Multi-sound-velocity photoacoustic image reconstruction method and device based on implicit neural representation

Abstract

The invention provides a multi-sound-velocity photoacoustic image reconstruction method and device based on implicit neural representation, and relates to the technical field of image processing. The method comprises the following steps: adjusting parameters of an implicit neural representation network: obtaining first initial sound pressure distribution and first sound velocity distribution through the implicit neural representation network based on space coordinates in an imaging area, obtaining a predicted detector signal based on the first initial sound pressure distribution and the first sound velocity distribution through a differentiable acoustic forward propagation model, and minimizing the difference between the predicted detector signal and the detector measurement signal to update parameters of the implicit neural representation network; iteration is carried out in this way until a preset convergence condition is met; and using the implicit neural representation network after parameter adjustment to obtain second initial sound pressure distribution based on the space coordinates in the imaging area. According to the invention, photoacoustic image reconstruction can be realized without a sound velocity distribution diagram or a pairing truth value.

Description

Technical Field

[0001] The present disclosure relates to the field of image processing technology, and in particular to a multi-speed photoacoustic image reconstruction method and device based on implicit neural representation. Background Art

[0002] Photoacoustic imaging (PAI), an emerging hybrid medical imaging modality, combines the high contrast of optical imaging with the deep tissue penetration of ultrasound imaging, showing significant potential for application in biomedical fields such as brain functional imaging and early tumor detection. Its imaging principle involves irradiating biological tissue with a pulsed laser. Absorbers within the tissue, such as hemoglobin, absorb the light energy, generating transient thermal expansion and stimulating an ultrasonic signal (i.e., a photoacoustic signal). This signal is received by a detector array, and a reconstruction algorithm is used to invert and reconstruct an image of the light absorption distribution within the tissue. However, the practical application of photoacoustic imaging, particularly in complex scenarios such as transcranial imaging, faces a key technical bottleneck: the inherent acoustic heterogeneity of biological tissue. For example, different tissue types, such as fat, muscle, soft tissue, and skull bone, exhibit significant differences in acoustic properties such as sound velocity and density.

[0003] Current mainstream photoacoustic image reconstruction algorithms, such as Delay-and-Sum (DAS), Filtered Back Projection (FBP), and Time Reversal (TR), are generally based on a key assumption for computational simplicity: acoustic waves propagate at a constant average velocity in a homogeneous medium. This simplifying assumption is applicable in soft tissues with relatively uniform acoustic properties. However, in scenarios involving interfaces with high acoustic impedance differences, such as those involving sound waves traveling through the skull, significant refraction, reflection, and scattering occur when photoacoustic signals pass through multiple layers of tissue with vastly different acoustic properties, such as the skull. This can lead to significant deviations in the actual propagation path and propagation time of the acoustic waves from the results calculated under the homogeneous medium assumption. At this time, traditional reconstruction algorithms based on the constant sound velocity assumption will inevitably introduce severe image distortion. For example, the position, shape, and size of target structures (such as blood vessels and lesions) in the reconstructed image will be significantly distorted, and the spatial resolution and contrast of the image will be severely degraded. For example, the recognition of subtle structures such as tiny blood vessels will be significantly reduced. At the same time, non-real structures such as arc artifacts or strip artifacts caused by signal misfocusing will also be generated. Therefore, a photoacoustic image reconstruction method that can effectively compensate for or correct the influence of acoustic heterogeneity is needed to improve the imaging quality of photoacoustic imaging in complex scenes. Summary of the Invention

[0004] In view of this, the present disclosure provides a multi-speed photoacoustic image reconstruction method and device based on implicit neural representation.

[0005] According to a first aspect of the present disclosure, a multi-speed photoacoustic image reconstruction method based on implicit neural representation is provided, the method comprising: Adjusting parameters of an implicit neural representation network, the adjusting the parameters of the implicit neural representation network comprising: obtaining a first initial sound pressure distribution and a first sound velocity distribution based on spatial coordinates within an imaging region through the implicit neural representation network, simulating a propagation process of a sound pressure wave excited by the first initial sound pressure distribution in a medium defined by the first sound velocity distribution through a differentiable acoustic forward propagation model to obtain a predicted detector signal, and minimizing a difference between the predicted detector signal and a detector measurement signal to update the parameters of the implicit neural representation network; iterating in this manner until a preset convergence condition is satisfied; The second initial sound pressure distribution is obtained based on the spatial coordinates within the imaging area using the implicit neural representation network with adjusted parameters.

[0006] In some embodiments of the first aspect of the present disclosure, the adjusting the parameters of the implicit neural representation network also includes: obtaining the first density distribution while obtaining the first initial sound pressure distribution and the first sound speed distribution; simulating the propagation process of the sound pressure wave excited by the first initial sound pressure distribution in the medium defined by the first sound speed distribution through the differentiable acoustic forward propagation model to obtain the predicted detector signal, specifically: simulating the propagation process of the sound pressure wave excited by the first initial sound pressure distribution in the medium defined by the first sound speed distribution and the first density distribution through the differentiable acoustic forward propagation model to obtain the predicted detector signal.

[0007] In some embodiments of the first aspect of the present disclosure, the method further includes: pre-training the implicit neural representation network before adjusting the parameters of the implicit neural representation network; wherein, pre-training the implicit neural representation network includes: using the DAS algorithm to preliminarily reconstruct the detector measurement signal to obtain a third initial sound pressure distribution, using the third initial sound pressure distribution as the supervision target of the initial boost distribution, initializing the density distribution and the sound speed distribution to predetermined constants respectively and constraining their physical consistency through the differentiable acoustic forward propagation model, and executing the implicit neural representation network for pre-training.

[0008] In some embodiments of the first aspect of the present 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 the mean square error between the predicted detector signal and the detector measurement signal, using the mean square error to perform back propagation to obtain the gradient and updating the parameters of the implicit neural representation network based on the gradient.

[0009] In some embodiments of the first aspect of the present disclosure, the method further includes: obtaining a synthetic detector signal based on the second initial sound pressure distribution, a predetermined third sound speed distribution and a predetermined third density distribution through a differentiable acoustic forward model; and reconstructing the synthetic detector signal using a DAS algorithm to obtain a fourth initial sound pressure distribution.

[0010] In some embodiments of the first aspect of the present disclosure, the method further includes: using an implicit neural representation network with adjusted parameters to obtain a second initial sound pressure distribution based on spatial coordinates within the imaging area, while obtaining a second sound speed distribution and / or a second density distribution.

[0011] In some implementations of the first aspect of the present disclosure, the implicit neural representation network is a multi-layer perceptron MLP with position encoding, and the differentiable acoustic forward propagation model is a j-Wave simulator implemented based on a JAX library.

[0012] According to a second aspect of the present disclosure, a multi-speed photoacoustic image reconstruction device based on implicit neural representation is provided, the multi-speed photoacoustic image reconstruction device based on implicit neural representation comprising: a parameter fine-tuning unit, configured to adjust parameters of an implicit neural representation network, wherein adjusting the parameters of the implicit neural representation network comprises: obtaining a first initial sound pressure distribution and a first sound velocity distribution based on spatial coordinates within an imaging region through the implicit neural representation network, simulating a propagation process of a sound pressure wave excited by the first initial sound pressure distribution in an inhomogeneous medium defined by the first sound velocity distribution through a differentiable acoustic forward propagation model to obtain a predicted detector signal, and minimizing a difference between the predicted detector signal and a detector measurement signal to update the parameters of the implicit neural representation network; and iterating in this manner until a preset convergence condition is satisfied; The reconstruction execution unit is used to obtain a second initial sound pressure distribution based on the spatial coordinates in the imaging area using the implicit neural representation network with adjusted parameters.

[0013] 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.

[0014] According to a fourth aspect of the present disclosure, a computer-readable storage medium storing a program is provided, wherein the program includes instructions, which, when executed by one or more processors of a computing device, cause the computing device to perform the above-mentioned method.

[0015] As can be seen from the above technical solution, the embodiments of the present disclosure can achieve joint implicit estimation and compensation of the initial sound pressure distribution (i.e., photoacoustic image) and spatially varying acoustic parameters (sound velocity, density) through a single neural network without the need to pre-acquire a sound velocity distribution map or paired true values, significantly improving the reconstruction quality of photoacoustic images, including structural fidelity, detail clarity, and artifact suppression capabilities, especially the quality of photoacoustic image reconstruction in complex acoustic environments such as transcranial imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0017] Figure 1 A flowchart of a multi-speed photoacoustic image reconstruction method based on implicit neural representation provided by an embodiment of the present disclosure; Figure 2 Schematic diagram of a single execution process of adjusting INR network parameters involved in an embodiment of the present disclosure; Figure 3 A schematic diagram of the process of implementing transcranial photoacoustic imaging reconstruction using the method provided in an embodiment of the present disclosure; Figure 4 A comparison chart of the reconstruction results on a two-dimensional human brain digital phantom using the method of an embodiment of the present disclosure, the true initial sound pressure distribution, and the reconstruction results on a two-dimensional human brain digital phantom using related technologies; Figure 5 A schematic structural diagram of a multi-speed photoacoustic image reconstruction device based on implicit neural representation provided by an embodiment of the present disclosure; Figure 6 A schematic structural block diagram of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present disclosure in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present disclosure without making any creative efforts shall fall within the scope of protection of the present disclosure.

[0019] The terms used in the embodiments of the present disclosure are for the purpose of describing specific embodiments only and are not intended to limit the present disclosure. The singular forms "a," "an," "the," 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.

[0020] As used herein, the words "if," "if," and the like may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0021] In related technologies, the following are the main technical solutions for reconstructing photoacoustic images: 1) Correction method based on a priori sound velocity maps: A sound velocity distribution map of the imaging area is obtained in advance using other imaging modalities such as ultrasound transmission tomography (UTT), computed tomography (CT), or magnetic resonance imaging (MRI). This map is then used to correct the reconstruction algorithm. The disadvantages of this method are that it requires additional hardware, increasing the system's hardware complexity and cost. Furthermore, registration errors between multimodal images can affect correction accuracy.

[0022] 2) Joint reconstruction of sound velocity and photoacoustic image: This approach treats both the sound velocity map and the initial sound pressure map as unknown quantities and solves them simultaneously within an iterative optimization framework. This method is highly nonlinear and non-convex, prone to local optima, sensitive to initial guesses, computationally expensive, and difficult to converge.

[0023] 3) Supervised learning-based methods: Deep neural networks such as U-net are used to learn the mapping from low-quality images with sound velocity distortion to high-quality ground-truth images. This method relies heavily on large-scale, high-quality paired training datasets (i.e., distorted images and their corresponding undistorted ground-truth images). In real-world scenarios, obtaining such paired ground-truth data is extremely difficult or even impossible, which limits its generalization and reliability in practical applications.

[0024] Therefore, there is an urgent need for a new method that does not rely on external sound velocity maps or paired true value data, can effectively deal with the problem of sound velocity heterogeneity and reconstruct high-quality photoacoustic images.

[0025] In view of this, the embodiments of the present disclosure provide the following multi-sound-speed photoacoustic image reconstruction method and device based on implicit neural representation, which effectively corrects image distortion and artifacts caused by sound speed inhomogeneity (especially the skull) without the need to pre-acquire a sound speed distribution map or paired true values. Through a single neural network, the initial sound pressure distribution (i.e., photoacoustic image) and spatially varying acoustic parameters (sound speed, density) are jointly implicitly estimated and compensated, significantly improving the reconstruction quality of photoacoustic images, including structural fidelity, detail clarity, and artifact suppression capabilities, especially the quality of photoacoustic image reconstruction in complex acoustic environments such as transcranial imaging.

[0026] Figure 1 The flowchart of the multi-speed photoacoustic image reconstruction method based on implicit neural representation provided by the embodiment of the present disclosure is shown. Figure 1 , the method of the embodiment of the present disclosure may include the following steps: Step 101: Adjusting parameters of an implicit neural representation (INR) network. Adjusting the parameters of the INR network includes: obtaining a first initial sound pressure distribution and a first sound velocity distribution based on spatial coordinates within an imaging region through the INR network; simulating the propagation of a sound pressure wave excited by the first initial sound pressure distribution in a non-uniform medium defined by the first sound velocity distribution through a differentiable acoustic forward propagation model to obtain a predicted detector signal; and minimizing the difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network. This process is repeated until a preset convergence condition is met. Step 102 : Using the INR network with adjusted parameters, a second initial sound pressure distribution is obtained based on the spatial coordinates within the imaging area.

[0027] The method of the disclosed embodiment utilizes a single neural network to parameterize an unknown physical field without the need for pre-acquisition of a sound velocity distribution map or paired true value supervision data. It then combines this with a differentiable physical model for end-to-end optimization, thereby achieving joint implicit estimation and compensation of the initial sound pressure distribution (i.e., the photoacoustic image) and spatially varying acoustic parameters.

[0028] In the disclosed embodiment, the INR network may be a multilayer perceptron (MLP) with position encoding. The input of the INR network is the spatial coordinate r(x, y) within the imaging area, and the output of the INR network is the values ​​of multiple physical fields corresponding to each spatial coordinate within the imaging area. The multiple physical fields may be the initial sound pressure distribution, the sound speed distribution, and the density distribution, respectively. In one example, the multiple physical fields output by the INR network may be: the normalized initial sound pressure distribution p0(r) and the sound speed distribution c(r). In another example, for complex scenarios such as transcranial imaging, the multiple physical fields may include: the initial sound pressure p0(r), the sound speed c(r), and the density ρ(r).

[0029] The disclosed embodiments use a single INR network shared parameters to simultaneously represent the initial sound pressure field and the acoustic medium field, which establishes implicit constraints between the physical fields and achieves a compact representation of the model parameters.

[0030] Exemplarily, the INR network can be constructed as a multilayer perceptron with a total of 8 layers and 512 neurons in each layer, and the activation function is ReLU. A position encoding module is set before the input layer of the multilayer perceptron, and the position encoding module can be used to map the spatial coordinate r(x, y) to a high-dimensional feature space to enhance the representation capability of high-frequency details. The output layer of the multilayer perceptron can be set with 3 neurons, which correspond to the initial sound pressure distribution p0, the sound speed distribution c and the density distribution ρ respectively, and the outputs of these 3 neurons are constrained to a reasonable physical range through activation functions such as tanh and sigmoid and linear transformations. The reasonable physical range can be pre-set. For example, the physical range of the sound speed distribution can be set to [1450,3500] in m / s.

[0031] In the disclosed embodiments, the differentiable acoustic forward propagation model may be, but is not limited to, a j-Wave simulator implemented using the JAX library. Specifically, the differentiable acoustic forward propagation model may use the differentiable simulator Fsim, employing j-Wave, with initial settings consistent with the detector geometry and temporal sampling rate used.

[0032] Figure 2 FIG. 1 is a schematic diagram showing a single execution process of adjusting the INR network parameters in step 101 . Figure 2 In the forward model, the forward model is a differentiable acoustic forward propagation model, F θ represents the INR network, the coordinate r is the spatial coordinate within the imaging area, p0 represents the initial sound pressure distribution, c represents the sound velocity distribution, P pred represents the predicted detector signal, P means Represents the detector measurement signal, and Loss represents the difference between the predicted detector signal and the detector measurement signal. Figure 2 , the spatial coordinate r within the imaging area is input to the INR network F θ To generate the physical field (p0, c), and then obtain the predicted detector signal P through the differentiable forward model F pred , predict the detector signal P pred With the detector measuring signal P meas The loss L is calculated, and the parameters θ of the INR network are updated through backpropagation. This closed-loop optimization process is repeated multiple times until a predetermined convergence condition is reached. In specific applications, this convergence condition can be flexibly set. For example, the convergence condition may be: the number of iterations reaches a preset value; or the loss value falls below a preset threshold.

[0033] In some examples, in step 101, a first density distribution can be obtained simultaneously with the first initial sound pressure distribution and the first sound velocity distribution. Simulating the propagation of sound pressure waves excited by the first initial sound pressure distribution in the medium defined by the first sound velocity distribution using a differentiable acoustic forward propagation model to obtain a predicted detector signal specifically includes: simulating the propagation of sound pressure waves excited by the first initial sound pressure distribution in the medium defined by the first sound velocity distribution and the first density distribution using a differentiable acoustic forward propagation model to obtain a predicted detector signal. This method is therefore applicable to complex scenarios such as cranial brain imaging.

[0034] In step 101, the differentiable acoustic forward propagation model uses the physical field (p0(r), c(r), ρ(r)) generated by the INR network or the physical field (p0(r), c(r)) as input, and numerically simulates the propagation process of the sound pressure wave excited by the initial sound pressure p0(r) in the inhomogeneous medium based on the acoustic wave equation. The acoustic properties of the medium are defined by the sound velocity c(r) and density ρ(r), and outputs the simulated time series signal at the detector array position. The time series signal at the detector array position is the predicted detector signal Ppred.

[0035] In step 101, without requiring any ground truth image, the parameters of the INR network are updated by minimizing the difference between the predicted detector signal and the detector measurement signal. Specifically, the parameters of the implicit neural representation network are updated by minimizing the difference between the predicted detector signal and the detector measurement signal. This includes calculating the mean squared error between the predicted and measured detector signals, performing backpropagation using the mean squared error to determine the gradient, and updating the parameters of the implicit neural representation network based on the gradient.

[0036] Specifically, a first loss function is predefined to measure the difference between the predicted detector signal and the detector measurement signal. For example, the first loss function may be a mean square error (MSE) as shown in the following equation (1).

[0037] L(θ) = ||Ppred -P meas ||²(1) Where L(θ) represents the loss between the predicted detector signal and the detector measurement signal, P pred represents the predicted detector signal, P meas Indicates the detector measurement signal.

[0038] Specifically, a gradient descent optimization algorithm such as the Adam optimizer can be used to update the parameters θ of the INR network according to the first loss function. Here, since the entire process is end-to-end differentiable, the gradient of each step can be efficiently calculated through automatic differentiation.

[0039] As can be seen from the above, the embodiment of the present disclosure only uses the original measurement signal P meas The optimization of the INR network can be achieved as the optimization target, and the unsupervised adjustment of the INR network parameters can be realized, thus getting rid of the dependence on paired true value data.

[0040] Furthermore, before step 101, the method of the embodiment of the present disclosure may further include: pre-training the INR network. Specifically, pre-training the INR network includes: using the DAS algorithm to preliminarily reconstruct the detector measurement signal to obtain a third initial sound pressure distribution, using the third initial sound pressure distribution as a supervisory target for the initial boost pressure distribution, initializing the density distribution and sound velocity distribution to predetermined constants and constraining their physical consistency through a differentiable acoustic forward propagation model, and executing an implicit neural representation network for pre-training.

[0041] When pre-training the INR network, a second loss function can be used. The second loss function is the sum of the reconstruction loss and the physical consistency loss. The reconstruction loss can be the mean square error between the third initial sound pressure distribution and the predicted initial sound pressure distribution currently generated by the INR. The physical consistency loss can be the mean square error between the predicted detector signal and the detector measurement signal obtained by the differentiable acoustic forward propagation model with the predicted initial sound pressure distribution currently generated by the INR, the density distribution set to a predetermined constant, and the sound speed distribution set to a predetermined constant as input.

[0042] Specifically, first use DAS to measure the detector signal P meas A rough, distorted initial sound pressure distribution is obtained through preliminary reconstruction. This initial sound pressure distribution is the aforementioned third initial sound pressure distribution. This rough, distorted initial sound pressure distribution is then used as the supervisory target for the initial sound pressure distribution p0(r) predicted by the INR network. The sound velocity distribution c(r) is set to a predetermined constant c, and the density distribution ρ(r) is set to a predetermined constant ρ. The INR network is pre-trained for a small number of iterations to obtain the initial INR network parameters. During this pre-training process, the spatial coordinate r within the imaging area is fixed.

[0043] Pre-training can provide a more reasonable starting point for the unsupervised optimization of the subsequent step 101, which helps to accelerate the convergence of step 101 and avoid falling into undesirable local optimality.

[0044] In step 102, all coordinate points r of the imaging area are input into the INR network optimized in step 101 to obtain a second initial acoustic pressure distribution, which serves as the reconstructed photoacoustic image. Simultaneously, the INR network can also be used to obtain a second acoustic velocity distribution and a second density distribution, which can serve as adjuncts to the reconstructed photoacoustic image.

[0045] Furthermore, after step 102, the method may further include: obtaining a synthetic detector signal based on the second initial sound pressure distribution, a predetermined third sound velocity distribution, and a predetermined third density distribution using a differentiable acoustic forward model; and reconstructing the synthetic detector signal using a DAS algorithm to obtain a fourth initial sound pressure distribution. This allows post-processing of the reconstructed photoacoustic image to remove distortion and artifacts, further improving the quality of the reconstructed photoacoustic image.

[0046] Specifically, the learned initial sound pressure field (i.e., the second initial sound pressure distribution) is extracted from the optimized INR network. The non-uniform acoustic parameters learned by the network (i.e., the second sound velocity distribution and the second density distribution) are ignored. An ideal, uniform virtual acoustic medium (i.e., the medium defined by the third sound velocity distribution and the third density distribution) is set. The initial sound pressure field is used as the sound source. A forward simulation is performed again in this ideal uniform medium using the differentiable acoustic forward model to generate a set of theoretically undistorted synthetic detector signals. Finally, the synthetic detector signals are reconstructed using a standard reconstruction algorithm such as DAS to obtain a fourth initial sound pressure distribution, which is the corrected high-quality reconstructed photoacoustic image. The third sound velocity distribution and the third density distribution can be pre-set. For example, the third sound velocity distribution can be pre-set to 1540 m / s, and the third density distribution can be pre-set to 1000 kg / m³.

[0047] Therefore, by utilizing the learned sound source information and ideal propagation model, the residual artifacts introduced by the complex acoustic propagation path are effectively "cleaned", further improving the clarity and fidelity of the reconstructed photoacoustic image.

[0048] The method of the disclosed embodiment utilizes the actual measured photoacoustic signal to reconstruct the photoacoustic image, eliminating the need for paired ground-truth images for supervised training. This significantly reduces the need for data such as ground-truth images while improving the quality of photoacoustic image reconstruction, and can be directly applied to imaging scenarios where ground-truth images are difficult to obtain. Furthermore, the disclosed embodiment uses an algorithmic approach to estimate and compensate for changes in sound velocity, eliminating the need for additional hardware to measure the sound velocity map, as required by methods such as UTT. This maintains the simplicity of the photoacoustic imaging system, reduces costs, and reduces potential registration errors.

[0049] The disclosed embodiment uses the INR network to represent the entire physical field as the weights of a small neural network, achieving high data compression, a compact model and continuous representation, and can query images at any resolution, avoiding the limitations brought by pixel / voxel discretization.

[0050] Simulation experiment results show that the photoacoustic image reconstructed by the method of the embodiment of the present disclosure is significantly superior to traditional reconstruction algorithms in objective indicators such as peak signal-to-noise ratio (PSNR) and structural similarity (SSIM). Especially in complex scenarios such as transcranial imaging, it can effectively remove geometric distortion and artifacts and clearly restore fine structures such as intracranial blood vessels.

[0051] In addition, the disclosed embodiments can also provide multi-physical field information. In addition to reconstructing high-quality photoacoustic images, they can also simultaneously output a sound velocity distribution map and a density distribution map that are consistent with the physical scene, which provides additional valuable information for tissue characterization and quantitative analysis.

[0052] The specific implementation process of the embodiment of the present disclosure is described in detail below using the two-dimensional simulation of transcranial photoacoustic imaging as an example.

[0053] Before the reconstruction process, the following data preparation is performed to obtain a measurement signal that includes the skull distortion effect, which is an example of the detector measurement signal mentioned above: A two-dimensional human brain phantom was constructed, including structures such as the scalp, skull, cerebrospinal fluid, brain parenchyma, and blood vessels. Different optical and acoustic parameters were assigned to different tissues. Optical parameters included but were not limited to absorption coefficient and scattering coefficient, while acoustic parameters included sound velocity and density. The actual initial sound pressure distribution p generated in the two-dimensional human brain phantom after pulsed laser irradiation was calculated through Monte Carlo optical simulation. 0GT , the actual initial sound pressure distribution p 0GT As the sound source, the acoustic parameters of the two-dimensional human brain numerical phantom are used as the medium, and the acoustic wave propagation is simulated by acoustic simulation tools such as k-Wave and the measurement signal P received by the ring detector array, which includes the skull distortion effect, is recorded. meas .

[0054] Figure 3The figure shows a schematic diagram of the process of implementing transcranial photoacoustic imaging reconstruction using the method provided by the embodiment of the present disclosure. Figure 3 The transcranial photoacoustic imaging reconstruction process may include the following steps a1 to a3: Step a1, initialization, obtain the initial INR network F θ ; For the measurement signal P meas Perform a standard DAS reconstruction, assuming a uniform sound speed of 1540 m / s (i.e., the third sound speed distribution mentioned above), and obtain the rough initial sound pressure distribution p 0DAS , the rough initial sound pressure distribution p 0DAS is an example of the aforementioned third initial sound pressure distribution. θ Pre-training is performed, specifically: fix the spatial coordinates r(x,y) within the imaging area (r∈R 2 ), with a rough initial sound pressure distribution p 0DAS As the supervision target of the initial sound pressure distribution p0, the sound speed distribution c is set to a constant "1540", and the density distribution ρ is set to a constant "1000" (i.e., the third density distribution mentioned above). The Adam optimizer is used to perform about 1000 iterations to obtain the initialized INR network parameters.

[0055] Step a2, unsupervised optimization, obtain the INR network F after adjusting the parameters θ* ; Based on the initialized INR network parameters, perform unsupervised optimization of the INR network, iterating 10,000 times or other times to obtain the adjusted INR network parameters. Specifically, each iteration of the unsupervised optimization performs the following steps: Step b1, input the spatial coordinates r(x, y) within the imaging area into the INR network to obtain the predicted physical field (p 0pred , c pred ,ρ pred ), p 0pred represents the first initial sound pressure distribution, c pred represents the first sound velocity distribution, ρ pred Represents the first density distribution; Step b2, input the predicted physical field into the forward model F sim , i.e., the j-Wave simulator (i.e., the differentiable acoustic forward propagation model), calculates the predicted detector signal P pred ; Step b3, calculate the predicted detector signal P pred With the real measurement signal P meas The mean square error loss Loss between them; step b4, calculate the gradient of the mean square error loss Loss with respect to the INR network parameters by automatic differentiation; step b5, use the Adam optimizer to update the INR network parameters according to the gradient.

[0056] Step a3: post-processing to obtain the fourth initial sound pressure distribution p final .

[0057] Specifically, from the INR network F after adjusting the parameters θ* Extract the learned initial sound pressure field p0* (i.e., the second initial sound pressure distribution in the previous text) and call the forward model F sim The initial sound pressure field p0* is used as the sound source in a uniform medium (i.e., c ideal =1540m / s,ρ ideal =1000kg / m³) and forward propagation is performed to obtain the synthesized distortion-free signal P synth (i.e., the synthesized detector signal mentioned above), the synthesized undistorted signal P synth Perform a DAS reconstruction to obtain the fourth initial sound pressure distribution p final , the fourth initial sound pressure distribution p final This is the reconstructed photoacoustic image.

[0058] The results are evaluated as follows: the fourth initial sound pressure distribution p final and the actual initial sound pressure distribution p 0GT By comparison, quantitative analysis shows that the fourth initial sound pressure distribution p final The peak signal-to-noise ratio (PSNR) reaches 33.42dB, and the structural similarity index (SSIM) reaches 0.939, which are significantly higher than the results of DAS (17.57dB, 0.629), TR (24.10dB, 0.718) and supervised U-net (31.38dB, 0.948). This proves that the method provided by the embodiments of the present disclosure has good effectiveness and superiority compared with related technologies.

[0059] Figure 4 The results of the reconstruction using the method of the embodiment of the present disclosure on the two-dimensional human brain digital phantom are shown, and the actual initial sound pressure distribution p is shown. 0GT , and a comparison of the reconstruction results of related technologies (DAS, TR, U-net) on a two-dimensional human brain digital phantom. Figure 4In the figure, from left to right are: the reconstruction results of the Delay-and-Sum (DAS) algorithm on a two-dimensional human brain digital phantom, the reconstruction results of the Time Reversal Algorithm (TR) on a two-dimensional human brain digital phantom, the reconstruction results of the supervised U-shaped Convolutional Neural Network (U-Net) on a two-dimensional human brain digital phantom, the reconstruction results of the method of the embodiment of the present disclosure on a two-dimensional human brain digital phantom, and the ideal ground truth (GT) image without skull distortion (i.e., the actual initial sound pressure distribution p 0GT ).pass Figure 4 It can be intuitively seen that the method of the embodiment of the present disclosure is superior to the related art in terms of structural fidelity and artifact suppression.

[0060] Figure 5 FIG2 shows a schematic diagram of the structure of a multi-speed photoacoustic image reconstruction device based on implicit neural representation provided by an embodiment of the present disclosure, wherein the multi-speed photoacoustic image reconstruction device based on implicit neural representation is applied to an electronic device 600. Figure 5 The multi-speed photoacoustic image reconstruction device 500 based on implicit neural representation may include: A parameter fine-tuning unit 501 is configured to adjust parameters of the implicit neural representation network. Adjusting the parameters of the implicit neural representation network includes: obtaining a first initial sound pressure distribution and a first sound velocity distribution based on spatial coordinates within the imaging region through the implicit neural representation network; simulating the propagation process of a sound pressure wave excited by the first initial sound pressure distribution in a non-uniform medium defined by the first sound velocity distribution through a differentiable acoustic forward propagation model to obtain a predicted detector signal; and minimizing the difference between the predicted detector signal and the detector measurement signal to update the parameters of the implicit neural representation network; and iterating until a preset convergence condition is met. The reconstruction execution unit 502 is configured to obtain a second initial sound pressure distribution based on the spatial coordinates within the imaging area using the implicit neural representation network after adjusting parameters.

[0061] Furthermore, the multi-speed photoacoustic image reconstruction device 500 based on implicit neural representation can also include: a pre-training unit 503, used to pre-train the implicit neural representation network; wherein, the pre-training of the implicit neural representation network includes: using the DAS algorithm to preliminarily reconstruct the detector measurement signal to obtain a third initial sound pressure distribution, using the third initial sound pressure distribution as the supervision target of the initial boost distribution, initializing the density distribution and the sound speed distribution to predetermined constants respectively and constraining their physical consistency through a differentiable acoustic forward propagation model, and executing the implicit neural representation network for pre-training.

[0062] Furthermore, the multi-speed photoacoustic image reconstruction device 500 based on implicit neural representation can also include: a post-processing unit 504, which is used to obtain a synthetic detector signal based on the second initial sound pressure distribution, the predetermined third sound speed distribution and the predetermined third density distribution through a differentiable acoustic forward model, and reconstruct the synthetic detector signal using the DAS algorithm to obtain a fourth initial sound pressure distribution.

[0063] For further technical details regarding the multi-speed photoacoustic image reconstruction device 500 based on implicit neural representation, please refer to the section on the multi-speed photoacoustic image reconstruction method based on implicit neural representation above and will not be repeated here. In specific applications, the multi-speed photoacoustic image reconstruction device 500 based on implicit neural representation can be implemented using the electronic device 600 described below or as software within the electronic device 600.

[0064] In addition, an embodiment of the present disclosure also provides a computer-readable storage medium having a computer program stored thereon, the program including instructions that, when executed by one or more processors of a computing device, perform the steps of the aforementioned multi-speed photoacoustic image reconstruction method based on implicit neural representation.

[0065] Figure 6 Schematic diagram of the structure of the electronic device provided by the embodiment of the present disclosure is shown. 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 shown in the above embodiments of the present disclosure and / or the program units corresponding to each unit in the device.

[0066] 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.

[0067] 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.).

[0068] 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 1The program instructions / units corresponding to the multi-speed photoacoustic image reconstruction method 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 embodiment. Figure 1 The programs, instructions and units corresponding to the multi-speed photoacoustic image reconstruction method based on implicit neural representation are shown.

[0069] 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.

[0070] 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.

[0071] Over time and with the advancement of technology, the meaning of "medium" has become increasingly broad. The dissemination of computer programs is no longer limited to tangible media and can also be directly downloaded from the Internet. Any combination of one or more computer-readable storage media can be used. Computer-readable storage media can be, but are not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or components, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media 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 of the foregoing. In this document, a 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 component.

[0072] The technical solutions provided by the present disclosure are described in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present disclosure. The descriptions of the above embodiments are intended only to help understand the methods and core concepts of the present disclosure. Furthermore, those skilled in the art will appreciate that variations in the specific implementation methods and scope of application may occur based on the concepts of the present disclosure. In summary, the contents of this specification should not be construed as limiting the present disclosure.

[0073] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. Any modifications, equivalent substitutions, etc. made within the spirit and principles of the present disclosure should be included in the scope of protection of the present disclosure.

Claims

1. A multi-speed photoacoustic image reconstruction method based on implicit neural representation, characterized in that: The method comprises: Adjusting parameters of an implicit neural representation network, the adjusting the parameters of the implicit neural representation network comprising: obtaining a first initial sound pressure distribution and a first sound velocity distribution based on spatial coordinates within an imaging region through the implicit neural representation network, simulating a propagation process of a sound pressure wave excited by the first initial sound pressure distribution in a medium defined by the first sound velocity distribution through a differentiable acoustic forward propagation model to obtain a predicted detector signal, and minimizing a difference between the predicted detector signal and a detector measurement signal to update the parameters of the implicit neural representation network; iterating in this manner until a preset convergence condition is satisfied; The second initial sound pressure distribution is obtained based on the spatial coordinates within the imaging area using the implicit neural representation network with adjusted parameters.

2. The method according to claim 1, characterized in that The adjusting the parameters of the implicit neural representation network further includes: obtaining a first density distribution while obtaining the first initial sound pressure distribution and the first sound speed distribution; The method of simulating the propagation process of the sound pressure wave excited by the first initial sound pressure distribution in the medium defined by the first sound velocity distribution by using a differentiable acoustic forward propagation model to obtain the predicted detector signal is specifically as follows: simulating the propagation process of the sound pressure wave excited by the first initial sound pressure distribution in the medium defined by the first sound velocity distribution and the first density distribution by using a differentiable acoustic forward propagation model to obtain the predicted detector signal.

3. The method according to claim 2, characterized in that The method further includes: before adjusting the parameters of the implicit neural representation network, pre-training the implicit neural representation network; Among them, pre-training the implicit neural representation network includes: using the DAS algorithm to preliminarily reconstruct the detector measurement signal to obtain a third initial sound pressure distribution, using the third initial sound pressure distribution as the supervision target of the initial boost distribution, initializing the density distribution and the sound speed distribution to predetermined constants and constraining their physical consistency through the differentiable acoustic forward propagation model, and executing the implicit neural representation network for pre-training.

4. The method according to claim 1, wherein 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 the mean square error between the predicted detector signal and the detector measurement signal, using the mean square error to perform back propagation to obtain the gradient, and updating the parameters of the implicit neural representation network based on the gradient.

5. The method according to claim 1, characterized in that: The method further comprises: obtaining a synthetic detector signal based on the second initial sound pressure distribution, a predetermined third sound speed distribution, and a predetermined third density distribution through a differentiable acoustic forward model; The synthetic detector signal is reconstructed using a DAS algorithm to obtain a fourth initial sound pressure distribution.

6. The method according to claim 1, wherein Also includes: The implicit neural representation network with adjusted parameters is used to obtain a second initial sound pressure distribution based on the spatial coordinates in the imaging area, and at the same time, a second sound speed distribution and / or a second density distribution is obtained.

7. The method according to claim 1, characterized in that The implicit neural representation network is a multi-layer perceptron MLP with position encoding, and the differentiable acoustic forward propagation model is a j-Wave simulator implemented based on the JAX library.

8. A multi-speed photoacoustic image reconstruction device based on implicit neural representation, characterized in that: The multi-speed photoacoustic image reconstruction device based on implicit neural representation includes: a parameter fine-tuning unit, configured to adjust parameters of an implicit neural representation network, wherein adjusting the parameters of the implicit neural representation network comprises: obtaining a first initial sound pressure distribution and a first sound velocity distribution based on spatial coordinates within an imaging region through the implicit neural representation network, simulating a propagation process of a sound pressure wave excited by the first initial sound pressure distribution in an inhomogeneous medium defined by the first sound velocity distribution through a differentiable acoustic forward propagation model to obtain a predicted detector signal, and minimizing a difference between the predicted detector signal and a detector measurement signal to update the parameters of the implicit neural representation network; and iterating in this manner until a preset convergence condition is satisfied; The reconstruction execution unit is used to obtain a second initial sound pressure distribution based on the spatial coordinates in the imaging area using the implicit neural representation network with adjusted parameters.

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, the program comprising instructions, which, when executed by one or more processors of a computing device, cause the computing device to perform the method according to any one of claims 1 to 7.

Images