Photoacoustic chromatography absorption coefficient reconstruction method and device based on implicit neural representation

Abstract

The invention provides a photoacoustic chromatography absorption coefficient reconstruction method and device based on implicit neural representation, and relates to the technical field of image processing. The method provided by the embodiment of the invention comprises the following steps: acquiring first absorption coefficient distribution by utilizing a detector measurement signal based on an SBDC method, and embedding the first absorption coefficient distribution into an INR network as prior information; and optimizing parameters of the implicit neural representation network by using a finite element method forward solver and a PAT physical model, and evaluating absorption coefficient values at each spatial position in the photoacoustic imaging area by using the INR network after parameter optimization to generate continuous third absorption coefficient distribution. According to the invention, the absorption coefficient reconstruction precision of application scenes such as deep tissues and the like can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of image processing, and particularly relates to a photoacoustic tomography absorption coefficient reconstruction method and device based on implicit neural representation. BACKGROUND

[0002] The goal of quantitative photoacoustic tomography (qPAT) is to reconstruct the absorption coefficient distribution of the tissue, and the relationship between the absorption coefficient distribution and the initial photoacoustic image (i.e., the photoacoustic initial sound pressure image) depends on the light fluence (LF). However, since the absorption coefficient reconstruction process involves complex light transport equation inversion, the nonlinearity and the underdetermined problem of the absorption coefficient reconstruction process lead to instability of the solution, making the absorption coefficient reconstruction process extremely challenging.

[0003] In practical applications, unknown light scattering in deep tissue can introduce errors in light fluence simulation and initial pressure map reconstruction, further exacerbate the instability of the solution caused by the underdetermined problem, and cause the reconstructed absorption coefficient distribution to deviate from the real physiological parameters, so that the real physiological parameters cannot be accurately reflected, and the application of the absorption coefficient reconstruction is limited.

[0004] In order to improve the reconstruction accuracy, a segmentation-based direct correction method (SBDC), a supervised learning model based on a deep neural network, and other methods have been proposed. These related technologies have problems such as a large demand for training data and low absorption coefficient reconstruction accuracy in scenarios such as deep tissue. SUMMARY

[0005] Therefore, the present disclosure provides a photoacoustic tomography absorption coefficient reconstruction method and device based on implicit neural representation.

[0006] According to a first aspect of the present disclosure, a photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation is provided, and the method comprises: obtaining a first absorption coefficient distribution based on a SBDC method using a detector measurement signal, and embedding the first absorption coefficient distribution as prior information into an implicit neural representation network; optimizing the parameters of the implicit neural representation network, the optimizing the parameters of the implicit neural representation network comprising iteratively performing the following steps until a preset convergence condition is satisfied: evaluating absorption coefficient values at each spatial location within the photoacoustic imaging region using the implicit neural representation network to generate a continuous second absorption coefficient distribution, calculating a light fluence field under the second absorption coefficient distribution by a finite element method forward solver to obtain a light fluence distribution, simulating a predicted detector signal based on the light fluence distribution and the second absorption coefficient distribution using a photoacoustic tomography physical model, minimizing a difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network; evaluating absorption coefficient values at each spatial location within the photoacoustic imaging region using the implicit neural representation network with the optimized parameters to generate a continuous third absorption coefficient distribution.

[0007] In some embodiments of the first aspect of the present disclosure, the minimizing a difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network comprises: calculating a loss value based on the predicted detector signal and the detector measurement signal based on a predetermined target loss function, performing backpropagation to calculate gradients in the implicit neural representation network using the loss value, and updating the parameters of the implicit neural representation network based on the gradients.

[0008] In some embodiments of the first aspect of the present disclosure, the target loss function is as follows:

[0009] wherein, represents a loss value, represents a detector measurement signal, represents a predicted detector signal, r represents a grid coordinate, represents the second absorption coefficient distribution, represents a light fluence distribution, represents a dot product.

[0010] In some embodiments of the first aspect of the present disclosure, the method further comprises: before the evaluating absorption coefficient values at each spatial location within the photoacoustic imaging region using the implicit neural representation network to generate a continuous second absorption coefficient distribution, encoding each spatial location of the photoacoustic imaging region using a hash encoding method.

[0011] In some embodiments of the first aspect of the present disclosure, optimizing the parameters of the implicit neural representation network further includes: before using the implicit neural representation network to evaluate the absorption coefficient value at each spatial position in the photoacoustic imaging area to generate a continuous second absorption coefficient distribution, discretizing the spatial domain of the reconstructed object into a two-dimensional grid, where the spatial coordinates of a grid node in the two-dimensional grid represent a spatial position within the photoacoustic imaging area.

[0012] In some embodiments of the first aspect of the present disclosure, the optimizing the parameters of the implicit neural representation network further includes: in a preset interval iteration or one or more randomly selected rounds of iteration, adaptively refining the two-dimensional grid according to the second absorption coefficient distribution obtained from the previous round of reconstruction, so that the grid node density of the two-dimensional grid is positively correlated with its absorption coefficient value.

[0013] In some implementations of the first aspect of the present disclosure, the implicit neural representation network is a multilayer perceptron.

[0014] According to a second aspect of the present disclosure, a device for reconstructing photoacoustic tomography absorption coefficients based on implicit neural representation is provided, the device comprising: a priori embedding unit, configured to obtain a first absorption coefficient distribution using a detector measurement signal based on an SBDC method, and embed the first absorption coefficient distribution as priori information into an implicit neural representation network; a parameter optimization unit, configured to optimize the parameters of the implicit neural representation network, comprising iteratively performing the following steps until a preset convergence condition is satisfied: using the implicit neural representation network to evaluate absorption coefficient values ​​at each spatial position within a photoacoustic imaging region to generate a continuous second absorption coefficient distribution, calculating a light flux field under the second absorption coefficient distribution using a finite element method forward solver to obtain a light flux distribution, using a photoacoustic tomography physical model to simulate a predicted detector signal based on the light flux distribution and the second absorption coefficient distribution, and minimizing a difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network; A reconstruction execution unit is used to evaluate the absorption coefficient value at each spatial position in the photoacoustic imaging area using the implicit neural representation network with optimized parameters to generate a continuous third absorption coefficient distribution.

[0015] According to a third aspect of the present disclosure, an electronic device is provided, comprising: one or more processors and a memory storing a program, wherein the program comprises instructions, and when the instructions are executed by the processor, the processor executes the above method.

[0016] According to a fourth aspect of the present disclosure, a computer readable storage medium storing a program is provided, the program comprising instructions which, when executed by one or more processors, cause the processors to perform the method described above.

[0017] From the above technical solutions, it can be seen that the embodiments of the present disclosure can effectively improve the absorption coefficient reconstruction accuracy in application scenarios such as deep tissue without training data, with low cost and good physical interpretability. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present disclosure, and other drawings can also be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0019] Figure 1 A flowchart of the method for photoacoustic tomography absorption coefficient reconstruction based on implicit neural representation provided by the embodiments of the present disclosure is shown. Figure 2 A specific implementation process example diagram of the method for photoacoustic tomography absorption coefficient reconstruction based on implicit neural representation provided by the embodiments of the present disclosure is shown. Figure 2 A specific implementation process example diagram of step 102 in the method for photoacoustic tomography absorption coefficient reconstruction based on implicit neural representation provided by the embodiments of the present disclosure is shown. Figure 2 A flowchart example diagram of prior embedding in the INR network in the method for photoacoustic tomography absorption coefficient reconstruction based on implicit neural representation provided by the embodiments of the present disclosure is shown. Figure 2 A two-dimensional grid adaptive refinement example diagram in the method for photoacoustic tomography absorption coefficient reconstruction based on implicit neural representation provided by the embodiments of the present disclosure is shown. Figure 3 A qualitative result comparison diagram of the method for reconstructing absorption coefficient image on photoacoustic digital brain image provided by the embodiments of the present disclosure and related technologies is shown. Figure 4 A qualitative result comparison diagram of the method for reconstructing absorption coefficient image on phantom data provided by the embodiments of the present disclosure and related technologies is shown. Figure 5 A structural diagram of the device for photoacoustic tomography absorption coefficient reconstruction based on implicit neural representation provided by the embodiments of the present disclosure is shown. Figure 6 A schematic structural block diagram of an electronic device provided by the embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present disclosure will be clearly and completely described with reference to the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the scope of the present disclosure.

[0021] The terms used in the embodiments of the present disclosure are merely for the purpose of describing specific embodiments, and are not intended to limit the present disclosure. The singular forms "a", "an" and "the" used in the embodiments of the present disclosure and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0022] Depending on the context, the word "if" as used herein can be interpreted to mean "when" or "upon" or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if it is determined" or "if [a stated condition or event] is detected" can be interpreted to mean "upon determining" or "in response to determining" or "upon detecting [the stated condition or event]" or "in response to detecting [the stated condition or event]".

[0023] The related art and its defects are briefly described below.

[0024] 1) Segmentation-Based Direct Correction (SBDC): The absorption coefficient distribution map is obtained by dividing the PAT image by the light fluence (LF) map. This method requires manual setting of the initial optical parameters of the segmentation region, which is complex and dependent on experience. Although subsequent automatic segmentation algorithms for this method were proposed to reduce human intervention, the low soft tissue contrast of the PAT image leads to insufficient segmentation accuracy, and the assumption that the absorption coefficient distribution is uniform within the segmented region also leads to errors in the estimation, resulting in low accuracy of the absorption coefficient reconstruction.

[0025] 2) Absorption coefficient reconstruction based on deep neural network: A large amount of training data is collected in advance to train an end-to-end deep neural network to obtain a supervised learning model based on the deep neural network, and the absorption coefficient is reconstructed by the supervised learning model based on the deep neural network. This method requires a large amount of training data, and it is difficult and costly to obtain a large number of accurately labeled LF maps in actual qPAT, which limits the application of this method in application scenarios such as deep tissue and reduces the reconstruction accuracy.

[0026] In view of this, the embodiments of the present disclosure provide the following method and device for reconstructing the absorption coefficient of photoacoustic tomography based on implicit neural representation, which embeds the absorption coefficient distribution reconstructed based on the SBDC method as prior information into the INR network, then optimizes the parameters of the INR network by combining the finite element method (FEM) forward solver and the photoacoustic tomography (PAT) physical model, and finally uses the optimized INR network to reconstruct the absorption coefficient distribution. The embodiments of the present disclosure do not require training data and can achieve absorption coefficient reconstruction in an unsupervised manner. Experimental verification shows that the embodiments of the present disclosure can accurately restore the absorption coefficient, and the accuracy of absorption coefficient reconstruction is significantly improved compared to related technologies in quantitative and qualitative evaluations.

[0027] Figure 1 Schematic diagram of the process of photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation provided by the embodiment of the present disclosure. Figure 1 The method for reconstructing photoacoustic tomography absorption coefficients based on implicit neural representation according to an embodiment of the present disclosure may include the following steps: Step 101: obtaining a first absorption coefficient distribution using a detector measurement signal based on the SBDC method, and embedding the first absorption coefficient distribution as prior information into an INR network; Step 102, optimizing parameters of an implicit neural representation network, including: using an implicit neural representation (INR) network to evaluate absorption coefficient values ​​at various spatial locations within the photoacoustic imaging region to generate a continuous second absorption coefficient distribution; calculating a luminous flux field under the second absorption coefficient distribution using an FEM forward solver to obtain a luminous flux distribution; simulating a predicted detector signal based on the luminous flux distribution and the second absorption coefficient distribution using a PAT physical model; and minimizing the difference between the predicted detector signal and the measured detector signal to update the parameters of the INR network, iterating until a preset convergence condition is met. Step 103 : Using the INR network with optimized parameters, the absorption coefficient value is evaluated at each spatial position in the photoacoustic imaging region to generate a continuous third absorption coefficient distribution.

[0028] In the embodiment of the present disclosure, the INR network may be, but is not limited to, a multilayer perceptron (MLP). For example, the INR network may be a three-layer MLP, where the dimension of the first layer is the same as the high-dimensional feature vector. The dimensions are consistent, the second layer is the hidden layer, contains 64 neurons, uses ReLU as the activation function, and the third layer is the output layer, contains 64 neurons, uses Sigmoid as the activation function. represents a real vector, is a set of real numbers, representing a vector Each element in is a real number, and the vector The total dimension is L × F, where L represents the number of feature positions and F represents the feature dimension of each position. The values ​​of L and F can be preset. The dimension of the first layer is the same as that of the high-dimensional feature vector. The dimensions of the image are consistent, enabling high-dimensional mapping. This high-dimensional mapping allows the INR network to learn high-frequency information such as detailed edges in the image. The use of a three-layer MLP INR network can also help speed up training and rationally utilize memory resources. In specific applications, the structure of the INR network can be flexibly adjusted based on actual needs and application scenarios. The specific structure of the INR network is not limited in the embodiments of this disclosure.

[0029] In the embodiment of the present disclosure, the detector measurement signal may include a PAT image, and the predicted detector signal may be a predicted PAT image obtained by the INR network, the FEM forward solver and the PAT physical model at each spatial position in the photoacoustic imaging area.

[0030] In the embodiment of the present disclosure, the reconstruction object may also be referred to as target tissue, which may be but is not limited to deep tissue or others, and the embodiment of the present disclosure does not impose any limitation on this.

[0031] Figure 2 An example diagram of the specific implementation process of the photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation provided by an embodiment of the present disclosure is shown. Figure 2 In the figure, r(x,y) represents a spatial position of the photoacoustic imaging area, and F θ and F θ (r) represents the INR network, Φ(r) represents the luminous flux distribution, 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.

[0032] In step 101, the first absorption coefficient distribution obtained based on SBDC is used as prior information to be embedded into the INR network, which can help the INR network obtain preliminary structural information and help accelerate the convergence speed of the INR network.

[0033] Figure 2 b shows a schematic diagram of the process of prior embedding in the INR network. Figure 2 b. Reconstruct the first absorption coefficient distribution x using the detector measurement signal (e.g., the real PAT image) from the SBDC SBDC , the first absorption coefficient distribution x SBDCAs prior information embedded into the INR network to provide the INR network with initial structural information of the reconstruction object, thereby, the structural information (for example, the structural information of deep tissue) of the reconstruction object can be obtained before the INR optimization and embedded into the INR network, so as to avoid overfitting of the PAT image and prevent the INR network from obtaining a degenerated reconstruction solution, further improve the absorption coefficient reconstruction accuracy on the premise of accelerating the convergence speed of the INR optimization process.

[0034] Further, before step 102, the method of the embodiments of the present disclosure can further include: discretizing 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 position in the photoacoustic imaging region. Thus, the spatial coordinates of each grid node in the two-dimensional grid can be taken as the spatial positions in the photoacoustic imaging region as the input of the INR network. Wherein, the spatial domain of the reconstruction object can also be referred to as the target tissue spatial domain, that is, the biological anatomical region in the PAT image whose absorption coefficient needs to be reconstructed.

[0035] In the embodiments of the present disclosure, by optimizing the parameters of the INR network, the INR network can learn a continuous function that maps the spatial position to the absorption coefficient, rather than the traditional discrete pixel method. In step 102, by introducing a FEM forward solver in the INR network optimization process to approximate the distribution of optical absorption, the continuous function learned by the INR network that maps the spatial position to the absorption coefficient can better conform to the real situation of the optical absorption of the reconstruction object such as deep tissue, thereby improving the physical interpretability, reducing the deviation of the reconstructed absorption coefficient distribution from the real situation, and further improving the absorption coefficient reconstruction accuracy.

[0036] Figure 2 a shows the specific implementation process of step 102. Referring to Figure 2 a, the prior information x SBDC is embedded into the INR network, the spatial domain of the reconstruction object is discretized into a two-dimensional grid, and the spatial coordinates r = (x, y) (r e 2 ) of the grid nodes in the two-dimensional grid are input to the INR network as the spatial positions of the photoacoustic imaging region, the INR network is used to reconstruct the absorption coefficient to realize modeling based on the continuous mapping of the spatial coordinates r and output a second absorption coefficient distribution, the light flux distribution is obtained by the forward finite element solving (i.e., the FEM forward solver f) on the second absorption coefficient distribution, the light 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 a predicted PAT image, and the network parameters of the INR network are updated by minimizing the difference between the predicted detector signal and the detector measurement signal (for example, the real PAT image p(r).

[0037] In step 102, the convergence condition can be flexibly set according to requirements. For example, the convergence condition can be set to, but not limited to, that 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.

[0038] In step 102, the parameters of the implicit neural representation network are updated by minimizing the difference between the predicted detector signal and the detector measurement signal, including: calculating the loss value based on the predetermined target loss function based on the predicted detector signal and the detector measurement signal, using the loss value to perform gradient calculation by back propagation in the implicit neural representation network, and updating the parameters of the implicit neural representation network based on the gradient.

[0039] Specifically, the parameters of the INR network are optimized by minimizing the target loss function shown in formula (1) as follows.

[0040] (1) wherein, represents the loss value, represents the detector measurement signal, represents the predicted detector signal, r represents the grid coordinate, represents the second absorption coefficient distribution, represents the light flux distribution, f represents the FEM forward solver, represents the dot product.

[0041] In step 102, the loss value calculated by minimizing the target loss function given by formula (1) penalizes the difference between the predicted detector signal calculated by the INR network, the FEM forward solver and the PAT physical model and the detector measurement signal p(r), thereby optimizing the parameters of the INR network.

[0042] In order to estimate the absorption coefficient value at each spatial position of the photoacoustic imaging region using the INR network, before generating the continuous second absorption coefficient distribution using the INR network in step 102, a hash coding method can be used to encode each spatial position of the photoacoustic imaging region to enhance the fitting ability of the INR network for nonlinear scattered light intensity, thereby accelerating the optimization process of the INR network. Specifically, each spatial position of the photoacoustic imaging region can be the spatial coordinates of the grid nodes in the two-dimensional grid corresponding to the spatial domain of the reconstructed object, and the spatial coordinates of all grid nodes in the two-dimensional grid can be encoded by the hash coding method and then input into the INR network for absorption coefficient reconstruction.

[0043] Considering that the qPAT resolution depends on the accuracy of the forward model, a high-resolution solution usually requires solving the relevant equations on a fine grid, but the problem of exponential growth of degrees of freedom in the related technical method causes the solving on the fine grid to be likely to cause an unbearable memory overhead. To solve this problem, an adaptive grid refinement technique can be introduced in the optimization process of step 102, a connection is established between the neural representation and the geometric representation, and efficient optimization of the INR network is realized, so as to balance between the calculation cost and the accuracy of the absorption coefficient reconstruction.

[0044] Specifically, step 102 can further include: in a preset interval iteration or a randomly selected one or more iterations, the two-dimensional grid is adaptively refined according to the second absorption coefficient distribution obtained by the last round of reconstruction, so that the grid node density of the two-dimensional grid is positively correlated with the absorption coefficient value thereof.

[0045] Here, the preset interval iteration can include each iteration, every n iterations in an interval, and n is an integer greater than 1, and the value of n can be pre-set. The randomly selected iteration can be randomly selected by a random algorithm, and also includes a case of determining whether to perform adaptive refinement of the two-dimensional grid in the current round according to the situation of the last round of iteration (for example, the loss value size, etc.).

[0046] In some examples, the adaptive refinement of the two-dimensional grid according to the second absorption coefficient distribution obtained by the last round of reconstruction so that the grid node density of the two-dimensional grid is positively correlated with the absorption coefficient value thereof can include: determining whether each grid node belongs to a high absorption region, a medium absorption region or a low absorption region according to the absorption coefficient value of each grid node in the two-dimensional grid in the second absorption coefficient distribution obtained by the last round of reconstruction, thereby dividing the two-dimensional grid into a high absorption region, a medium absorption region and a low absorption region, and refining the grid (for example, reducing the grid size, etc.) for the high absorption region and the medium absorption region, and the existing grid size can be maintained for the low absorption region. In this way, the grid node density of the two-dimensional grid can be positively correlated with the absorption coefficient value thereof, so that the INR network performs higher density reconstruction for the high absorption region and the medium absorption region in the subsequent optimization process, thereby further improving the absorption coefficient reconstruction accuracy.

[0047] Among them, the high absorption region, the medium absorption region and the low absorption region can be defined by the absorption coefficient value interval, the lower limit of the absorption coefficient value of the high absorption region is greater than or equal to the upper limit of the absorption coefficient value of the medium absorption region, and the lower limit of the absorption coefficient value of the medium absorption region is greater than or equal to the upper limit of the absorption coefficient value of the low absorption region. For example, the high absorption region can be defined as a region with an absorption coefficient value interval of [a, b], where a represents the lower limit of the absorption coefficient value of the high absorption region, and b is the upper limit of the absorption coefficient value of the high absorption region. Similarly, the medium absorption region and the low absorption region can be defined by the absorption coefficient value interval.

[0048] In specific applications, the sub-ranges of the absorption region can be flexibly set as needed. For example, the sub-ranges of the absorption region can be two sub-ranges of high absorption region and low absorption region, or four or more sub-ranges, in addition to the three sub-ranges of high absorption region, medium absorption region, and low absorption region. The embodiments of the present disclosure do not limit this.

[0049] In some examples, the adaptive refinement of the two-dimensional grid according to the second absorption coefficient distribution obtained in the last round of reconstruction can include that, in the iterative reconstruction process, the spatial gradient amplitude of the absorption coefficient distribution reconstructed in the current round (i.e., the kth round) is calculated , the two-dimensional grid is locally encrypted according to a preset cell size function shown in formula (2) below, and the next round of reconstruction is performed on the encrypted grid. In this way, the grid resolution of the high gradient region is improved, and the grid node density of the two-dimensional grid is positively correlated with the absorption coefficient value thereof.

[0050] (2) wherein, is a preset minimum size of the grid cell, a is an adjustable scaling factor, h(r) represents the size of the grid cell at the spatial position r, ∈[0.05, 0.5]mm, and k represents the current round identifier.

[0051] In specific applications, in the iterative process of step 102, the two-dimensional grid can be dynamically refined by using the solution of the previous round in the iteration through an adaptive mesh refinement toolkit (ameshref, Adaptive Mesh Enhancement Suite - High Refinement). Figure 2 c shows the process of adaptive mesh refinement. Referring to Figure 2 c, the absorption coefficient distribution obtained in the previous round of reconstruction is used to guide the adaptive mesh refinement, and denser grid division is performed in the high absorption region to enhance the detail resolution.

[0052] In step 103, the third absorption coefficient distribution generated by the optimized INR network is the final reconstructed absorption coefficient distribution.

[0053] Figure 3 shows the qualitative results of reconstructing the absorption coefficient image on the photoacoustic digital brain image by different methods. Figure 3 In the figure, 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 of the embodiments of the present disclosure, and GT represents the true absorption coefficient distribution. ByFigure 3 As can be seen, SBDC only captures general structures and has low accuracy in reconstructing absorption coefficients in deep tissue. Although TSIA can reconstruct most absorption coefficients well, it also lacks precision in deep tissue regions. The U-shaped network (U-net) and EAP reconstruction network (EAPNet) suffer from global inaccuracies. The optimized INR network significantly outperforms these methods in terms of precision and accuracy in recovering absorption coefficients, especially in deep tissue reconstruction, where the precision and accuracy are significantly higher than those of other methods.

[0054] Figure 4 The qualitative results of different methods for reconstructing absorption coefficient images on phantom data are shown. Figure 4 From left to right in the figure are: PAT image, SBDC reconstruction result, TSIA reconstruction result, EAPNet reconstruction result, Ours represents the reconstruction result of the optimized INR network of the embodiment of the present disclosure, and Reference represents the reference absorption coefficient distribution. Figure 4 The optimized INR network performs well in recovering the absorption coefficients of various substances. Although EAPNet can satisfactorily recover the absorption coefficient of each component, the error map shows that it suffers from inaccurate edge reconstruction. The results of SBDC and TSIA are less than satisfactory, as there is no significant difference between them due to the simple model segmentation and uniform absorption coefficient within each component. Intuitively, the optimized INR network significantly outperforms the above methods in terms of absorption coefficient reconstruction accuracy and edge accuracy, further verifying that the method provided by the embodiments of the present disclosure can effectively improve the accuracy of absorption coefficient reconstruction.

[0055] Figure 5 FIG2 shows a schematic diagram of the structure of a device for reconstructing absorption coefficients of photoacoustic tomography based on implicit neural representation provided by an embodiment of the present disclosure. The device for reconstructing absorption coefficients of photoacoustic tomography based on implicit neural representation is applied to an electronic device 600. Figure 5 The device 500 for reconstructing photoacoustic tomography absorption coefficients based on implicit neural representation may include: A priori embedding unit 501 is configured to obtain a first absorption coefficient distribution using a detector measurement signal based on an SBDC method, and embed the first absorption coefficient distribution as priori information into an implicit neural representation network; The parameter optimization unit 502 is configured to optimize the parameters of the implicit neural representation network, including iteratively performing the following steps until a preset convergence condition is met: evaluating the absorption coefficient value at each spatial position in the photoacoustic imaging region using the implicit neural representation network to generate a continuous second absorption coefficient distribution, calculating the optical flux field under the second absorption coefficient distribution by a finite element method forward solver to obtain an optical flux distribution, simulating a predicted detector signal based on the optical 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 measured signal to update the parameters of the implicit neural representation network. The reconstruction execution unit 503 is configured to evaluate the absorption coefficient value at each spatial position in the photoacoustic imaging region using the implicit neural representation network with the optimized parameters to generate a continuous third absorption coefficient distribution.

[0056] Further, the photoacoustic tomographic absorption coefficient reconstruction device based on the implicit neural representation 500 can further include an encoding unit 504 configured to encode each spatial position of the photoacoustic imaging region using a hash encoding method.

[0057] Further, the photoacoustic tomographic absorption coefficient reconstruction device based on the implicit neural representation 500 can further include a gridding unit 505 configured to discretize the reconstruction object spatial domain into a two-dimensional grid, and the spatial coordinates of a grid node in the two-dimensional grid represent a spatial position in the photoacoustic imaging region.

[0058] Further, the photoacoustic tomographic absorption coefficient reconstruction device based on the implicit neural representation 500 can further include an adaptive refinement unit 506 configured to adaptively refine the two-dimensional grid according to the second absorption coefficient distribution obtained in the previous round of reconstruction in a preset interval of iterations or a randomly selected one or more rounds of iterations, so that the grid node density of the two-dimensional grid is positively correlated with the absorption coefficient value thereof.

[0059] For other technical details of the photoacoustic tomographic absorption coefficient reconstruction device based on the implicit neural representation 500, refer to the foregoing photoacoustic tomographic absorption coefficient reconstruction method based on the implicit neural representation, which will not be described here. In specific applications, the photoacoustic tomographic absorption coefficient reconstruction device based on the implicit neural representation 500 can be implemented by the electronic device 600 below, or can be implemented as software in the electronic device 600.

[0060] In addition, the embodiments of the present disclosure also provide a computer readable storage medium having a computer program stored thereon, the program including instructions which, when executed by one or more processors of a computing device, perform the steps of the foregoing photoacoustic tomographic absorption coefficient reconstruction method based on the implicit neural representation.

[0061] Figure 6 A structural schematic diagram of an electronic device provided by the embodiments of the present disclosure is shown. Referring toFigure 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 shown in the above embodiments of the present disclosure and / or the program units corresponding to each unit in the device.

[0062] The various components are interconnected using various buses and may be mounted on a common motherboard or in other ways as needed. Processor 601 may process instructions for execution within the electronic device, including instructions stored in or on memory for displaying graphical information of a user interface on an external input / output device (such as a display device coupled to the interface). In other embodiments, multiple processors and / or multiple buses may be used with multiple memories and multiple storage devices, if desired.

[0063] The processor 601 may include one or more single-core processors or multi-core processors. The processor 601 may include any combination of general-purpose processors or dedicated processors (such as an image processor, an application processor, a baseband processor, etc.).

[0064] The memory 602 is a computer-readable storage medium provided by the present disclosure, which can be used to store non-transient software programs, non-transient computer executable programs and units, such as the embodiment of the present disclosure. Figure 1 The program instructions / units corresponding to the method for photoacoustic tomography absorption coefficient reconstruction based on implicit neural representation are shown. The processor 601 executes the non-transient software program, instructions and units stored in the memory 602 to perform the above method embodiments. Figure 1 The programs, instructions and units corresponding to the photoacoustic tomography absorption coefficient reconstruction method based on implicit neural representation are shown.

[0065] The electronic device 600 may further include an input device 603 and an output device 604. The processor 601, the memory 602, the input device 603 and the output device 604 may be connected via a bus or other means. Figure 6 The bus connection is taken as an example.

[0066] The programs (also referred to as software, software applications, or code) described above include machine instructions for a programmable processor and may be implemented using an object-oriented programming language, assembly, or machine language.

[0067] With the development of time and technology, the medium meaning is more and more extensive, and the propagation path of the computer program is no longer limited to a tangible medium, but can also be directly downloaded from a network, etc. Any combination of one or more computer-readable storage media can be used. The computer-readable storage medium can use, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device or instrument, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: portable computer disk, hard disk, 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 device, magnetic storage device, or any suitable combination of the above. In this document, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or instrument.

[0068] The above describes the technical solutions provided by the present disclosure in detail. The principles and implementation manners of the present disclosure are described by applying specific examples. The above examples are only used to help understand the method and core idea of the present disclosure; at the same time, for those skilled in the art, according to the idea of the present disclosure, the specific implementation manner and application range will be changed. In summary, the content of the specification should not be understood as a limitation of the present disclosure.

[0069] The above only describes the preferred embodiments of the present disclosure and does not limit the present disclosure. Any modification, equivalent replacement, etc. made within the spirit and principles of the present disclosure should be included in the protection scope of the present disclosure.

Claims

1. A method for reconstructing absorption coefficients of photoacoustic tomography based on implicit neural representation, characterized in that: The method comprises: A first absorption coefficient distribution is obtained by using a detector measurement signal based on the SBDC method, and the first absorption coefficient distribution is embedded into an implicit neural representation network as prior information; Optimizing parameters of the implicit neural representation network, wherein optimizing the parameters of the implicit neural representation network comprises iteratively performing the following steps until a preset convergence condition is satisfied: using the implicit neural representation network to evaluate absorption coefficient values ​​at each spatial position within a photoacoustic imaging region to generate a continuous second absorption coefficient distribution, calculating a light flux field under the second absorption coefficient distribution using a finite element method forward solver to obtain a light flux distribution, using a photoacoustic tomography physical model to simulate based on the light flux distribution and the second absorption coefficient distribution to obtain a predicted detector signal, and minimizing a 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 after optimizing parameters is used to evaluate the absorption coefficient value at each spatial position in the photoacoustic imaging area to generate a continuous third absorption coefficient distribution.

2. The method according to claim 1, characterized in that The 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 calculate 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 objective loss function is as follows: in, represents the loss value, represents the detector measurement signal, represents the predicted detector signal, r represents the grid coordinate, represents the second absorption coefficient distribution, represents the luminous flux distribution," ” denotes the dot product.

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

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

6. The method according to claim 5, characterized in that The optimizing the parameters of the implicit neural representation network further includes: in a preset interval iteration or one or more randomly selected iterations, adaptively refining the two-dimensional grid according to the second absorption coefficient distribution obtained from the previous round of 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 multi-layer perceptron.

8. A device for reconstructing absorption coefficients of photoacoustic tomography based on implicit neural representation, characterized in that: The device for reconstructing photoacoustic tomography absorption coefficient based on implicit neural representation includes: a priori embedding unit, configured to obtain a first absorption coefficient distribution using a detector measurement signal based on an SBDC method, and embed the first absorption coefficient distribution as priori information into an implicit neural representation network; a parameter optimization unit, configured to optimize the parameters of the implicit neural representation network, comprising iteratively performing the following steps until a preset convergence condition is satisfied: using the implicit neural representation network to evaluate absorption coefficient values ​​at each spatial position within a photoacoustic imaging region to generate a continuous second absorption coefficient distribution, calculating a light flux field under the second absorption coefficient distribution using a finite element method forward solver to obtain a light flux distribution, using a photoacoustic tomography physical model to simulate a predicted detector signal based on the light flux distribution and the second absorption coefficient distribution, and minimizing a difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network; A reconstruction execution unit is used to evaluate the absorption coefficient value at each spatial position in the photoacoustic imaging area 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: One or more processors and a memory storing a program, wherein the program includes instructions, and when the instructions are executed by the processor, the processor performs the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a program, wherein the program comprises instructions, and when the instructions are executed by one or more processors, the processors are caused to perform the method according to any one of claims 1 to 7.

Images