Photoacoustic tomography method and apparatus for reconstructing absorption coefficient based on implicit neural representation

Abstract

The present disclosure provides a photoacoustic tomography absorption coefficient reconstruction method and device based on implicit neural representation, and relates to the technical field of image processing. The method of the present disclosure comprises: obtaining a first absorption coefficient distribution based on an SBDC method using a detector measurement signal and embedding the INR network as prior information; optimizing the parameters of the implicit neural representation network using a finite element method forward solver and a PAT physical model, and using the INR network after optimization of the parameters to evaluate the absorption coefficient value at each spatial position in the photoacoustic imaging area to generate a continuous third absorption coefficient distribution. The present disclosure can effectively improve the absorption coefficient reconstruction accuracy in application scenarios such as deep tissue.

Description

Technical Field

[0001] This disclosure relates to the field of image processing technology, and in particular to a method and apparatus for reconstructing photoacoustic absorption coefficients based on implicit neural representation. Background Technology

[0002] The goal of quantitative photoacoustic tomography (qPAT) is to reconstruct the absorption coefficient distribution of a tissue, and the relationship between this distribution and the initial photoacoustic image (i.e., the initial photoacoustic sound pressure image) depends on the illumination distribution (LF). However, the absorption coefficient reconstruction process is extremely challenging because it involves complex inversion of the light transport equations, and its nonlinear characteristics and underdetermined problems lead to instability of the solutions.

[0003] In practical applications, unknown light scattering in deep tissues introduces errors in the simulation of illumination distribution and the reconstruction of the initial pressure map, further exacerbating the instability of the solution caused by the underdetermined problem. This results in the reconstructed absorption coefficient distribution deviating from the true physiological parameters, failing to accurately reflect the true physiological parameters, and limiting the application of absorption coefficient reconstruction.

[0004] To improve reconstruction accuracy, methods such as segmentation-based direct correction (SBDC) and supervised learning models based on deep neural networks have been proposed. However, these related technologies have problems such as large training data requirements and low reconstruction accuracy of absorption coefficients in scenarios such as deep tissues. Summary of the Invention

[0005] In view of this, the present disclosure provides a method and apparatus for reconstructing photoacoustic absorption coefficients based on implicit neural representation.

[0006] According to a first aspect of this disclosure, a method for reconstructing photoacoustic tomography absorption coefficients based on implicit neural representations is provided, the method comprising:

[0007] The first absorption coefficient distribution is obtained by using the detector to measure the signal based on the SBDC method, and the first absorption coefficient distribution is embedded as prior information into the implicit neural representation network.

[0008] Optimizing the parameters of the implicit neural representation network includes iteratively performing the following steps until a preset convergence condition is met: using the implicit neural representation network to evaluate absorption coefficient values ​​at various spatial locations within the photoacoustic imaging region to generate a continuous second absorption coefficient distribution; calculating the luminous flux field under the second absorption coefficient distribution using a finite element method forward solver to obtain the luminous flux distribution; simulating the predicted detector signal based on the luminous flux distribution and the second absorption coefficient distribution using a photoacoustic tomography physical model; and minimizing the difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network.

[0009] The implicit neural representation network with optimized parameters is used to evaluate absorption coefficient values ​​at various spatial locations within the photoacoustic imaging region to generate a continuous third absorption coefficient distribution.

[0010] In some embodiments of the first aspect of this disclosure, minimizing the difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network includes: calculating a loss value based on the predicted detector signal and the detector measurement signal based on a predetermined target loss function, using the loss value to perform backpropagation in the implicit neural representation network to obtain a gradient, and updating the parameters of the implicit neural representation network based on the gradient.

[0011] In some embodiments of the first aspect of this disclosure, the target loss function is as follows:

[0012]

[0013] in, Indicates the loss value. This indicates the detector's measurement signal. This represents the predictor signal, where r represents the grid coordinates. This represents the distribution of the second absorption coefficient. Indicates the distribution of luminous flux, " " indicates dot product.

[0014] In some embodiments of the first aspect of this disclosure, the method further includes: before using the implicit neural representation network to evaluate the absorption coefficient values ​​at the spatial locations of the photoacoustic imaging region to generate a continuous second absorption coefficient distribution, encoding each spatial location of the photoacoustic imaging region using a hash encoding method.

[0015] In some embodiments of the first aspect of this disclosure, optimizing the parameters of the implicit neural representation network further includes: before using the implicit neural representation network to evaluate absorption coefficient values ​​at various spatial locations within the photoacoustic imaging region to generate a continuous second absorption coefficient distribution, discretizing the spatial domain of the reconstructed object into a two-dimensional grid, wherein the spatial coordinates of a grid node in the two-dimensional grid represent a spatial location within the photoacoustic imaging region.

[0016] In some embodiments of the first aspect of this disclosure, optimizing the parameters of the implicit neural representation network further includes: adaptively refining the two-dimensional mesh according to the second absorption coefficient distribution obtained from the previous reconstruction during iterations at preset intervals or in one or more randomly selected iterations, such that the mesh node density of the two-dimensional mesh is positively correlated with its absorption coefficient value.

[0017] In some embodiments of the first aspect of this disclosure, the implicit neural representation network is a multilayer perceptron.

[0018] According to a second aspect of this disclosure, a photoacoustic tomography absorption coefficient reconstruction device based on implicit neural representation is provided, the photoacoustic tomography absorption coefficient reconstruction device based on implicit neural representation comprising:

[0019] The prior embedding unit is used to obtain the first absorption coefficient distribution by using the detector measurement signal based on the SBDC method, and embed the first absorption coefficient distribution as prior information into the implicit neural representation network.

[0020] The parameter optimization unit is used to optimize the parameters of the implicit neural representation network, including iteratively executing the following steps until a preset convergence condition is met: using the implicit neural representation network to evaluate the absorption coefficient values ​​at each spatial location within the photoacoustic imaging region to generate a continuous second absorption coefficient distribution; calculating the light flux field under the second absorption coefficient distribution using a finite element method forward solver to obtain the light flux distribution; simulating the predicted detector signal based on the light flux distribution and the second absorption coefficient distribution using a photoacoustic tomography physical model; and minimizing the difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network.

[0021] The reconstructed execution unit is used to evaluate the absorption coefficient values ​​at various spatial locations within the photoacoustic imaging region using the implicit neural representation network with optimized parameters to generate a continuous third absorption coefficient distribution.

[0022] According to a third aspect of this disclosure, an electronic device is provided, comprising: one or more processors and a memory storing a program, the program including instructions that, when executed by the processor, cause the processor to perform the methods described above.

[0023] According to a fourth aspect of this disclosure, a computer-readable storage medium storing a program is provided, the program including instructions that, when executed by one or more processors, cause the processors to perform the methods described above.

[0024] As can be seen from the above technical solutions, the embodiments disclosed herein can effectively improve the accuracy of absorption coefficient reconstruction in application scenarios such as deep tissues without the need for training data, and are low in cost while having good physical interpretability. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A flowchart of a photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation provided in this disclosure embodiment;

[0027] Figure 2 This diagram illustrates a specific implementation process of the photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation provided in this embodiment of the disclosure. Figure 2 Figure a is an example of the specific implementation process of step 102 in the photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation provided in the embodiments of this disclosure. Figure 2 b is a flowchart illustrating the prior embedding process in the INR network of the photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation provided in this embodiment of the present disclosure. Figure 2 c is an example diagram of adaptive refinement of two-dimensional mesh in the photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation provided in the embodiments of this disclosure;

[0028] Figure 3 This is a schematic diagram comparing the qualitative results of reconstructing absorption coefficient images on photoacoustic digital brain images with embodiments of this disclosure and related technologies;

[0029] Figure 4 This is a schematic diagram comparing the qualitative results of reconstructing absorption coefficient images on phantom data using the methods and related techniques of the embodiments of this disclosure;

[0030] Figure 5 A schematic diagram of the structure of the photoacoustic tomography absorption coefficient reconstruction device based on implicit neural representation provided in the embodiments of this disclosure;

[0031] Figure 6 A schematic structural block diagram of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0032] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0033] The terminology used in the embodiments of this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of this disclosure. The singular forms “a,” “the,” and “the” as used in the embodiments of this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0034] Depending on the context, words such as "if," "when," etc., used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrases "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0035] The following is a brief explanation of the relevant technologies and their shortcomings.

[0036] 1) Segmentation-Based Direct Correction (SBDC): This method obtains the absorption coefficient distribution map by dividing the PAT image by the light flux distribution (LF) map. However, this method requires manually setting the initial optical parameters of the segmented region, making it complex and experience-dependent. Although subsequent automatic segmentation algorithms have been proposed to reduce human intervention, the low soft tissue contrast in the PAT image leads to insufficient segmentation accuracy. Furthermore, the assumption of a uniform absorption coefficient distribution within the segmented region introduces estimation errors, resulting in low accuracy in absorption coefficient reconstruction.

[0037] 2) Absorption Coefficient Reconstruction Based on Deep Neural Networks: This method trains an end-to-end deep neural network using a large amount of pre-collected training data to obtain a supervised learning model based on deep neural networks. Absorption coefficients are then reconstructed using this supervised learning model. This method requires a large amount of training data, but obtaining a large number of accurately labeled LF maps in actual qPAT is costly and difficult, limiting its application in scenarios such as deep tissues and resulting in lower reconstruction accuracy.

[0038] In view of this, the present disclosure provides the following method and apparatus for reconstructing photoacoustic tomography absorption coefficients based on implicit neural representation. The absorption coefficient distribution reconstructed based on the SBDC method is embedded as prior information into an INR network. Subsequently, the parameters of the INR network are optimized using a finite element method (FEM) forward solver and a photoacoustic tomography (PAT) physical model. Finally, the optimized INR network is used to reconstruct the absorption coefficient distribution. This disclosure requires no training data and can achieve absorption coefficient reconstruction in an unsupervised manner. Experimental verification shows that this disclosure can accurately recover the absorption coefficient, and the accuracy of absorption coefficient reconstruction in both quantitative and qualitative evaluations is significantly improved compared to related technologies.

[0039] Figure 1 This diagram illustrates a flowchart of a photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation provided in an embodiment of this disclosure. See also... Figure 1 The photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation in this disclosure may include the following steps:

[0040] Step 101: Based on the SBDC method, the first absorption coefficient distribution is obtained by using the detector measurement signal, and the first absorption coefficient distribution is embedded as prior information into the INR network.

[0041] Step 102, optimizing the parameters of the implicit neural representation network, includes: using the implicit neural representation (INR) network to evaluate the absorption coefficient values ​​at each spatial location within the photoacoustic imaging region to generate a continuous second absorption coefficient distribution; calculating the luminous flux field under the second absorption coefficient distribution using the FEM forward solver to obtain the luminous flux distribution; using the PAT physical model to simulate the predicted detector signal based on the luminous flux distribution and the second absorption coefficient distribution; minimizing the difference between the predicted detector signal and the detector measurement signal to update the parameters of the INR network; and iterating in this way until the preset convergence condition is met.

[0042] Step 103: Use the optimized INR network to evaluate the absorption coefficient values ​​at each spatial location within the photoacoustic imaging region to generate a continuous third absorption coefficient distribution.

[0043] In this embodiment of the disclosure, the INR network can be, but is not limited to, a multilayer perceptron (MLP). For example, the INR network can be a three-layer MLP, with the first layer having a dimension of [missing information - likely a specific dimension or feature vector]. The dimensions are consistent; the second layer is a hidden layer containing 64 neurons, using ReLU as the activation function; the third layer is the output layer, containing 64 neurons, using Sigmoid as the activation function. Represent a real number vector. Let be the set of real numbers, and let represent vectors. Each element in the vector is a real number. The total dimension is L × F, where L represents the number of feature locations and F represents the feature dimension of each location. The values ​​of L and F can be preset. The first layer's dimension and the high-dimensional feature vector... Consistent dimensions enable high-dimensional mapping, allowing the INR network to learn high-frequency information such as detail edges in images. Employing a three-layer MLP INR network also helps accelerate training and optimize memory resource utilization. In specific applications, the structure of the INR network can be flexibly adjusted according to actual needs and application scenarios; this disclosure does not limit the specific structure of the INR network.

[0044] In this embodiment of the disclosure, the detector measurement signal may include a PAT image, and the predicted detector signal may be a predicted PAT image obtained by passing through an INR network, an FEM forward solver, and a PAT physical model at each spatial location within the photoacoustic imaging area.

[0045] In this embodiment of the disclosure, the object to be reconstructed may also be referred to as the target tissue, which may be, but is not limited to, deep tissue or others. This embodiment of the disclosure does not impose any restrictions on this.

[0046] Figure 2 The diagram illustrates a specific implementation process of the photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation provided in this embodiment of the present disclosure. Figure 2 In the diagram, r(x,y) represents a spatial location of the photoacoustic imaging region, and F... θ and F θ (r) represents the INR network, Φ(r) represents the light flux distribution, and μ α (r) represents the absorption coefficient distribution of the INR network output, p(r) represents the detector measurement signal, Loss represents the loss between the predicted detector signal and the detector measurement signal, and x SDBC Represents prior information.

[0047] In step 101, the first absorption coefficient distribution obtained based on SBDC is used as prior information to embed into the INR network, which can help the INR network obtain preliminary structural information and help accelerate the convergence speed of the INR network.

[0048] Figure 2 b illustrates the prior embedding process in the INR network. See also Figure 2b. Reconstruct the first absorption coefficient distribution x from the SBDC using detector measurement signals (e.g., real PAT images). SBDC The first absorption coefficient distribution x SBDC As prior information, it is embedded into the INR network to provide the INR network with the initial structural information of the object to be reconstructed. Thus, the structural information of the object to be reconstructed (e.g., the structural information of deep tissues) can be obtained and embedded into the INR network before the INR network is optimized, thereby avoiding overfitting of the PAT image and preventing the INR network from obtaining a degenerate reconstruction solution. This further improves the reconstruction accuracy of the absorption coefficient while accelerating the convergence speed of the INR optimization process.

[0049] Furthermore, prior to step 102, the method of this embodiment may further include: discretizing the spatial domain of the reconstructed object into a two-dimensional grid, wherein the spatial coordinates of a grid node in the two-dimensional grid represent a spatial location within the photoacoustic imaging region. Thus, the spatial coordinates of each grid node in the two-dimensional grid can be used as the input to the INR network as each spatial location within the photoacoustic imaging region. The spatial domain of the reconstructed object can also be referred to as the target tissue spatial domain, i.e., the biological anatomical region in the PAT image where the absorption coefficient needs to be reconstructed.

[0050] In this embodiment, by optimizing the parameters of the INR network, the INR network can learn a continuous function that maps spatial location to absorption coefficients, rather than the traditional discrete pixel approach. In step 102, by introducing an FEM forward solver during the INR network optimization process to approximate the distribution of optical absorption, the continuous function that maps spatial location to absorption coefficients learned by the INR network better matches the actual optical absorption of reconstructed objects such as deep tissues. This improves physical interpretability, reduces the deviation between the reconstructed absorption coefficient distribution and the actual situation, and further improves the accuracy of absorption coefficient reconstruction.

[0051] Figure 2 a illustrates the specific implementation process of step 102. See also Figure 2 a, prior information x SBDC After being embedded into the INR network, the spatial domain of the reconstructed object is discretized into a two-dimensional mesh, and the spatial coordinates of the mesh nodes in this two-dimensional mesh are r = (x,y) (r∈ 2The spatial location of the photoacoustic imaging region is input to the INR network. The INR network is used to reconstruct the absorption coefficient to realize the modeling of the spatial coordinates r based on continuous mapping, thereby outputting the second absorption coefficient distribution. The second absorption coefficient distribution is solved by forward finite element method (i.e., FEM forward solver f) to obtain the luminous flux distribution. The luminous flux distribution and the second absorption coefficient distribution are used to obtain the predicted detector signal through the PAT physical model. The predicted detector signal can be the predicted PAT image. Then, the network parameters of the INR network are updated by minimizing the difference between the predicted detector signal and the detector measurement signal (e.g., the real PAT image p(r)).

[0052] In step 102, the convergence conditions can be flexibly set according to requirements. For example, the convergence conditions can be set to, but are not limited to: the number of iterations does not exceed the preset maximum number of iterations, and the loss value is less than the preset loss value threshold.

[0053] In step 102, minimizing the difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network includes: calculating a loss value based on the predicted detector signal and the detector measurement signal based on a predetermined target loss function, using the loss value to perform backpropagation in the implicit neural representation network to obtain the gradient, and updating the parameters of the implicit neural representation network based on the gradient.

[0054] Specifically, the parameters of the INR network are optimized by minimizing the target loss function shown in equation (1).

[0055] (1)

[0056] in, Indicates the loss value. This indicates the detector's measurement signal. This represents the predictor signal, where r represents the grid coordinates. This represents the distribution of the second absorption coefficient. The light flux distribution is represented by f, which represents the forward FEM solver. " " indicates dot product.

[0057] In step 102, the difference between the predicted detector signal and the detector measurement signal p(r) calculated by the loss value calculated by minimizing the target loss function given by formula (1) and the difference between the predicted detector signal and the detector measurement signal p(r) calculated by the joint calculation of the INR network, the FEM forward solver and the PAT physical model is realized, thereby optimizing the parameters of the INR network.

[0058] To estimate the absorption coefficient value at each spatial location in the photoacoustic imaging region using the INR network, a hash encoding method can be used to encode each spatial location in the photoacoustic imaging region before generating a continuous second absorption coefficient distribution using the INR network in step 102. This enhances the INR network's ability to fit the nonlinear scattered light intensity, thereby accelerating the optimization process of the INR network. Specifically, each spatial location in the photoacoustic imaging region can be the spatial coordinates of grid nodes in a two-dimensional grid corresponding to the spatial domain of the reconstructed object. The spatial coordinates of all grid nodes in this two-dimensional grid can be encoded using a hash encoding method before being input into the INR network for absorption coefficient reconstruction.

[0059] Considering that the resolution of qPAT depends on the accuracy of the forward model, high-resolution solutions typically require solving related equations on a fine mesh. However, related techniques suffer from an exponential increase in degrees of freedom, making the memory overhead of solving on fine meshes potentially prohibitive. To address this issue, an adaptive mesh refinement technique can be introduced during the optimization process in step 102. This establishes a connection between the neural representation and the geometric representation, enabling efficient optimization of the INR network and achieving a balance between computational cost and the accuracy of absorption coefficient reconstruction.

[0060] Specifically, step 102 may further include: in the iteration at a preset interval or in one or more randomly selected iterations, adaptively refining the two-dimensional mesh according to the second absorption coefficient distribution obtained from the previous reconstruction, so that the mesh node density of the two-dimensional mesh is positively correlated with its absorption coefficient value.

[0061] Here, the preset interval of iteration can include each iteration, or every n intervals, where n is an integer greater than 1, and the value of n can be preset. The randomly selected iteration can be randomly selected by a random algorithm, or it can include determining whether to perform 2D mesh adaptive refinement in the current round based on the situation of the previous round of iteration (e.g., the size of the loss value).

[0062] In some examples, the 2D mesh is adaptively refined based on the second absorption coefficient distribution obtained from the previous reconstruction, making the mesh node density of the 2D mesh positively correlated with its absorption coefficient value. This can include: determining whether each mesh node belongs to a high-absorption region, medium-absorption region, or low-absorption region based on the absorption coefficient value of each mesh node in the 2D mesh obtained from the previous reconstruction, thereby dividing the 2D mesh into high-absorption, medium-absorption, and low-absorption regions. The mesh is then refined for the high-absorption and medium-absorption regions (e.g., reducing the mesh size), while the existing mesh size can be maintained for the low-absorption regions. This makes the mesh node density of the 2D mesh positively correlated with its absorption coefficient value, so that the INR network can reconstruct the high-absorption and medium-absorption regions with higher density in subsequent optimization processes, thereby further improving the accuracy of absorption coefficient reconstruction.

[0063] The high absorption region, medium absorption region, and low absorption region can be defined by the range of absorption coefficient values. The lower limit of the absorption coefficient value in the high absorption region is greater than or equal to the upper limit of the absorption coefficient value in the medium absorption region, and the lower limit of the absorption coefficient value in the medium absorption region is greater than or equal to the upper limit of the absorption coefficient value in the low absorption region. For example, the high absorption region can be defined as the region with absorption coefficient values ​​in the range [a, b], where 'a' represents the lower limit of the absorption coefficient value in the high absorption region, and 'b' represents the upper limit of the absorption coefficient value in the high absorption region. Similarly, the medium absorption region and the low absorption region can be defined by the range of absorption coefficient values.

[0064] In practical applications, the grading of the absorption region can be flexibly set as needed. For example, in addition to the aforementioned three grades of high absorption region, medium absorption region, and low absorption region, the absorption region can also be divided into two grades of high absorption region and low absorption region, or even four or more grades. This disclosure does not impose any limitations on this.

[0065] In some examples, the 2D mesh is adaptively refined based on the second absorption coefficient distribution obtained from the previous reconstruction, so that the mesh node density is positively correlated with its absorption coefficient value. This can include: calculating the reconstructed absorption coefficient distribution of the current round (i.e., the k-th round) during the iterative reconstruction process. Spatial gradient magnitude The two-dimensional mesh is locally refined according to the preset unit size function shown in equation (2), and the next round of reconstruction is performed on the refined mesh. As a result, the mesh resolution in the high gradient region can be improved, and the mesh node density of the two-dimensional mesh is positively correlated with its absorption coefficient value.

[0066] (2)

[0067] in, Here, α is the preset minimum grid cell size, α is an adjustable scaling factor, and h(r) represents the grid cell size at spatial location r. ∈[0.05,0.5]mm, k represents the current round identifier.

[0068] In practical applications, during the iteration process of step 102, you can start with a coarse mesh and use the Adaptive Mesh Enhancement Suite - High Refinement to dynamically refine the two-dimensional mesh using the solution from the previous round during the iteration. Figure 2 c illustrates the adaptive mesh refinement process. See also Figure 2 c. Using the absorption coefficient distribution obtained from the previous reconstruction to guide adaptive mesh refinement, a denser mesh is generated in the high absorption region to enhance the detail resolution.

[0069] In step 103, the third absorption coefficient distribution generated by the optimized INR network is the final reconstructed absorption coefficient distribution.

[0070] Figure 3 Qualitative results of reconstructing absorption coefficient images on photoacoustic digital brain images using different methods are shown. Figure 3 In the image, from left to right: PAT image, SBDC reconstruction result, TSIA reconstruction result, Unet reconstruction result, EAPNet reconstruction result, Ours represents the reconstruction result of the optimized INR network according to this embodiment, and GT represents the true absorption coefficient distribution. Figure 3 It can be seen that SBDC only captures general structures and has low accuracy in reconstructing absorption coefficients in deep tissues. Although TSIA can reconstruct most absorption coefficients well, it also lacks accuracy in deep tissue regions. U-shaped networks (U-net) and EAP reconstruction networks (EAPNet, Extractor-Attention-Predictor Network) have global inaccuracies. The optimized INR network significantly outperforms the above methods in terms of accuracy and precision in recovering absorption coefficients, especially in the reconstruction of deep tissues.

[0071] Figure 4 Qualitative results of reconstructing absorption coefficient images on phantom data using different methods are shown. Figure 4 From left to right: PAT image, SBDC reconstruction result, TSIA reconstruction result, EAPNet reconstruction result, Ours represents the reconstruction result of the optimized INR network in this embodiment, and Reference represents the reference absorption coefficient distribution. Figure 4 The optimized INR network performs excellently in recovering the absorption coefficients of various substances. Although EAPNet can satisfactorily recover the absorption coefficient of each part, the error plot shows that it has problems with inaccurate edge reconstruction. The results of SBDC and TSIA are not satisfactory, as there is no significant difference between them due to the simple model segmentation and the uniform absorption coefficient within each part. Intuitively, the optimized INR network is significantly better than the above methods in terms of accuracy and edge accuracy in absorption coefficient reconstruction, further verifying that the method provided in this embodiment can effectively improve the accuracy of absorption coefficient reconstruction.

[0072] Figure 5A schematic diagram of a photoacoustic tomography absorption coefficient reconstruction device based on implicit neural representation provided in an embodiment of this disclosure is shown. This device is applied to an electronic device 600. See also Figure 5 The photoacoustic tomography absorption coefficient reconstruction device 500 based on implicit neural representation may include:

[0073] The prior embedding unit 501 is used to obtain the first absorption coefficient distribution by using the detector measurement signal based on the SBDC method, and embed the first absorption coefficient distribution as prior information into the implicit neural representation network.

[0074] The parameter optimization unit 502 is used to optimize the parameters of the implicit neural representation network, including iteratively executing the following steps until a preset convergence condition is met: using the implicit neural representation network to evaluate the absorption coefficient values ​​at each spatial location within the photoacoustic imaging region to generate a continuous second absorption coefficient distribution; calculating the light flux field under the second absorption coefficient distribution using a finite element method forward solver to obtain the light flux distribution; simulating the predicted detector signal based on the light flux distribution and the second absorption coefficient distribution using a photoacoustic tomography physical model; and minimizing the difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network.

[0075] The reconstruction execution unit 503 is used to evaluate the absorption coefficient values ​​at various spatial locations within the photoacoustic imaging region using an implicit neural representation network with optimized parameters to generate a continuous third absorption coefficient distribution.

[0076] Furthermore, the photoacoustic tomography absorption coefficient reconstruction device 500 based on implicit neural representation may also include: an encoding unit 504, used to encode each spatial location of the photoacoustic imaging region using a hash encoding method.

[0077] Furthermore, the photoacoustic tomography absorption coefficient reconstruction device 500 based on implicit neural representation may also include: a gridding unit 505, used to discretize the spatial domain of the reconstruction object into a two-dimensional grid, wherein the spatial coordinates of a grid node in the two-dimensional grid represent a spatial location within the photoacoustic imaging region.

[0078] Furthermore, the photoacoustic tomography absorption coefficient reconstruction device 500 based on implicit neural representation may also include: an adaptive refinement unit 506, used to adaptively refine the two-dimensional mesh according to the second absorption coefficient distribution obtained in the previous round of reconstruction during iterations at preset intervals or in one or more randomly selected rounds of iterations, so that the mesh node density of the two-dimensional mesh is positively correlated with its absorption coefficient value.

[0079] Further technical details regarding the photoacoustic tomography absorption coefficient reconstruction device 500 based on implicit neural representation can be found in the preceding section on photoacoustic tomography absorption coefficient reconstruction methods based on implicit neural representation, and will not be repeated here. In specific applications, the photoacoustic tomography absorption coefficient reconstruction device 500 based on implicit neural representation can be implemented by the electronic device 600 described below, or it can be implemented as software within the electronic device 600.

[0080] In addition, embodiments of this disclosure also provide a computer-readable storage medium storing a computer program thereon, the program including instructions that, when executed by one or more processors of a computing device, perform the steps of the aforementioned photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation.

[0081] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure is shown. See also... Figure 6 The electronic device 600 may include one or more processors 601, and a memory 602 storing one or more programs, which are executed by the one or more processors 601 to implement the method flow and / or program units corresponding to each unit in the apparatus shown in the above embodiments of this disclosure.

[0082] The various components are interconnected via different buses and can be mounted on a common motherboard or otherwise as required. Processor 601 can process instructions executed within the electronic device, including instructions stored in or on memory to display graphical information of a user interface on an external input / output device (such as a display device coupled to an interface). In other embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired.

[0083] Processor 601 may include one or more single-core or multi-core processors. Processor 601 may include any combination of general-purpose processors or special-purpose processors (such as graphics processors, application processors, baseband processors, etc.).

[0084] Memory 602 is the computer-readable storage medium provided in this disclosure, which can be used to store non-transitory software programs, non-transitory computer-executable programs, and units, such as those in the embodiments of this disclosure. Figure 1 The program instructions / units corresponding to the photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation are shown. Processor 601 executes non-transient software programs, instructions, and units stored in memory 602, thereby performing operations such as those described in the above method embodiments. Figure 1 The program, instructions, and units corresponding to the photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation are shown.

[0085] The electronic device 600 may further include an input device 603 and an output device 604. The processor 601, memory 602, input device 603, and output device 604 can be connected via a bus or other means. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0086] The aforementioned programs (also known as software, software applications, or code) include machine instructions for a programmable processor and can be implemented using object-oriented programming languages, assembly language, or machine language.

[0087] With the development of time and technology, the meaning of "medium" has become increasingly broad. The dissemination of computer programs is no longer limited to tangible media; they can also be downloaded directly from the network. Any combination of one or more computer-readable storage media can be used. Computer-readable storage media can be, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, device, or apparatus.

[0088] The technical solutions provided in this disclosure have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this disclosure. The descriptions of the embodiments above are only for the purpose of helping to understand the methods and core ideas of this disclosure. Furthermore, those skilled in the art will recognize that, based on the ideas of this disclosure, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this disclosure.

[0089] The above description is merely a preferred embodiment of this disclosure and is not intended to limit this disclosure. Any modifications or equivalent substitutions made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for reconstructing photoacoustic absorption coefficients based on implicit neural representation, characterized in that, The method includes: The first absorption coefficient distribution is obtained by using the detector measurement signal based on the SBDC method. The first absorption coefficient distribution is then embedded as prior information into the implicit neural representation network. The detector measurement signal includes PAT images. Optimizing the parameters of the implicit neural representation network includes iteratively performing the following steps until a preset convergence condition is met: using the implicit neural representation network to evaluate absorption coefficient values ​​at various spatial locations within the photoacoustic imaging region to generate a continuous second absorption coefficient distribution; calculating the luminous flux field under the second absorption coefficient distribution using a finite element method forward solver to obtain the luminous flux distribution; simulating the predicted detector signal based on the luminous flux distribution and the second absorption coefficient distribution using a photoacoustic tomography physical model; and minimizing the difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network. The implicit neural representation network with optimized parameters is used to evaluate absorption coefficient values ​​at various spatial locations within the photoacoustic imaging region to generate a continuous third absorption coefficient distribution.

2. The method according to claim 1, characterized in that, Minimizing the difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network includes: calculating a loss value based on the predicted detector signal and the detector measurement signal based on a predetermined target loss function; using the loss value to perform backpropagation in the implicit neural representation network to obtain the gradient; and updating the parameters of the implicit neural representation network based on the gradient.

3. The method according to claim 2, characterized in that, The target loss function is as follows: in, Indicates the loss value. This indicates the detector's measurement signal. This represents the predictor signal, where r represents the grid coordinates. This represents the distribution of the second absorption coefficient. Indicates the distribution of luminous flux. " " indicates dot product.

4. The method according to claim 1, characterized in that, The method further includes: before using the implicit neural representation network to evaluate the absorption coefficient value at the spatial location of the photoacoustic imaging region to generate a continuous second absorption coefficient distribution, encoding each spatial location of the photoacoustic imaging region using a hash encoding method.

5. The method according to claim 1, characterized in that, The optimization of the parameters of the implicit neural representation network further includes: before using the implicit neural representation network to evaluate the absorption coefficient values ​​at each spatial location within the photoacoustic imaging region to generate a continuous second absorption coefficient distribution, discretizing the spatial domain of the reconstructed object into a two-dimensional grid, wherein the spatial coordinates of a grid node in the two-dimensional grid represent a spatial location within the photoacoustic imaging region.

6. The method according to claim 5, characterized in that, The optimization of the parameters of the implicit neural representation network further includes: in the iteration at a preset interval or in one or more randomly selected iterations, adaptively refining the two-dimensional grid according to the second absorption coefficient distribution obtained from the previous reconstruction, so that the grid node density of the two-dimensional grid is positively correlated with its absorption coefficient value.

7. The method according to claim 1, characterized in that, The implicit neural representation network is a multilayer perceptron.

8. A photoacoustic tomography absorption coefficient reconstruction device based on implicit neural representation, characterized in that, The photoacoustic tomography absorption coefficient reconstruction device based on implicit neural representation includes: A priori embedding unit is used to obtain a first absorption coefficient distribution using detector measurement signals based on the SBDC method, and to embed the first absorption coefficient distribution as prior information into a hidden neural representation network. The detector measurement signals include PAT images. The parameter optimization unit is used to optimize the parameters of the implicit neural representation network, including iteratively executing the following steps until a preset convergence condition is met: using the implicit neural representation network to evaluate the absorption coefficient values ​​at each spatial location within the photoacoustic imaging region to generate a continuous second absorption coefficient distribution; calculating the light flux field under the second absorption coefficient distribution using a finite element method forward solver to obtain the light flux distribution; simulating the predicted detector signal based on the light flux distribution and the second absorption coefficient distribution using a photoacoustic tomography physical model; and minimizing the difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network. The reconstructed execution unit is used to evaluate the absorption coefficient values ​​at various spatial locations within the photoacoustic imaging region using the implicit neural representation network with optimized parameters to generate a continuous third absorption coefficient distribution.

9. An electronic device, characterized in that, include: A memory for storing one or more processors and programs, the programs comprising instructions that, when executed by the processor, cause the processor to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a program, the program comprising instructions that, when executed by one or more processors, cause the processors to perform the method as described in any one of claims 1 to 7.

Images