Photoacoustic image reconstruction method, device, equipment and medium

Abstract

The invention discloses a photoacoustic image reconstruction method and device, equipment and a medium, and relates to the field of photoacoustic imaging. The method comprises the following steps: acquiring time-varying sound pressure data of a target biological tissue at a plurality of preset positions; determining a photoacoustic wave equation; embedding the photoacoustic wave equation into a loss function of the physical information neural network to obtain a total loss function; constructing a data sample set by taking a space-time coordinate formed by a preset position and time as input and taking the sound pressure data changing along with time as a label; based on the total loss function, training a physical information neural network by using the data sample set to obtain a trained physical information neural network; and inputting the space-time coordinates at the initial moment into the trained physical information neural network, and outputting sound pressure data at the initial moment so as to obtain initial sound pressure distribution. According to the invention, accurate reconstruction of the photoacoustic image can be realized.

Description

Technical Field

[0001] This application relates to the field of photoacoustic imaging, and particularly to a photoacoustic image reconstruction method, apparatus, device, and medium. Background Art

[0002] Photoacoustic Computed Tomography (PACT) is a rapidly developing biomedical imaging technology that has received extensive attention in recent years. Photoacoustic image reconstruction plays a crucial role in PACT. PACT combines the high contrast of optical imaging and the high resolution of ultrasound. As an emerging non-invasive biomedical imaging modality, it has great potential in the frontier exploration of medical imaging and clinical application research.

[0003] Existing PACT methods mainly include Time Reversal (TR), Filtered Back-Projection (FBP), and Deep Learning (DL).

[0004] The time reversal method is based on the principle of time-reversal symmetry of the wave equation, and realizes image reconstruction by inversely solving the forward problem of sound field propagation. Its physical implementation approach is: after time-reversing the received acoustic pressure signal in the time domain as the boundary condition, it propagates backward in the numerical model to the computational domain, and reconstructs the source distribution through the acoustic energy convergence effect. This method utilizes the reversible characteristic of wave propagation, and was initially applied to the research of precise ultrasound focusing in inhomogeneous media, with the dual advantages of adaptability to complex acoustic media and compatibility with irregular detection geometries. Specifically, the time reversal method can utilize the known limited field of view data, restore the source point information through the backward propagation of the signal, and effectively perform image reconstruction.

[0005] Generally speaking, due to its relatively simple operation, high measurement accuracy, and fast calculation speed, the TR method has become the most commonly used photoacoustic image reconstruction method at present. However, the reconstruction result of this method is still affected by factors such as the number and bandwidth of detectors and boundary conditions. The reconstructed image is prone to artifacts and is very sensitive to noise.

[0006] The basic idea of the filtered backprojection method is to first filter the received photoacoustic data to remove noise and enhance the signal. Then, the filtered data is converted into a two-dimensional or three-dimensional image of the target object using the backprojection algorithm. The FBP algorithm has high computational efficiency, is simple to implement, and can quickly generate high-quality images when there is complete projection angle coverage. However, the applicability of this method is limited by physical assumptions. The FBP algorithm assumes that the sound speed in the medium is uniform, while the acoustic heterogeneity of biological tissues may cause wavefront distortion, resulting in a decrease in imaging quality. In addition, the FBP algorithm has high requirements for data integrity and requires a long scanning time to obtain sufficient projection data, which to a certain extent limits its application in real-time imaging.

[0007] Deep learning has been widely applied to various image processing and pattern recognition tasks, especially showing excellent performance in extracting temporal and spatial features during wave propagation. Through the characteristics of automated learning and data-driven, neural networks can effectively capture complex patterns in photoacoustic wave propagation. However, the training of deep learning usually relies on a large amount of high-quality data sets, usually requiring the training data to be paired, and it is relatively difficult to obtain a large amount of paired data in practice.

[0008] Reference images for network training in DL methods, such as full-bandwidth photoacoustic images, are often difficult to obtain, and the network can only be trained using simulated data sets. This training method based on simulated and phantom data has certain differences from the actual situation of real human tissues, resulting in limitations in the generalization and interpretability of the model when applied to real humans. In addition, the black-box characteristics of deep learning models make them difficult to explain and analyze, further limiting their potential in clinical translation. Therefore, although deep learning has certain advantages in photoacoustic computed tomography, its limitations also significantly hinder its wide application.

[0009] The above PACT methods can be summarized into the following problems (1)-(5).

[0010] (1) Traditional PACT methods rely on full-angle detection and densely sampled signals. However, this is very difficult to achieve in practice, leading to the ill-posed problem of reconstructing images from incomplete projections.

[0011] (2) Traditional PACT methods, such as those based on time reversal method, filtered backprojection method, etc., are limited by the inherent limitations of their algorithms, and artifacts are likely to appear in the reconstructed images. For example, the conclusion that the impulse response of the FBP algorithm is a delta function is only valid in ideal cases, such as unbounded plane, unbounded cylinder, and closed spherical imaging geometries and infinite detector bandwidth. However, such ideal cases can never be fully satisfied in real experiments. Deviating from these ideal cases will contaminate the PSF (Point Spread Function), ultimately reducing the image quality.

[0012] (3) The traditional PACT method is very sensitive to noise and it is difficult to reconstruct a smooth sound pressure distribution map. Therefore, it is necessary to develop an inversion algorithm for the spatial distribution of sound pressure with strong anti-noise ability.

[0013] (4) At present, the photoacoustic tomography technology based on deep learning can effectively improve the quality of image reconstruction. However, the performance of the DL method highly depends on the quality and quantity of training data. Usually, it is required that the training data is paired, and it is difficult to obtain a large amount of paired data in practice.

[0014] (5) In the DL method, reference images for network training, such as full-bandwidth photoacoustic images, are often difficult to obtain, and the network can only be trained using simulated data sets. Since the experimental conditions are usually much more complex than simulations, training the network only with simulated data sets will lead to a decline in reconstruction performance and poor robustness. At the same time, the training method that completely relies on data-driven has defects such as poor interpretability and poor generalization.

[0015] In summary, the existing PACT methods cannot accurately reconstruct photoacoustic images. SUMMARY OF THE INVENTION

[0016] The purpose of this application is to provide a photoacoustic image reconstruction method, device, equipment and medium, which can achieve accurate reconstruction of photoacoustic images.

[0017] To achieve the above purpose, this application provides the following solutions.

[0018] In the first aspect, this application provides a photoacoustic image reconstruction method, including: obtaining the sound pressure data of the target biological tissue changing with time at multiple preset positions; determining the photoacoustic wave equation; embedding the photoacoustic wave equation into the loss function of the physics-informed neural network to obtain the total loss function; constructing a data sample set with the spatio-temporal coordinates composed of the preset positions and time as the input and the sound pressure data changing with time as the label; training the physics-informed neural network based on the total loss function using the data sample set to obtain a trained physics-informed neural network; inputting the spatio-temporal coordinates at the initial moment into the trained physics-informed neural network, and outputting the sound pressure data at the initial moment, so as to obtain the initial sound pressure distribution.

[0019] Optionally, determining the photoacoustic wave equation specifically includes: establishing three coupled acoustic equations as , and ; where is the spatial position, and respectively represent the abscissa and ordinate of the spatial position, is the velocity of the particle at the spatial position at time , is the particle mass density, is the gradient operator; is the spatial position of the particle mass density at, is the spatial position of the sound speed at, ; is the coefficient of thermal expansion, is at time located at the spatial position of the sound pressure data; is the specific heat capacity at constant pressure, at time the particle is located at the spatial position of the temperature, represents the thermal energy deposited per unit volume per unit time; According to three coupled acoustic equations, the photoacoustic wave equation is determined as: ; In the formula, is the Laplace operator.

[0020] Optionally, the initial conditions satisfied by the photoacoustic wave equation are: and ; In the formula, is the sound pressure data at the spatial position at the initial moment, is the sound pressure data at the spatial position obtained by the photoacoustic effect generated by pulsed laser excitation.

[0021] Optionally, embed the photoacoustic wave equation into the loss function of the physics-informed neural network to obtain the total loss function, specifically including: Establish a data-driven loss function as: ; In the formula, is the data-driven loss, is the mean square error, is the sound pressure data of the target biological tissue in the historical spatio-temporal region, is the sound pressure data output by the physics-informed neural network; Convert the photoacoustic wave equation into a homogeneous wave equation: ; According to the homogeneous wave equation, establish a physics-information loss function as: ; In the formula, is the physics-information loss; According to the initial conditions satisfied by the photoacoustic wave equation, establish an initial-condition loss function as: ; In the formula, is the initial-condition loss; According to the data-driven loss function, the physics-information loss function and the initial-condition loss function, obtain the total loss function as: ; In the formula, is the total loss function, is the weight of the physical information loss, is the weight of the data-driven loss, is the weight of the initial condition loss.

[0022] Optionally, the physics-informed neural network includes: an input layer, an output layer, and multiple hidden layers; the input layer includes 3 input parameters; each hidden layer includes 100 neurons; the output layer includes 1 output parameter.

[0023] Optionally, when training the physics-informed neural network using a data sample set: at every preset number of rounds, multiple samples are extracted from the entire spatio-temporal domain using the Latin hypercube sampling method for training.

[0024] Optionally, the physics-informed neural network is a pre-trained physics-informed neural network.

[0025] In a second aspect, the present application provides a photoacoustic image reconstruction device, including: a data acquisition module, an equation determination module, a loss function establishment module, a sample set construction module, a training module, and an application module.

[0026] The data acquisition module is configured to acquire the acoustic pressure data of the target biological tissue changing with time at multiple preset positions. The equation determination module is configured to determine the photoacoustic wave equation. The loss function establishment module is configured to embed the photoacoustic wave equation into the loss function of the physics-informed neural network to obtain the total loss function. The sample set construction module is configured to construct a data sample set with the spatio-temporal coordinates composed of the preset positions and time as the input and the acoustic pressure data changing with time as the label. The training module is configured to train the physics-informed neural network based on the total loss function using the data sample set to obtain a trained physics-informed neural network. The application module is configured to input the spatio-temporal coordinates at the initial moment into the trained physics-informed neural network and output the acoustic pressure data at the initial moment, so as to obtain the initial acoustic pressure distribution.

[0027] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the photoacoustic image reconstruction method described in any one of the above.

[0028] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the photoacoustic image reconstruction method described in any one of the above.

[0029] According to the specific embodiments provided by the present application, the present application has the following technical effects.

[0030] The present application provides a photoacoustic image reconstruction method, apparatus, device and medium, which applies a physics-informed neural network to photoacoustic computed tomography. The physics-informed neural network can extract more spatio-temporal features of photoacoustic waves, and in the case of limited view and sparse sampling, it can maximize the avoidance of estimation errors caused by incomplete projection data. Compared with traditional deep learning methods, encoding the photoacoustic wave equation as a loss function enables the physics-informed neural network to no longer learn data features in a chaotic manner, but to solve parameters under the guidance of physical laws. At the same time, the physics-informed neural network no longer simply relies on the stacking of a large amount of training data, greatly reducing the requirement for the amount of training data. Since the physics-informed neural network combines physical information equations and has a good filtering effect on data, it has good anti-noise ability compared with traditional methods, thus enabling the accurate reconstruction of photoacoustic images. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0032] Figure 1 It is a schematic flowchart of a photoacoustic image reconstruction method provided by an embodiment of the present application.

[0033] Figure 2 It is a schematic flowchart of the training of the physics-informed neural network provided by another embodiment of the present application.

[0034] Figure 3 It is a schematic diagram of the functional modules of a photoacoustic image reconstruction apparatus provided by an embodiment of the present application.

[0035] Figure 4 It is a schematic diagram of the structure of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some, rather than all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present application belong to the scope of protection of the present application.

[0037] To make the above objects, features and advantages of the present application more obvious and understandable, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0038] As an emerging method, Physics-Informed Neural Networks (PINN) combines the powerful modeling ability of deep learning and the constraints of physical equations, and is gradually becoming a powerful tool for solving tissue acoustic inversion problems. By encoding the governing equations of physical problems, such as partial differential equations, into a fully connected neural network, PINN incorporates physical laws as part of the neural network training. It not only retains the strong learning ability of deep learning models but also combines physical laws, making PINN more robust and physically interpretable when facing complex problems. PINN has gradually become an effective tool for solving forward and inverse problems, especially in cases involving complex multi-physics field couplings, showing strong application prospects.

[0039] Since PINN can more accurately capture the physical characteristics in the wave process through the constraints of physical equations, this method is expected to overcome many limitations in the existing technologies. Specifically, when dealing with tissues with strong heterogeneity, PINN can more accurately describe the complex situations in photoacoustic wave propagation, thus providing a new solution for the accurate estimation of the initial acoustic pressure distribution in photoacoustic imaging. With the in-depth research of the PINN method in this field, it has the potential to significantly improve the clinical application value of photoacoustic computed tomography technology and promote its wide application in precision medicine.

[0040] In view of the above, in an exemplary embodiment, as Figure 1 shown, a photoacoustic image reconstruction method is provided, including the following steps 101 to step 106.

[0041] Step 101: Obtain the acoustic pressure data of the target biological tissue changing with time at multiple preset positions.

[0042] Step 102: Determine the photoacoustic wave equation.

[0043] Step 103: Embed the photoacoustic wave equation into the loss function of the physics-informed neural network to obtain the total loss function.

[0044] Step 104: Construct a data sample set with the spatio-temporal coordinates composed of the preset positions and time as the input and the acoustic pressure data changing with time as the label.

[0045] Step 105: Based on the total loss function, use the data sample set to train the physics-informed neural network to obtain a trained physics-informed neural network.

[0046] Step 106: Input the spatio-temporal coordinates at the initial moment into the trained physics-informed neural network, and output the acoustic pressure data at the initial moment, thereby obtaining the initial acoustic pressure distribution.

[0047] By implementing the above steps 101 to 106, applying PINN to photoacoustic computed tomography and embedding the photoacoustic wave equation into PINN, the problem of poor image quality in sparse sampling reconstruction of existing PACT methods is solved, and at the same time, the PINN of this application has stronger anti-noise ability.

[0048] In another exemplary embodiment of this application, in the above step 101, a pulsed laser is used to emit laser pulses to the target biological tissue. The target biological tissue expands due to heat and generates ultrasonic signals, and an ultrasonic transducer is used to collect the ultrasonic signals. The data collected by the ultrasonic transducer is the acoustic pressure signal that changes with time collected at the spatial position of the ultrasonic transducer. Using the position and time of the ultrasonic transducer, spatio-temporal coordinates can be formed.

[0049] In another exemplary embodiment of this application, the generation and propagation of photoacoustic waves can be mathematically modeled using the photoacoustic wave equation. Then the above step 102 can be replaced by the following steps 201 to 202.

[0050] Step 201: Establish three coupled acoustic equations as follows: Linearized equation of motion: ; Linearized continuity equation: ; Thermoelastic equation: ; Where, is the spatial position, and respectively represent the abscissa and ordinate of the spatial position, is the velocity of the particle at the spatial position at time , is the particle mass density, is the gradient operator, is the particle mass density at the spatial position , is the sound speed at the spatial position , ; is the coefficient of thermal expansion, is the acoustic pressure data at the spatial position at time ; is the specific heat capacity at constant pressure, is the temperature of the particle at the spatial position at time , represents the heat energy deposited per unit volume per unit time.

[0051] Step 202: Determine the photoacoustic wave equation according to three coupled acoustic equations as follows: ; wherein, is the Laplace operator.

[0052] Exemplarily, the initial conditions satisfied by the photoacoustic wave equation are: ; ; wherein, is the sound pressure data at the spatial position at the initial moment, is the sound pressure data at the spatial position generated by the photoacoustic effect excited by the pulsed laser.

[0053] In another exemplary embodiment of the present application, the PINN network structure designed in the present application is as shown in Figure 2 . The PINN includes three input parameters ( , and ), 8 hidden layers and an output parameter . Each hidden layer includes 100 neurons. The output of the PINN is trained to learn to fit the input data and solve the photoacoustic wave equation. By fitting the input data (the sound pressure data collected by the ultrasonic transducer), the PINN will infer the propagation process of the photoacoustic wave in the entire space-time region (i.e., the sound pressure distribution in the entire space-time region). Setting to 0, the sound pressure distribution at the initial moment can be obtained, which is the ultimate goal of the photoacoustic image reconstruction in the present application.

[0054] In another exemplary embodiment of the present application, the weights and biases of the network will be updated through the training process of the backpropagation algorithm to achieve the purpose of minimizing the loss function. In order to enable the fully connected feedforward neural network to learn the complex mapping relationship between the input parameters and the output parameters, Tanh is selected as the activation function, and its continuity and smoothness characteristics help to accurately simulate the behavior of the physical field while ensuring the numerical stability of the model during the training process. sin or the like can also be used as the activation function. The total loss function of the PINN is composed of a data-driven loss function, a physical information loss function, and an initial condition loss function. Then, the above step 103 can be replaced by the following steps 301 to 305.

[0055] Step 301: Establish the data-driven loss function as: ; wherein, is the data-driven loss, is the mean square error, is the sound pressure data of the target biological tissue in the historical spatio-temporal region, is the sound pressure data output by the physics-informed neural network.

[0056] Step 302: Convert the photoacoustic wave equation into a homogeneous wave equation: .

[0057] According to Duhamel's principle, the inhomogeneous photoacoustic wave equation is equivalent to the homogeneous wave equation, plus the initial conditions, as follows: ; is the time parameter, representing the laser pulse heating time; represents the sound pressure generated by the pulsed laser excitation to produce the photoacoustic effect, that is, the initial sound pressure generated after laser heating.

[0058] The sound pressure generated by the pulsed laser excitation to produce the photoacoustic effect can be expressed by the following formula: ; where, represents the initial distribution of the optical absorption density. represents the Gruneisen constant.

[0059] Due to the time parameter of the pulsed laser excitation is generally in the nanosecond level, therefore, for solving the photoacoustic wave equation, it can be ignored; the photoacoustic wave equation to be solved is converted into the following formula: .

[0060] Embed this equation into the physics-informed neural network to solve the objective function: .

[0061] Step 303: According to the homogeneous wave equation, establish the physics-informed loss function as: ; in the formula, is the physics-informed loss.

[0062] Step 304: According to the initial conditions satisfied by the photoacoustic wave equation, establish the initial condition loss function as: ; in the formula, is the initial condition loss.

[0063] Step 305: According to the data-driven loss function, the physics-informed loss function and the initial condition loss function, obtain the total loss function as: ; in the formula, is the total loss function, is the weight of the physics-informed loss, is the weight of the data-driven loss, The weight for the initial condition loss. , and are used to balance the influence of the three-part loss function on network training to ensure their effective convergence during training.

[0064] In another exemplary embodiment of the present application, to improve the computational efficiency and ensure the representativeness of the training data, when training the physics-informed neural network using a data sample set, multiple samples are extracted from the entire spatio-temporal domain at every preset number of epochs using the Latin Hypercube Sampling (LHS) method for training. The Latin Hypercube Sampling method helps to maximize the coverage of the samples and reduce the sampling bias by ensuring uniform sampling in each dimension. This way of sampling training not only improves the utilization efficiency of the data but also accelerates the network training process.

[0065] Exemplarily, the PINN is set to extract 20,000 points from the entire spatio-temporal domain for training at every 1000 epochs. Specifically, referring to Figure 2 , the acoustic pressure data is obtained by collecting ultrasonic signals through ultrasonic transducers, and the positions and acquisition times of the ultrasonic transducers form spatio-temporal coordinates , and are used as the first batch input of the physics-informed neural network. The physics-informed neural network outputs , the input data-driven loss . The data points in the second batch input of the physics-informed neural network are extracted from the entire spatio-temporal domain ( ) at every 1000 epochs for 20,000 points for training. The sampling process uses the Latin Hypercube Sampling method. The second batch output of the physics-informed neural network is input into the physics-informed loss function. The used for the initial condition loss function and the physics-informed loss function are not the same, not the acoustic pressure at the same position. It can be seen that the input of the data sample set is the spatio-temporal domain position of the collected data, while the input of the physics-informed loss function is the spatio-temporal coordinates collected within the entire spatio-temporal domain, so that the acoustic pressure at any position can be obtained.

[0066] To terminate the training when the total loss function reaches convergence, the total number of iterations is defined to be equal to 5×10 4 , the learning rate is equal to 10 -4 , = = 0.1, = 1. When the data noise is large, = = = 1, it has a better filtering effect on data. PINN uses the Adam optimizer to update parameters to minimize the total loss function. When or is reached, the calculation is stopped. is the initial total loss ratio. The spatio-temporal coordinates at the initial moment are input into the trained physics-informed neural network, and the network output is the sound pressure data at the initial moment. The method of this application cleverly utilizes the similarity between the backpropagation characteristics of the neural network and the chain rule of derivative, and effectively uses the characteristics of automatic differentiation to accurately solve the partial differential equation (photoacoustic wave equation).

[0067] In another exemplary embodiment of this application, during the training of PINN, the weights and biases of the network are randomly initialized, which may be a time-consuming process. To better apply the method of this application to clinical practice and accelerate the network training speed, transfer learning is used to accelerate the calculation. PINN is pre-trained on existing similar data, and the pre-training process is the same as the formal training process, forcing the network to satisfy the solution process of the photoacoustic wave equation. Then, the data sample set of this application is input into the pre-trained model for training, so that the data sample set can find the gradient descent direction at the fastest speed after being input into the pre-trained model. After testing, it is found that the pre-trained PINN can reduce the training time by more than 80% without reducing the imaging quality, greatly accelerating its inversion solution speed, and can meet the clinical use requirements to the greatest extent.

[0068] Compared with traditional photoacoustic imaging methods, PINN can extract more spatio-temporal characteristics of photoacoustic waves. In the case of limited view and sparse sampling, it can largely avoid the estimation error caused by incomplete projection data. In addition, since PINN combines the physical information equation and has a good filtering effect on data, it has good anti-noise ability compared with traditional methods. Compared with traditional deep learning methods, this application encodes the physical information equation as a loss function, so that the network does not learn data features disorderly, but guides by physical laws to solve parameters. This method of embedding physical information makes the neural network no longer simply rely on the stacking of a large amount of training data, greatly reducing the demand for the amount of training data, and even can achieve the accurate solution of single-case data. At the same time, this application can view the solution process of each internal parameter, greatly improving the interpretability of deep learning. Therefore, the method of this application has application advantages far beyond the prior art.

[0069] In another exemplary embodiment of this application, the PINN network structure of this application can also use B-spline functions. The parameters of the B-spline functions can be updated through the training process, and the activation function corresponding to each neuron will change continuously during the training process, and the solution of the partial differential equation can also be achieved.

[0070] In step 106 above, any spatio-temporal coordinate in the spatio-temporal space at the initial moment ( ) is input into the trained physics-informed neural network, and the sound pressure data of any spatio-temporal coordinate in the spatio-temporal space at the initial moment is output. The sound pressure data of all spatio-temporal coordinates in the spatio-temporal space can form the initial sound pressure distribution. Since the image reconstruction of photoacoustic imaging is to invert the initial sound pressure distribution from the detected ultrasonic signals through mathematical algorithms, after obtaining the initial sound pressure distribution in step 106 above, the photoacoustic image reconstruction is achieved.

[0071] Based on the same inventive concept, an embodiment of the present application also provides a photoacoustic image reconstruction device for implementing the photoacoustic image reconstruction method involved above. The implementation solutions provided by this device to solve problems are similar to the implementation solutions described in the above method. Therefore, the specific limitations in one or more embodiments of the photoacoustic image reconstruction device provided below can refer to the limitations on the photoacoustic image reconstruction method in the above text, and will not be repeated here.

[0072] In an exemplary embodiment, as Figure 3 shown, a photoacoustic image reconstruction device is provided, including: a data acquisition module, an equation determination module, a loss function establishment module, a sample set construction module, a training module, and an application module.

[0073] The data acquisition module is configured to acquire the sound pressure data of the target biological tissue changing with time at multiple preset positions. The equation determination module is configured to determine the photoacoustic wave equation. The loss function establishment module is configured to embed the photoacoustic wave equation into the loss function of the physics-informed neural network to obtain the total loss function. The sample set construction module is configured to construct a data sample set with the spatio-temporal coordinates composed of the preset positions and time as the input and the sound pressure data changing with time as the label. The training module is configured to train the physics-informed neural network based on the total loss function using the data sample set to obtain a trained physics-informed neural network. The application module is configured to input the spatio-temporal coordinates at the initial moment into the trained physics-informed neural network and output the sound pressure data at the initial moment, so as to obtain the initial sound pressure distribution.

[0074] In an exemplary embodiment, a computer device is provided. This computer device can be a server or a terminal, and its internal structure diagram can be as Figure 4As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the initial sound pressure distribution. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a photoacoustic image reconstruction method.

[0075] Those skilled in the art can understand that Figure 4 the structure shown in the figure is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0076] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0077] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties.

[0078] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memories can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0079] The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0080] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0081] In this text, specific examples are used to illustrate the principles and implementation manners of the present application. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A photoacoustic image reconstruction method, characterized in that, Comprising: Obtaining the acoustic pressure data of the target biological tissue changing with time at multiple preset positions; Determining the photoacoustic wave equation; Embedding the photoacoustic wave equation into the loss function of the physics-informed neural network to obtain the total loss function; Constructing a data sample set with the spatio-temporal coordinates composed of the preset positions and time as the input and the acoustic pressure data changing with time as the label; Training the physics-informed neural network based on the total loss function by using the data sample set to obtain a trained physics-informed neural network; Inputting the spatio-temporal coordinates at the initial moment into the trained physics-informed neural network and outputting the acoustic pressure data at the initial moment, thereby obtaining the initial acoustic pressure distribution.

2. The photoacoustic image reconstruction method according to claim 1, wherein Determining the photoacoustic wave equation, specifically including: Three coupled acoustic equations are established as , and ; where is the spatial position, and represent the abscissa and ordinate of the spatial position respectively, is the velocity of the particle at spatial position at time , is the particle mass density, is the gradient operator; is the particle mass density at spatial position , is the speed of sound at spatial position , ; is the coefficient of thermal expansion, is the acoustic pressure data at spatial position at time ; is the specific heat capacity at constant pressure, is the temperature of the particle at spatial position at time , represents the thermal energy deposited per unit volume per unit time; Based on three coupled acoustic equations, the photoacoustic wave equation is determined as follows: ; where is the Laplace operator.

3. The photoacoustic image reconstruction method according to claim 2, wherein The initial conditions satisfied by the photoacoustic wave equation are: ； ； In the formula, is the sound pressure data at the initial moment located at the spatial position . is the sound pressure data at the spatial position obtained by the photoacoustic effect generated by pulsed laser excitation.

4. The photoacoustic image reconstruction method according to claim 3, wherein Embedding the photoacoustic wave equation into the loss function of the physics-informed neural network to obtain the total loss function, specifically including: The data-driven loss function is established as follows: ; where is the data-driven loss, is the mean square error, is the sound pressure data of the target biological tissue in the historical spatio-temporal region, is the sound pressure data output by the physics-informed neural network; Convert the photoacoustic wave equation into a homogeneous wave equation: ; Based on the homogeneous wave equation, a physical information loss function is established as follows: ; where is the physical information loss; According to the initial conditions satisfied by the photoacoustic wave equation, an initial condition loss function is established as follows: ; where is the initial condition loss; According to the data-driven loss function, the physical-information loss function, and the initial-condition loss function, the total loss function is obtained as follows: ; where is the total loss function, is the weight of the physical-information loss, is the weight of the data-driven loss, is the weight of the initial-condition loss.

5. The photoacoustic image reconstruction method according to claim 1, wherein The physics-informed neural network includes: an input layer, an output layer and multiple hidden layers; The input layer includes 3 input parameters; Each hidden layer includes 100 neurons; The output layer includes 1 output parameter.

6. The photoacoustic image reconstruction method according to claim 1, wherein When training the physics-informed neural network by using the data sample set: Every preset number of rounds, multiple samples are extracted from the entire spatio-temporal domain by using the Latin hypercube sampling method for training.

7. The photoacoustic image reconstruction method according to claim 1, wherein The physics-informed neural network is a pre-trained physics-informed neural network.

8. An optoacoustic image reconstruction device, characterized in that Comprising: A data acquisition module, configured to obtain the acoustic pressure data of the target biological tissue changing with time at multiple preset positions; An equation determination module, configured to determine the photoacoustic wave equation; A loss function establishment module, configured to embed the photoacoustic wave equation into the loss function of the physics-informed neural network to obtain the total loss function; A sample set construction module, configured to construct a data sample set with the spatio-temporal coordinates composed of the preset positions and time as the input and the acoustic pressure data changing with time as the label; A training module, configured to train the physics-informed neural network based on the total loss function by using the data sample set to obtain a trained physics-informed neural network; An application module, configured to input the spatio-temporal coordinates at the initial moment into the trained physics-informed neural network and output the acoustic pressure data at the initial moment, thereby obtaining the initial acoustic pressure distribution.

9. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the photoacoustic image reconstruction method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the photoacoustic image reconstruction method according to any one of claims 1-7.

Images