Shear wave physical information inversion method and related device based on external excitation

Abstract

The present application discloses a shear wave physical information inversion method based on external excitation drive and related devices, which relate to the field of physical information reaction technology. The method is specifically as follows: low-frequency mechanical vibration is induced on the surface of the object to be measured by means of an external excitation device, and the shear wave echo signal is acquired based on the ultrafast ultrasonic acquisition sequence and vibration detection is performed. Subsequently, the discrete time and space coordinates corresponding to the measured data of the pure velocity field are input into the shear wave physical information inversion model for iterative inversion, and the two-dimensional inversion result of the shear modulus can be obtained. The shear wave physical information inversion model constructed here contains physical information neural networks that are related to time and space and space respectively. The present application constructs a two-dimensional inversion model by introducing a physical information neural network, successfully breaking through the limitation of traditional instantaneous elastic imaging relying only on one-dimensional sampling, significantly reducing the impact of human operation on the results, and greatly reducing equipment costs and operating thresholds.

Description

Technical Field

[0001] The present application relates to the technical field of physical information inversion, and in particular to a shear wave physical information inversion method based on external excitation drive and related devices. Background Art

[0002] Ultrasound elastography is a medical imaging technique specifically designed to assess the elastic properties of biological tissues. Its fundamental principle is that different biological tissues (both normal and diseased) possess varying elastic moduli. When subjected to external forces or alternating vibrations, tissues with varying elastic moduli will produce varying degrees of strain, which manifests morphologically as tissue deformation. By capturing signal fragments from a region of interest over a specific time period and utilizing comprehensive analysis techniques such as autocorrelation, these signals are converted into grayscale or color-coded images, thereby visually displaying the tissue's stiffness distribution.

[0003] In the field of computational mechanics, traditional numerical methods such as the finite difference method (FDM) and the finite element method (FEM) have been widely used for numerical solutions to wave equations and for modeling direct problems. However, these methods face a dual challenge when solving nonlinear, high-dimensional inverse problems. First, the inversion process requires repeated solutions to the direct problem, resulting in an exponential increase in computational complexity. Second, factors such as the prevalence of noise interference in actual measurement data, spatial discrete sampling, and uncertainty in boundary conditions make traditional numerical methods prone to ill-conditioned solution spaces, severely limiting their value in clinical applications.

[0004] In recent years, deep learning methods have gradually become an effective means of extracting feature representations from high-dimensional observational data. Despite this, purely data-driven models still have significant limitations: First, the annotation of medical imaging data relies on ex vivo tissue mechanical testing or numerical simulation, which leads to high data acquisition costs and the risk of domain shift. Second, existing feature extraction methods lack physical constraints. When the distribution of test data exceeds the range of the training set, the generalization ability of the model decreases sharply, and may even produce predictions that violate the laws of physics. Therefore, there is an urgent need to deeply integrate physical conservation laws with deep learning architectures. By embedding prior knowledge, the network can be guided to learn feature representations that conform to physical laws, thereby improving the robustness and extrapolation ability of the model under limited data conditions. Summary of the Invention

[0005] The purpose of this application is to provide a shear wave physical information inversion method and related devices based on external excitation driving, which can improve the robustness and extrapolation ability of the inversion model under limited data conditions.

[0006] To achieve the above objectives, this application provides the following solutions:

[0007] In a first aspect, the present application provides a shear wave physical information inversion method based on external excitation driving, comprising the following steps:

[0008] Low-frequency mechanical vibration is generated on the surface of the object under test through an external excitation device. At the same time, shear wave echo signals are acquired based on the ultrafast ultrasonic acquisition sequence, and vibration detection is performed on the shear wave echo signals to obtain the measured data of the pure velocity field of the object under test. The external excitation device includes a signal generator, a power amplifier and a customized vibrator.

[0009] The linear elastic wave control equation is converted into the physical information driven loss term of the physical information neural network to construct a shear wave physical information inversion model; the shear wave physical information inversion model includes a time-space related physical information neural network and a space-related physical information neural network; the time-space related physical information neural network is used to use discrete time coordinates and discrete space coordinates as model inputs, and output stream function prediction results and Lagrange multiplier prediction results; the partial derivative of the stream function prediction results is taken to obtain the axial velocity field prediction results; the space-related physical information neural network is used to use discrete space coordinates as model inputs, and output the shear modulus prediction results; the loss function of the shear wave physical information inversion model calculates the total loss using the stream function prediction results, the pure velocity field measured data and the shear modulus prediction results.

[0010] The discrete spatial coordinates and discrete time coordinates corresponding to the measured data of the pure velocity field are used as the first input data, and the discrete spatial coordinates corresponding to the measured data of the pure velocity field are used as the second input data, and are input together into the shear wave physical information inversion model. The measured data of the pure velocity field are used as labels of the time-space related physical information neural network for iterative inversion to obtain the shear wave physical information inversion results; the loss function of the shear wave physical information inversion model includes data-driven loss terms and physical information-driven loss terms; the shear wave physical information inversion results are the shear modulus inversion results corresponding to each discrete spatial coordinate; the shear modulus inversion results are the shear modulus prediction results output when the model meets the iteration exit conditions after several iterations.

[0011] Optionally, low-frequency mechanical vibration is generated on the surface of the object under test by an external excitation device, and shear wave echo signals are acquired based on an ultrafast ultrasonic acquisition sequence. Vibration detection is performed on the shear wave echo signals to obtain pure velocity field measured data of the object under test, specifically including:

[0012] Low-frequency mechanical vibration is generated on the surface of the object being measured through an external excitation device, and the displacement field estimation method based on the phase change of the in-phase component and orthogonal component of the ultrasonic echo signal is used to estimate the displacement field of the shear wave echo signal to obtain tissue micro-displacement field data.

[0013] The frequency domain 2D directional filtering technology is used to perform direction filtering on the tissue micro-displacement field data to obtain the pure velocity field measured data of the measured object; the pure velocity field measured data is used as the data-driven loss term of the physical information neural network.

[0014] Optionally, the displacement field is estimated according to the following formula:

[0015] .

[0016] in, is the average phase shift relative to the center frequency, c is the propagation speed of ultrasound in the object being measured, is the center frequency of the transducer, is the pulse repetition frequency, M is the number of sampling points in a single frame, N is the number of frames collected continuously, and Respectively represent n Frame No. m The in-phase component and quadrature component of each sampling point.

[0017] Optionally, a frequency domain 2D directional filtering technique is used to perform directional filtering on the tissue micro-displacement field data to obtain the measured data of the pure velocity field of the object being measured, specifically including:

[0018] The displacement data in the tissue micro-displacement field data are subjected to fast Fourier transform to obtain frequency domain displacement data.

[0019] The frequency domain displacement data is multiplied by the mask function, and the frequency domain components that meet the shear wave velocity characteristics are retained to obtain the filtered frequency domain displacement data.

[0020] The filtered frequency domain displacement data is restored to the time domain using inverse Fourier transform to obtain the measured data of the pure velocity field of the measured object.

[0021] Optionally, the mask function is as follows:

[0022] .

[0023] in, is the mask function, is the spatial wave number, is the angular frequency, is the target speed, To allow deviation.

[0024] The filtered frequency domain displacement data is restored to the time domain according to the following formula:

[0025] .

[0026] in, is the measured value of the pure velocity field of the object being measured, is the frequency domain displacement data after filtering, i is the imaginary unit, t is the discrete time coordinate.

[0027] Optionally, the loss function of the shear wave physical information inversion model is as follows:

[0028] .

[0029] in, is the value of the loss function of the shear wave physical information inversion model, is the physical information driven loss term, is the data-driven loss term, and are the weights of the physical information driven loss term and the data driven loss term, respectively.

[0030] The physical information driven loss term is calculated as follows:

[0031] .

[0032] .

[0033] in, μ is the shear modulus, v x and v y They are x Direction and y Speed ​​data of direction, x and y is the discrete space coordinate, t is the discrete time coordinate, p is the Lagrange multiplier p The partial derivative of 0 with respect to time, , ρ is the density of the object being measured.

[0034] The data-driven loss term is calculated as follows:

[0035] .

[0036] in, for y The velocity data in the direction is equivalent to the measured data of the pure velocity field of the object being measured. .

[0037] In a second aspect, the present application provides a shear wave physical information inversion system based on external excitation drive, comprising:

[0038] The ultrasonic application and vibration detection module is used to generate low-frequency mechanical vibration on the surface of the object under test through an external excitation device, and at the same time obtain shear wave echo signals based on the ultrafast ultrasonic acquisition sequence, and perform vibration detection on the shear wave echo signals to obtain the pure velocity field measured data of the object under test; the external excitation device includes a signal generator, a power amplifier and a customized vibrator.

[0039] The shear wave inversion model construction module is used to convert the linear elastic wave control equation into the physical information driven loss term of the physical information neural network to construct the shear wave physical information inversion model; the shear wave physical information inversion model includes the time-space related physical information neural network and the space-related physical information neural network; the time-space related physical information neural network is used to use discrete time coordinates and discrete space coordinates as model inputs, and output the stream function prediction results and the Lagrange multiplier prediction results; the partial derivative of the stream function prediction results is taken to obtain the axial velocity field prediction results; the space-related physical information neural network is used to use discrete space coordinates as model inputs, and output the shear modulus prediction results; the loss function of the shear wave physical information inversion model calculates the total loss based on the stream function prediction results, the pure velocity field measured data and the shear modulus prediction results.

[0040] The shear wave physical information inversion module is used to input the discrete spatial coordinates and discrete time coordinates corresponding to the pure velocity field measured data as the first input data, and the discrete spatial coordinates corresponding to the pure velocity field measured data as the second input data, and input them together into the shear wave physical information inversion model, and use the pure velocity field measured data as the label of the time-space related physical information neural network for iterative inversion to obtain the shear wave physical information inversion result; the loss function of the shear wave physical information inversion model includes data-driven loss terms and physical information-driven loss terms; the shear wave physical information inversion result is the shear modulus inversion result corresponding to each discrete spatial coordinate; the shear modulus inversion result is the shear modulus prediction result output when the model meets the iteration exit condition after several iterations.

[0041] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the shear wave physical information inversion method based on external excitation drive described above.

[0042] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the shear wave physical information inversion method based on external excitation driving as described above.

[0043] In a fifth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the shear wave physical information inversion method based on external excitation driving as described above.

[0044] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0045] The present application provides a shear wave physical information inversion method and related device based on external excitation drive. In this method, low-frequency mechanical vibration is generated on the surface of the object to be measured by an external excitation device, and the shear wave echo signal is obtained based on the ultrafast ultrasonic acquisition sequence and vibration detection is performed; then the discrete time and space coordinates corresponding to the measured data of the pure velocity field are input into the shear wave physical information inversion model for iterative inversion, and the shear wave physical information inversion result, that is, the two-dimensional inversion result of the shear modulus, can be obtained; the shear wave physical information inversion model constructed by the present application includes a time-space related physical information neural network and a space-related physical information neural network, which are respectively used to predict the output stream function prediction result and the Lagrange multiplier prediction result and the shear modulus prediction result. The loss function of the model includes a data-driven loss term and a physical information-driven loss term. Compared with traditional transient elastic imaging (TE) technology, the present application constructs a two-dimensional elastic modulus inversion model by introducing a physical information neural network (PINN), which completely breaks through the limitation of TE relying only on one-dimensional sampling. Furthermore, traditional TE relies on the operator to manually locate the ROI, which carries the risk of subjective bias. However, this application uses PINN to automatically learn the wave propagation trajectory, significantly reducing the impact of human operation on the results. This is particularly robust in obese patients (acoustic attenuation causes a 20% failure rate in traditional TE measurements) and deep organ detection. Furthermore, this application demonstrates significant advantages in hardware compatibility and imaging performance. For example, shear wave elastography (SWE) based on acoustic radiation force pulses relies on high-frequency focused ultrasound probes, with an imaging depth typically less than 6 cm and stringent requirements for probe pressure control, severely limiting its application in deep liver and pelvic organs. This application, however, uses a low-frequency probe with a penetration depth of over 8 cm. Combined with external oscillator excitation, it can excite stable shear waves without the need for a dedicated transducer, significantly reducing equipment costs and operational barriers. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present application 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 application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1A flowchart of a shear wave physical information inversion method based on external excitation driving is provided in one embodiment of the present application.

[0048] Figure 2 This is a flowchart of step A1 in a shear wave physical information inversion method based on external excitation driving provided in one embodiment of the present application.

[0049] Figure 3 This is a flowchart of step A12 in a shear wave physical information inversion method based on external excitation driving provided in one embodiment of the present application.

[0050] Figure 4 A schematic diagram of a technical route for inverting and iterating a shear wave physical information inversion model in a shear wave physical information inversion method based on external excitation drive provided in one embodiment of the present application.

[0051] Figure 5 A schematic diagram of the functional modules of a shear wave physical information inversion system based on external excitation drive provided in one embodiment of the present application.

[0052] Figure 6 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0054] Tissue biomechanical properties are widely recognized as crucial parameters for tissue characterization. Many pathological and physiological changes in clinical practice can lead to changes in tissue mechanical properties. Therefore, assessing changes in the stiffness of biological soft tissues can aid in the diagnosis and treatment of numerous diseases. Ultrasound elastography, as a noninvasive and reliable technique, has been widely adopted in clinical practice.

[0055] Currently, there are several problems with ultrasonic shear wave elastography technology, including:

[0056] (1) Insufficient imaging depth: Although high-frequency probes (≥5 MHz) improve the resolution of superficial tissues, ultrasound attenuation increases significantly with depth, resulting in a decrease in the shear wave signal-to-noise ratio in deep tissues (such as the deep liver and pelvic organs, depth > 8 cm), a decrease in the accuracy of elastic modulus measurement, and difficulty in accurately assessing the elastic characteristics of deep lesions.

[0057] (2) High requirements for probe hardware: A high-precision ultrasonic probe (such as a wide-band sensor) is required to quickly capture the time-space information of shear wave propagation. The hardware design is complex and the cost is high.

[0058] (3) Dependence on operator technique: The force, angle, and stability of the probe pressure directly affect the excitation and propagation of shear waves. When scanning irregular organs, the probe position needs to be manually adjusted to avoid blood vessels, bones, and other structures. Differences in technique can easily lead to significant fluctuations in the elasticity value of the same lesion.

[0059] (4) Low resolution: Spatial resolution decreases with depth, and the ability to identify the elastic boundaries of tiny lesions (<5 mm) is insufficient. Ultrasonic speckle noise and tissue heterogeneity (such as fat particles and fibrous septa) cause the elastic map to have a "salt and pepper effect", which requires algorithmic smoothing and may mask local elastic details.

[0060] Ultrasound transient elastography (TEE) is currently listed as the preferred noninvasive method for liver fibrosis assessment by authoritative guidelines such as the European Association for the Study of the Liver (EASL) and the American Association for the Study of Liver Diseases (AASLD). It is explicitly recommended as an alternative to liver biopsy for patients without contraindications. However, it only provides average elasticity values ​​within a one-dimensional ROI and cannot locate or assess focal lesions. Tissue heterogeneity, such as inflammation, necrosis, or tumors, can distort elasticity values. Although it has irreplaceable advantages in assessing diffuse liver lesions, its ability to assess focal lesions, extrahepatic organs, and complex pathological scenarios is significantly limited, requiring supplementation with multimodal imaging techniques (such as SWE and MRI elastography) and comprehensive clinical judgment.

[0061] In view of the problems existing in current ultrasound elastography technology, this application aims to propose an inversion model that integrates physical information networks, so that it can accurately invert the mechanical properties of biological tissues using limited data without the need to establish a data set.

[0062] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0063] The shear wave physical information inversion method based on external excitation driving provided in the embodiment of the present application, in an exemplary embodiment, Figure 1 As shown, the following steps are included:

[0064] A1. An external excitation device generates low-frequency mechanical vibration on the surface of the object being measured. Simultaneously, shear wave echo signals are acquired based on an ultrafast ultrasonic acquisition sequence. Vibration detection is performed on the shear wave echo signals to obtain the measured data of the pure velocity field of the object being measured. The external excitation device includes a signal generator, a power amplifier, and a customized vibrator.

[0065] As an exemplary embodiment, the external excitation device builds a controllable shear wave excitation platform based on the electrical connection architecture of customized vibrators, signal generators and power amplifiers. Among them, the signal generator serves as an excitation source, outputting a single-cycle sinusoidal pulse electrical signal, the frequency of which can be adjusted within a multi-band range, and the typical operating frequency includes key frequency bands such as 200Hz. After the electrical signal is amplified by the power amplifier, it drives the customized vibrator to generate low-frequency mechanical vibrations on the surface of the object under test. The vibration propagates into the interior of the object through the interface coupling effect, thereby inducing the generation of shear waves. The external excitation device is highly controllable. By flexibly setting the frequency, output amplitude and waveform parameters of the signal generator, combined with the gain matching mechanism of the power amplifier, shear waves of specific wavelengths can be accurately generated according to the needs of different detection scenarios.

[0066] The accompanying acquisition device utilizes a Versonics programmable ultrasound platform, which achieves high-precision acquisition of shear wave IQ signals based on an ultrafast ultrasound acquisition sequence. By dynamically adjusting imaging frame rate and resolution parameters, the platform can be optimized based on actual testing requirements, enabling dynamic monitoring and precise analysis of shear wave propagation characteristics.

[0067] Compared to shear wave elastography based on acoustic radiation force impulse (ARFI) excitation, this solution does not require specialized hardware such as high-end ultrasound probes. By optimizing the imaging depth through the PINN algorithm, the system significantly reduces its reliance on operator expertise, significantly improving the clinical applicability and accessibility of the technology.

[0068] A2. The linear elastic wave control equation is converted into the physical information driven loss term of the physical information neural network to construct the shear wave physical information inversion model; the shear wave physical information inversion model includes the time-space related physical information neural network and the space-related physical information neural network; the time-space related physical information neural network is used to use discrete time coordinates and discrete space coordinates as model inputs, and output the stream function prediction results and the Lagrange multiplier prediction results; the partial derivative of the stream function prediction results is taken to obtain the axial velocity field prediction results; the space-related physical information neural network is used to use discrete space coordinates as model inputs, and output the shear modulus prediction results; the loss function of the shear wave physical information inversion model calculates the total loss using the stream function prediction results, the pure velocity field measured data and the shear modulus prediction results.

[0069] This approach leverages the unique physical constraints of PINN to significantly enhance the system's robustness to noise interference during ultrasound signal processing. Experiments have shown that this method can achieve higher spatial resolution in the detection of tiny foreign bodies (such as early-stage tumor lesions), significantly outperforming traditional numerical calculation methods.

[0070] A3. The discrete spatial coordinates and discrete time coordinates corresponding to the measured data of the pure velocity field are used as the first input data, and the discrete spatial coordinates corresponding to the measured data of the pure velocity field are used as the second input data, and are input together into the shear wave physical information inversion model. The measured data of the pure velocity field are used as labels of the time-space related physical information neural network for iterative inversion to obtain the shear wave physical information inversion results; the loss function of the shear wave physical information inversion model includes data-driven loss terms and physical information-driven loss terms; the shear wave physical information inversion results are the shear modulus inversion results corresponding to each discrete spatial coordinate; the shear modulus inversion results are the shear modulus prediction results output when the model meets the iteration exit conditions after several iterations.

[0071] In an exemplary embodiment, Figure 2 As shown, step A1 includes the following steps:

[0072] A11. Generate low-frequency mechanical vibrations on the surface of the object being measured through an external excitation device, and use a displacement field estimation method based on the phase changes of the in-phase and quadrature components of the ultrasonic echo signal to estimate the displacement field of the shear wave echo signal to obtain tissue micro-displacement field data.

[0073] To achieve ultrafast ultrasonic reconstruction speed in shear wave elastography, this embodiment adopts a displacement field estimation method based on the phase changes of the in-phase component I and the orthogonal component Q of the ultrasonic echo signal. This method achieves high-precision reconstruction of tissue micro-displacement by analyzing the phase characteristics of the IQ signals between adjacent ultrasonic frames. Specifically, a one-dimensional autocorrelation algorithm is used to determine the relative velocity between reference frames by calculating the average phase shift relative to the center frequency. Specifically, in this embodiment, the displacement field is estimated according to the following formula:

[0074] .

[0075] in, is the average phase shift relative to the center frequency, c is the propagation speed of ultrasound in the object being measured, is the center frequency of the transducer, is the pulse repetition frequency, M is the number of sampling points in a single frame, N is the number of frames collected continuously, and Respectively represent n Frame No. m The in-phase component and quadrature component of each sampling point.

[0076] This step effectively suppresses random noise interference by statistically averaging the phase difference of the shear wave IQ signal, achieving micron-level displacement resolution and kilohertz-level imaging frame rate. It is particularly suitable for monitoring the propagation characteristics of shear waves in biological soft tissues and provides a data basis for high-precision reconstruction of tissue elastic modulus.

[0077] A12. Use frequency domain 2D directional filtering technology to perform direction filtering on the tissue micro-displacement field data to obtain the pure velocity field measured data of the measured object; the pure velocity field measured data is used as the data-driven loss item of the physical information neural network.

[0078] In this embodiment, if Figure 3 As shown, step A12 specifically includes the following steps:

[0079] A121. Perform fast Fourier transform on the displacement data in the tissue micro-displacement field data to obtain frequency domain displacement data. Due to the complex boundary conditions at the excitation site, this solution uses frequency domain 2D directional filtering technology to address the reflection and refraction interference caused by the complex excitation boundary. The two-dimensional directional filtering technology performs fast Fourier transform FFT on all points at a specific depth. The phase estimation in the previous step obtains In fact, it is the displacement field of all points in the entire two-dimensional plane at a specific depth z over a period of time, that is, First, the relative speed between the reference frames Perform fast Fourier transform (FFT) to convert to frequency domain representation ;in, yes The displacement set of all points with different x coordinates when y is fixed at a certain value.

[0080] A122. Multiply the frequency domain displacement data by the mask function, retain the frequency domain components that meet the shear wave velocity characteristics, and obtain the filtered frequency domain displacement data. Specifically, the mask function is shown as follows:

[0081] .

[0082] in, is the mask function, is the spatial wave number, is the angular frequency, is the target speed, is the allowable deviation. and Multiply to retain the frequency domain components that conform to the shear wave velocity characteristics and obtain the filtered frequency domain displacement data .

[0083] A123. Use the inverse Fourier transform to restore the filtered frequency domain displacement data to the time domain to obtain the measured value of the pure velocity field of the measured object. In this embodiment, the filtered frequency domain displacement data can be restored to the time domain according to the following formula:

[0084] .

[0085] in, is the measured value of the pure velocity field of the object being measured, is the frequency domain displacement data after filtering, i is the imaginary unit, t is a discrete time coordinate. Through the inverse Fourier transform operation, the frequency domain 2D directional filtering The pure velocity field data is reconstructed from the frequency-wavenumber domain to the time domain.

[0086] The above step A12 utilizes the dispersion characteristics of shear waves and the selective suppression capability of directional filtering operators to precisely control the energy distribution of interference waves in the frequency domain, effectively filtering out frequency domain components of interference waves such as linear interference and random noise, and significantly improving the signal-to-noise ratio and resolution of velocity field data.

[0087] Specifically in this embodiment, before constructing the inversion model, it is first necessary to complete the derivation of the wave equation that conforms to the data. Specifically, the two-dimensional dynamic equilibrium equation is shown as follows:

[0088] .

[0089] .

[0090] Here we consider a linear elastic model that satisfies incompressibility and isotropy in two-dimensional space, and its constitutive equation is expressed as:

[0091] .

[0092] .

[0093] .

[0094] in, 、 denote normal stress and shear stress respectively, and the subscripts j and k The value of x 、 y 、 z ; For strain, is the shear modulus, is the Lagrange multiplier associated with the incompressible constraint. Combining the above formulas, the governing equation of linear elastic waves in inhomogeneous media represented by displacement can be derived:

[0095] .

[0096] Due to the incompressible constraint, the volumetric strain is 0:

[0097] .

[0098] The mixed partial derivatives in the linear elastic wave equation can be combined and simplified:

[0099] .

[0100] For this embodiment, the particle velocity is often used. Therefore, the partial derivative of the simplified wave equation with respect to time can be calculated to obtain the linear elastic wave governing equation represented by velocity, which can be summarized as follows:

[0101] .

[0102] in, , and in order to ensure the incompressibility condition, that is, the divergence of the velocity field is 0, that is ; In order to naturally meet this condition, a stream function is introduced Satisfy the following formula:

[0103] .

[0104] in, v y The physical meaning represented is consistent with the measured data of the pure velocity field after direction filtering.

[0105] The shear wave physical information inversion model consists of two sub-models constructed by physical information neural networks, which are used to solve the spatial correlation and spatiotemporal correlation parameters respectively. Its architecture is as follows: Figure 4 As shown. The time-space related physical information neural network (NN1) takes the time-space coordinates as input parameters and the output parameter is the stream function and Lagrange multipliers p , where the stream function output by NN1 is The speed data in different directions can be obtained by taking partial derivatives and The spatial-related physical information neural network (NN2) only takes the spatial coordinates as input parameters and outputs the shear modulus. The outputs of NN1 and NN2 are iteratively learned through inversion to fit the input data and solve the pre-set physical constraints (i.e., the governing equations of linear elastic waves). While NN1 fits the input data, NN2 infers the spatial distribution of the shear modulus in the wave propagation region.

[0106] In order for the shear modulus to be correctly inverted, it is necessary to ensure that the input shear wave enters the ROI from one side and leaves the ROI from the other side, such as Figure 4 As shown in Figure 1, the shear wave propagates from the left side of the ROI to its right side. A fully connected feedforward structure is adopted for NN1 and NN2. The specific neural network structure parameters are shown in Table 1.

[0107] Table 1 Neural network structure parameters

[0108]

[0109] The number of layers of NN1 and NN2 is 6 (one input layer, one output layer and four hidden layers), and each hidden layer contains 80 neurons. The weights and biases will be updated through the inversion iterative process of the backpropagation algorithm to minimize the loss function L. In this embodiment, Tanh is selected as the activation function. Its continuity and smoothness characteristics help to accurately simulate the behavior of the physical field and ensure the numerical stability of the model during the inversion iterative process. Since the output parameters of NN2 are There are only positive values, so the final output of NN2 will pass through the Softplus activation function to ensure that the output meets the requirements.

[0110] As an exemplary embodiment, when the model parameters are updated during the inversion iteration process, the loss function of the shear wave physical information inversion model is shown as follows:

[0111] .

[0112] in, is the value of the loss function of the shear wave physical information inversion model, is the physical information driven loss term, is the data-driven loss term, and are the weights of the physical information driven loss term and the data driven loss term, respectively, which are used to balance the influence of the two parts of the loss function to ensure that they converge effectively during the inversion iteration process, especially in the experimental data.

[0113] The physical information driven loss term is calculated as follows:

[0114] .

[0115] .

[0116] in, μ is the shear modulus, v x and v y They are x Direction and y Speed ​​data of direction, x and y is the discrete space coordinate, t is the discrete time coordinate, p is the Lagrange multiplier p The partial derivative of 0 with respect to time, , ρ is the density of the object being measured.

[0117] The data-driven loss term is calculated as follows:

[0118] .

[0119] in, for y The velocity data in the direction is equivalent to the measured data of the pure velocity field of the object being measured. .

[0120] To improve computational efficiency and ensure representativeness of the training data, the model inversion iteration process is set to randomly sample 10,000 points from the spatiotemporal space every 1,000 epochs. The sampling process uses the Latin Hypercube Sampling (LHS) method, which helps maximize sample coverage and reduce sampling bias by ensuring uniform sampling in each dimension. This strategy not only improves data utilization efficiency but also accelerates the network inversion iteration process.

[0121] In order to stop the iterative inversion when the loss function converges, such as Figure 4 As shown, define the maximum number of iterations N 0=8×10 4 and learning rate , the iteration exit condition is when the number of Epochs N achieve N 0 or current total loss With the initial total loss If the ratio is less than the preset value ε, the calculation is stopped. To ensure the comparability of the results, the same weights and biases are loaded into the network for initialization, and the Adam optimizer is used for parameter update to accelerate convergence and improve training stability.

[0122] This scheme constructs a collaborative framework of frequency-domain two-dimensional directional filtering technology and PINN inversion model, and effectively suppresses interference factors in the shear wave propagation process through the closed-loop optimization strategy of "signal preprocessing-elastic parameter reconstruction". This mechanism ensures that the imaging system can still maintain high-precision elastic parameter reconstruction capabilities under complex boundary conditions. At the same time, based on the PINN physical information neural network, it innovatively integrates the physical equations of elastic mechanics with the data-driven learning paradigm. This method effectively overcomes the large error problem existing in traditional technologies in the inversion of elastic modulus of heterogeneous tissues such as fatty liver and liver fibrosis, and realizes high-precision quantitative evaluation of the elastic distribution of focal lesion areas. In addition, by introducing an adaptive weight distribution strategy, the PINN algorithm can dynamically adjust the imaging parameters according to the characteristics of different tissue regions, further optimize the two-dimensional elastic distribution imaging effect, and make the imaging results more in line with actual clinical needs.

[0123] The shear wave physical information inversion method based on external excitation driving provided in the above embodiments of the present application innovatively applies the PINN technology to ultrasonic transient elastic imaging, overcoming many application limitations of many traditional elastic imaging methods.

[0124] Compared to traditional transient elastography (TE) technology, this application introduces a physical information neural network (PINN) to construct a two-dimensional elastic modulus inversion model, completely overcoming the limitation of TE's reliance on one-dimensional sampling. Traditional TE, constrained by the homogeneous medium assumption, suffers from elastic modulus inversion errors exceeding 20% ​​for heterogeneous tissues such as fatty liver and fibrosis, and is unable to locate the specific hardness characteristics of focal lesions. However, this application uses ultrafast ultrasound to acquire spatial-temporal propagation maps, combined with the elastic mechanics wave equation embedded in the PINN, to accurately capture shear wave variations in heterogeneous tissues. Furthermore, traditional TE relies on the operator to manually locate the ROI (e.g., the S5 / S8 segments of the right lobe of the liver), which carries the risk of subjective bias. However, this application uses PINN to automatically learn wave propagation trajectories, significantly reducing the impact of human intervention on the results. This method demonstrates enhanced robustness, particularly in obese patients (where acoustic attenuation leads to a 20% failure rate in traditional TE measurements) and for deep organ examinations.

[0125] Compared to shear wave elastography (SWE) based on acoustic radiation force impulse (ARFI), this application demonstrates significant advantages in hardware compatibility and imaging performance. ARFI technology relies on a high-frequency focused ultrasound probe, typically with an imaging depth of less than 6 cm and stringent probe pressure control requirements (pressure deviation >10 g / cm² results in elasticity fluctuations >15%), severely limiting its application in deep liver and pelvic organs. This application, on the other hand, utilizes a low-frequency probe, enabling penetration depths exceeding 8 cm. Combined with external transducer excitation (e.g., 200 Hz low-frequency vibration), this method eliminates the need for a dedicated transducer to generate stable shear waves, significantly reducing equipment cost and operational requirements. At the algorithmic level, the proposed PINN model inherently suppresses ultrasound speckle noise and motion artifacts through physical equation constraints, maintaining minimal inversion error even at low signal-to-noise ratios. This "lightweight hardware + intelligent algorithm" design not only addresses the complex and limited depth limitations of ARFI equipment but also provides a clinically applicable elastic imaging solution that combines data-driven flexibility with the rigor of mechanical laws through physically interpretable inversion.

[0126] Based on the same inventive concept, the embodiment of the present application also provides a system for implementing the above-mentioned shear wave physical information inversion method based on external excitation. The solution provided by the system is similar to the solution described in the above-mentioned method. In an exemplary embodiment, Figure 5 As shown, a shear wave physical information inversion system based on external excitation driving is provided, comprising:

[0127] The ultrasonic application and vibration detection module is used to generate low-frequency mechanical vibration on the surface of the object under test through an external excitation device, and at the same time obtain shear wave echo signals based on the ultrafast ultrasonic acquisition sequence, and perform vibration detection on the shear wave echo signals to obtain the pure velocity field measured data of the object under test; the external excitation device includes a signal generator, a power amplifier and a customized vibrator.

[0128] The shear wave inversion model construction module is used to convert the linear elastic wave control equation into the physical information driven loss term of the physical information neural network to construct the shear wave physical information inversion model; the shear wave physical information inversion model includes the time-space related physical information neural network and the space-related physical information neural network; the time-space related physical information neural network is used to use discrete time coordinates and discrete space coordinates as model inputs, and output the stream function prediction results and the Lagrange multiplier prediction results; the partial derivative of the stream function prediction results is taken to obtain the axial velocity field prediction results; the space-related physical information neural network is used to use discrete space coordinates as model inputs, and output the shear modulus prediction results; the loss function of the shear wave physical information inversion model calculates the total loss based on the stream function prediction results, the pure velocity field measured data and the shear modulus prediction results.

[0129] The shear wave physical information inversion module is used to input the discrete spatial coordinates and discrete time coordinates corresponding to the pure velocity field measured data as the first input data, and the discrete spatial coordinates corresponding to the pure velocity field measured data as the second input data, and input them together into the shear wave physical information inversion model, and use the pure velocity field measured data as the label of the time-space related physical information neural network for iterative inversion to obtain the shear wave physical information inversion result; the loss function of the shear wave physical information inversion model includes data-driven loss terms and physical information-driven loss terms; the shear wave physical information inversion result is the shear modulus inversion result corresponding to each discrete spatial coordinate; the shear modulus inversion result is the shear modulus prediction result output when the model meets the iteration exit condition after several iterations.

[0130] certainly, Figure 5 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different functions. Figure 5 One or at least two components of the system shown.

[0131] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a shear wave physical information inversion method based on external excitation drive provided in the above embodiment can be implemented.

[0132] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0133] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0134] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0135] In an exemplary embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0136] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0137] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0138] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0139] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0140] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A shear wave physical information inversion method based on external excitation driving, characterized in that: include: Low-frequency mechanical vibration is generated on the surface of the object under test through an external excitation device, and shear wave echo signals are acquired based on an ultrafast ultrasonic acquisition sequence. Vibration detection is performed on the shear wave echo signals to obtain pure velocity field measured data of the object under test; the external excitation device includes a signal generator, a power amplifier and a customized vibrator; The linear elastic wave control equation is converted into a physical information driven loss term of a physical information neural network to construct a shear wave physical information inversion model; the shear wave physical information inversion model includes a time-space related physical information neural network and a space-related physical information neural network; the time-space related physical information neural network is used to take discrete time coordinates and discrete space coordinates as model inputs, and output stream function prediction results and Lagrange multiplier prediction results; the partial derivative of the stream function prediction result is obtained to obtain the axial velocity field prediction result; the space-related physical information neural network is used to take discrete space coordinates as model inputs, and output the shear modulus prediction result; the loss function of the shear wave physical information inversion model calculates the total loss using the stream function prediction result, the pure velocity field measured data and the shear modulus prediction result; The discrete spatial coordinates and discrete time coordinates corresponding to the measured data of the pure velocity field are used as first input data, and the discrete spatial coordinates corresponding to the measured data of the pure velocity field are used as second input data, and are inputted into the shear wave physical information inversion model together. The measured data of the pure velocity field are used as labels of a spatiotemporal related physical information neural network for iterative inversion to obtain a shear wave physical information inversion result; the loss function of the shear wave physical information inversion model includes a data-driven loss term and a physical information-driven loss term; The shear wave physical information inversion result is the shear modulus inversion result corresponding to each discrete space coordinate; the shear modulus inversion result is the shear modulus prediction result output when the model meets the iteration exit condition after several iterations.

2. The shear wave physical information inversion method based on external excitation driving according to claim 1 is characterized in that: An external excitation device generates low-frequency mechanical vibration on the surface of the object being measured. At the same time, a shear wave echo signal is acquired based on an ultrafast ultrasonic acquisition sequence. Vibration detection is performed on the shear wave echo signal to obtain the pure velocity field measured data of the object being measured, specifically including: Low-frequency mechanical vibration is generated on the surface of the object being measured by an external excitation device, and the displacement field estimation method based on the phase change of the in-phase component and the orthogonal component of the ultrasonic echo signal is used to estimate the displacement field of the shear wave echo signal to obtain tissue micro-displacement field data; The tissue micro-displacement field data is directionally filtered using frequency domain 2D directional filtering technology to obtain pure velocity field measured data of the object being measured; the pure velocity field measured data is used as the data-driven loss term of the physical information neural network.

3. The shear wave physical information inversion method based on external excitation driving according to claim 2 is characterized in that: The displacement field is estimated according to the following formula: ； in, is the average phase shift relative to the center frequency, c is the propagation speed of ultrasound in the object being measured, is the center frequency of the transducer, is the pulse repetition frequency, M is the number of sampling points in a single frame, N is the number of frames collected continuously, and Respectively represent n Frame No. m The in-phase component and quadrature component of each sampling point.

4. The shear wave physical information inversion method based on external excitation driving according to claim 2 is characterized in that: The frequency domain 2D directional filtering technology is used to perform directional filtering on the tissue micro-displacement field data to obtain the pure velocity field measured data of the measured object, specifically including: Performing fast Fourier transform on the displacement data in the tissue micro-displacement field data to obtain frequency domain displacement data; Multiplying the frequency domain displacement data by a mask function, retaining the frequency domain components that conform to the shear wave velocity characteristics, and obtaining filtered frequency domain displacement data; The filtered frequency domain displacement data is restored to the time domain using inverse Fourier transform to obtain the pure velocity field measured data of the measured object.

5. The shear wave physical information inversion method based on external excitation driving according to claim 4 is characterized in that: The mask function is as follows: ； in, is the mask function, is the spatial wave number, is the angular frequency, is the target speed, To allow deviation; The filtered frequency domain displacement data is restored to the time domain according to the following formula: ； in, is the measured data of the pure velocity field of the object being measured, is the frequency domain displacement data after filtering, i is the imaginary unit, t is the discrete time coordinate.

6. The shear wave physical information inversion method based on external excitation driving according to claim 2 is characterized in that: The loss function of the shear wave physical information inversion model is shown as follows: ； in, is the value of the loss function of the shear wave physical information inversion model, is the physical information driven loss term, is the data-driven loss term, and are the weights of the physical information driven loss term and the data driven loss term respectively; The physical information driven loss term is calculated as follows: ； ； in, μ is the shear modulus, v x and v y They are x Direction and y Speed ​​data of direction, x and y is the discrete space coordinate, t is the discrete time coordinate, p is the Lagrange multiplier p The partial derivative of 0 with respect to time, , ρ is the density of the object being measured; The data-driven loss term is calculated as follows: ； in, for y The velocity data in the direction is equivalent to the measured data of the pure velocity field of the object being measured. .

7. A shear wave physical information inversion system based on external excitation drive, characterized in that: include: The ultrasonic application and vibration detection module is used to generate low-frequency mechanical vibration on the surface of the object under test through an external excitation device, and simultaneously obtain shear wave echo signals based on the ultrafast ultrasonic acquisition sequence, and perform vibration detection on the shear wave echo signals to obtain the measured data of the pure velocity field of the object under test; the external excitation device includes a signal generator, a power amplifier and a customized vibrator; A shear wave inversion model construction module is used to convert the linear elastic wave control equation into a physical information driven loss term of a physical information neural network, and construct a shear wave physical information inversion model; the shear wave physical information inversion model includes a time-space related physical information neural network and a space-related physical information neural network; the time-space related physical information neural network is used to use discrete time coordinates and discrete space coordinates as model inputs, and output stream function prediction results and Lagrange multiplier prediction results; the partial derivative of the stream function prediction result is obtained to obtain an axial velocity field prediction result; the space-related physical information neural network is used to use discrete space coordinates as model inputs, and output a shear modulus prediction result; the loss function of the shear wave physical information inversion model calculates the total loss using the stream function prediction result, the pure velocity field measured data and the shear modulus prediction result; A shear wave physical information inversion module is used to input the discrete spatial coordinates and discrete time coordinates corresponding to the pure velocity field measured data as first input data, and the discrete spatial coordinates corresponding to the pure velocity field measured data as second input data, and input them together into the shear wave physical information inversion model, and use the pure velocity field measured data as labels of the time-space related physical information neural network for iterative inversion to obtain the shear wave physical information inversion result; the loss function of the shear wave physical information inversion model includes a data-driven loss term and a physical information-driven loss term; The shear wave physical information inversion result is the shear modulus inversion result corresponding to each discrete space coordinate; the shear modulus inversion result is the shear modulus prediction result output when the model meets the iteration exit condition after several iterations.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the shear wave physical information inversion method based on external excitation driving as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the shear wave physical information inversion method based on external excitation driving according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the shear wave physical information inversion method based on external excitation driving according to any one of claims 1 to 6 is implemented.

Images