Coronary artery plaque rupture risk assessment method based on non-local physical information neural network

Abstract

The invention relates to a coronary artery plaque rupture risk assessment method based on a non-local physical information neural network, and belongs to the field of artificial intelligence, and the method comprises the following steps: S1, introducing a near-field dynamics theory, reconstructing a mechanical constitutive equation of plaque rupture, and deducing a crack propagation criterion; s2, constructing a non-local physical information neural network, taking geometric structure information and material attributes of the coronary artery plaque as input, and taking a displacement field and a stress field of the plaque as output; taking the plaque mechanics basic equation as a physical constraint condition, and adding a loss function; s3, constructing a data set; s4, training the physical information neural network by using the training set, and adjusting network parameters by using a stochastic gradient descent optimization algorithm to minimize a loss function; optimizing and testing the model; and S5, inputting a coronary artery medical image number to be evaluated into the trained non-local physical information neural network, and outputting a rupture risk evaluation result of the coronary artery plaque.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence, and relates to a method for assessing the risk of coronary plaque rupture based on a non-local physics-informed neural network. Background Art

[0002] Coronary plaque rupture is one of the main causes of acute cardiovascular events (such as acute myocardial infarction). Accurately assessing the rupture risk of coronary plaques is of crucial significance for the prevention, diagnosis, and treatment of cardiovascular diseases.

[0003] Currently, methods for assessing the risk of coronary plaque rupture mainly rely on medical imaging techniques (such as angiography, intravascular ultrasound, and optical coherence tomography, etc.) and clinical indicators (such as blood lipid levels, blood pressure, etc.). However, most of these methods can only provide morphological information of plaques and are difficult to accurately reflect the internal mechanical state and rupture mechanism of plaques. In recent years, numerical simulation methods based on physical models have gradually been applied to the assessment of the risk of coronary plaque rupture. These methods establish a mechanical model of the plaque and simulate the stress distribution of the plaque under hemodynamic effects to evaluate the rupture risk of the plaque. However, numerical simulation methods face difficulties such as mesh re - division and long calculation time, while deep learning methods also face problems such as lack of training data and poor interpretability.

[0004] Physics - informed neural networks improve the physical consistency and interpretability of the model by embedding physical equation constraints, and reduce the demand for the scale of training data. However, traditional physics - informed neural networks rely on automatic differentiation techniques and are prone to problems such as gradient failure and convergence difficulties in discontinuous regions such as cracks, affecting the accuracy and convergence of the results. In addition, the scale of clinical imaging data of coronary plaques is limited and the acquisition cost is high, which affects the generalization ability of the model. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method for assessing the risk of coronary plaque rupture based on a non - local physics - informed neural network. By introducing the peridynamics theory to construct a non - local physics - informed neural network, it avoids the failure of automatic differentiation on the crack surface in traditional physics - informed neural networks, realizes the collaborative drive of physics and data, and improves the accuracy and reliability of the assessment of the risk of coronary plaque rupture.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] A method for assessing the risk of coronary plaque rupture based on a non - local physics - informed neural network, comprising the following steps:

[0008] S1: Reconstruct the mechanical constitutive equation of the plaque: Introduce the peridynamics theory, reconstruct the mechanical constitutive equation of plaque rupture, and derive the crack propagation criterion;

[0009] S2: Construct a non-local physical information neural network: Take the geometric structure information and material properties of the coronary artery plaque as inputs, and the displacement field and stress field of the plaque as outputs; Incorporate the reconstructed basic mechanical equation of the plaque as a physical constraint condition into the loss function of the neural network;

[0010] S3: Collect high-resolution images of coronary artery plaques, use image segmentation algorithms to reconstruct the geometric structure of coronary artery plaques from medical image data, and use the peridynamics numerical method to solve the mechanical state of coronary artery plaques; Compose a data set with the mechanical information obtained by traditional numerical methods;

[0011] S4: Model training and optimization: Divide the collected data set into a training set, a validation set, and a test set; Use the training set to train the physical information neural network, and adopt the stochastic gradient descent optimization algorithm to adjust the network parameters to minimize the loss function; During the training process, use the validation set to evaluate and optimize the model; Use the test set to test the model;

[0012] S5: Risk assessment of coronary artery plaque rupture: Input the coronary artery medical image data to be evaluated into the trained non-local physical information neural network, and output the risk assessment result of coronary artery plaque rupture.

[0013] Furthermore, the reconstruction of the mechanical constitutive equation of the plaque in step S1 specifically includes the following steps:

[0014] S11: The mechanical behavior of the main vascular components of the coronary artery plaque conforms to an isotropic hyperelastic material and is described by an improved Mooney-Rivlin model:

[0015] Ψ=c1(I1 - 3)+D1[exp(D2(I1 - 3)) - 1] (1)

[0016]

[0017] where Ψ represents the strain energy function, S and σ represent the second Piola-Kirchhoff stress tensor and the Cauchy stress tensor respectively, C represents the right Cauchy-Green deformation tensor, I1, I2, and I3 represent the first, second, and third invariants of the right Cauchy-Green deformation tensor respectively, F is the deformation gradient tensor, c1, D1, and D2 represent material properties respectively, tr(*) represents the trace of the tensor, and det(*) represents the determinant;

[0018] In continuum mechanics, the motion equation of the plaque is expressed as:

[0019]

[0020] Where t represents time, ρ s and u are the density and displacement of the plaque material, respectively, and b represents the body force;

[0021] S12: By introducing the integral operator in peridynamics to replace the differential operator, the integral form of the motion equation of the patch material is expressed as:

[0022]

[0023]

[0024] in, represents the influence domain of material point x, g 20 (ξ),g 02 (ξ) and g 11 (ξ) represents the peridynamic operator, ξ = x′ − x represents the relative position between material points x′ and x, σ(x′) and σ(x) represent the stresses at material points x′ and x, respectively, and dV′ represents the volume of material point x′;

[0025] The second Piola-Kirchhoff stress tensor and deformation gradient tensor in the constitutive equation of plaque mechanical behavior are expressed as:

[0026]

[0027] in represents the right Cauchy-Green deformation tensor, represents the deformation gradient tensor, Y <ξ> represents the relative position of the material point x′ and x after deformation, K represents the shape tensor, represents the tensor product;

[0028] S13: Derivation of the fracture criterion for plaque rupture: Write the scalar-valued coefficients used to describe the bond rupture state as a function of time t:

[0029]

[0030] Among them, w ξ represents the bond strain energy, w c represents the critical bond strain energy, h is the thickness, G0 is the critical energy release rate, and δ represents the radius of the influence domain;

[0031] Damage variable at material point x It is defined as the ratio of the number of broken bonds to the number of intact bonds within its domain:

[0032]

[0033] Among them, μ(ξ, t) characterizes the fracture state of the bond, taking values of 0 or 1, and the damage variable ranges from [0, 1].

[0034] Furthermore, for the construction of the non-local physical information neural network in step S2, the geometric structure information and material properties of the coronary artery plaque are used as inputs, and the displacement field and stress field of the plaque are used as outputs, which specifically includes:

[0035] Construct a deep neural network As the core of the model, where the input variables are the spatial coordinates x = (x, y, z) and time t, and the output variables are the displacement field u(x, t) and stress field σ(x, t). Let represent the input layer, represent the output layer; the connections between neural networks are recursively defined as:

[0036] Input layer:

[0037] Hidden layer:

[0038] Output layer:

[0039] Among them, σ represents the activation function, W l and b l respectively represent the weights and biases of the l-th layer in the neural network, W L and b L respectively represent the weights and biases of the output layer, respectively represent the input space, parameter space, and output space.

[0040] Furthermore, the reconstructed basic equations of plaque mechanics are used as physical constraint conditions and added to the loss function of the neural network, which specifically includes: constructing the loss function of the neural network, including a data loss term and a physical consistency loss term; the physical consistency loss term includes an initial condition loss term, a boundary condition loss term, and a physical equation loss term;

[0041] Data loss term is used to measure the difference between the model output and the observed data, expressed as:

[0042]

[0043] where N d is the number of sampling points in the measurement data, u i and are respectively the displacement prediction value and the reference value of the neural network at the sampling point i;

[0044] Initial condition loss term is:

[0045]

[0046] where N i is the number of sampling points in the initial condition, and u int is the reference value of the initial condition;

[0047] The boundary condition loss term is defined as

[0048]

[0049] where N b is the number of sampling points in the boundary condition, and p i and p0 are the pressure prediction value of the neural network and the boundary condition pressure value, respectively;

[0050] The physical equation loss term that satisfies the control mechanical equation is defined as:

[0051]

[0052] where, N p is the number of sampling points in the control equation;

[0053] The loss function of the neural network is expressed as the sum of each loss term:

[0054]

[0055] where, λ i is the weight of the corresponding loss term and is used to adjust the relative importance of the loss term.

[0056] Furthermore, in step S4, the network parameters are adjusted by using the stochastic gradient descent optimization algorithm, which specifically includes:

[0057] In the pre-balancing stage, the parameters of all neural networks are initialized; the weight coefficients of different loss terms are assigned;

[0058] In the dynamic adjustment stage, calculate each loss term The gradient of with respect to the neural network parameter θ

[0059]

[0060] where represents the L2 norm of the gradient , k represents the summation index, and n represents the total number of loss functions;

[0061] Finally, use the Adam algorithm to minimize the loss function Optimizing the parameters of the neural network model:

[0062]

[0063] Furthermore, in step S4, a transfer learning strategy is adopted to train the neural network, including:

[0064] Generating diverse plaque morphologies using a parametric coronary plaque geometry model, and using the displacement and stress fields obtained from simulation calculations as pre-training data;

[0065] Training a non-local physics-informed neural network model on the simulated data to enable it to learn general plaque mechanical characteristics;

[0066] Freezing some network layers and only fine-tuning the top layer network, and updating the parameters using clinical OCT image data;

[0067] Finally, testing and validating the model in a clinical real scenario.

[0068] Furthermore, in step S5, after the model training is completed, the accuracy and stability of the model are evaluated by comparing the clinical image plaque rupture results and comparing the numerical simulation results using multiple evaluation metrics.

[0069] The beneficial effects of the present invention are as follows:

[0070] Failure of the differential operator on the crack surface: By introducing the peridynamics theory to reconstruct the basic equations of plaque mechanics, the present invention avoids the problem of the failure of the differential operator on the crack surface in traditional physics-informed neural networks, can more accurately describe the mechanical state inside the plaque, and improves the accuracy of rupture risk assessment.

[0071] Fusion of physical information and data-driven: The non-local physics-informed neural network constructed by the present invention combines the physical model and the data-driven method, and adds physical constraint conditions during the network training process, so that the physical laws learned by the network are consistent with the actual situation, and the generalization ability and reliability of the model are improved.

[0072] High clinical application value: The coronary plaque rupture risk assessment algorithm of the present invention can provide more accurate diagnostic basis for clinicians, helps to early detect the risk of coronary plaque rupture, formulate personalized treatment plans, and reduce the incidence and mortality of cardiovascular diseases.

[0073] Other advantages, objectives and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be learned from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following specification. Description of the Drawings

[0074] To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be described in detail and preferably below in conjunction with the accompanying drawings, where:

[0075] Figure 1 Schematic flow chart of the method for evaluating the mechanical state of coronary artery plaques based on the non-local physical information neural network described in Embodiment 1;

[0076] Figure 2 Schematic diagram of the peridynamics theory; where (a) is the action range of the non-local integral domain, and (b) is the characterization of damage and rupture;

[0077] Figure 3 Schematic diagram of the non-local physical information neural network structure; where (a) is the geometric shape after plaque reconstruction; (b) is the basic framework of the non-local physical information neural network, and (c) is the output result of the neural network. Specific embodiments

[0078] The following uses specific specific examples to illustrate the implementation manners of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0079] It should be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner. Therefore, only the components related to the present invention are shown in the diagrams, rather than being drawn according to the number, shape, and size of the components in actual implementation. The types, quantities, and proportions of the components in actual implementation can be arbitrarily changed, and the component layout type may also be more complex.

[0080] In the following description, a large number of details are explored to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0081] Embodiment 1:

[0082] As Figure 1 shown, the present invention provides a method for evaluating the mechanical state of coronary artery plaques based on a non-local physical information neural network, including the following steps:

[0083] Step 1: Reconstruct the constitutive equation of plaque mechanics: Introduce the peridynamics theory, reconstruct the constitutive equation of plaque rupture mechanics, and derive the crack propagation criterion to avoid the stress singularity on the crack surface of the original continuum mechanics theory.

[0084] Step 2: Construct a non-local physical information neural network: Take the geometric structure information and material properties of coronary artery plaques as inputs, and take the displacement field and stress field of the plaques as outputs. The neural network structure includes an input layer, a hidden layer, and an output layer. The hidden layer uses multiple non-linear activation functions to enhance the expression ability of the network. Incorporate the reconstructed basic equation of plaque mechanics as a physical constraint condition into the loss function of the neural network to ensure that the physical laws learned by the network are consistent with the actual situation.

[0085] Step 3: Collect high-resolution images of coronary artery plaques, use image segmentation algorithms to reconstruct the geometric structure of coronary artery plaques from medical image data, and use the peridynamics numerical method to accurately solve the mechanical state of coronary artery plaques. The mechanical information obtained by traditional numerical methods constitutes a high-quality data set.

[0086] Step 4: Model training and optimization: Divide the collected data set into a training set, a validation set, and a test set, with proportions of 70%, 15%, and 15% respectively. Use the training set to train the physical information neural network, and use the stochastic gradient descent optimization algorithm to adjust the network parameters to minimize the loss function. During the training process, use the validation set to evaluate and optimize the model, and improve the performance and generalization ability of the model by adjusting hyperparameters such as the network structure and learning rate.

[0087] Step 5: Risk assessment of coronary artery plaque rupture: Input the coronary artery medical image data to be evaluated into the trained non-local physical information neural network, and the network outputs the risk assessment result of coronary artery plaque rupture, providing a mechanical basis for formulating personalized treatment plans.

[0088] Example 2:

[0089] In this example, the method for evaluating the mechanical state of coronary artery plaques based on a non-local physical information neural network specifically includes the following steps:

[0090] Step 1: Reconstruct the constitutive equation of plaque mechanics

[0091] The mechanical behaviors of the main vascular components of coronary artery plaques (including fibrous caps, lipid cores, and vessel walls) conform to isotropic hyperelastic materials and are described by an improved Mooney-Rivlin model:

[0092] Ψ = c1(I1 - 3) + D1[exp(D2(I1 - 3)) - 1] (1)

[0093]

[0094] Where Ψ represents the strain energy function, S and σ represent the second Piola - Kirchhoff stress tensor and the Cauchy stress tensor respectively, C represents the right Cauchy - Green deformation tensor, I1, I2 and I3 represent the first, second and third invariants of the right Cauchy - Green deformation tensor respectively, F is the deformation gradient tensor, tr(*) represents the trace of a tensor, det(*) represents the determinant, c1, D1 and D2 represent material properties respectively, and the material parameters in the patch mechanics constitutive are shown in Table 1 for details.

[0095] Table 1

[0096]

[0097] Therefore, the motion equation of the patch in continuum mechanics can be expressed as:

[0098]

[0099] Where, t represents time, ρ s and u are the density and displacement of the patch material respectively, and b represents the body force.

[0100] In the mechanical analysis of patch rupture, due to the generation and propagation of cracks, discontinuity is introduced into the displacement field, resulting in the failure of the continuity assumption of the differential operator in the constitutive equation. Therefore, when using the continuum mechanics theory to solve discontinuous problems such as patch rupture, problems such as convergence difficulties caused by stress singularity are faced. To overcome this defect, as Figure 2 shown in (a) - (b) below, the present invention introduces the integral operator in peridynamics to replace the differential operator, thus avoiding the singularity of the differential operator at the discontinuity.

[0101] Therefore, the integral form of the motion equation of the patch material is expressed as:

[0102]

[0103] Where, represents the influence domain of the material point x, g 20 (ξ), g 02 (ξ) and g 11 (ξ) represent the peridynamics operators, ξ = x′ - x represents the relative position between the material points x′ and x, σ(x′) and σ(x) represent the stresses of the material points x′ and x respectively, and dV′ represents the volume of the material point x′.

[0104] Similarly, the second Piola-Kirchhoff stress tensor and the deformation gradient tensor in the constitutive equation of the plaque mechanical behavior can be expressed as:

[0105]

[0106] where denotes the right Cauchy-Green deformation tensor, denotes the deformation gradient tensor, Y <ξ> represents the relative position of the material points x′ and x after deformation, K represents the shape tensor, denotes the tensor product.

[0107] Derive the fracture criterion for plaque rupture:

[0108] The peridynamics theory considers the damage and failure of materials by taking into account the interaction between removed material points. When the interaction between a pair of material points exceeds a predetermined value, the interaction between the two material points will disappear. Therefore, the scalar coefficient used to describe the bond fracture state can be written as a function of time t:

[0109]

[0110] where, w ξ denotes the bond strain energy, w c denotes the critical bond strain energy, h is the thickness, G0 is the critical energy release rate, and δ represents the radius of the influence domain.

[0111] The damage variable at the material point x is defined as the ratio of the number of broken bonds to the number of intact bonds within its domain:

[0112]

[0113] where, μ(ξ,t) characterizes the fracture state of the bond, taking values of 0 or 1, and the damage variable ranges from [0,1].

[0114] Step 2: Construction of the non-local physical information neural network

[0115] The core of constructing the non-local physical information neural network model lies in using the basic mechanical equations as the loss function or constraint conditions of the neural network, so that the neural network not only minimizes the data error during training but also satisfies the physical laws.

[0116] As shown in (a), (b), (c) of Figure 3 firstly, construct a deep neural network as the core of the model, where the input variables are the spatial coordinates x = (x, y, z) and time t, and the output variables are the displacement field u(x,t) and the stress field σ(x,t), where Represents the input layer, Represents the output layer. The connections between neural networks are recursively defined as:

[0117] Input layer:

[0118] Hidden layer:

[0119] Output layer:

[0120] where σ represents the activation function, W l and b l represent the weights and biases of the l-th layer in the neural network respectively, and W L and b L represent the weights and biases of the output layer respectively, represent the input space, parameter space, and output space respectively.

[0121] Then, the loss function of the neural network is constructed, which mainly consists of a data loss term and a physical consistency loss term. The physical consistency loss term includes an initial condition loss term, a boundary condition loss term, and a physical equation loss term.

[0122] Data residual term Measures the difference between the model output and the observed data (such as numerical simulation results):

[0123]

[0124] Consider the initial condition residual term

[0125]

[0126] Apply a static blood pressure of p0 = 100 mmHg on the lumen surface and apply a fixed displacement constraint at the outer membrane boundary. Therefore, the boundary condition residual term Can be defined as

[0127]

[0128] The physical residual term that satisfies the control mechanical equation

[0129]

[0130] where N d , N p , N b and N i are the number of sampling points in the measurement data, control equation, boundary condition, and initial condition respectively, u i and They are the displacement prediction value and reference value of the neural network at sampling point i, respectively, u int is the reference value of the initial condition and boundary condition, p i and p0 are the pressure prediction value of the neural network and the boundary condition pressure value, respectively.

[0131] Therefore, the loss function of the neural network is expressed as the sum of each residual term:

[0132]

[0133] where λ i is the weight of the corresponding loss term and is used to adjust the relative importance of the loss term.

[0134] Step 3: Use the peridynamics numerical method to accurately solve the mechanical state of the coronary plaque. During the solution process, fully consider factors such as the dynamic characteristics of blood flow, the material properties of the blood vessel wall, and the geometric shape of the plaque to obtain accurate stress-strain related information and form a high-quality training dataset.

[0135] Step 4: Training of the non-local physical information neural network model

[0136] To effectively coordinate physics and data-driven and enhance the generalization ability of the model, this project plans to use the adaptive weight adjustment method and transfer learning strategy to train the neural network. Since different loss terms (such as boundary conditions, control equations, etc.) usually have different scales and importance, the gradient-based adaptive weight adjustment method dynamically adjusts the weights of these loss terms to improve the training effect and convergence speed of the model.

[0137] First, in the pre-balancing stage, initialize all parameters of the neural network, including weights and biases. Then, assign the weight coefficients of different loss terms. In the dynamic adjustment stage, calculate the gradient of each loss term with respect to the neural network parameters θ (including weights and biases)

[0138]

[0139] where represents the L2 norm of the gradient and k represents the summation index, and n represents the total number of loss functions.

[0140] Finally, use the Adam algorithm to minimize the loss function to optimize the parameters of the neural network model:

[0141]

[0142] The transfer learning strategy first uses a parameterized coronary plaque geometric model to generate diverse plaque morphologies (such as 200 cases), and the displacement and stress fields obtained through simulation calculations are used as pre-training data. Then, a non-local physics-informed neural network model is trained on the simulated data to enable it to learn general plaque mechanical characteristics. After that, some network layers are frozen, and only the top layer of the network is fine-tuned. The parameters are updated using clinical OCT image data (such as 50 cases) to improve the model's adaptability to real plaque structures. Finally, the model is tested and verified in a clinical real scenario.

[0143] Step 5: After the model training is completed, the accuracy and stability of the model are evaluated by comparing the plaque rupture results in clinical images and comparing the numerical simulation results using various evaluation metrics (such as mean square error, mean absolute error, coefficient of determination, etc.). The coronary medical image data to be evaluated is input into the trained physics-informed neural network, and the neural network outputs the plaque rupture risk assessment result (for example, if the stress of the plaque fibrous cap is greater than 300 kPa, it is considered that the plaque has a rupture risk). The doctor formulates a corresponding treatment plan based on the evaluation result.

[0144] In the above embodiments, the mention of "this embodiment" in the specification means that the specific features, structures, or characteristics described in connection with the embodiment are included in at least some embodiments, but not necessarily all embodiments. Multiple occurrences of "this embodiment" do not necessarily all refer to the same embodiment.

[0145] In the above embodiments, although the present invention has been described in connection with specific embodiments of the present invention, many substitutions, modifications, and variations of these embodiments will be apparent to those of ordinary skill in the art based on the previous description. For example, other storage structures (such as dynamic RAM (DRAM)) can be used in the discussed embodiments. The embodiments of the present invention are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims.

[0146] This embodiment also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements any one of the methods in this embodiment.

[0147] This embodiment also provides an electronic terminal, including: a processor and a memory;

[0148] The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory so that the terminal executes any one of the methods in this embodiment.

[0149] For the computer-readable storage medium in this embodiment, those of ordinary skill in the art can understand that all or part of the steps for implementing the above method embodiments can be completed by hardware related to a computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps included in the above method embodiments; and the aforementioned storage medium includes: ROM, RAM, magnetic disk, or optical disk and other various media that can store program codes.

[0150] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication with each other. The memory is used to store a computer program, the communication interface is used for communication, and the processor and the transceiver are used to run the computer program so that the electronic terminal executes each step of the above method.

[0151] In this embodiment, the memory may include a random access memory (Random Access Memory, abbreviated as RAM), and may also include a non-volatile memory, such as at least one disk memory.

[0152] The above-mentioned processor may be a general-purpose processor, including a central processing unit (Central Processing Unit, abbreviated as CPU), a network processor (Network Processor, abbreviated as NP), etc.; it may also be a digital signal processor (Digital Signal Processing, abbreviated as DSP), an application specific integrated circuit (Application Specific Integrated Circuit, abbreviated as ASIC), a field-programmable gate array (Field-Programmable Gate Array, abbreviated as FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0153] The present invention can be used in many general-purpose or special-purpose computing system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and so on.

[0154] The present invention may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including storage devices.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention may be modified or equivalently replaced without departing from the spirit and scope of the present technical solution, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for assessing the risk of coronary artery plaque rupture based on a non-local physical information neural network, characterized in that: Including the following steps: S1: Reconstruct the mechanical constitutive equation of the plaque: Introduce the peridynamics theory, reconstruct the mechanical constitutive equation of plaque rupture, and derive the crack propagation criterion; S2: Construct a non-local physical information neural network: Take the geometric structure information and material properties of the coronary artery plaque as inputs, and take the displacement field and stress field of the plaque as outputs; Incorporate the reconstructed basic mechanical equation of the plaque as a physical constraint condition into the loss function of the neural network; S3: Collect high-resolution images of coronary artery plaques, use image segmentation algorithms to reconstruct the geometric structure of coronary artery plaques from medical image data, and use the peridynamics numerical method to solve the mechanical state of coronary artery plaques; Compose a data set with the mechanical information obtained by traditional numerical methods; S4: Model training and optimization: Divide the collected data set into a training set, a validation set, and a test set; Use the training set to train the physical information neural network, and adopt the stochastic gradient descent optimization algorithm to adjust the network parameters to minimize the loss function; During the training process, use the validation set to evaluate and optimize the model; Use the test set to test the model; S5: Risk assessment of coronary artery plaque rupture: Input the coronary artery medical image data to be evaluated into the trained non-local physical information neural network, and output the risk assessment result of coronary artery plaque rupture.

2. The method for evaluating the risk of coronary plaque rupture based on a non-local physics-informed neural network according to claim 1, wherein: The step of reconstructing the mechanical constitutive equation of the plaque described in step S1 specifically includes the following steps: S11: The mechanical behavior of the main vascular components of the coronary artery plaque conforms to an isotropic hyperelastic material and is described by an improved Mooney-Rivlin model: Where Ψ represents the strain energy function, S and σ represent the second Piola-Kirchhoff stress tensor and the Cauchy stress tensor respectively, C represents the right Cauchy-Green deformation tensor, I1, I2, and I3 represent the first, second, and third invariants of the right Cauchy-Green deformation tensor respectively, F is the deformation gradient tensor, c1, D1, and D2 represent material properties respectively, tr(*) represents the trace of a tensor, and det(*) represents the determinant; In continuum mechanics, the motion equation of the plaque is expressed as: where t represents time, ρ s and u are the density and displacement of the plaque material respectively, and b represents the body force; S12: Introduce the integral operator in peridynamics to replace the differential operator, and the integral form of the motion equation of the plaque material is expressed as: Among them, represents the influence domain of material point x, g 20 (ξ), g 02 (ξ) and g 11 (ξ) represent the peridynamic operator, ξ = x ′ -x represents the relative position between material point x ′ and x, σ(x ′ ) and σ(x) represent the stresses of material points x ′ and x respectively, and dV′ represents the volume of material point x ′ ; In the constitutive equation of the mechanical behavior of the plaque, the second Piola-Kirchhoff stress tensor and the deformation gradient tensor are expressed as: wherein represents the right Cauchy - Green deformation tensor, represents the deformation gradient tensor, Y <ξ> represents the relative position of the material points x′ and x after deformation, and K represents the shape tensor, represents the tensor product; S13: Derive the fracture criterion for plaque rupture: Write the scalar coefficient used to describe the bond fracture state as a function of time t: where w ξ represents the bond strain energy, w c represents the critical bond strain energy, h is the thickness, G0 is the critical energy release rate, and δ represents the radius of the influence domain; Damage variable at material point x is defined as the ratio of the number of broken bonds to the number of intact bonds within its domain: Among them, μ(ξ,t) represents the fracture state of the bond, taking values of 0 or 1, and the damage variable ranges from [0, 1].

3. The method for evaluating the risk of coronary plaque rupture based on a non-local physical information neural network according to claim 1, wherein: The step of constructing a non-local physical information neural network described in step S2, taking the geometric structure information and material properties of the coronary artery plaque as inputs and the displacement field and stress field of the plaque as outputs, specifically includes: Constructing a deep neural network As the core of the model, where the input variables are the spatial coordinates x = (x, y, z) and time t, and the output variables are the displacement field u(x, t) and the stress field σ(x, t). Let denote the input layer, denote the output layer; the connections between neural networks are recursively defined as: Input layer: Hidden layer: Output layer: Among them, σ represents the activation function, W l and b l represent the weights and biases of the l-th layer in the neural network respectively, W L and b L represent the weights and biases of the output layer respectively respectively represent the input space, the parameter space, and the output space.

4. The method for evaluating the risk of coronary plaque rupture based on the non-local physical information neural network according to claim 1, wherein: The step of incorporating the reconstructed basic mechanical equation of the plaque as a physical constraint condition into the loss function of the neural network specifically includes: The loss function of the neural network includes a data loss term and a physical consistency loss term; The physical consistency loss term includes an initial condition loss term, a boundary condition loss term, and a physical equation loss term; Data loss term Used to measure the difference between the model output and the observed data, expressed as: Where N d is the number of sampling points in the measurement data, and u i and are the displacement prediction value and the reference value of the neural network at the sampling point i, respectively; Initial condition loss term is as follows: where N i is the number of sampling points in the initial condition, and u int is the reference value of the initial condition; Boundary condition loss term is defined as: where N b is the number of sampling points in the boundary condition, p i and p0 are the pressure prediction value of the neural network and the boundary condition pressure value respectively; Physical equation loss term satisfying the control mechanical equation It is defined as: Among them, N p is the number of sampling points in the control equation; The loss function of the neural network is expressed as the sum of each loss term: Among them, λ i is the weight of the corresponding loss term and is used to adjust the relative importance of the loss term.

5. The method for evaluating the risk of coronary plaque rupture based on a non-local physics-informed neural network according to claim 4, wherein: In step S4, adjusting the network parameters by using the stochastic gradient descent optimization algorithm specifically includes: Initializing the parameters of all neural networks in the pre-balancing stage; allocating the weight coefficients of different loss terms; During the dynamic adjustment phase, calculate each loss term The gradient of the neural network parameters θ Adjust the weight of each loss term according to the magnitude of the gradient norm: where represents the L2 norm of the gradient , k represents the summation index, and n represents the total number of loss functions; Finally, the Adam algorithm is used to minimize the loss function to optimize the parameters of the neural network model:

6. The method for evaluating the risk of coronary plaque rupture based on a non-local physics-informed neural network according to claim 1, characterized in that: In step S4, training the neural network by using the transfer learning strategy, including: Generating diverse plaque morphologies with a parametric coronary plaque geometry model, and using the displacement and stress fields calculated by simulation as pre-training data; Training the non-local physical information neural network model on the simulated data to enable it to learn the general plaque mechanical characteristics; Freezing some network layers, only fine-tuning the top layer network, and updating the parameters with clinical OCT image data; Finally, testing and validating the model in the clinical real scenario.

7. The method for evaluating the risk of coronary plaque rupture based on a non-local physics-informed neural network according to claim 1, wherein: In step S5, after the model training is completed, the accuracy and stability of the model are evaluated by comparing the plaque rupture results in the clinical images and comparing the numerical simulation results with multiple evaluation indicators.

Images