A sound velocity-adaptive photoacoustic tomography method and system based on spherical harmonic probes
By parameterizing the sound pressure distribution and sound velocity field using the spherical harmonic probe method, the phase distortion and computational resource bottleneck caused by acoustic heterogeneity in photoacoustic imaging are solved, achieving efficient and accurate photoacoustic tomography.
Patent Information
- Application Number
- CN202611282438.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-24
- Publication Date
- 2026-11-17
AI Technical Summary
Existing photoacoustic computed tomography technology is limited by phase distortion caused by acoustic heterogeneity, high complexity of explicit sound velocity modeling, and bottlenecks in computing resources and storage space, which affect imaging quality and real-time performance.
A sound velocity adaptive photoacoustic tomography method based on spherical harmonic probes is adopted. The initial sound pressure distribution of the region to be reconstructed is parameterized as a set of three-dimensional Gaussian elements. A three-dimensional regular grid is constructed and spherical harmonic probes are deployed. The path-average sound velocity and flight time are calculated, and the reconstruction is performed in combination with the analytical propagation model of sound waves.
It achieves accurate compensation for changes in sound speed with spatial position and direction without increasing computational complexity, solves the phase distortion problem in photoacoustic imaging, and improves imaging quality and efficiency.
Smart Images

Figure CN123118901A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of three-dimensional reconstruction, and more specifically, to a sound velocity adaptive photoacoustic tomography method and system based on a spherical harmonic probe. Background Technology
[0002] Photoacoustic computed tomography (PACT) is an emerging, multi-scale, multi-functional biomedical imaging technique. Its core physical mechanism involves irradiating biological tissue with short-pulse lasers. Light-absorbing substances within the tissue (such as hemoglobin and melanin) absorb photon energy and undergo thermoelastic expansion, radiating high-frequency ultrasound waves. These signals are captured by an array of ultrasound transducers placed around the tissue, and the initial sound pressure distribution within the tissue is reconstructed using algorithms. This allows for structural and functional imaging that combines high optical contrast with deep ultrasound penetration.
[0003] However, in actual clinical applications, image quality is severely limited by the following existing technological shortcomings: Phase distortion due to acoustic heterogeneity: Traditional analytical reconstruction algorithms (such as Universal Backprojection (UBP)) and existing sparse characterization methods (such as SlingBAG) are based on the ideal assumption of "globally uniform sound velocity." However, the acoustic properties of real biological tissues (such as skull, fat, and glands) are highly heterogeneous. Sound waves experience different sound velocities along their propagation path, causing a deviation between the actual time to the transducer (ToF) and the theoretically calculated value. This deviation manifests spatially as interference misalignment of the backprojected signal, ultimately leading to image defocus, anatomical distortion, and severe fringe artifacts, directly affecting the accuracy of clinical diagnosis.
[0004] The high complexity of explicit sound velocity modeling: Existing sound velocity compensation methods typically require obtaining structural segments of tissue through auxiliary modalities (such as X-ray CT or ultrasound) and assigning empirical sound velocity values to different regions. This not only significantly increases system cost and operational complexity but also depends on the registration accuracy of multimodal images. Any deviation in the segmentation edges introduces new reconstruction errors. For example, patent application CN103142216A discloses a multi-layer medium sound velocity calculation method that relies on preset medium geometric boundaries and initial sound velocities for reconstruction. However, this method still has drawbacks: it heavily relies on prior knowledge of tissue boundaries, cannot adaptively compensate for the heterogeneity of sound velocities in continuous space, and is prone to image defocusing.
[0005] Bottlenecks in computational resources and storage space: Iterative methods such as full-wave joint inversion (FWI) attempt to solve wave equations on dense grids. To meet the sampling requirements of high-frequency acoustic waves, the number of grid points increases cubically with spatial scale. For 3D high-resolution imaging, the memory overhead often exceeds the limitations of commercial GPUs, and the computation time per iteration is long, failing to meet the imaging requirements for real-time feedback. Summary of the Invention
[0006] In view of one of the deficiencies in the prior art, the purpose of this application is to provide a sound velocity adaptive photoacoustic tomography method and system based on a spherical harmonic probe.
[0007] A first aspect of this application provides a sound velocity-adaptive photoacoustic tomography method based on a spherical harmonic probe, comprising: The initial sound pressure distribution of the region to be reconstructed is parameterized as a set of three-dimensional Gaussian elements; a three-dimensional regular grid covering the region to be reconstructed is constructed, and spherical harmonic probes are deployed at the nodes of the three-dimensional grid, with each spherical harmonic probe storing spherical harmonic coefficients; For each of the three-dimensional Gaussian elements to the transducer, spatial interpolation is performed on the spherical harmonic coefficients at the vertices of the three-dimensional mesh surrounding the three-dimensional Gaussian elements, and the path-average sound speed and flight time are calculated in combination with the directional spherical harmonic basis functions corresponding to the propagation direction. Substitute the average sound speed along the path and the flight time into the analytical propagation model of sound waves to calculate the theoretical sound pressure signal at each transducer. The theoretical sound pressure signal is compared and inverted with the actual acquired signal to output a reconstructed initial sound pressure distribution image, thus completing photoacoustic tomography.
[0008] Optionally, parameterizing the initial sound pressure distribution of the region to be reconstructed as a set of three-dimensional Gaussian elements includes: The initial sound pressure distribution is characterized as N discrete three-dimensional Gaussian elements, each of which carries optimizable parameters including: center position coordinates, initial pressure amplitude and spatial standard deviation. The center position coordinates are used to locate the spatial position of the photoacoustic source; the initial pressure amplitude is used to characterize the local initial sound pressure intensity; and the spatial standard deviation is used to control the spatial range of the sound source.
[0009] Optionally, the construction of a three-dimensional regular mesh covering the region to be reconstructed, and the deployment of spherical harmonic probes at the nodes of the three-dimensional mesh, each of the spherical harmonic probes storing spherical harmonic coefficients, including: The region to be reconstructed is discretized into a three-dimensional regular grid; A spherical harmonic probe is configured at each node of the three-dimensional regular mesh; Each of the spherical harmonic probes stores spherical harmonic coefficients of order 0 to 2, forming a 9-dimensional parameter vector.
[0010] Optionally, the step of spatially interpolating the spherical harmonic coefficients at the mesh vertices surrounding the three-dimensional Gaussian elements for the propagation direction to the transducer, and calculating the path-averaged sound speed and flight time in conjunction with the directional spherical harmonic basis functions corresponding to the propagation direction, includes: Identify the three-dimensional mesh cell surrounding the center position of the three-dimensional Gaussian element, and extract the 8 vertices of the three-dimensional mesh cell; Perform trilinear interpolation on the spherical harmonic coefficients stored by the spherical harmonic probes at the eight vertices to generate a local spherical harmonic coefficient vector at the center position of the three-dimensional Gaussian element; Based on the unit direction vector of the three-dimensional Gaussian element pointing to the target transducer, construct the corresponding 9-dimensional real spherical harmonic basis function vector; The dot product of the local spherical harmonic coefficient vector and the 9-dimensional real spherical harmonic basis function vector is calculated to obtain the path-average sound speed in the propagation direction. The sound wave flight time is then calculated based on the Euclidean distance between the center position and the target transducer and the path-average sound speed.
[0011] Optionally, substituting the path-average sound speed and the flight time into the analytical propagation model of sound waves to calculate the theoretical sound pressure signal at each transducer includes: A forward projection formula is constructed based on the far-field asymptotic approximation of the three-dimensional acoustic wave equation, serving as the analytical propagation model of the acoustic wave. Substitute the path-average sound speed and the flight time into the propagation delay term of the forward projection formula to calculate the temporal sound pressure contribution of each of the three-dimensional Gaussian elements to the target transducer. The time-domain sound pressure contributions of all the three-dimensional Gaussian elements are superimposed and summed to output the theoretical time-domain sound pressure signal at each transducer.
[0012] Optionally, the step of comparing and inverting the theoretical sound pressure signal with the actual acquired signal to output a reconstructed initial sound pressure distribution image, thereby completing photoacoustic tomography, includes: Calculate the mean square error between the theoretical sound pressure signal and the actual acquired signal, and construct a global loss function; The gradient of the loss function with respect to the three-dimensional Gaussian parameters and the spherical harmonic probe parameters is obtained based on the automatic differentiation technique; Based on the gradient, the three-dimensional Gaussian meta-parameters and the spherical harmonic probe parameters are updated synchronously using a gradient optimization algorithm, and the gradient calculation and parameter update steps are executed iteratively. After each iteration, it is checked whether the global loss function meets the preset convergence condition. When the global loss function meets the preset convergence condition, the iteration is terminated, and the converged three-dimensional Gaussian meta-parameters are mapped and output as the initial sound pressure distribution image to complete photoacoustic tomography.
[0013] Optionally, the iterative update process further includes an adaptive density control step, specifically: During each iteration of backpropagation, the gradient norm corresponding to each three-dimensional Gaussian element is extracted as the local reconstruction error, and its spatial standard deviation is read as the spatial scale. When the local reconstruction error is greater than a preset error threshold and the spatial scale is greater than a preset scale threshold, the three-dimensional Gaussian primitive is split into multiple sub-Gaussian primitives, and the parameter distribution of the original primitive is inherited. The absolute value of the gradient of the global loss function with respect to the initial pressure magnitude of the three-dimensional Gaussian element is calculated as the contribution. When the contribution is lower than the preset contribution threshold, the three-dimensional Gaussian element is removed from the set.
[0014] A second aspect of this application provides a sound velocity-adaptive photoacoustic tomography system based on a spherical harmonic probe, comprising: The joint characterization construction module parameterizes the initial sound pressure distribution of the region to be reconstructed into a three-dimensional Gaussian element set; constructs a three-dimensional regular grid covering the region to be reconstructed, and deploys spherical harmonic probes at the nodes of the three-dimensional grid. Each spherical harmonic probe stores spherical harmonic coefficients to construct a path-averaged sound velocity field. The sound speed and flight time calculation module performs spatial interpolation on the spherical harmonic coefficients at the vertices of the three-dimensional mesh surrounding the three-dimensional Gaussian elements for the propagation direction from each of the three-dimensional Gaussian elements to the transducer, and calculates the path-average sound speed and flight time in combination with the directional spherical harmonic basis functions corresponding to the propagation direction. The acoustic forward projection module substitutes the average sound speed along the path and the flight time into the analytical propagation model of sound waves to calculate the theoretical sound pressure signal at each transducer. The joint optimization and reconstruction module compares and inverts the theoretical sound pressure signal with the actual acquired signal, outputs the reconstructed initial sound pressure distribution image, and completes photoacoustic tomography.
[0015] A third aspect of this application provides a terminal including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, can be used to execute any of the spherical harmonic probe-based sound velocity adaptive photoacoustic tomography methods described in the present invention.
[0016] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can be used to perform any of the sound velocity adaptive photoacoustic tomography methods based on spherical harmonic probes described in the present invention.
[0017] The acoustic velocity adaptive photoacoustic tomography method based on a spherical harmonic probe provided in this application highly abstracts the complex sound wave propagation process by discretizing and parameterizing the anisotropic path-averaged sound velocity field using a spherical harmonic probe. This method can explicitly capture and compensate for changes in sound velocity with spatial location and sound wave emission direction without increasing physical computational complexity, thus solving the most critical phase distortion problem in photoacoustic imaging.
[0018] Other technical effects resulting from the additional features will be further illustrated in the corresponding embodiments. Attached Figure Description
[0019] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating a sound velocity-adaptive photoacoustic tomography method based on a spherical harmonic probe according to an exemplary embodiment; Figure 2 This is a schematic diagram of the overall framework of a sound velocity adaptive photoacoustic tomography method based on a spherical harmonic probe, according to an exemplary embodiment. Figure 3 This is a schematic diagram illustrating the definition and representation of the average sound velocity field according to an exemplary embodiment; Figure 4 This is a comparison chart of simulation data experimental results according to an exemplary embodiment; Figure 5 This is a comparison diagram of experimental results of phantom data according to an exemplary embodiment; Figure 6 The half-width and height measurement data of the reconstructed detail portion shown according to an exemplary embodiment; Figure 7 This is a comparison diagram of experimental results of sparse sampling of phantom data according to an exemplary embodiment; Figure 8 This is a schematic diagram of the structure of a sound velocity adaptive photoacoustic tomography system based on a spherical harmonic probe, according to an exemplary embodiment. Detailed Implementation
[0020] The present application will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application, and these all fall within the protection scope of the present application. Parts not described in detail in the following embodiments can be implemented using existing technology.
[0021] Sound waves experience different sound velocities along their propagation path, causing a discrepancy between the actual time-of-flight (ToF) to the transducer and the theoretically calculated value. This discrepancy manifests spatially as interference misalignment of the back-projected signal, ultimately leading to image defocus, anatomical distortion, and severe fringe artifacts, directly affecting the accuracy of clinical diagnosis. To address these issues, this application provides a sound velocity-adaptive photoacoustic tomography method based on a spherical harmonic probe to resolve these problems.
[0022] Reference Figure 1 and Figure 2 As shown in one embodiment of this application, a sound velocity-adaptive photoacoustic tomography method based on a spherical harmonic probe is provided, comprising the following steps: S1, parameterize the initial sound pressure distribution of the region to be reconstructed into a three-dimensional Gaussian element set; construct a three-dimensional regular grid covering the region to be reconstructed, and deploy spherical harmonic probes at the nodes of the three-dimensional grid. Each spherical harmonic probe stores spherical harmonic coefficients to construct the path-averaged sound velocity field. Among them, the three-dimensional regular mesh constitutes a fixed spatial reference system for the region to be reconstructed, and the position coordinates of the three-dimensional Gaussian elements are defined and located based on the spatial reference system.
[0023] S2, for each 3D Gaussian element to the transducer, spatial interpolation is performed on the spherical harmonic coefficients at the vertices of the 3D mesh surrounding the 3D Gaussian element, and the path average sound speed and flight time are calculated by combining the directional spherical harmonic basis functions corresponding to the propagation direction. S3, Substitute the path-average sound speed and flight time into the analytical propagation model of sound waves to calculate the theoretical sound pressure signal at each transducer; S4 compares and inverts the theoretical sound pressure signal with the actual acquired signal, outputs the reconstructed initial sound pressure distribution image, and completes photoacoustic tomography.
[0024] The above embodiments, by modeling anisotropic sound velocities using spherical harmonic probes, can accurately correct time-of-flight shifts caused by uneven sound velocities in biological tissues.
[0025] Traditional UBP methods require manual pre-setting of sound velocity values and tissue boundaries before reconstruction (usually aided by CT or ultrasound delineation), which is not only time-consuming and labor-intensive, but also results in severe misalignment of the reconstructed image if the delineation is off-target or incorrect. Furthermore, traditional algorithms assume sound waves travel in straight lines, ignoring the actual bending and delay of sound waves caused by the inhomogeneity of human tissue. This application completely simplifies this process: given only an initial sound velocity value, the algorithm can automatically infer the complex sound velocity distribution within the body during iterative calculations, achieving sound velocity field calibration. No additional CT / ultrasound scans are required throughout the process, eliminating tedious image alignment steps. Simultaneously, the algorithm strictly follows the actual physical laws of sound waves in forward deduction, ensuring that the reconstruction results accurately reflect the acoustic properties of the tissue, rather than being a rigid mathematical fit.
[0026] When the number of probes is small (sparse sampling), traditional methods based on fixed grids (voxels) are prone to radial stripe artifacts. This is because voxels are like rigid "pixel blocks," and when the data density is insufficient, high-frequency details will alias and become corrupted. This application uses three-dimensional Gaussian elements (which can be understood as smooth "spots") to represent the image. The Gaussian function is naturally smooth and continuous, and mathematically equivalent to a built-in "softening filter," which can automatically filter out high-frequency noise that easily causes stripes. This is like replacing rigid pixel mosaic with soft brushstrokes; even with sparse probe signals, it can effectively eliminate messy stripes and stably output high-quality images with clear edges and clean backgrounds.
[0027] To efficiently obtain acoustic propagation parameters along the path from any photoacoustic source to the transducer, a dual-path characterization initialization and geometric prior loading are required before S1. Known transducer position coordinates are received as spatial geometric priors. The physical field of the region to be reconstructed is initialized: a constant initialization strategy is used for the acoustic medium characterization (i.e., the spherical harmonic probe mesh), for example, uniformly assigning all mesh nodes the zeroth-order spherical harmonic coefficients corresponding to the average sound velocity of biological soft tissue, while initializing higher-order coefficients to zero, thus constructing a smooth initial sound velocity field basis; a random initialization strategy is used for the photoacoustic source characterization (i.e., the three-dimensional Gaussian primitive set), for example, uniformly and randomly scattering points within the convex hull of the space to be reconstructed to generate primitive center coordinates, and assigning randomly distributed initial pressure amplitudes and spatial standard deviations. It should be noted that random initialization can easily lead to optimization divergence or getting trapped in local optima. However, this application initializes the Gaussian element acoustic pulse width to be 5-10 times the transducer sampling interval and the size to be 1 / 5-1 / 3 of the reconstruction region, fitting only the low-frequency contour of the signal to avoid local traps; it uses time-of-flight error to directly map the position to correct the gradient, providing a clear physical update direction; and it combines S402 adaptive density control to dynamically prune redundant primitives and split high-error regions, making the parameter distribution quickly fit the real structure. These multiple aspects avoid getting trapped in local optima. Based on this, in some specific embodiments of this application, S1, the initial sound pressure distribution of the region to be reconstructed is parameterized as a three-dimensional Gaussian element set; a three-dimensional regular mesh covering the region to be reconstructed is constructed, and spherical harmonic probes are deployed at the nodes of the three-dimensional mesh. Each spherical harmonic probe stores spherical harmonic coefficients, which can be achieved through the following steps: S101, Gaussian parameterization of the initial sound pressure distribution.
[0028] Abandoning the traditional fixed voxel mesh generation, the initial sound pressure distribution of the area to be reconstructed is... Parameterized as a set of N discrete and freely movable three-dimensional Gaussian elements .
[0029] in: This represents the center spatial coordinates of the i-th primitive, which determines the source location of the photoacoustic signal; This represents the initial pressure amplitude, characterizing the light absorption intensity and grayscale of the reconstructed image in that region; The spatial standard deviation controls the influence range and structural refinement of this primitive.
[0030] This representation method enables the model to have spatial adaptive capabilities similar to "point clouds" in subsequent optimization, and can accurately fit irregular anatomical geometric shapes such as blood vessels and tumor edges.
[0031] S102, construct a five-dimensional path-averaged sound velocity field.
[0032] Instead of directly modeling the microscopic pointwise sound velocity, a continuous five-dimensional scalar field is defined. The independent variables of this field consist of the following five dimensions: Three-dimensional spatial coordinates ( x , y , z ): Represents the spatial location of any photoacoustic source point within the imaging area; Two-dimensional propagation direction ( θ , ): Represents the azimuth and elevation angles of a unit direction vector pointing from this point to the transducer surface in spherical coordinates.
[0033] For any combination in the above five-dimensional space, the field directly outputs from the point. Along direction The average speed of sound along the straight path that propagates to the sensor surface.
[0034] This physical definition implicitly compensates for the complex wave path integral process caused by acoustic heterogeneity in the average sound velocity, and significantly reduces the computational dimensionality and storage overhead of sound velocity modeling by explicitly encoding the dependence of position and direction.
[0035] S103, Discretization and parameterization of sound velocity field based on spherical harmonic probe.
[0036] First, the three-dimensional spatial dimension of the five-dimensional field is discretized, and the imaging area is divided into a three-dimensional regular volume grid. The physical coordinates of each grid node (i,j,k) uniquely correspond to a spatial slice in the five-dimensional field. Secondly, the two-dimensional directional dimension of the five-dimensional field is parameterized and encoded, and a spherical harmonic probe is deployed at each grid node. Utilizing the orthogonality and completeness of real spherical harmonic functions on a unit sphere, the continuous function of sound velocity varying with direction at the node is expanded into a series.
[0037] Extracting spherical harmonic basis functions of order D=0,1,2, resulting in a total of 1+3+5=9 coefficients, forms a 9-dimensional parameter vector.
[0038] Therefore, a five-dimensional continuous field Mapped to discrete data structures: The direction of sound velocity at any spatial point x is uniquely represented by the 9-dimensional coefficient vector of the grid node where it is located. The specific sound velocity value in any direction ω is obtained in real time by substituting ω into a 9-dimensional spherical harmonic basis function and performing a dot product operation with a 9-dimensional parameter vector. This mapping mechanism achieves a high-fidelity discrete representation of the five-dimensional sound velocity field with extremely low memory overhead.
[0039] The above embodiments complete the initial characterization of the complex sound velocity distribution of biological tissues with extremely low memory overhead, providing a parameter basis for subsequent differentiable acoustic projection.
[0040] The above embodiments highly abstract the complex sound wave propagation process by constructing an "anisotropic path-averaged sound velocity field." This modeling approach enables the explicit capture and compensation of sound velocity variations with spatial location and sound wave emission direction without increasing physical computational complexity, thereby solving the core phase distortion problem in photoacoustic imaging.
[0041] To efficiently obtain acoustic propagation parameters along the path from any photoacoustic source to the transducer, in some specific embodiments of this application, S2, for each Gaussian element to the transducer, spatial interpolation is performed on the spherical harmonic coefficients at the grid vertices surrounding the Gaussian element, and the path-averaged sound velocity and flight time are calculated in combination with the directional spherical harmonic basis functions corresponding to the propagation direction. Figure 3 As shown, the following steps can be taken: S201, spatial domain sound velocity feature interpolation.
[0042] For any Gaussian element's center position First, locate the cubic mesh cell surrounding the location and its 8 vertex nodes.
[0043] Read the pre-stored spherical harmonic coefficients at the 8 vertex nodes and perform trilinear interpolation to generate a local spherical harmonic coefficient vector at position x. This step smoothly transitions the anisotropic characteristics of sound velocity from discrete grid nodes to any source point in continuous space, obtaining a sound velocity direction encoding vector with spatial continuity.
[0044] S202, Angular sound speed projection and flight time calculation.
[0045] Based on the unit propagation direction vector of the Gaussian element pointing to the target transducer Analyze the corresponding 9-dimensional real spherical harmonic basis function vector. ; Local spherical harmonic coefficient vector With basis function vectors By performing a dot product operation, the path-average sound speed of this specific propagation path can be calculated. Here, the path-average sound speed is the five-dimensional scalar field. The specific value.
[0046] Based on the Euclidean distance between the center position of the Gaussian element and the position on the transducer surface. And the path-average sound speed, calculate the sound wave flight time .
[0047] It is worth noting that the time of flight is a key time delay parameter for sound waves to travel from the sound source to the transducer. In this application, explicitly calculating the time of flight serves a dual purpose: Physical alignment: It is the foundation for determining the arrival time of signals and achieving precise phase matching in analytical models; Computational Acceleration: Based on the local characteristics of Gaussian signals, the effective signal is only distributed within a short time window centered on the time of flight. By pre-calculating the time of flight, the algorithm can quickly filter out Gaussian primitives that "contribute at the current moment," skipping redundant calculations of a large number of non-contributing primitives, thereby significantly improving reconstruction efficiency.
[0048] Three-dimensional photoacoustic imaging generates massive amounts of data, and traditional grid-based iterative methods are computationally extremely expensive when processing large-scale voxels. The above embodiment, through this acoustic medium characterization interpolation strategy, greatly compresses the parameter space, enabling high-resolution three-dimensional autofocus reconstruction to run smoothly on ordinary consumer-grade GPUs.
[0049] The above embodiments provide an efficient representation of the sound velocity in the reconstructed region. A spherical harmonic probe grid is used to represent the path-averaged sound velocity field, discretizing and representing the continuous field with minimal parameters, resulting in high storage efficiency. By first spatial interpolating and then angular interpolating, the average sound velocity at any location and in any direction can be quickly calculated. Furthermore, this step-by-step interpolation strategy avoids generating a large number of intermediate calculation results, making it computationally efficient. During optimization, the gradient of the Gaussian kernel-averaged sound velocity is directly backpropagated to the spherical harmonic probe grid through the interpolation operator, resulting in high optimization efficiency.
[0050] To explicitly incorporate acoustic heterogeneity compensation into the signal synthesis process, in some specific embodiments of this application, S3 constructs an analytical acoustic model and performs forward projection. This model serves as the core computational engine, such as... Figure 2 As shown, there are three inputs: the geometrically calibrated transducer position, the acoustic medium representation (spherical harmonic probe field) constructed by S1, and the photoacoustic source representation (three-dimensional Gaussian element set) generated by S1. The model internally couples the ray propagation approximation with the spherical harmonic direction interpolation algorithm, substituting the path-averaged sound velocity and flight time into the analytical sound wave propagation model to calculate the theoretical sound pressure signal at each transducer.
[0051] Specifically, S3 can adopt the following steps: S301, using the far-field asymptotic analytical solution derived from the three-dimensional acoustic wave equation under the assumption of a homogeneous medium, calculates the transducer. Expected sound pressure at the location That is, the formula for forward projection is: in, No. m Each transducer at time t The theoretical sound pressure signal (expected sound pressure value) at the location; Represents the total number of three-dimensional Gaussian elements; i is the Gaussian element index. i =1,2,…, N ; m is the transducer channel index; Indicates the first i The initial pressure amplitude of a three-dimensional Gaussian element characterizes the light absorption intensity of the region corresponding to that element. Indicates the first i The center position of the three-dimensional Gaussian element and the first m The Euclidean distance between each transducer; This indicates the result calculated in step S2, the first... i The first Gaosky Yuan to the first m The path-average sound velocity in the propagation direction of each transducer; t represents the signal acquisition time variable; The equivalent spatial broadening parameter of the i-th three-dimensional Gaussian element is used to control the temporal width and spatial resolution of the acoustic pulse. This represents the sound wave propagation delay correction term. When this term is zero, it indicates that the sound wave front has just arrived at the transducer, at which point the sound pressure contribution reaches its peak. Here, the far-field asymptotic approximation is for the sound radiation field of a single three-dimensional Gaussian element, approximating its wavefront at the transducer location as a spherical wave, rather than treating the entire photoacoustic source as a far-field point source.
[0052] In step S302, the adaptive path average sound velocity obtained in step S2 is substituted into the propagation term in the analytical formula of S301. This means that for the same photoacoustic source, the sound waves emitted towards different transducers will experience different propagation delays and waveform evolutions, thus mathematically and explicitly modeling the wavefront distortion phenomenon caused by non-uniform media.
[0053] The above embodiments construct a reconstruction architecture with physical interpretability. By constructing a three-dimensional Gaussian kernel acoustic wave propagation model based on physical constraints and incorporating acoustic heterogeneity into the signal calculation process, the reconstruction process has stronger physical interpretability compared to the direct projection of UBP.
[0054] The above embodiments use a smooth Gaussian kernel function to represent the reconstructed scene, which can sample voxels of arbitrary resolution without producing artifacts, and its spatial resolution is significantly better than traditional voxel-based methods.
[0055] In one specific embodiment of this application, step S4 performs signal preprocessing, closed-loop optimization, and voxelization output. First, the input transducer signal is subjected to frequency domain filtering (e.g., bandpass filtering or wavelet thresholding denoising) to suppress low-frequency baseline drift and high-frequency noise, outputting a clean time-domain reference signal. Then, the theoretical sound pressure signal is compared and inverted with the actual acquired signal to output a reconstructed initial sound pressure distribution image. S4 can include the following sub-steps: S401, Constructing the Loss Function and Multi-Parameter Collaborative Update The mean square error between the theoretical sound pressure signal and the actual signal acquired by the transducer is calculated, and a global mean square loss function (MSE) is constructed. , Actual acquired signals, This is the theoretical sound pressure signal; Backpropagation is performed based on this loss function. The gradient of the loss function with respect to the 3D Gaussian meta-parameters and the spherical harmonic probe parameters is calculated using automatic differentiation. The Gaussian source parameters are then updated synchronously using a gradient descent optimization algorithm. Harmonic velocity field parameters This enables joint optimization of image representation and sound velocity field.
[0056] S402, Adaptive Density Control.
[0057] S4021: Calculate the gradient L2 norm of the corresponding parameters of each three-dimensional Gaussian primitive as the local reconstruction error; extract the spatial standard deviation of each three-dimensional Gaussian primitive as the spatial scale; calculate the absolute value of the initial pressure amplitude of each primitive as the contribution index.
[0058] S4022, when the local reconstruction error exceeds a preset splitting threshold and the spatial scale exceeds a preset lower limit (this preset lower limit is determined based on the transducer sampling frequency and the average sound velocity of the medium, and is typically set to 0.5 to 1.0 times the minimum resolvable spatial distance), the primitive is split into two sub-primitives. Specific execution: Position offset: Offset by 0.5 to 0.8 times (preferably 0.6 times) of the original spatial standard deviation along the original gradient normalization direction and in the opposite direction. Scale reduction: Reduce the standard deviation of the sub-element space to 1 / 1.6 to 1 / 2 of the original standard deviation (preferably 1 / 1.6). Amplitude allocation: Based on the principle of energy conservation in three-dimensional Gaussian integrals, the total integral energy of the two sub-fundamentals is made equal to that of the original fundamental elements, while keeping the sound pressure field energy constant. S4023, Low-contribution primitive removal: Primitives are removed if any of the following conditions are met: Amplitude too low: The absolute value of the initial pressure amplitude is less than 0.1% of the global maximum amplitude; Too small a scale: If the spatial standard deviation is too small, the width of the acoustic pulse will be less than half of the transducer sampling interval, which is considered redundant noise.
[0059] In S4021, the preset splitting threshold is set using a dynamic feedback strategy based on the number of target primitives. The target total number of primitives is set in advance based on the imaging size, resolution, and memory capacity. During the iteration process, the current total number of primitives is counted. If the current total number is less than the target total number, the splitting threshold is lowered to promote splitting. If the current total number is greater than the target total number, the splitting threshold is raised to suppress splitting. This feedback mechanism keeps the number of primitives dynamically stable near the target total number, balancing reconstruction details and computational efficiency.
[0060] S403, Convergence Determination and Result Output The iteration terminates when the loss function meets the preset convergence condition. The converged discrete photoacoustic source representation (a set of three-dimensional Gaussian primitives) is mapped into a regular three-dimensional matrix through a voxelization rendering pipeline (such as rasterization resampling based on Gaussian kernels or Alpha Blending projection), and finally outputs a high-resolution initial sound pressure distribution image to complete photoacoustic tomography.
[0061] The above embodiments achieve prior-free self-focusing reconstruction. The closed-loop optimization framework relies solely on photoacoustic temporal pressure signals, simultaneously inverting image and sound velocity parameters through a differentiable forward model and automatic differentiation technology, without requiring external anatomical priors such as CT or ultrasound. In scenarios with strong acoustic heterogeneity, such as the skull and breast, the system can automatically compensate for propagation delay deviations, achieving high-precision structural focusing.
[0062] The above embodiments improve the computational efficiency of 3D reconstruction. The spherical harmonic probe characterization and trilinear interpolation strategy compress the high-dimensional sound velocity field into low-dimensional discrete parameters. Combined with the sparsity characteristics of the 3D Gaussian characterization, this significantly reduces the parameter space dimension and memory overhead during the optimization process. This method effectively overcomes the computational bottleneck of traditional mesh iteration methods in large-scale voxel computation, enabling high-resolution 3D self-focusing reconstruction to run efficiently on conventional computing devices.
[0063] Based on the same technical concept, other embodiments of this application provide a sound velocity adaptive photoacoustic tomography system 100 based on a spherical harmonic probe, such as... Figure 8 As shown, it includes: The joint characterization construction module 110 parameterizes the initial sound pressure distribution of the region to be reconstructed into a three-dimensional Gaussian element set; constructs a three-dimensional regular grid covering the region to be reconstructed, and deploys spherical harmonic probes at the nodes of the three-dimensional grid. Each spherical harmonic probe stores spherical harmonic coefficients to construct the path-averaged sound velocity field. The sound speed and flight time calculation module 120 performs spatial interpolation of the spherical harmonic coefficients at the vertices of the three-dimensional mesh surrounding the three-dimensional Gaussian elements for the propagation direction from each three-dimensional Gaussian element to the transducer, and calculates the path-average sound speed and flight time by combining the directional spherical harmonic basis functions corresponding to the propagation direction. The acoustic forward projection module 130 substitutes the path average sound speed and flight time into the sound wave analytical propagation model to calculate the theoretical sound pressure signal at each transducer. The joint optimization and reconstruction module 140 compares and inverts the theoretical sound pressure signal with the actual acquired signal, outputs the reconstructed initial sound pressure distribution image, and completes photoacoustic tomography.
[0064] The specific implementation techniques of each module / unit in the above examples of this application can be referred to the steps of the sound velocity adaptive photoacoustic tomography method based on spherical harmonic probe in the above embodiments, and will not be repeated here.
[0065] The preferred features in the above embodiments can be used individually in any embodiment, or in any combination thereof, provided they do not conflict with each other. Furthermore, parts not described in detail in the embodiments can be implemented using existing technologies.
[0066] The following examples and comparative examples will be used to further illustrate this application in order to better understand the above-mentioned technical solutions. It should be understood that the following are only some examples and are not intended to limit this application.
[0067] Example 1: Sound velocity self-correction and cerebral vascular reconstruction in transcranial photoacoustic imaging In transcranial photoacoustic imaging, a key challenge lies in the significant difference in sound velocity between the rigid skull and the soft brain tissue. This typically results in a substantial phase delay as sound waves penetrate the skull, leading to severe blurring and structural misalignment in the reconstructed brain images. By applying the self-focusing framework described in this application, the imaging system can automatically sense and compensate for the distortion effects of the skull on sound waves without requiring any prior skull CT images or auxiliary information. Figure 4 The qualitative comparison results of the simulation data show that the traditional dual-velocity UBP method highly relies on the accurate injection of additional structural priors, while the method in this embodiment adaptively learns the sound velocity distribution without prior knowledge, effectively eliminating phase cancellation and geometric distortion artifacts caused by non-uniform sound velocities. The structural fidelity comprehensively surpasses that of the prior-dependent dual-velocity UBP. During the reconstruction process, the system automatically optimizes the sound velocity properties at various points in space based on the acquired signal characteristics, enabling the originally divergent photoacoustic signals to accurately converge to the actual vascular anatomical locations. Ultimately, this framework can reconstruct the three-dimensional structure of the intricate microvascular network deep within the cerebral cortex, providing high-sharp visual evidence for brain function research and clinical diagnosis, and significantly improving the reliability of transcranial imaging.
[0068] Example 2: Efficient Reconstruction in Large-Field Breast Tumor Screening In wide-field breast tumor screening applications, the complex tissue composition within the breast creates a highly heterogeneous acoustic environment. These local differences in sound velocity often lead to significant structural distortions and blurred details in reconstructed images, particularly in deeper or peripheral regions, severely impacting the accurate assessment of lesion characteristics. By applying the sound velocity adaptive scheme of this application, the imaging system can intelligently identify and compensate for the influence of different tissue components on the sound wave propagation path, thereby achieving precise phase alignment across the entire three-dimensional imaging field of view. Figure 5 and Figure 6 The reconstruction results of the physical phantoms in complex acoustic environments are shown in the comparison. The method in this embodiment corrects acoustic aberrations under completely blind resolution conditions, and its structural sharpness and feature separation are significantly better than the single-velocity UBP and SlingBAG methods. Furthermore, the reconstruction effect is close to that of dual-velocity UBP, which relies on precise sound velocity priors. To quantify and verify the sharpness advantage, Figure 6 Simultaneously supplementing the full width at half maximum (FWHM) data of the reconstructed detailed structures by various methods, the method in this application exhibited the lowest FWHM value under the same test data, confirming its significant advantage in suppressing defocus blur. This compensation mechanism significantly enhances the visualization of the microvascular network within the breast and the tumor margin structure, enabling the realistic reconstruction of vascular connectivity and spatial distribution characteristics. Ultimately, this provides clinicians with a high-fidelity tissue anatomy map, facilitating more accurate identification of early-stage micro-tumors and assessment of their blood supply status, greatly improving the accuracy of photoacoustic imaging in breast cancer diagnosis.
[0069] Example 3: Robustness Verification of High-Resolution Reconstruction and Sparse Sampling Experiments demonstrate that the sharpness and spatial resolution of the reconstructed images obtained in this application are significantly superior to the traditional UBP and SlingBAG methods, effectively solving the defocusing problem in complex scenes such as transcranial and breast imaging. Addressing the inherent limitations of traditional methods in resolution and artifact control, Figure 7 This demonstrates the physical phantom data sampled from sparse viewpoints (retaining 12.5%). Comparison of reconstruction performance at 50% channel measurement. Traditional UBP and dual-velocity UBP methods are based on discrete voxel grids, with computational speed proportional to the number of voxels. Spatial resolution is limited by computational resources, and staircase artifacts are easily introduced during magnification. The SlingBAG method uses a staircase kernel function with multiple solid spheres superimposed to fit a Gaussian distribution, inevitably introducing banded artifacts when the reconstruction result is magnified. In contrast, this application uses a smooth Gaussian kernel function to characterize the sound source, and its frequency domain exponential attenuation characteristic is equivalent to a built-in low-pass filter. The reconstruction result can be resampled to voxels of arbitrary spatial resolution without producing aliasing artifacts. Figure 6As shown, relying on the inherent sparsity constraint of explicit Gaussian characterization, the method in this embodiment successfully suppressed the fringe artifacts caused by spatial aliasing under extreme undersampling conditions, maintaining a high degree of structural integrity. The results are combined from simulation and physical phantom experiments. Figures 4-7 This demonstrates that, without requiring external prior knowledge, this application achieves high-fidelity structural reconstruction, resistance to undersampling artifacts, and lossless output at arbitrary resolution, fully verifying its engineering applicability and algorithm robustness in complex clinical scenarios.
[0070] Based on the same technical concept, in other embodiments of this application, a terminal is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it can be used to execute the above-described sound velocity adaptive photoacoustic tomography method based on spherical harmonic probe.
[0071] Based on the same technical concept, in other embodiments of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, can be used to perform the above-described sound velocity adaptive photoacoustic tomography method based on a spherical harmonic probe.
[0072] Optionally, the memory is used to store programs; the memory may include volatile memory, such as random-access memory (RAM), such as static random-access memory (SRAM), double data rate synchronous dynamic random-access memory (DDR SDRAM), etc.; the memory may also include non-volatile memory, such as flash memory. The memory is used to store computer programs (such as application programs and functional modules that implement the above methods), computer instructions, etc., and the aforementioned computer programs and computer instructions can be partitioned and stored in one or more memories. Furthermore, the aforementioned computer programs, computer instructions, data, etc., can be accessed by the processor.
[0073] The aforementioned computer programs, computer instructions, etc., can be stored in partitions within one or more memory locations. Furthermore, the aforementioned computer programs, computer instructions, data, etc., can be accessed by a processor.
[0074] A processor is used to execute a computer program stored in memory to implement the various steps of the methods involved in the above embodiments. For details, please refer to the relevant descriptions in the preceding method embodiments.
[0075] The processor and memory can be separate structures or integrated structures. When the processor and memory are separate structures, they can be coupled together via a bus.
[0076] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0077] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0078] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0079] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0080] The foregoing has described some specific embodiments of this application. It should be understood that this application is not limited to the specific embodiments described above, and those skilled in the art can make various modifications or variations within the scope of the claims, which do not affect the substantive content of this application. The above-described preferred features can be used in any combination without conflict.
Claims
1. A sound velocity-adaptive photoacoustic tomography method based on a spherical harmonic probe, characterized in that, include: The initial sound pressure distribution of the region to be reconstructed is parameterized as a set of three-dimensional Gaussian elements. A three-dimensional regular grid covering the region to be reconstructed is constructed, and spherical harmonic probes are deployed at the nodes of the three-dimensional grid. Each spherical harmonic probe stores spherical harmonic coefficients. For each of the three-dimensional Gaussian elements to the transducer, spatial interpolation is performed on the spherical harmonic coefficients at the vertices of the three-dimensional mesh surrounding the three-dimensional Gaussian elements, and the path-average sound speed and flight time are calculated in combination with the directional spherical harmonic basis functions corresponding to the propagation direction. Substitute the average sound speed along the path and the flight time into the analytical propagation model of sound waves to calculate the theoretical sound pressure signal at each transducer. The theoretical sound pressure signal is compared and inverted with the actual acquired signal to output a reconstructed initial sound pressure distribution image, thus completing photoacoustic tomography.
2. The method according to claim 1, characterized in that, The parameterization of the initial sound pressure distribution of the region to be reconstructed into a three-dimensional Gaussian element set includes: The initial sound pressure distribution is characterized as N discrete three-dimensional Gaussian elements, each of which carries optimizable parameters including: center position coordinates, initial pressure amplitude and spatial standard deviation. The center position coordinates are used to locate the spatial position of the photoacoustic source; the initial pressure amplitude is used to characterize the local initial sound pressure intensity; and the spatial standard deviation is used to control the spatial range of the sound source.
3. The method according to claim 1, characterized in that, The process involves constructing a three-dimensional regular mesh covering the region to be reconstructed, and deploying spherical harmonic probes at the nodes of the three-dimensional mesh. Each spherical harmonic probe stores spherical harmonic coefficients, including: The region to be reconstructed is discretized into a three-dimensional regular grid; A spherical harmonic probe is configured at each node of the three-dimensional regular mesh; Each of the spherical harmonic probes stores spherical harmonic coefficients of order 0 to 2, forming a 9-dimensional parameter vector.
4. The method according to claim 1, characterized in that, The process involves spatially interpolating the spherical harmonic coefficients at the mesh vertices surrounding the three-dimensional Gaussian elements along their propagation direction to the transducer, and then calculating the path-averaged sound speed and flight time using the directional spherical harmonic basis functions corresponding to the propagation direction. This includes: Identify the three-dimensional mesh cell surrounding the center position of the three-dimensional Gaussian element, and extract the 8 vertices of the three-dimensional mesh cell; Perform trilinear interpolation on the spherical harmonic coefficients stored by the spherical harmonic probes at the eight vertices to generate a local spherical harmonic coefficient vector at the center position of the three-dimensional Gaussian element; Based on the unit direction vector pointing to the target transducer from the three-dimensional Gaussian element, construct the corresponding 9-dimensional real spherical harmonic basis function vector; The dot product of the local spherical harmonic coefficient vector and the 9-dimensional real spherical harmonic basis function vector is calculated to obtain the path-average sound speed in the propagation direction. The sound wave flight time is then calculated based on the Euclidean distance between the center position and the target transducer and the path-average sound speed.
5. The method according to claim 1, characterized in that, The step of substituting the path-average sound speed and the flight time into the analytical propagation model of sound waves to calculate the theoretical sound pressure signal at each transducer includes: A forward projection formula is constructed based on the far-field asymptotic approximation of the three-dimensional acoustic wave equation, serving as the analytical propagation model of the acoustic wave. Substitute the path-average sound speed and the flight time into the propagation delay term of the forward projection formula to calculate the temporal sound pressure contribution of each of the three-dimensional Gaussian elements to the target transducer. The time-domain sound pressure contributions of all the three-dimensional Gaussian elements are superimposed and summed to output the theoretical time-domain sound pressure signal at each transducer. The forward projection formula is expressed as follows: ; For the first m Each transducer at time t The theoretical sound pressure signal; N This represents the total number of three-dimensional Gaussian elements; For the first i The initial pressure amplitude of each Gaussian element; For the first i The first Gaosky Yuan and the first m Euclidean distance between transducers; The average speed of sound along the path; For the first i Spatial standard deviation of Gaussian elements.
6. The method according to claim 1, characterized in that, The step of comparing and inverting the theoretical sound pressure signal with the actual acquired signal to output a reconstructed initial sound pressure distribution image, thereby completing photoacoustic tomography, includes: Calculate the mean square error between the theoretical sound pressure signal and the actual acquired signal, and construct a global loss function; The gradient of the loss function with respect to the three-dimensional Gaussian parameters and the spherical harmonic probe parameters is obtained based on the automatic differentiation technique; Based on the gradient, the three-dimensional Gaussian meta-parameters and the spherical harmonic probe parameters are updated synchronously using a gradient optimization algorithm, and the gradient calculation and parameter update steps are executed iteratively. After each iteration, it is checked whether the global loss function meets the preset convergence condition. When the global loss function meets the preset convergence condition, the iteration is terminated, and the converged three-dimensional Gaussian meta-parameters are mapped and output as the initial sound pressure distribution image to complete photoacoustic tomography.
7. The method according to claim 6, characterized in that, The iterative update process also includes an adaptive density control step, specifically: During each iteration of backpropagation, the gradient norm corresponding to each three-dimensional Gaussian element is extracted as the local reconstruction error, and its spatial standard deviation is read as the spatial scale. When the local reconstruction error is greater than a preset error threshold and the spatial scale is greater than a preset scale threshold, the three-dimensional Gaussian primitive is split into multiple sub-Gaussian primitives, and the parameter distribution of the original primitive is inherited. The absolute value of the gradient of the global loss function with respect to the initial pressure magnitude of the three-dimensional Gaussian element is calculated as the contribution. When the contribution is lower than the preset contribution threshold, the three-dimensional Gaussian element is removed from the set.
8. A sound velocity adaptive photoacoustic tomography system based on a spherical harmonic probe, characterized in that, include: The joint characterization construction module parameterizes the initial sound pressure distribution of the region to be reconstructed into a three-dimensional Gaussian element set; constructs a three-dimensional regular grid covering the region to be reconstructed, and deploys spherical harmonic probes at the nodes of the three-dimensional grid. Each spherical harmonic probe stores spherical harmonic coefficients to construct a path-averaged sound velocity field. The sound speed and flight time calculation module performs spatial interpolation on the spherical harmonic coefficients at the vertices of the three-dimensional mesh surrounding the three-dimensional Gaussian elements for the propagation direction from each of the three-dimensional Gaussian elements to the transducer, and calculates the path-average sound speed and flight time in combination with the directional spherical harmonic basis functions corresponding to the propagation direction. The acoustic forward projection module substitutes the average sound speed along the path and the flight time into the analytical propagation model of sound waves to calculate the theoretical sound pressure signal at each transducer. The joint optimization and reconstruction module compares and inverts the theoretical sound pressure signal with the actual acquired signal, outputs the reconstructed initial sound pressure distribution image, and completes photoacoustic tomography.
9. A terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it can be used to perform the method of any one of claims 1-7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program can be used to perform the method of any one of claims 1-7.
Citation Information
Patent Citations
Method for calculating multilayer medium sound velocity based on photoacoustic imaging technology
CN103142216A
