A method for synchronously quantifying concentration of magnetic particles and viscoelasticity of tissue

Abstract

This application discloses a method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity. The method includes: forming an external time-varying magnetic field; applying the time-varying magnetic field to magnetic particles in a medium, causing the magnetic particles to vibrate and transmit the vibration displacement to surrounding tissues; acquiring raw ultrasound data of the vibration of the magnetic particles and the surrounding tissues; calculating the measured axial vibration displacement of the magnetic particles and the surrounding tissues based on the raw ultrasound data; and iteratively solving the problem under dual-loss constraints using a physical information neural network, combined with the measured axial vibration displacement, to simultaneously output quantitative imaging results of magnetic particle concentration, tissue viscosity, and tissue shear modulus. The dual-loss constraints include data loss constraints and physical equation loss constraints. This application can simultaneously quantitatively image two key types of information: magnetic particle concentration and tissue viscoelasticity, and improves quantitative accuracy and robustness.

Description

Technical Field

[0001] This application relates to the technical field of imaging analysis, specifically to a method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity. Background Technology

[0002] Currently, magnetic particles are materials with a magnetic core that achieve specific functions through surface coating or functionalization. Due to their excellent biocompatibility and tunable physicochemical properties, they exhibit broad application potential in modern medicine. These particles can be used in various applications such as drug delivery, molecular tracing, imaging, and targeted therapy. However, to fully utilize these functions, it is necessary to accurately determine their quantitative concentration distribution within tissues and the viscoelastic distribution of the affected tissues (i.e., the mechanical properties of the tissue, including viscosity and shear modulus, and their spatial distribution). Determining the quantitative concentration distribution of magnetic particles can not only optimize the design and delivery strategies of drug carriers but also quantify the enrichment efficiency and clearance mechanisms of drugs in target tissues, providing a scientific basis for dose control and efficacy evaluation. Simultaneously, obtaining the viscoelastic distribution of the affected tissues helps reveal the transport behavior and therapeutic response characteristics of particles in different tissue environments, thereby improving imaging accuracy, enhancing therapeutic effects, and providing reliable data support for multimodal biomedical research.

[0003] In related technologies, magnetic particle imaging methods mainly focus on magnetic particle imaging and magneto-ultrasound. These techniques can spatially represent the distribution of magnetic particles and reflect their enrichment level through signal intensity. Under certain conditions, they can also estimate the quantitative concentration distribution of magnetic particles. However, most existing concentration estimation methods are limited to acquiring magnetic particle information, and in their modeling of tissue mechanical properties, they typically ignore the viscous component of the tissue, considering only elastic parameters and assuming their uniform distribution as a constant. If the spatial distribution or concentration of magnetic particles is imaged while ignoring or oversimplifying the tissue viscoelastic distribution, the coupling relationship between tissue state and particle distribution cannot be fully assessed, thus reducing quantitative accuracy and robustness, and affecting clinical diagnosis and treatment outcomes.

[0004] Therefore, there is an urgent need in this field for a technical solution that can overcome the above-mentioned defects and achieve synchronous, quantitative, and high-precision imaging of magnetic particle concentration and tissue viscoelasticity. Summary of the Invention

[0005] The purpose of this application is to provide [something] for improving the accuracy and robustness of quantitative imaging analysis procedures.

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

[0007] In a first aspect, this application provides a method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity. The method includes: forming an external time-varying magnetic field; acting on magnetic particles in a medium through the time-varying magnetic field to cause the magnetic particles to vibrate and transmit the vibration displacement to surrounding tissues; acquiring raw ultrasound data of the vibration of the magnetic particles and the surrounding tissues; calculating the measured axial vibration displacement of the magnetic particles and the surrounding tissues based on the raw ultrasound data; and iteratively solving the problem under dual-loss constraints using a physical information neural network, combined with the measured axial vibration displacement, to simultaneously output quantitative imaging results of magnetic particle concentration, tissue viscosity, and tissue shear modulus; wherein the dual-loss constraints include data loss constraints and physical equation loss constraints.

[0008] For example, the formation of the external time-varying magnetic field includes: generating a signal of a specified waveform through a signal generator, and driving a coil to generate the external time-varying magnetic field after the signal is amplified by a power amplifier; or, driving a coil to generate the external time-varying magnetic field by instantaneous discharge of a capacitor energy storage circuit.

[0009] For example, the raw ultrasound data includes amplitude information and phase information. Calculating the axial vibration displacement of the magnetic particle and the surrounding tissue based on the raw ultrasound data includes: calculating the axial vibration displacement by combining the amplitude information and the phase information according to a displacement tracking algorithm.

[0010] For example, the calculation of axial vibration displacement by combining the amplitude information and the phase information is shown in the following formula:

[0011]

[0012] In the formula, To measure the axial vibration displacement, The speed of sound in the surrounding tissues, To demodulate the center frequency of the previous ultrasound radio frequency data, and These are the amplitude information and phase information, respectively. For axial sample range, and This indicates two adjacent sampling times.

[0013] For example, the physical information neural network includes a first network and a second network. The first network includes a displacement prediction network, which takes spatial coordinates and discrete time series as input features and outputs intermediate variables and Lagrange multipliers related to incompressible constraints. By taking the lateral and axial derivatives of the intermediate variables respectively, the predicted lateral vibration displacement and the predicted axial vibration displacement are calculated as prediction data. The second network includes a parameter inversion network, which takes spatial coordinates as input features and outputs magnetic particle concentration, tissue viscosity, and tissue shear modulus.

[0014] For example, the first network has an input data dimension of 3, an output data dimension of 2, a number of hidden layers of 9, and 120 neurons in each hidden layer; the second network has an input data dimension of 2, an output data dimension of 3, a number of hidden layers of 5, and 60 neurons in each hidden layer.

[0015] For example, the step of iteratively solving the problem based on the physical information neural network and the axial vibration displacement under dual loss constraints, and simultaneously outputting quantitative imaging results of magnetic particle concentration, tissue viscosity, and tissue shear modulus, includes: inputting spatial coordinates and discrete time series into the physical information neural network, performing forward propagation to obtain predicted data and output results under the current network parameters; calculating the total loss based on the predicted data, the output results, and the measured axial vibration displacement; determining whether a termination condition is met; if the termination condition is met, outputting the magnetic particle concentration, tissue viscosity, and tissue shear modulus calculated by the current parameter inversion network as the final output results; if the termination condition is not met, performing backpropagation based on the total loss, updating the parameters of the physical information neural network, and performing the next iteration; wherein the termination condition is: the ratio of the total loss to the initial total loss is less than a preset threshold, or the number of iterations reaches a preset maximum number of iterations.

[0016] For example, the total loss is the sum of data loss and physical equation loss, wherein the equation loss includes lateral partial differential equation loss terms and axial partial differential equation loss terms.

[0017] For example, the data loss is as follows:

[0018]

[0019] The loss of the physical equations is as follows:

[0020]

[0021]

[0022] in, Indicates data loss. This represents the loss term in the transverse partial differential equation. This represents the loss term in the axial partial differential equation. These represent the horizontal and axial directions, respectively. It is a discrete time series. The Lagrange multiplier associated with the incompressible constraint, The concentration of magnetic particles. Indicates the viscosity of the tissue. Indicates the tissue shear modulus. This indicates the predicted lateral vibration displacement. This indicates the predicted axial vibration displacement. This represents the measured axial vibration displacement. Indicates permeability, , is the magnetic susceptibility Represents the total volume. and magnetic particle concentration The relationship is: , This represents the volume of a single magnetic particle. This represents the volume fraction of the magnetic core. It is the magnetic induction intensity vector. This represents the L2 norm.

[0023] Secondly, this application provides a computer device, including: 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 above-described method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity.

[0024] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0025] The method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity in this application obtains measured axial vibration displacement through acquisition and calculation. Combined with the measured axial vibration displacement, iterative solutions are performed under the dual constraints of data loss and physical equation loss. This allows for the simultaneous output of quantitative spatial distribution maps of magnetic particle concentration, tissue viscosity, and tissue shear modulus on the same spatiotemporal reference, providing comprehensive multi-parameter quantitative data. Because the physical equation loss constraint embeds the physical laws describing tissue mechanical behavior, it overcomes the limitation of existing technologies that oversimplify tissue viscoelasticity to a constant. Guided by both physical laws and measured data, the inversion process can more realistically reflect the complex mechanical properties of tissue, thereby significantly improving the quantitative accuracy and reliability (robustness) of the final obtained magnetic particle concentration and viscoelastic parameters. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart of the method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity in the embodiments of this application.

[0028] Figure 2 This is a schematic diagram of the network model architecture in an embodiment of this application.

[0029] Figure 3 This is a comparison diagram of the output results in the embodiments of this application and related technologies. Detailed Implementation

[0030] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0031] like Figure 1 As shown in the embodiments of this application, a method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity is provided, the method comprising the following steps:

[0032] S110. An external time-varying magnetic field is formed, which acts on the magnetic particles in the medium to cause the magnetic particles to vibrate and transmit the vibration displacement to the surrounding tissue.

[0033] S120. Acquire the original ultrasound data of the magnetic particle and surrounding tissue vibration, and calculate the measured axial vibration displacement of the magnetic particle and surrounding tissue based on the original ultrasound data.

[0034] S130. Based on the physical information neural network and combined with the measured axial vibration displacement, iterative solutions are performed under the dual loss constraint condition to simultaneously output quantitative imaging results of magnetic particle concentration, tissue viscosity and tissue shear modulus; among which, the dual loss constraint condition includes data loss constraint and physical equation loss constraint.

[0035] The method in this embodiment obtains the measured axial vibration displacement through acquisition and calculation. Combined with the measured axial vibration displacement, iteratively solves the problem under the dual constraints of data loss and physical equation loss. This allows for the simultaneous output of quantitative spatial distribution maps of magnetic particle concentration, tissue viscosity, and tissue shear modulus on the same spatiotemporal reference, providing a comprehensive multi-parameter quantitative basis. Since the physical equation loss constraint embeds the physical laws describing the mechanical behavior of tissue, it overcomes the limitation of existing technologies that oversimplify tissue viscoelasticity to a constant. Guided by both physical laws and measured data, the inversion process can more realistically reflect the complex mechanical properties of tissue, thereby significantly improving the quantitative accuracy and reliability (robustness) of the final obtained magnetic particle concentration and viscoelastic parameters.

[0036] For example, the generation of the external time-varying magnetic field in step S110 can be achieved in two ways, including: generating a signal of a specified waveform through a signal generator, which is then amplified by a power amplifier to drive the coil to generate the external time-varying magnetic field; or, driving the coil to generate the external time-varying magnetic field by instantaneously discharging the energy through a capacitor storage circuit.

[0037] Regardless of the method used to generate the time-varying magnetic field, it will act on the magnetic particles in the medium (e.g., biological tissue containing magnetic particles), causing them to produce minute vibrational displacements, which are gradually transmitted to the surrounding tissue. The resulting vibrational response depends not only on the strength and frequency of the magnetic field but also on the physical properties of the medium (e.g., viscoelasticity). This application needs to consider the influence of the viscoelasticity of the medium on the vibrational response to improve the accuracy of the final results.

[0038] For example, in step S120 above, the raw ultrasound data includes amplitude information and phase information. Specifically, an ultrasound device based on multi-angle composite ultrafast plane wave imaging technology is used, and signal synchronization is performed with an external magnetic field excitation device (used to generate a time-varying magnetic field) to ensure precise synchronization between external magnetic field excitation and ultrasound acquisition. Raw ultrasound IQ data containing information on magnetic particles and vibrations of surrounding tissues is acquired, where the in-phase component (I) represents the amplitude information of the signal, and the quadrature component (Q) represents the phase information of the signal. In this embodiment, the axial vibration displacement is calculated based on a displacement tracking algorithm, combining the amplitude information and the phase information.

[0039] The acquired raw ultrasonic IQ data were subjected to an autocorrelation algorithm for axial vibration displacement tracking. The algorithm estimates the average differential displacement of the signal by calculating the average phase shift of the signal relative to the center frequency. The measured axial vibration displacement was calculated by combining amplitude and phase information, as described above, as shown in the following formula:

[0040] (1)

[0041] In the formula, To measure the axial vibration displacement, The speed of sound in the surrounding tissues, To demodulate the center frequency of the previous ultrasound radio frequency data, and These are the amplitude information and phase information, respectively. The range of axial samples represents the total number of axial samples. and This indicates two adjacent sampling times.

[0042] After calculating the measured axial vibration displacement, it is combined with discrete spatial coordinates and discrete time series to form a dataset, the data format of which is as follows: .in, These represent discrete spatial coordinates, representing the horizontal and axial directions respectively; Represents a discrete time series; This represents the measured axial vibration displacement obtained through data collection and calculation.

[0043] After obtaining the collected data, it is used as input to execute step S130 of the method. Based on the physical information neural network and combined with the measured axial vibration displacement, iterative solution is performed under the dual loss constraint condition, and quantitative imaging results of magnetic particle concentration, tissue viscosity and tissue shear modulus are output simultaneously.

[0044] For example, the physical information neural network includes a first network and a second network. Table 1 shows a parameter comparison table for the first and second networks. The first network includes a displacement prediction network, which takes spatial coordinates and discrete time series as input features and outputs intermediate variables and Lagrange multipliers related to incompressible constraints. By calculating the lateral and axial derivatives of the intermediate variables, the predicted lateral vibration displacement and the predicted axial vibration displacement are calculated. The second network includes a parameter inversion network, which takes spatial coordinates as input features and outputs magnetic particle concentration, tissue viscosity, and tissue shear modulus.

[0045]

[0046] Table 1

[0047] As shown in the table above, the first network takes spatial coordinates x and z and a discrete time series t as input, and outputs the Lagrange multiplier related to intermediate variables and incompressible constraints. The network's predicted lateral and axial vibration displacements are calculated using these intermediate variables. The second network takes spatial coordinates x and z as input and outputs magnetic particle concentration, tissue viscosity, and tissue shear modulus. The magnetic particle concentration, tissue viscosity, and tissue shear modulus are output as two-dimensional spatial distribution data, which can be used to generate quantitative imaging maps for easy observation and analysis. Based on the input discrete coordinates and the network model, image information representing multiple corresponding information is generated. The first network has an input data dimension of 3, an output data dimension of 2, 9 hidden layers, and 120 neurons per hidden layer. The second network has an input data dimension of 2, an output data dimension of 3, 5 hidden layers, and 60 neurons per hidden layer. Both the first and second networks use the same activation function and initialization method.

[0048] It should be noted that the activation function chosen in this application is the Tanh function, but other activation functions such as ReLU can also achieve similar functionality. The network optimizer chosen is Adam, but other optimizers such as SGD can also achieve similar functionality. Slight modifications to the network structure, such as adding or removing a layer, or varying the number of neurons, such as adding 10 or removing 5, can also achieve similar functionality. Regarding the network structure, this application uses a traditional multilayer perceptron (MLP), where only the weights of neurons are changed during training, while the activation function remains unchanged. In some embodiments, a variable activation function strategy is employed, specifically a B-spline function, where the parameters can be updated during training. This ensures that the activation function corresponding to each neuron continuously changes during training, enabling the solution of partial differential equations.

[0049] like Figure 2 The diagram shown is a schematic of the network model architecture in an embodiment of this application. The following is a combination of... Figure 2 The network model in the embodiments of this application will be described in detail:

[0050] The system consists of two parallel neural networks: the first network is a displacement prediction network, and the second network is a parameter inversion network. The two networks are tightly coupled through shared physical equations and a unified loss function, forming a synergistic whole. Figure 2 The diagram illustrates the direction of data flow and how physical information is injected into the system as a constraint.

[0051] For the first network, its inputs are spatial coordinates (x, z) and a discrete time series t, and its output is intermediate variables. Through the By taking the partial derivatives with respect to z and x respectively, the predicted lateral vibration displacement u can be generated. x And predicting axial vibration displacement u z This ensures the generation of the displacement field (u). x u z Under the condition of incompressibility:

[0052] (2)

[0053] because and Then we have:

[0054] (3)

[0055] The output of the first network also includes the Lagrange multiplier associated with the incompressibility constraint. This is the hydrostatic pressure field introduced by the incompressible constraint. Since biological tissues are mainly composed of water and structural materials, water has extremely low compressibility, and the rigidity of the tissue results in very small volume changes; therefore, it can be considered incompressible. It directly participates in the calculation of subsequent physical equations as part of the stress. The core function of the first network is to infer a complete vibration displacement field that conforms to physical laws (incompressible) based on the input spatiotemporal coordinates, including the axial vibration displacement u that can be obtained through measurement and calculation. z (The actual axial vibration displacement has been obtained through observation and calculation in the previous text) and the lateral vibration displacement u, which is difficult to measure directly. x .

[0056] For the second network, the input is spatial coordinates (x, z), because it is assumed that the physical parameters to be determined (magnetic particle concentration, tissue viscosity, tissue shear modulus) are static fields that vary with spatial location but not with time. Its output is the magnetic particle concentration. Tissue viscosity Tissue shear modulus The second network serves to establish a mapping from spatial location to the inherent physical properties and particle distribution of the tissue. It is the direct source for ultimately obtaining quantitative imaging results.

[0057] Figure 2 The paper also demonstrates how the loss term of the physical equation is constructed, mainly including the following process: combining the outputs of the two networks, i.e., the prediction of network one... u x 、u z 、 , Magnetic particle concentration predicted by the second network Tissue viscosity Tissue shear modulus These physical quantities are then substituted into a pre-defined momentum-conserving wave equation based on the Kelvin-Voigt viscoelastic model.

[0058] The residuals on both sides of the wave equation are calculated to force the network to approach zero during training, thus obeying the laws of physics. Data loss is calculated by comparing the axial displacement u predicted by the first network. z The axial displacement obtained by actual measurement and calculation using ultrasound The differences between the data are used to construct the model (e.g., using the L2 norm). This ensures that the network's predictions do not deviate from the actual measurement data. The ultimate goal of system optimization is to minimize the total loss. The total loss function incorporates both data-driven and physics-model-driven constraints, ensuring that the final solved parameters conform to both the measured data and fundamental physical laws.

[0059] For example, based on the Kelvin-Voigt viscoelastic model and combining the above equations (2)-(3), the component form of the solid mechanics wave equation under the incompressible condition can be expressed by the following equations:

[0060] (4)

[0061] (5)

[0062] Among them, when the permeability is In a non-magnetic medium, the total volume is The magnetic force vector on superparamagnetic iron oxide can be expressed as:

[0063] (6)

[0064] The components in the x and z directions can be represented as:

[0065] (7)

[0066] (8)

[0067] in, It is magnetic susceptibility. It is the volume fraction of the magnetic core. It is the magnetic induction intensity vector. In the embodiments of this application, the magnetic force on the magnetic particle is mainly concentrated along its axis ( (direction), while horizontal ( Components of direction It is approximately zero. Therefore, the embodiments of this application only consider the axial magnetic force component. .

[0068] For example, in step S130 above, based on the physical information neural network and combined with axial vibration displacement, iterative solutions are performed under dual loss constraints to simultaneously output quantitative imaging results of magnetic particle concentration, tissue viscosity, and tissue shear modulus, including:

[0069] S131. Input the spatial coordinates and discrete time series into the physical information neural network, perform forward propagation, and obtain the prediction data and output results under the current network parameters.

[0070] S132. Calculate the total loss based on the predicted data, output results, and measured axial vibration displacement.

[0071] S133. Determine if the termination condition is met. If the termination condition is met, output the magnetic particle concentration, tissue viscosity, and tissue shear modulus calculated by the current parameter inversion network as the final output results. If the termination condition is not met, backpropagate based on the total loss, update the parameters of the physical information neural network, and proceed to the next iteration. The termination condition is: the ratio of the total loss to the initial total loss is less than a preset threshold, or the number of iterations reaches the preset maximum number of iterations.

[0072] The total loss is the sum of the data loss and the physical equation loss, where the physical equation loss includes the loss terms of the transverse partial differential equation and the axial partial differential equation; the data loss is the L2 norm of the difference between the predicted axial vibration displacement and the axial vibration displacement.

[0073] Combining equations (4), (5), and (8) from the previous text, the loss term is identified as follows:

[0074] (9)

[0075] (10)

[0076] (11)

[0077] in, Represents data loss. This represents the loss term in the transverse partial differential equation. This represents the loss term in the axial partial differential equation. These represent the horizontal and axial directions, respectively. It is a discrete time series. The Lagrange multiplier associated with the incompressible constraint, The concentration of magnetic particles. Indicates the viscosity of the tissue. Indicates the tissue shear modulus. This indicates the predicted lateral vibration displacement. To predict axial vibration displacement, This indicates the measured axial vibration displacement obtained from the calculations mentioned above, which can be considered the actual value of the axial vibration displacement. Indicates permeability, , is the magnetic susceptibility Represents the total volume, and the total volume and magnetic particle concentration The relationship is: , This represents the volume of a single magnetic particle. This represents the volume fraction of the magnetic core. It is the magnetic induction intensity vector. This represents the L2 norm.

[0078] In this embodiment, a loss threshold is preset based on the actual usage scenario, and the total loss value is calculated as follows: Confirm the initial total loss. and preset threshold .

[0079] If satisfied and If so, proceed to the next round, and repeat the above process continuously. If the condition is met... or The magnetic particle concentration, tissue viscosity, and tissue shear modulus output in this round will then be used as the final output. N Indicates the number of iterations. N 0 This represents the preset maximum number of iterations. During the iteration process described above, based on the dual constraints of the acquired data and the physical equations, the magnetic particle concentration, tissue viscosity, and tissue shear modulus output by the second network gradually converge from initially random and meaningless values ​​to the true magnetic particle concentration, tissue viscosity, and tissue shear modulus that meet the conditions. Since the second network outputs corresponding magnetic particle concentration, tissue viscosity, and tissue shear modulus values ​​for each discrete two-dimensional spatial coordinate input, simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity is achieved.

[0080] like Figure 3 The figure shows a comparison between the output results of this application's embodiments and related technologies. As can be seen from the figure, this application performs simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity under the same spatiotemporal reference, taking viscoelasticity into account instead of treating it as a constant as in related technologies. The resulting results have better accuracy and robustness.

[0081] It should be noted that the method proposed in this invention is a general framework applicable to various viscoelastic models (such as the Kelvin-Voigt model, Maxwell model, etc.) and different types of magnetic particles (such as superparamagnetic iron oxide, ferrite magnetic nanoparticles, etc.). The network structure diagram shown in this paper, using the Kelvin-Voigt model and superparamagnetic iron oxide particles as examples, illustrates the application of this method under specific conditions. This case is only one example, intended to illustrate the implementation of this method, and does not represent all its application scenarios. This method can be extended to other viscoelastic models and particle systems.

[0082] This application also provides a computer device, including: 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 above-described method.

[0083] 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, data stored, data displayed, 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 the relevant data must comply with relevant regulations.

[0084] The method in this embodiment of the application, by inputting the collected axial vibration displacement into a physical information neural network and iteratively solving it under the dual constraints of data loss and physical equation loss, can synchronously output quantitative spatial distribution maps of magnetic particle concentration, tissue viscosity, and tissue shear modulus on the same spatiotemporal reference, providing a comprehensive multi-parameter quantitative basis. Since the physical equation loss constraint embeds the physical laws describing the mechanical behavior of tissue, it overcomes the limitation of the prior art that oversimplifies tissue viscoelasticity to a constant. Guided by the physical laws and measured data, the inversion process can more realistically reflect the complex mechanical properties of tissue, thereby significantly improving the quantitative accuracy and reliability (robustness) of the final obtained magnetic particle concentration and viscoelastic parameters.

[0085] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0086] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.

[0087] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity, characterized in that, The method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity includes: An external time-varying magnetic field is formed, which acts on magnetic particles in a medium, causing the magnetic particles to vibrate and transmit the vibration displacement to the surrounding tissue; the formation of the external time-varying magnetic field includes: generating a signal of a specified waveform through a signal generator, and driving a coil to generate the external time-varying magnetic field after the signal is amplified by a power amplifier; or, driving a coil to generate the external time-varying magnetic field by instantaneous discharge of a capacitor energy storage circuit. Acquire raw ultrasound data of the magnetic particle and the surrounding tissue during vibration, the raw ultrasound data including amplitude information and phase information; calculate the measured axial vibration displacement of the magnetic particle and the surrounding tissue based on the raw ultrasound data, specifically including: calculating the measured axial vibration displacement by combining the amplitude information and the phase information according to a displacement tracking algorithm; Based on the physical information neural network and combined with the measured axial vibration displacement, iterative solutions are performed under dual loss constraints to simultaneously output quantitative imaging results of magnetic particle concentration, tissue viscosity, and tissue shear modulus; wherein, the dual loss constraints include data loss constraints and physical equation loss constraints.

2. The method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity according to claim 1, characterized in that, The measured axial vibration displacement is calculated by combining the amplitude information and the phase information, as shown in the following formula: ； In the formula, To measure the axial vibration displacement, The speed of sound in the surrounding tissues, To demodulate the center frequency of the previous ultrasound radio frequency data, and These are the amplitude information and phase information, respectively. For axial sample range, and This indicates two adjacent sampling times.

3. The method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity according to claim 1, characterized in that, The physical information neural network includes a first network and a second network. The first network includes a displacement prediction network, which takes spatial coordinates and discrete time series as input features and outputs intermediate variables and Lagrange multipliers related to incompressible constraints. By taking the lateral and axial derivatives of the intermediate variables respectively, the predicted lateral vibration displacement and the predicted axial vibration displacement are calculated as prediction data. The second network includes a parameter inversion network, which takes spatial coordinates as input features and magnetic particle concentration, tissue viscosity, and tissue shear modulus as output results.

4. The method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity according to claim 3, characterized in that, The first network has an input data dimension of 3, an output data dimension of 2, 9 hidden layers, and 120 neurons in each hidden layer; the second network has an input data dimension of 2, an output data dimension of 3, 5 hidden layers, and 60 neurons in each hidden layer.

5. The method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity according to claim 3, characterized in that, The method, based on a physical information neural network and combined with the measured axial vibration displacement, iteratively solves the problem under dual loss constraints, simultaneously outputting quantitative imaging results of magnetic particle concentration, tissue viscosity, and tissue shear modulus, including: Spatial coordinates and discrete time series are input into the physical information neural network and forward propagation is performed to obtain the prediction data and output results under the current network parameters. The total loss is calculated based on the predicted data, the output results, and the measured axial vibration displacement. Determine whether the termination condition is met. If the termination condition is met, output the magnetic particle concentration, tissue viscosity, and tissue shear modulus calculated by the current parameter inversion network as the final output result. If the termination condition is not met, backpropagation is performed based on the total loss to update the parameters of the physical information neural network and proceed to the next iteration. The termination condition is: the ratio of the total loss to the initial total loss is less than a preset threshold, or the number of iterations reaches the preset maximum number of iterations.

6. The method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity according to claim 5, characterized in that, The total loss is the sum of data loss and physical equation loss, wherein the physical equation loss includes lateral partial differential equation loss terms and axial partial differential equation loss terms.

7. The method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity according to claim 6, characterized in that, The data loss is as follows: ； The loss of the physical equations is as follows: ； ； in, Indicates data loss. This represents the loss term in the transverse partial differential equation. This represents the loss term in the axial partial differential equation. These represent the horizontal and axial directions, respectively. It is a discrete time series. The Lagrange multiplier associated with the incompressible constraint, The concentration of magnetic particles. Indicates the viscosity of the tissue. Indicates the tissue shear modulus. This indicates the predicted lateral vibration displacement. This indicates the predicted axial vibration displacement. This represents the measured axial vibration displacement. Indicates permeability, , is the magnetic susceptibility Represents the total volume. and magnetic particle concentration The relationship is: , This represents the volume of a single magnetic particle. This represents the volume fraction of the magnetic core. It is the magnetic induction intensity vector. This represents the L2 norm.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the method for simultaneous quantitative imaging of magnetic particle concentration and tissue viscoelasticity as described in any one of claims 1-7.

Images