemodynamic parameter calculation. In addition, by constructing the boundary condition term of the loss function in the physical information neural network, not only the calculation of the hemodynamic parameters of a single blood vessel in the target area is completed, but also the calculation and analysis of the hemodynamic parameters of the complex vascular network in the target area can be realized, providing a powerful tool for monitoring the development of tumor microvessels and peripheral blood vessels.

[0037] Furthermore, the preprocessed velocity vectors obtained by ultrasonic super-resolution tracking are assimilated with hemodynamic simulation to obtain an auxiliary data set, which effectively reduces the gap between the training data set and the actual blood flow conditions, helps to improve the generalization ability of the physical information neural network model and its friendliness to real experimental data, and accelerates the convergence of the network. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of the method for analyzing blood flow velocity field and pressure field with ultra-temporal and spatial resolution according to the present invention;

[0039] Figure 2 It is a framework diagram of the ultra-temporal and spatial resolution blood flow velocity field and pressure field analysis method of the present invention;

[0040] Figure 3 This is a diagram showing the assimilation results of the super-resolution velocity vector and hemodynamic simulation data of the present invention.

[0041] Figure 4 This is a result diagram of the super-resolution blood flow velocity field of the target blood vessel of the present invention.

[0042] Figure 5This is a result diagram of the super-resolution blood flow pressure field of the target blood vessel of the present invention. DETAILED DESCRIPTION

[0043] The present invention will be further described in detail below in conjunction with the accompanying drawings, which are intended to explain the present invention rather than to limit it.

[0044] See also Figure 1 A fast ultra-temporal and ultra-resolution blood flow velocity field and pressure field analysis method includes collecting blood flow data and performing super-resolution imaging, generating training data sets, preprocessing data, constructing physical information neural networks, and iterative optimization of network models. The specific steps include:

[0045] Step 1: Acquire contrast-enhanced ultrasound blood flow data of the target object.

[0046] Specifically, the ultrasound contrast agent is injected into the target object (the target object includes: phantom blood vessels, animal blood vessels and human blood vessels), so that the microbubbles follow the blood flow and circulate to the target position, and a transducer of suitable frequency is used to emit a plane wave or a divergent wave and a focused wave with an angle equal to or greater than an angle, and the ultrasound echo signals of the contrast microbubbles in the blood vessels of the target area are collected to reflect the blood flow movement, and contrast-enhanced ultrasound blood flow image data is obtained through beam synthesis.

[0047] Step 2: For the contrast-enhanced ultrasound blood flow image data generated in step 1, the singular value decomposition (SVD) algorithm is used to filter out the tissue signal in the target area and retain the blood flow signal. Then, the ultrasound blood flow data after SVD is processed using a multi-scale statistical feature positioning algorithm and a tracking algorithm that minimizes the total matching distance between adjacent frames to obtain an ultrasound super-resolution vascular structure map and a blood flow velocity map.

[0048] The SVD method is an important matrix decomposition method in linear algebra, especially in dealing with matrix rank, feature space and least squares problems in numerical linear algebra. In the field of ultrasonic medical image processing, SVD is used to process contrast-enhanced ultrasonic imaging data, decomposing the ultrasonic signal of the target area into the form of several matrix multiplications to express the characteristic information in the ultrasonic image. In ultrasonic echo data, it is assumed that the tissue is stationary and the blood flow is flowing. Therefore, the tissue has high spatiotemporal coherence, which is manifested in the ultrasonic image as the consistency of a large number of spatial pixels on the time scale, represented by the larger eigenvalues ​​and corresponding eigenvectors in the matrix, while the spatiotemporal coherence of the blood flow echo signal is low, representing that the blood flow is the smaller eigenvalue in the matrix and its corresponding eigenvector. In addition, there is still some noise in the ultrasonic echo signal that needs to be filtered out, so in actual operation, the tissue signal is filtered out by setting a bidirectional threshold method while achieving denoising. Specifically, the SVD method can be described as:

[0049] S f =SVI f V * =UΔ f V *

[0050] Among them, S f is the data after SVD filtering; I f is a matrix filter, that is, the first several diagonal elements of the diagonal unit matrix are set to 0 to remove the eigenvalues ​​corresponding to tissue motion, so it is a truncated unit diagonal matrix, corresponding to the threshold diagonal matrix for removing singular values ​​of tissue motion; S is the input combined contrast-enhanced ultrasound data, U and V are orthogonal matrices, * represents conjugate transpose, Δ f Is a diagonal matrix. Similarly, since the noise signal also has a low temporal and spatial correlation, it is mixed with the blood flow information in a smaller eigenvalue. Therefore, a bidirectional threshold is set to filter out tissue signals and noise signals at the same time.

[0051] The multi-scale statistical feature positioning algorithm uses a multi-scale Hessian matrix to perform convolution to calculate eigenvalues, construct regional energy and microbubble spot features to determine microbubble candidate groups and separate overlapping microbubbles, then model the multi-scale kernel and directional kernel, and retain the region with a statistical feature confidence higher than the threshold as the accurate microbubble signal position according to the set threshold.

[0052] Specifically, first, the eigenvalues ​​and vectors of the Hessian matrix are determined, and based on this, the regional energy and speckle characteristics are calculated to separate overlapping microbubbles, which can be expressed as:

[0053]

[0054] Among them, λ n,j is the jth normalized eigenvector in the Hessian matrix The corresponding eigenvalue, n is the calculation scale. According to the Hessian matrix, we can calculate the regional energy to distinguish the background and noise, and the calculation of the microbubble spot characteristics can be used to separate overlapping microbubbles. The combination of the above two constitutes a microbubble separation filter. Finally, the maximum response value of the microbubble separation filter within the specified range is obtained, which is the separated microbubble. It can be expressed as:

[0055]

[0056] Among them, Fr e and Fr. b They are regional energy filtering and microbubble spot feature filtering, γ e and γ bIt is used to control the sensitivity of the above filtering. Filter is a microbubble separation filter, which separates overlapping microbubbles by calculating the maximum value of the microbubble separation filter at each pixel.

[0057] Next, the microbubble morphology is modeled using multi-scale kernels and directional kernels, the ultrasound image is convolved, and the area with statistical feature confidence higher than the threshold is retained as the accurate microbubble signal location according to the custom threshold, where the statistical feature confidence includes the amplitude and gradient feature confidence. Finally, the weighted average method is used for all maximum values ​​in the same cluster to achieve sub-pixel positioning of microbubbles.

[0058] The tracking algorithm that minimizes the total matching distance between adjacent frames is achieved by converting the data association problem in multi-target (localized microbubble) tracking into a problem of solving the maximum matching problem of a bipartite graph. The bipartite graph is the search box in the adjacent frames. Assume that the set of all search boxes in the current frame is (X n ,Y n ), the set of all search boxes in the previous frame or the next frame (X n-1 ,Y n-1 ), different search boxes in the same frame are not related to each other, and the search boxes between adjacent frames are connected to each other, so that the microbubble positioning points in the search boxes between adjacent frames are almost perfectly matched, and the microbubble movement is tracked by minimizing the total matching distance. This process includes the appearance of new targets (microbubbles), the disappearance of old targets (microbubbles), and the matching of microbubble identifiers between adjacent frames and the current frame.

[0059] Step 3: Based on the super-resolution vessel resolution structure map and blood flow velocity map, the geometric features of the vessel boundary in the super-resolution structure map are extracted, including morphological information such as vessel position and diameter. The spatial coordinate set of the vessel in the time series is generated according to the vessel morphological information. The above two constitute the coordinate data set;

[0060] At the same time, the instantaneous velocity vector of the target blood vessel or vascular network is obtained according to the super-resolution blood flow velocity, including the two-dimensional velocity value and the spatial position corresponding to the velocity value. Then, the velocity vector is preprocessed, and since the super-resolution velocity vectors in the target area are less distributed, the super-resolution blood flow velocity vectors in a short time set are stacked to construct a high-resolution instantaneous velocity vector, and the obtained high-resolution instantaneous velocity vector is data assimilated with the hemodynamic simulation value to obtain a complete and uniformly distributed instantaneous blood flow velocity vector. The above operations constitute an auxiliary data set, which is used as a data-driven auxiliary network to accelerate convergence during the physical information neural network training process.

[0061] Data preprocessing is standardized median filtering and boundary value solver smoothing interpolation.

[0062] The standardized median filtering method is an outlier detection method that is adaptive to local flow conditions. The filtering method evaluates the fluctuation relative to the median velocity in a 3*3, 5*5 or 7*7 neighborhood around the center vector, and then processes the median of the fluctuation using normalization of the median test.

[0063] The boundary value solver smooth interpolation method is used to replace missing data using interpolation method for the location where outliers are removed. Similarly, in 3*3, 5*5 or 7*7 neighborhoods, the boundary value solver smooth interpolation method is used. In areas with more missing data, it calculates the mean value that tends to the boundary speed and fills the missing data with the mean value.

[0064] See also Figure 3 , the high-resolution instantaneous velocity vector is data assimilated with the hemodynamic simulation value. Specifically, the position and number of velocity vectors are identified in the custom search area, and the weight coefficient of weighted average is set according to the distance of the velocity vector from the search center position in the search area, so as to obtain the velocity vector at the center position. For the positions where the velocity vector cannot be identified in the search area, the hemodynamic simulation value is used to fill in and form an auxiliary data set. It is worth noting that the data dimension of the auxiliary data set is consistent with the data dimension of the coordinate data set.

[0065] Step 4: A training data set is constructed based on the coordinate data set and auxiliary data set obtained in step 3. Furthermore, a physical information neural network for fast calculation of ultra-temporal and spatial resolution blood flow velocity field and pressure field is constructed using two-dimensional Navier-Stokes equations (NS equations).

[0066] The physical information neural network mainly includes the following parts: data mapping network and physical residual network encoded with NS equations.

[0067] The data mapping network includes an input layer, a hidden layer and an output layer. The input layer is used to input training data, mainly a set of spatial coordinates in a time series. The blood flow velocity and pressure are mapped through the hidden layer, which is the output layer of the data mapping network. The hidden layer includes at least two network channels. The physical residual network encoded by the NS equation is mainly constructed based on the NS equation, where the NS equation is expressed as:

[0068]

[0069] In the above formula, u represents fluid velocity, p represents pressure, t represents time, and ρ represents fluid density. Expressed as dynamic viscosity. The left side of the equation represents the inertial force, and the right side of the equation corresponds to the pressure, viscous force and external force acting on the fluid.

[0070] The physical information neural network for optimizing hemodynamic parameters is constructed using the two-dimensional NS equations, where the physical residual network is as follows:

[0071]

[0072] e3=e1+e2

[0073] In the above formula, e1, e2 and e3 represent the residual network corresponding to the basic parameters, assuming that the blood flow is an incompressible fluid. is the gradient operator, which represents the sum of the first-order derivatives in all directions in space. is the Laplace operator, which represents the sum of the second-order derivatives in all directions in space, U represents the two-dimensional velocity vector, Expressed as dynamic viscosity.

[0074] Step 5: Construct a loss function of the physical information neural network structure, iteratively optimize the physical information neural network structure to obtain the optimal network model for calculating the ultra-temporal and spatial resolution blood flow velocity field and pressure field.

[0075] Specifically, the loss function of the physical information neural network is constructed by adding the mean square error of the data assimilated by the high-resolution instantaneous velocity vector and the hemodynamic simulation and the predicted output value, the mean square error of the physical residual network, and the mean square error of the given velocity of the super-resolution vascular boundary position point and its predicted value. In the iterative training process, the Adam optimizer and L-BFGS are selected to train the neural network, and the activation function selects sin(x) to initialize the network parameters, and sin(x) meets the convergence requirements of the network model. At the same time, the convergence speed of the neural network is further accelerated by adjusting the fixed learning rate or setting the adaptive learning rate, the number of iterations, and the size of the training data set selected for one training. Finally, the optimal model of the physical information neural network for estimating the super-resolution blood flow velocity field and pressure field is obtained.

[0076] In the physical information neural network, the boundary condition term of the loss function is constructed by extracting the given velocity of the boundary position point and the mean square error of its predicted value according to the geometric characteristics of the blood vessel boundary. This can limit the blood flow range of the target area, especially the range of complex blood flow in the entire vascular network, and further obtain the corresponding blood flow velocity vector and its spatial position point.

[0077] The loss function constructed for the physical information neural network can be specifically expressed as:

[0078] L=L D +L E +L B

[0079]

[0080] Among them, LD , L E and L B They are the loss function terms corresponding to data-driven, physical residual network and super-resolution blood vessel boundary features. and β are the weights of the corresponding items, N represents the total number of data points, i represents the number of residual networks, and p represents the prediction result. Indicates the velocity value corresponding to the data point on the nth boundary.

[0081] Step 6. Use the training data set to train the physical information neural network. During the iterative training process, the training data set is input into the data mapping network, and the blood flow velocity and pressure are output after passing through the hidden layer. The output velocity and pressure values ​​are then input into the physical residual network for partial differential equation calculation. Furthermore, the results of the data mapping network and the physical residual network are passed to the Adam optimizer and the L-BFGS optimizer for gradient optimization, and the optimization results are fed back to the aforementioned network, thereby iteratively updating the network parameters to obtain the optimal model for calculating the ultra-temporal and spatial resolution blood flow velocity field and pressure field. Finally, the spatial coordinate set in the blood vessel under the time series is input into the aforementioned optimal model, which can realize the non-invasive and rapid calculation of the ultra-temporal and spatial resolution blood flow velocity field and pressure field without any constraints.

[0082] Figure 4 It is the ultra-temporal and spatial resolution blood flow velocity field map calculated based on the trained optimal model; Figure 5 This is the ultra-temporal and spatial resolution blood flow pressure field map calculated based on the trained optimal model.

[0083] On the other hand, the present invention also provides a super-temporal and spatial resolution blood flow velocity field and pressure field analysis system, including: an ultrasonic super-resolution imaging and feature extraction module, a data preprocessing module, a network model construction and optimization module, and a super-temporal and spatial resolution blood flow velocity field and pressure field calculation module.

[0084] The ultrasonic super-resolution imaging and feature extraction module is used to collect contrast-enhanced ultrasonic blood flow data, and uses singular value filtering, scale statistical feature positioning algorithm and tracking algorithm that minimizes the total matching distance between adjacent frames to obtain super-resolution vascular structure images and blood flow velocity images; further extract the geometric features of the vascular boundaries in the target area from the super-resolution imaging data, the set of spatial coordinate points in the time series within the blood vessel and the blood flow velocity vector obtained by super-resolution tracking.

[0085] The data preprocessing module is used to preprocess the velocity vectors obtained by ultrasound super-resolution tracking using the standardized median filtering method and the boundary value solver smoothing interpolation method, and to stack the super-resolution velocity vectors in the preprocessed short-time set. Finally, an auxiliary data set is constructed through data assimilation with the hemodynamic simulation as part of the training data set.

[0086] The network model construction and optimization module is used to construct a physical information neural network and construct the geometric features of the blood vessel boundaries in the target area as boundary conditions of the physical information neural network loss function; after iterative optimization, an optimal model of a physical information neural network for quickly estimating ultra-temporal and spatial resolution blood flow velocity fields and pressure fields is obtained.

[0087] The super-temporal and spatial resolution blood flow velocity field and pressure field calculation module is used to calculate the super-temporal and spatial resolution blood flow velocity field and pressure field based on the optimal model of the physical information neural network of the super-resolution blood flow velocity field and pressure field obtained by training, and inputs a set of spatial coordinate points in the time series within the blood vessel without the geometric characteristics of the blood vessel boundary, and calculates the super-temporal and spatial resolution blood flow velocity field and pressure field.

[0088] The boundary condition term in the loss function of the physical information neural network is obtained by calculating the mean square error between the position coordinate point of the blood vessel boundary and the corresponding velocity value according to the super-resolution blood vessel structure and the output value of the physical information network. The boundary condition term in the above loss function allows the constructed physical information neural network to not only complete the calculation of the super-temporal and spatial resolution blood flow velocity field and pressure field for a single blood vessel in the target area, but also calculate and analyze the super-temporal and spatial resolution hemodynamic parameters (velocity and pressure) in the complex blood vessel network in the target area.

[0089] The method obtains the geometric features of the blood vessel boundary and the set of spatial coordinate points in the blood vessel in time sequence according to the ultrasonic super-resolution image as the coordinate data set of the physical information neural network; and constructs an auxiliary data set by data assimilation of the velocity vector obtained by ultrasonic super-resolution tracking and the hemodynamic simulation; the auxiliary data set is combined with the coordinate data set after data preprocessing to train the physical information neural network, and after multiple iterative updates, the optimal model of the physical information neural network for estimating the super-resolution blood flow velocity field and pressure field is obtained; based on the optimal model of the physical information neural network of the super-resolution blood flow velocity field and pressure field, any spatial coordinate set in the blood vessel is used as the input of the network to predict the super-resolution blood flow velocity field and pressure field of the corresponding blood vessel without providing other constraints; the present invention solves the problem of low time and space resolution of the estimated velocity field in the process of ultrasonic super-resolution imaging, by stacking the velocity vectors under the short-time set and assimilating them with the hemodynamic simulation, and further using the physical information neural network for iterative optimization to establish a prediction model, so as to realize the calculation of the blood flow velocity field and pressure field with fast super-temporal and spatial resolution.

[0090] The above contents are only for explaining the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A method for analyzing blood flow velocity field and pressure field with ultra-temporal and spatial resolution, characterized in that: The following steps are involved: Step 1: According to the ultrasonic super-resolution vascular structure map and blood flow velocity map of the target object, the geometric features of the vascular boundary and the set of spatial coordinate points in the time sequence in the blood vessel are extracted to form a coordinate data set; A high-resolution instantaneous velocity vector is obtained according to the blood flow velocity map, and the high-resolution instantaneous velocity vector is data assimilated with the hemodynamic simulation value to construct an auxiliary data set; Step 2: construct a physical information neural network, and construct the loss function term corresponding to the physical residual network and data-driven according to the mean square error of the physical residual network, the mean square error of the intravascular velocity vector and its predicted value, and construct the boundary condition term of the loss function by extracting the given velocity of the boundary position point and the mean square error of its predicted value according to the geometric characteristics of the blood vessel boundary; Step 3: Use the coordinate data set and the auxiliary data set as training data sets, iteratively train the physical information neural network, and obtain the optimal model of the physical information neural network for estimating the super-resolution blood flow velocity field and pressure field after iterative optimization. Predict the instantaneous super-resolution blood flow velocity field and pressure field based on the optimal model of the physical information neural network.

2. The method for analyzing blood flow velocity field and pressure field with ultra-temporal and spatial resolution according to claim 1, characterized in that: In step 1, contrast-enhanced ultrasonic blood flow data of the target area blood vessels or blood vessel networks are extracted, and super-resolution imaging is performed on the ultrasonic blood flow data to obtain a super-resolution blood vessel structure map and a blood flow velocity map.

3. The method for analyzing blood flow velocity field and pressure field with ultra-temporal and spatial resolution according to claim 1, characterized in that: The method for obtaining the ultrasonic super-resolution vascular structure image and blood flow velocity image is as follows: The tissue signal in the contrast-enhanced ultrasound blood flow data of the target object is filtered to obtain the blood flow signal, and the super-resolution vascular structure map and blood flow velocity map are obtained by using a multi-scale statistical feature positioning algorithm and a tracking algorithm that minimizes the total matching distance between adjacent frames.

4. The method for analyzing blood flow velocity field and pressure field with ultra-temporal and spatial resolution according to claim 1, characterized in that: The method for constructing the auxiliary dataset described in step 1 is as follows: The super-resolution blood flow velocity vector within the set time set is stacked according to the blood flow velocity map to construct a high-resolution instantaneous velocity vector, which is then fused with the hemodynamic simulation through data assimilation to obtain an auxiliary data set.

5. The method for analyzing blood flow velocity field and pressure field with ultra-temporal and spatial resolution according to claim 4, characterized in that: The velocity vectors obtained by ultrasound super-resolution tracking were preprocessed using standardized median filtering and boundary value solver smoothing interpolation. The preprocessed super-resolution velocity vectors within a set time set were stacked and an auxiliary data set was constructed through data assimilation with hemodynamic simulation.

6. The method for analyzing blood flow velocity field and pressure field with ultra-temporal and spatial resolution according to claim 1, characterized in that: In step 3, the Navier-Stokes equations are encoded in the physical information neural network as a physical residual network as follows: e3=e1+e2 Among them, e1, e2 and e3 represent the residual network of the physical information network, θ is the dynamic viscosity of blood flow; u, v are dimensionless velocity vectors, p is dimensionless pressure, t is dimensionless time, is the gradient operator, is the Laplace operator U representing the two-dimensional velocity vector.

7. The method for analyzing blood flow velocity field and pressure field with ultra-temporal and spatial resolution according to claim 1, characterized in that: The training method of the physical information neural network is as follows: Input the training data set into the physical information neural network model to obtain the error function of the output value and the input value. Use the optimization algorithm to minimize the loss function and update the network model parameters. After multiple iterations of optimization, the optimal model is obtained. The chain rule is applied in physical information neural networks to solve differential functions using automatic differentiation, so that the physical control equations participate in the updating and optimization of the network model.

8. The method for analyzing blood flow velocity field and pressure field with ultra-temporal and spatial resolution according to claim 7, characterized in that: In the training process of physical information neural network, the optimizer is used to minimize the loss function, and the optimizers are Adam optimizer and L-BFGS optimizer.

9. The method for analyzing blood flow velocity field and pressure field with ultra-temporal and spatial resolution according to claim 1, characterized in that: The loss function of the physical information neural network is as follows: L=L D +L E +L B Among them, L D , L E and L B They are the loss function terms corresponding to data-driven, physical residual network and super-resolution blood vessel boundary features, and β are the weights of the corresponding items, N represents the total number of data points, i represents the number of residual networks, and p represents the prediction result. Indicates the velocity value corresponding to the data point on the nth boundary.

10. A system for executing the ultra-temporal and spatial resolution blood flow velocity field and pressure field analysis method according to any one of claims 1 to 9, characterized in that: include: A feature extraction module is used to extract the geometric features of the blood vessel boundary and the set of spatial coordinate points in the blood vessel in time sequence to form a coordinate data set based on the ultrasonic super-resolution blood vessel structure map and blood flow velocity map of the target object; A high-resolution instantaneous velocity vector is obtained according to the blood flow velocity map, and the high-resolution instantaneous velocity vector is data assimilated with the hemodynamic simulation value to construct an auxiliary data set; A network module is used to construct a physical information neural network, construct a loss function term corresponding to the physical residual network and data driving according to the mean square error of the physical residual network, the velocity vector and its predicted value, and construct a boundary condition term of the loss function by extracting the given velocity of the boundary position point and the mean square error of its predicted value according to the geometric characteristics of the blood vessel boundary; The prediction module is used to use the coordinate data set and the auxiliary data set as training data sets to iteratively train the physical information neural network. After iterative optimization, the optimal model of the physical information neural network for estimating the super-resolution blood flow velocity field and pressure field is obtained. Based on the optimal model, the instantaneous super-resolution blood flow velocity field and pressure field are predicted according to the set of spatial coordinate points in the intravascular time series.

Citation Information

Patent Citations

  • System and method for super-resolution imaging and multi-parameter detection of vascular mechanical characteristic

    CN101474083A

  • Device and method for estimating blood vessel blood flow velocity field through transcranial ultrasound

    CN112932539A

  • Method for evaluating curative effect of neoadjuvant chemotherapy of breast cancer by using ultrasonic positioning microscope

    CN116473592A

  • Method and system for reconstructing ultrasonic vessel velocity field and pressure field based on neural network

    CN117198542A

  • Blood flow visualizing diagnostic apparatus

    US20060025688A1

Cited By

  • Physical information neural network guided complex medium inversion imaging method and system

    CN120163018A

  • Physical information neural network guided complex media inversion imaging method and system

    CN120163018B

  • Agent model of vascular network exit boundary condition and construction method

    CN120277623A

  • A proxy model and construction method for vascular network outlet boundary conditions

    CN120277623B