Method and system for reconstructing a true distribution of tc99m-maa

Abstract

The present application relates to a method and system for reconstructing Tc99m-MAA true distribution, which reconstructs MAA true distribution by using contrast-enhanced computed tomography (CE-CT) segmentation of liver segments, tumor segments, arterial and capillary networks. The method and system use physical-based modeling and computational fluid dynamics (CFD) to calculate the true MAA distribution in the arterial network, then use it as the input of the Monte Carlo simulation, and reconstruct the virtual MAA image through the iterative fine adjustment program to approximate the obtained MAA image, make up for the contribution of the true distribution generated by the inflow of injected particles into the liver vascular network which is not considered by the traditional method, and improve the selective internal radiotherapy (SIRT) planning, break through various limitations of the conventional method for SIRT treatment planning, and solve the key demand of improving accuracy in SIRT planning.

Description

Technical Field

[0001] This invention relates to the field of medical imaging and nuclear medicine technology, and in particular to a method and system for reconstructing the true distribution of Tc99m-MAA. Background Technology

[0002] Tc99m-MAA is a radionuclide imaging enhancer and diagnostic imaging enhancer. Y90-selective internal radiotherapy (SIRT) is a key treatment for liver malignancies. The dosage and injection parameters of Y90 used depend on the predicted Tc99m-MAA distribution obtained from preoperative Tc99m-MAA injection and post-injection scans. The obtained Tc99m-MAA distribution prediction is considered to directly reflect the Y90 microsphere distribution. That is, the Tc99m-MAA dose and injection parameters corresponding to the target Tc99m-MAA distribution are indirectly used to confirm the Y90 dose and injection parameters in the treatment plan. However, due to the following reasons, there are significant differences between the MAA distribution and the actual Y90 microsphere distribution, and these differences may affect the consistency of their distribution in the hepatic artery.

[0003] 1. Differences in physical properties between MAA and Y90 microspheres: 1.1 Differences in particle size and distribution: MAA particles have a wider particle size distribution (10-150 μm), with 90% concentrated in the 10-90 μm range and an average of about 15 μm; Y90 microspheres have a more uniform particle size, with glass microspheres at about 20-30 μm and resin microspheres at about 32±2.5 μm. Smaller MAA particles may enter the pulmonary circulation through the hepatic sinusoidal capillaries, leading to an overestimation of the pulmonary shunt rate and affecting treatment decisions.

[0004] 1.2 Differences in particle number: The number of MAA injection particles (300,000-700,000) is significantly less than that of Y90 resin microspheres (approximately 50 million) or glass microspheres (approximately 4 million); the difference in particle number may lead to different distribution patterns at the microvascular level, especially in areas of tumor vascular heterogeneity.

[0005] 1.3 Density and Rheological Properties: The density of Y90 glass microspheres (~3.6 g / cm³) is significantly higher than that of MAA (~1.1 g / cm³); the density difference may affect the inertial deposition behavior of particles in blood flow, especially in vascular bifurcation or turbulent regions.

[0006] 1.4 Shape irregularity: MAA consists of irregularly aggregated albumin particles, while Y90 microspheres are precisely manufactured spheres; the difference in shape may affect the mechanical embedding efficiency of the particles in the microvascular bed.

[0007] 2. Impact on SIRT treatment planning: 2.1. Pulmonary shunt assessment: MAA may overestimate pulmonary shunt and should be combined with clinical judgment to avoid unnecessarily excluding patients suitable for SIRT; 2.2 Tumor Dose Prediction: Due to differences in physical properties, the prediction of tumor absorbed dose by MAA has a certain degree of uncertainty. Bland-Altman analysis shows that the 95% confidence interval can reach -100% to +320%. 2.3 Prediction of normal liver dose: In comparison, MAA is more reliable in predicting the absorbed dose in normal liver tissue, with a correlation coefficient usually >0.9; 2.4 Catheter position sensitivity: Studies have shown that when the difference in catheter tip position between the simulation and treatment phases is >5 mm, the tumor dose prediction error increases significantly.

[0008] Despite the physical differences between Tc99m-MAA and Y90 microspheres, MAA SPECT / CT remains the standard tool for SIRT treatment planning. To ensure accuracy, the catheter position should be kept as consistent as possible during simulation and treatment, away from major arterial bifurcations. MAA injection should be performed using slow bolus injection to match the microsphere injection kinetics during treatment. Differences can be reduced by adjusting injection parameters, pressure, and catheter position.

[0009] 3. Differences in imaging methods between 99mTc and Y90 radionuclides: 99mTc-MAA imaging is a standard single-photon nuclear medicine imaging method, while Y90 is bremsstrahlung SPECT imaging or Y-90 PET imaging (intra-pair generation).

[0010] 4. Injection and procedural factors: Injection speed, catheter position, vasospasm, and contrast agent interference can cause artifacts in MAA distribution. Y90 microspheres are for therapeutic infusion, while MAA is for diagnostic low-dose infusion; differences in dose and flow rate can amplify distribution deviations.

[0011] These differences limit the accuracy of treatment planning and dosimetric validation. Multiple differences in physics, blood flow, operation, and tumor biology between the two methods can prevent preoperative MAA from accurately simulating the Y90 distribution, ultimately leading to inaccurate tumor dosing, damage to normal organs, increased complications, and decreased efficacy. Summary of the Invention

[0012] To address the aforementioned issues, a method and system for reconstructing the true distribution of Tc99m-MAA are proposed. This method uses contrast-enhanced computed tomography (CE-CT) to segment and reconstruct the true distribution of technetium-99m large particle albumin (Tc99m-MAA) based on physical modeling. This helps to obtain the actual Y90 distribution during treatment and compensates for the fact that traditional methods fail to consider the contribution of the true distribution generated by the inflow of injected particles into the hepatic vascular network to treatment planning. This method can be used to improve the planning of selective internal radiotherapy (SIRT).

[0013] The technical solution of the present invention is: a method for reconstructing the true distribution of Tc99m-MAA, comprising the following steps: Step S1, data acquisition and modeling: acquiring contrast-enhanced computed tomography (CE-CT) images and Tc99m-MAA / CT images, segmenting the CE-CT images to obtain liver segments, lesions and vascular network structure models; Step S2, Registration: Perform deformable registration between the structural model obtained in step S1 and the Tc99m-MAA / CT image to achieve spatial alignment; Step S3, Computational Fluid Dynamics (CFD) Modeling: Based on the registered image, computational fluid dynamics is applied to physically model the flow dynamics of MAA particles in the vascular network and calculate the virtual real MAA distribution; the physical modeling process considers the vascular resistance coefficient, particle retention probability, injection pressure model and catheter position constraint physical parameters. Step S4, Monte Carlo simulation and reconstruction: Perform a Monte Carlo simulation of photon transmission on the physical model established in step S3 to reconstruct the virtual MAA image; Step S5, Iterative fine-tuning: The virtual real MAA distribution obtained in step S3 is sent into the Monte Carlo simulation. The reconstructed virtual MAA image is compared with the Tc99m-MAA / CT image obtained in step S1. The physical parameters in step S3 are updated according to the comparison results. Steps S3 to S4 are repeated until the convergence condition is met, and the imaging physical model of the reconstructed real MAA distribution is obtained. Step S6, Treatment plan generation: Apply the imaging physical model to the Tc99m-MAA planning data to generate a real MAA distribution map, and generate treatment planning parameters based on the real MAA distribution map.

[0014] Furthermore, in step S1, segmenting the CE-CT image includes: High-resolution contrast-enhanced CT images are acquired using a multi-phase acquisition protocol, which includes the arterial phase, portal venous phase, and delayed phase. By constructing an AI segmentation network using labeled data, the network automatically segments liver segments and lesions, the arterial network of the hepatic artery, portal vein, and hepatic venous system, as well as the capillary network based on perfusion patterns and anatomical constraints.

[0015] Furthermore, in step S3, the physical modeling includes: Hemodynamics is described using the Navier-Stokes equations, and mass conservation is ensured using the continuity equation. The particle transport process is modeled using the particle transport equation, which takes into account particle concentration, diffusion coefficient, decay constant and source term; The vascular resistance characteristics are described using a vascular resistance model, which is calculated based on blood viscosity, vessel length, and vessel radius.

[0016] Furthermore, the particle retention probability is calculated using a formula that takes into account particle diameter, blood vessel diameter, flow velocity, maximum flow velocity, and calibration parameters. The injection pressure model considers initial pressure, decay rate, and steady-state pressure to reflect actual injection conditions, including pressure decay over time. The catheter position constraint is represented by conditions that take into account the nominal catheter position and the maximum permissible deviation.

[0017] Furthermore, in step S5, updating the physical parameters from step S3 includes: The gradient descent optimization method was used to update the vascular resistance coefficient, particle retention probability, injection pressure model, and catheter position constraints to minimize the difference between the calculated distribution and the obtained distribution. The gradient of the objective function with respect to all physical parameters, including the vascular resistance coefficient, particle retention probability, injection pressure model, and catheter position constraints, is calculated.

[0018] Furthermore, the gradient descent optimization method also includes: An adaptive learning rate strategy is adopted, in which the learning rate is dynamically adjusted with the number of iterations; A momentum acceleration mechanism is introduced to speed up convergence and reduce oscillations, including a momentum update rule for the velocity vector; Regularization techniques are applied to prevent overfitting and ensure the physical reasonableness of the parameters. The regularization objective function includes the norm of the difference between the parameter and the nominal parameter value and the norm of the parameter gradient.

[0019] A system for reconstructing the true distribution of Tc99m-MAA includes: The data acquisition and modeling module is used to acquire contrast-enhanced computed tomography (CE-CT) images and Tc99m-MAA / CT images, segment the CE-CT images, and obtain liver segments, lesions, and vascular network structure models. The registration module is used to perform deformable registration between the structural model and the Tc99m-MAA / CT image to achieve spatial alignment; The CFD modeling module is used to physically model the flow dynamics of MAA particles in the vascular network based on the registered image and apply computational fluid dynamics to calculate the virtual real MAA distribution. The physical modeling process considers the vascular resistance coefficient, particle retention probability, injection pressure model and catheter position constraint physical parameters. The simulation and reconstruction module is used to perform Monte Carlo simulations of photon transport on the established physical model and reconstruct virtual MAA images; The iterative fine-tuning module is used to input the calculated virtual real MAA distribution into the Monte Carlo simulation, compare the reconstructed virtual MAA image with the acquired Tc99m-MAA / CT image, update the physical parameters according to the comparison results, and repeat the modeling, simulation and reconstruction process until the convergence condition is met to obtain the imaging physical model of the reconstructed real MAA distribution. The treatment planning generation module is used to apply the imaging physical model to Tc99m-MAA planning data, generate a real MAA distribution map, and generate treatment planning parameters based on the real MAA distribution map.

[0020] Preferably, the system further includes: Graphics processing unit (GPU) is used for real-time reconstruction and analysis; Deep learning algorithm module for automatic segmentation and analysis; A physical information neural network module is used to combine anatomical and functional information; The workflow management module is used for real-time workflow management of iterative processing cycles.

[0021] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method.

[0022] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method.

[0023] The beneficial effects of this invention are as follows: The method and system for reconstructing the true distribution of Tc99m-MAA in this invention reconstruct the true distribution of MAA by utilizing contrast-enhanced computed tomography (CE-CT) segmentation of liver segments, tumor segments, arteries, and capillary networks. This method and system use physics-based modeling and computational fluid dynamics (CFD) to calculate the true MAA distribution in the arterial network, then use this as input for Monte Carlo simulation, and reconstruct a virtual MAA image through an iterative fine-tuning procedure to approximate the obtained MAA image. This overcomes various limitations of conventional methods for SIRT treatment planning and solves the key need to improve accuracy in SIRT planning. Attached Figure Description

[0024] Figure 1 This is a flowchart of the method for reconstructing the true distribution of Tc99m-MAA according to the present invention. Detailed Implementation

[0025] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0026] This invention provides a method and system for reconstructing the true distribution of MAA using CE-CT and based on physical modeling and iterative adjustment. Specifically, it includes the following steps: Step 1, Data Acquisition and Modeling: Acquire high-resolution contrast-enhanced CT images and segment them. The acquired data includes segmented arteries, tumors, 8 liver segments, and Tc99m-MAA / CT images. The segmentation is implemented by training a CE-CT segmentation and vascular network model based on the data, as follows. CE-CT segmentation and vascular network modeling: High-resolution contrast-enhanced CT images were acquired using a multi-phase acquisition protocol (arterial phase, portal venous phase, and delayed phase), and an AI segmentation network was constructed using labeled data to achieve automatic segmentation of the following: automatic segmentation of liver segments (8 segments) and lesions; arterial network of the hepatic artery, portal vein, and hepatic vein systems; and connected capillary network based on perfusion patterns and anatomical constraints.

[0027] Step 2, Registration: Perform deformable registration on the CE-CT segmented structural image obtained in Step 1 and the acquired Tc99m-MAA / CT image to ensure spatial alignment.

[0028] Step 3, CFD Modeling: Based on the image registered in Step 2, computational fluid dynamics (CFD) is applied to physically model the flow dynamics of MAA particles in the vascular network, which is used to calculate the virtual real MAA distribution; The modeling process considers the vascular resistance coefficient in different vascular segments, the particle retention probability based on vascular diameter and flow velocity, the effect of injection pressure on particle distribution, catheter position constraints, and injection parameters.

[0029] Step 4, Monte Carlo simulation and reconstruction: Perform Monte Carlo simulation of photon transmission on the physical model established in Step 3, model the transmission of photons through the tissue, and consider: photon attenuation and scattering, detector response characteristics, noise characteristics of the imaging system, spatial resolution effects, etc., in order to perform accurate imaging physical modeling; reconstruct the virtual MAA image from the simulation results.

[0030] Step 5, Iterative Fine-tuning: Input the virtual real MAA distribution calculated in Step 3 into the Monte Carlo simulation and compare it with the MAA distribution obtained from the reconstructed virtual MAA image. Update the physical parameters of Step 3 based on the comparison results, including: vascular resistance coefficient, particle retention probability, injection pressure model, and catheter position constraint. Repeat Steps 3 and 4 for repeated iterations. Repeat the iterations from Step 3 to improve the accuracy of the imaging physical model. After iterative training, the final imaging physical model that can reconstruct the real MAA distribution is obtained.

[0031] Step 6, Treatment Planning: Input the Tc99m-MAA plan during treatment into the imaging physical model trained in Step 5 to generate a real MAA distribution map; provide a treatment plan based on the real MAA distribution map.

[0032] Updated physical parameters: 1. Vascular resistance coefficient: Adjusted for regional differences between the calculated and acquired distributions; differences in vascular resistance between different liver segments are considered; modified to better match the flow patterns observed in arterial and capillary networks; updated using gradient descent optimization to minimize distribution error.

[0033] 2. Particle retention probability: Modified based on vascular characteristics, including diameter, branching pattern, and flow velocity; based on the relationship between vascular diameter and flow velocity; adjusted to reflect the actual particle behavior in the vascular network; updated using statistical analysis of regional distribution patterns.

[0034] 3. Injection pressure model: Optimized to better match the observed distribution pattern; the effect of pressure on particle distribution throughout the vascular tree was considered; updated to reflect actual injection conditions, including pressure decay over time; modified with computational fluid dynamics (CFD) simulation and updated boundary conditions.

[0035] 4. Catheter position constraints: Updated to account for potential positioning errors and uncertainties; incorporates spatial uncertainties in intrahepatic artery catheter placement; adjusts for observed distribution patterns that may indicate variations in catheter tip position; updates are performed using probabilistic modeling of catheter position within feasible anatomical constraints.

[0036] Mathematical formula for iterative fine-tuning CFD physical modeling: The system employs computational fluid dynamics (CFD) to accurately model the flow dynamics of MAA particles in a vascular network. Hemodynamics is described by the Navier-Stokes equations, which are expressed as: u / t + (u· u = - p / ρ + ν ²u + f, Where u is the velocity field, p is the pressure, ρ is the blood density, ν is the kinematic viscosity, and f represents the external force. It is the Nabla Laplace operator. Meanwhile, the continuity equation... •u = 0 ensures the conservation of mass.

[0037] The particle transport process is governed by the particle transport equation: C / t + u· C = D ²C - λC + S, Modeling is performed where C is the particle concentration, D is the diffusion coefficient, λ is the decay constant, and S is the source term.

[0038] To accurately describe vascular resistance characteristics, the system uses a vascular resistance model: R_i = (8μL_i) / (πr_i 4 ), Where μ is blood viscosity, L_i is blood vessel length, and r_i is blood vessel radius.

[0039] The particle retention probability is expressed by the formula: P_lodge = 1 - exp(-α·d_p / d_v·v / v_max), Where d_p is the particle diameter, d_v is the blood vessel diameter, v is the flow velocity, v_max is the maximum flow velocity, and α is the calibration parameter.

[0040] The effect of injection pressure on particle distribution is demonstrated by an injection pressure model: P_inj(t) = P_0·exp(-βt) + P_steady, Where P_0 is the initial pressure, β is the decay rate, and P_steady is the steady-state pressure. The catheter position constraint is represented by the condition x_c ∈ Ω = {x | ‖x - x_nominal‖ ≤ Δx_max}, where x_nominal is the nominal catheter position and Δx_max is the maximum permissible deviation.

[0041] Gradient descent optimization method for parameter updates: The system employs a gradient descent optimization method to update physical parameters, minimizing the difference between the calculated distribution and the acquired distribution. First, the gradient of the objective function J(θ) with respect to all physical parameters, including the vascular resistance coefficient, particle retention probability, injection pressure model, and catheter position constraints, is calculated. The gradient is expressed as: J(θ) = [ J / R, J / P_lodge, J / P_inj, J / x_c].

[0042] Parameter updates follow these rules: θ^(k+1) = θ^(k) - η^(k)· J(θ^(k)), Where η^(k) is the learning rate when iterating k.

[0043] To improve optimization efficiency, the system adopts an adaptive learning rate strategy, where the learning rate is dynamically adjusted with the number of iterations. The calculation formula is: η^(k) = η0·(1 - k / K_max)^γ. Where η0 is the initial learning rate, K_max is the maximum number of iterations, and γ is the decay factor.

[0044] In addition, the system introduces a momentum acceleration mechanism to speed up convergence and reduce oscillations. The momentum update rule is as follows: v^(k+1) = β·v^(k) - η^(k)· J(θ^(k)) and θ^(k+1) = θ^(k) + v^(k+1), Where β is the momentum coefficient and v is the velocity vector.

[0045] To prevent overfitting and ensure the physical reasonableness of the parameters, the system applies regularization techniques. The regularization objective function is: J_reg(θ) = J(θ) + λ1‖θ - θ_nominal‖² + λ2‖ θ‖², Where λ1 and λ2 are regularization coefficients, and θ_nominal is the nominal parameter value.

[0046] Iterative process: After updating the physical parameters, the system executes an iterative loop to continuously optimize the accuracy of the reconstructed MAA distribution. First, the system returns to the physical modeling step, recalculating the virtual-to-real MAA distribution using the optimized parameters. During this process, the system applies an updated CFD model incorporating new vascular resistance and pressure parameters, combined with particle retention probabilities adjusted based on vascular characteristics, while simultaneously applying updated catheter position constraints to the injection model. Next, the system reconstructs the virtual MAA image, performing Monte Carlo simulations using the updated physical parameters to model photon transport and reconstruct the virtual MAA image, thereby generating a new virtual MAA image reflecting the updated parameter values. This iterative loop continues until convergence is met, ensuring optimal accuracy of the reconstructed real MAA distribution.

[0047] After each iteration, convergence criteria are evaluated to determine whether to continue the fine-tuning process. Convergence criteria include reaching the maximum number of iterations, parameter changes falling below a minimum threshold, or distribution errors reaching an acceptable level.

[0048] The technical implementation system includes: Graphics processing unit (GPU) accelerates processing for real-time reconstruction and analysis; Deep learning algorithms for automatic segmentation and analysis; Monte Carlo simulations for photon transport modeling and virtual MAA image reconstruction; Physical information neural networks are used to combine anatomical and functional information; Multi-resolution registration between CE-CT and MAA-CT data; An iterative fine-tuning algorithm is used to continuously improve the accuracy of the distribution; Adaptive parameter optimization based on validation feedback; Real-time workflow management for iterative processing cycles.

[0049] The method of this invention calculates the true distribution of MAA in arteries and connecting capillary networks by using CE-CT segmentation combined with Tc99m-MAA / CT images. It adopts a comprehensive workflow including: (1) registration of CE-CT and its segmentation with Tc99m-MAA / CT images; (2) calculation of virtual real MAA distribution using physical modeling based on computational fluid dynamics (CFD); (3) Monte Carlo simulation and then reconstruction of virtual MAA images; (4) iterative fine-tuning process to update physical parameters, including vascular resistance coefficient, particle retention probability, injection pressure model and catheter position constraints; and (5) treatment planning based on the real MAA distribution map.

[0050] The iterative fine-tuning process is a key feedback mechanism that systematically adjusts physical parameters based on regional differences between the calculated and acquired distributions. This approach compensates for differences in physical properties between MAA and Y90 microspheres, including differences in injection parameters such as pressure variations and catheter positioning, and provides robustness to uncertainties related to patient-specific vascular anatomy.

[0051] By iteratively fine-tuning to ensure optimal accuracy of the reconstructed true MAA distribution, this method compensates for differences in physical properties between MAA and Y90 microspheres and corrects for differences in injection parameters, including pressure variations and catheter positioning, thereby significantly improving the correlation between the predicted and actual Y90 distributions. This feedback loop is robust to uncertainties in patient-specific vascular anatomy and can accommodate individual patient variations in vascular resistance and flow patterns. The generated true MAA distribution map is used for optimized Y90 prescription, accurate dosimetry calculations, and improved patient selection and treatment outcome prediction. Precise treatment planning enhances patient safety, reduces radiation exposure to healthy tissues, and reduces operative time with automated analysis and planning tools, ultimately enabling reliable treatment planning for the Y90-SIRT procedure with improved accuracy and patient-specific optimization.

[0052] The embodiments described above merely illustrate specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method of reconstructing a Tc99m-MAA true distribution, characterized by, Includes the following steps: Step S1, Data Acquisition and Modeling: Acquire contrast-enhanced computed tomography (CE-CT) images and Tc99m-MAA / CT images. Segment the CE-CT images to obtain liver segments, lesions, and vascular network structure models. Step S2, Registration: Perform deformable registration between the structure models obtained in Step S1 and the Tc99m-MAA / CT images to achieve spatial alignment. Step S3, Computational Fluid Dynamics (CFD) Modeling: Based on the registered images, apply computational fluid dynamics to physically model the flow dynamics of MAA particles in the vascular network. Calculate the virtual real MAA distribution; the physical modeling process considers vascular resistance coefficient, particle retention probability, injection pressure model, and catheter position constraint physical parameters; Step S4, Monte Carlo simulation and reconstruction: Perform a Monte Carlo simulation of photon transmission on the physical model established in step S3 to reconstruct the virtual MAA image; Step S5, Iterative fine adjustment: Input the virtual real MAA distribution calculated in step S3 into the Monte Carlo simulation, compare the reconstructed virtual MAA image with the Tc99m-MAA / CT image obtained in step S1, update the physical parameters in step S3 according to the comparison results, and repeat steps S3 to S4 until the convergence condition is met to obtain the imaging physical model of the reconstructed real MAA distribution; Step S6, Treatment planning generation: Apply the imaging physical model to the Tc99m-MAA planning data to generate a real MAA distribution map, and generate treatment planning parameters based on the real MAA distribution map.

2. The method for reconstructing the true distribution of Tc99m-MAA according to claim 1, characterized in that, In step S1, segmenting the CE-CT image includes: acquiring high-resolution contrast-enhanced CT images using a multi-phase acquisition protocol, which includes the arterial phase, portal venous phase, and delayed phase; constructing an AI segmentation network using labeled data to automatically segment liver segments and lesions, the arterial network of the hepatic artery, portal vein, and hepatic venous system, and the connected capillary network based on perfusion patterns and anatomical constraints.

3. The method for reconstructing the true distribution of Tc99m-MAA according to claim 1, characterized in that, In step S3, the physical modeling includes: describing hemodynamics using the Navier-Stokes equations and ensuring mass conservation using the continuity equations; modeling the particle transport process using the particle transport equations, which consider particle concentration, diffusion coefficient, decay constant, and source term; and describing vascular resistance characteristics using a vascular resistance model, which is calculated based on blood viscosity, vascular length, and vascular radius.

4. The method for reconstructing the true distribution of Tc99m-MAA according to claim 3, characterized in that, The particle retention probability is calculated by a formula that takes into account particle diameter, vessel diameter, flow velocity, maximum flow velocity, and calibration parameters; the injection pressure model considers initial pressure, decay rate, and steady-state pressure to reflect actual injection conditions, including pressure decay over time; the catheter position constraint is represented by conditions that consider nominal catheter position and maximum permissible deviation.

5. The method for reconstructing the true distribution of Tc99m-MAA according to claim 1, characterized in that, In step S5, updating the physical parameters in step S3 includes: updating the vascular resistance coefficient, particle retention probability, injection pressure model, and catheter position constraint using a gradient descent optimization method to minimize the difference between the calculated distribution and the acquired distribution; and calculating the gradient of the objective function with respect to all physical parameters, including the vascular resistance coefficient, particle retention probability, injection pressure model, and catheter position constraint.

6. The method for reconstructing the true distribution of Tc99m-MAA according to claim 5, characterized in that, The gradient descent optimization method further includes: adopting an adaptive learning rate strategy, with the learning rate dynamically adjusted with the number of iterations; introducing a momentum acceleration mechanism to accelerate convergence and reduce oscillations, including a momentum update rule for the velocity vector; and applying regularization techniques to prevent overfitting and ensure the physical rationality of the parameters, with the regularization objective function including the norm of the difference between the parameter and the nominal parameter value and the norm of the parameter gradient.

7. A system for reconstructing the true distribution of Tc99m-MAA, characterized in that, include: The data acquisition and modeling module is used to acquire contrast-enhanced computed tomography (CE-CT) images and Tc99m-MAA / CT images, segment the CE-CT images, and obtain liver segments, lesions, and vascular network structure models; the registration module is used to perform deformable registration between the structure models and the Tc99m-MAA / CT images to achieve spatial alignment. The CFD modeling module is used to physically model the flow dynamics of MAA particles in the vascular network based on the registered image using computational fluid dynamics, and calculate the virtual-to-real MAA distribution. The physical modeling process considers vascular resistance coefficient, particle retention probability, injection pressure model, and catheter position constraint physical parameters. The simulation and reconstruction module performs Monte Carlo simulation of photon transmission on the established physical model to reconstruct the virtual MAA image. The iterative fine-tuning module feeds the calculated virtual-to-real MAA distribution into the Monte Carlo simulation, compares the reconstructed virtual MAA image with the acquired Tc99m-MAA / CT image, updates the physical parameters based on the comparison results, and repeats the modeling, simulation, and reconstruction process until convergence conditions are met, obtaining an imaging physical model of the reconstructed real MAA distribution. The treatment planning generation module applies the imaging physical model to Tc99m-MAA planning data to generate a real MAA distribution map, and generates treatment planning parameters based on the real MAA distribution map.

8. The system for reconstructing the true distribution of Tc99m-MAA according to claim 7, characterized in that, The system also includes: a graphics processing unit (GPU) for real-time reconstruction and analysis; a deep learning algorithm module for automatic segmentation and analysis; a physical information neural network module for combining anatomical and functional information; and a workflow management module for real-time workflow management of iterative processing cycles.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 6.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 6.

Images