A dynamic parameter estimation method based on artificial immune network and adaptive particle swarm optimization

Abstract

The application discloses a kinetic parameter estimation method based on artificial immune network and adaptive particle swarm optimization, and belongs to the technical field of biomedical data processing. The method combines the global search capability of the artificial immune network and the local fine optimization advantage of the adaptive particle swarm optimization, and ensures the stability and accuracy of parameter estimation through a multi-round verification mechanism. Specifically, the method comprises the following steps: data preprocessing, parameter boundary and initial value setting, global search based on the artificial immune network, local optimization based on the adaptive particle swarm optimization, multiple seed verification and statistical analysis, and the like. The application realizes adaptive optimization of the particle swarm optimization through linear decay inertia weight and dynamic adjustment of a learning factor, introduces a parameter standard deviation verification mechanism, effectively improves the precision and reliability of kinetic model parameter estimation, and is suitable for kinetic analysis of PET and other medical image data.

Description

Technical Field

[0001] This invention relates to the field of medical image analysis and parameter optimization technology, specifically to a dynamic parameter estimation method based on artificial immune network (AIN) and adaptive particle swarm optimization (APSO), belonging to the interdisciplinary field of machine learning and physiological dynamic parameter modeling and optimization. Background Technology

[0002] Hepatocellular carcinoma (HCC) is the most common type of liver cancer, accounting for approximately 90% of all liver cancer cases. Although malignant and benign tumors can be distinguished using traditional imaging techniques such as computed tomography (CT) and magnetic resonance imaging (MRI), HCC remains challenging in clinicopathological diagnosis and metabolic function analysis. Dynamic positron emission tomography / computed tomography (PET) technology, by acquiring image sequences at multiple time points and reconstructing the time-radioactivity curve (TAC) of in vivo radiotracer distribution over time, provides a more precise and dynamic imaging basis for quantitative analysis of HCC lesions and efficacy assessment.

[0003] In recent years, researchers have widely adopted optimization algorithms for pharmacokinetic modeling and parameter estimation of PET / CT data to reveal tissue metabolism and tracer transport characteristics. However, traditional parameter estimation methods often rely on single optimization strategies such as gradient descent or genetic search, focusing only on finding the local optimum with the minimum residual. They lack the ability to explore the global parameter space in high dimensions and are sensitive to initial value settings, easily getting trapped in local convergence, thus causing the estimation results to deviate from the true physiological meaning.

[0004] To overcome the aforementioned shortcomings, this invention proposes a kinetic parameter estimation method based on artificial immune networks and adaptive particle swarm optimization. This method combines the global search advantages of artificial immune algorithms with the local refinement capabilities of adaptive particle swarm optimization. It introduces dynamic learning factors and nonlinear inertial weight adjustment mechanisms during parameter estimation, achieving adaptive balance of the parameter search range and parallel evolution of the multiple solution space. This hybrid optimization strategy significantly improves the stability, accuracy, and physiological reliability of parameter estimation in PET / CT pharmacokinetic models, providing more reliable technical support for quantitative imaging analysis and personalized treatment evaluation of liver cancer. Currently, no research has combined artificial immune networks and particle swarm optimization in the field of pharmacokinetics and further improved upon this approach. Summary of the Invention

[0005] This invention provides a kinetic parameter estimation method based on artificial immune networks and adaptive particle swarm optimization for joint estimation of multiple parameters in the process of liver PET / CT kinetic modeling. Through this hybrid optimization strategy, the stability, accuracy and physiological reliability of the parameter estimation of PET / CT pharmacokinetic models can be significantly improved.

[0006] This invention combines the global search capability of artificial immune algorithms with the local fine-tuning optimization characteristics of adaptive particle swarm optimization (PSO). It adaptively adjusts the optimization strategy based on the search weights at different stages, enabling multi-scale exploration and local convergence control of the solution space. During optimization, an antibody population evolution mechanism and a linear decay strategy for particle swarm inertia weights are introduced to achieve a dynamic balance between parameter search direction and step size, effectively detecting and correcting parameter combinations that do not conform to physiological rationality. Through this hybrid optimization framework, this invention significantly improves the algorithm's global optimization capability and the stability of parameter estimation, obtaining pharmacokinetic parameter estimation results that are more consistent with physiological characteristics and have higher accuracy. The obtained parameters can effectively reflect the differences in metabolic kinetics between hepatocellular carcinoma and normal liver tissue, thus providing reliable technical support for the early diagnosis, quantitative analysis of lesions, and personalized treatment evaluation of liver cancer.

[0007] The technical solution of this invention is: a dynamic parameter estimation method based on artificial immune networks and adaptive particle swarm optimization, the specific steps of which are as follows:

[0008] Step 1: Based on the dual-input multi-compartment dynamics model, the tissue metabolic process is mathematically modeled, and combined with real medical imaging time-radioactivity curve data, an optimization objective function with the residual sum of squares as the core is constructed.

[0009] Step 2: Set initial parameter values ​​and physiologically reasonable boundaries to initialize the antibody population of the artificial immune network. Define hyperparameters such as antibody quantity, cloning rate, mutation rate, and inhibition threshold to provide a foundation for the global search of the artificial immune algorithm;

[0010] Step 3: An artificial immune algorithm is used to perform a global search for the kinetic parameters. Through clonal amplification, mutation evolution, and suppression screening mechanisms, a high-affinity candidate parameter set is generated to obtain the optimal candidate solution globally. This stage ensures that the algorithm can effectively escape local optima traps and obtain a representative group of candidate parameters.

[0011] Step 4: Using the candidate parameters output by the artificial immune algorithm as the initial particle swarm, an improved adaptive particle swarm optimization algorithm is used for local search. A balanced search is achieved through linearly decaying inertia weights and dynamically adjusted learning factors, combined with an early stopping mechanism to improve convergence speed and stability, resulting in more accurate parameter estimation results.

[0012] Step 5: Through a multi-random seed repeated mixing optimization process, the mean, standard deviation, and global optimal validation index of the parameters are calculated to ensure the stability and reliability of the results. Subsequently, goodness-of-fit calculations such as RMSE, AIC, and BIC are performed, and the independent samples t-test is used to analyze the significance of differences between normal tissues and tumor tissues. Visualized fitting curves and statistical result tables are output to achieve the quantification and verifiability of parameter estimation.

[0013] Furthermore, in Step 1, the objective function is defined by minimizing the root mean square error (RMSE) between the TAC value predicted by the computational model and the measured TAC value.

[0014] Furthermore, in Step 2, the antibody population of the artificial immune network is initialized, and patient data including time series, arterial and venous radioactivity concentration curves are loaded; the parameter boundary range is set according to the physiological characteristics of the tissue dynamics model, and the median of the boundary interval is used as a unified initial parameter; the parameter range is dynamically constrained through the boundary correction function to ensure the rationality of the search space; at the same time, the hyperparameters of the artificial immune algorithm are set, including cloning rate, mutation rate, inhibition threshold, similarity threshold, and maximum number of generations, to control the diversity and convergence speed of the antibody population, thereby providing an initial basis for parameter optimization for subsequent global search.

[0015] Furthermore, in Step 3, an artificial immune algorithm is used to perform a global search, generating a set of high-affinity candidate parameters through clonal amplification, mutation update, and inhibition screening mechanisms; the affinity evaluation function is used to select and evolve individual antibodies, realizing a global exploration of a multi-peak complex target space; this step can effectively avoid the algorithm getting stuck in local optima and improve the global convergence performance of the kinetic parameter search.

[0016] Furthermore, in Step 4, the candidate parameters output by the artificial immune algorithm are used as the initial particle swarm, and an adaptive particle swarm optimization algorithm is used for local fine-grained search. This algorithm achieves a balance between exploration and utilization by linearly decaying inertial weights and dynamically adjusting learning factors (c1, c2), and combines early stopping criteria to improve the convergence speed and stability of the algorithm, thereby obtaining high-precision tissue dynamics parameter estimation results.

[0017] Furthermore, in Step 4, the adaptive particle swarm optimization algorithm achieves a balance between global exploration and local utilization by linearly decaying inertia weights and dynamically adjusting learning factors (c1, c2). The inertia weights gradually decrease linearly from a large initial value during iteration to enhance the algorithm's early global search capability and improve local search accuracy in the later convergence phase. The learning factors c1 and c2 correspond to individual particle experience and swarm experience, respectively, and their values ​​are dynamically adjusted with iteration to strengthen individual exploration in the early search phase and enhance swarm aggregation in the convergence phase, thereby achieving adaptive adjustment of search direction and speed.

[0018] Furthermore, in Step 5, the mixed optimization process is repeatedly executed with multiple random seeds to calculate the mean, standard deviation, and improvement rate of the parameters in each optimization result, so as to verify the stability and global optimality of the parameter estimation results; further, the goodness-of-fit indices, including RMSE, AIC, and BIC, are calculated, and the differences in parameters between normal tissue and tumor tissue are statistically analyzed using an independent samples t-test; finally, the fitting curve and statistical results table are output to realize the visualization and quantitative verification of the kinetic parameter estimation.

[0019] The beneficial effects of this invention are:

[0020] 1. This invention utilizes a reversible dual-input three-compartment model to model the metabolic activity of liver tissue and combines it with real TAC data to determine the optimization target;

[0021] 2. This invention combines the global search mechanism of the artificial immune algorithm with the local fine-tuning search mechanism of the adaptive particle swarm optimization algorithm, thereby achieving a dynamic balance between global exploration and local convergence in parameter search. This effectively avoids the problem of traditional optimization algorithms easily getting trapped in local optima and significantly improves the global optimization capability.

[0022] 3. In the optimization process, this invention introduces a linear decay mechanism for inertial weights and a dynamic adjustment mechanism for learning factors (c1, c2). The search intensity and direction are automatically adjusted according to the iteration stage, enabling the algorithm to have strong global exploration capabilities in the early stage and achieve high-precision local fitting in the later stage. At the same time, combined with boundary constraints and physiological rationality detection mechanisms, unreasonable parameters are dynamically corrected and filtered, thereby ensuring that the obtained parameters are numerically stable and conform to tissue dynamics characteristics, improving the physiological reliability and stability of the fitting.

[0023] 4. The parameter results obtained by the proposed hybrid optimization framework can accurately reflect the differences in pharmacokinetic characteristics between hepatocellular carcinoma and background liver tissue, and significantly improve the quantitative discrimination ability of PET / CT images in early diagnosis, grading assessment and efficacy monitoring of liver cancer. Attached Figure Description

[0024] Figure 1This is a flowchart illustrating the present invention;

[0025] Figure 2 This is the reversible dual-input three-compartment model of the present invention. Detailed Implementation

[0026] Example 1: A method for estimating PET / CT dynamic parameters based on artificial immune networks and adaptive particle swarm optimization. In the implementation of this invention, the experimental data came from real PET images of 23 patients with hepatocellular carcinoma (HCC). Among them, 21 patients had a single tumor (including 1 follow-up case), 1 patient had 2 tumors, and another patient had 3 tumors. The diameter of the included tumors ranged from 1.6 cm to 17.0 cm, with an average diameter of approximately 6.75 cm.

[0027] In the implementation of this invention, the Ordered Subset Expectation-Maximization (OSEM) algorithm was used to reconstruct PET images. After the subjects received intravenous injection of the radiotracer 18F-FDG, the dynamic acquisition data for the first minute was divided and reconstructed into 12 frames, with a 5-second time interval between each frame; the data for the following 4 minutes was reconstructed into 4 frames, with a 60-second time interval between each frame; in addition, to more accurately analyze the metabolic kinetics of 18F-FDG in vivo, a static PET image of the liver at the 60-minute time point was selected for supplementary analysis. A total of 17 dynamic PET images were obtained. Time-series fusion and CT registration were performed on each frame to ensure spatial consistency and quantitative accuracy between PET and CT data. After all subjects completed PET / CT scans, experienced radiologists manually delineated regions of interest (ROIs) on the registered images, strictly avoiding large intrahepatic vessels and other high-radioactive areas during the delineation process to reduce interference from the standard uptake value (SUV) measurement. The maximum standard uptake value (SUV_max) of each frame was extracted from the obtained ROI regions, and a time-radioactivity curve (TAC) was constructed for subsequent kinetic modeling and parameter estimation analysis.

[0028] like Figure 2 As shown, this invention employs a reversible dual-input three-compartment dynamic model to simulate radioactive tracers. 18 The distribution and metabolic processes of F-FDG in liver tissue. This model can comprehensively reflect the dual input characteristics of the hepatic artery and portal vein, thus more accurately describing the dynamics of blood perfusion and glucose metabolism in liver tissue. Wherein f a C represents the proportion of hepatic arterial blood flow in total hepatic blood flow. a (t) and C v (t) represents the concentration of hepatic artery and portal vein blood, respectively. 18 C is a function of F-FDG concentration over time. i (t) represents the total blood input.18 The concentration function of F-FDG is obtained by proportionally weighted summation of the arterial and portal venous input functions:

[0029] (1)

[0030] In the organizational section, C f (t) represents the free state in liver tissue. 18 The relationship between F-FDG concentration and time, C p (t) represents the product formed after phosphorylation. 18 Relationship between F-FDG-6P concentration and time

[0031] The rate constant in the model is defined as follows:

[0032] K1 (ml·min) -1 ·ml -1 ):express 18 The transport rate constant of F-FDG from blood to liver tissue;

[0033] k2: indicates 18 The rate constant of F-FDG returning to the blood from liver tissue;

[0034] k3: Indicates that hexokinase in hepatocytes will... 18 F-FDG phosphorylation to 18 The rate constant of F-FDG-6P;

[0035] k4: indicates 18 F-FDG-6P is reconverted under the action of phosphatase to 18 The rate constant of F-FDG.

[0036] according to Figure 2 The reversible two-input three-compartment model shown can be used to derive the differential equation: (2) (3)

[0037] By solving this system of equations, the output function C, which describes the change of tracer concentration in the tissue over time, can be obtained. T (t), whose expression is:

[0038] (4)

[0039] Where vb is the blood volume fraction; parameters α1 and α2 are model eigenvalues ​​used to characterize the kinetic changes of tracers under different metabolic pathways;

[0040] (5) (6)

[0041] The parameter estimation of liver PET activity curve (TAC) data is based on the above-mentioned reversible two-input three-compartment model, and is performed by fitting C. T The difference between (t) and the measured PET curve is used to obtain the optimal set of kinetic parameters for subsequent metabolic analysis and tissue differentiation studies.

[0042] A dynamic parameter estimation method based on artificial immune networks and adaptive particle swarm optimization is provided, the specific steps of which are as follows:

[0043] Step 1: A reversible two-input three-compartment kinetic model was used to model the radiotracer metabolism process in liver tissue. Combined with real-time-radioactivity curves (TACs) extracted from PET / CT images, an objective function for parameter optimization was constructed. The objective function was optimized by minimizing the root mean square error (RMSE) between the model-predicted TAC values ​​and the measured TAC values.

[0044] Specifically, in Step 1, the objective function is defined as:

[0045] (7)

[0046] in:

[0047] θ=[K1,k2,k3,k4,f a ,v b [ ] represents the set of all dynamic parameters to be estimated;

[0048] C meas (t i () represents the measured TAC value extracted from n=17 frames of PET / CT images;

[0049] C model (t i ,θ) represents the theoretical TAC value calculated based on the above reversible two-input three-compartment model.

[0050] By minimizing the objective function, the optimal estimation of the model parameters can be achieved, providing an initial fitting basis for the subsequent hybrid optimization process.

[0051] Step 2: Initialize the antibody population of the artificial immune network. During the initialization phase of the artificial immune network, set the population size and hyperparameters, including: antibody quantity, cloning rate, mutation rate, similarity inhibition threshold, and inhibition strength factor.

[0052] The fitness function of the artificial immune network is evaluated based on the residual of the objective function. High-affinity antibodies are generated through clonal amplification operators, and Gaussian perturbation mutations are performed on them.

[0053] (8)

[0054] Where ℇ is the coefficient of variation, N(0, σ) 2 () has a mean of 0 and a variance of σ. 2 The normal distribution is randomly perturbed. High-affinity antibodies are retained in subsequent iterations, while low-affinity antibodies are suppressed or replaced, thus maintaining population diversity.

[0055] Finally, after the above initialization and antibody evolution operations, the artificial immune network can form a high-quality candidate solution set covering the global parameter space, providing a stable initial input for the subsequent adaptive particle swarm local fine-grained search, and improving the overall convergence and accuracy of the algorithm.

[0056] Step 3: A global search of the kinetic parameter space is performed using an artificial immune network algorithm. This algorithm simulates the adaptive evolutionary mechanism of antibody antigen recognition, clonal amplification, and mutation selection in the biological immune system. By continuously updating the antibody population within the parameter space, it achieves the search for the global optimum and preliminary optimization of parameter estimation. The global search phase of the artificial immune network aims to rapidly approach the optimal parameter region while maintaining the diversity of the solution space through population co-evolution, and to evaluate and screen individual antibodies using an affinity function.

[0057] First, for each antibody individual Ab i Calculate its fitness value. The fitness function is defined as the reciprocal of the objective function error and is used to measure the goodness of fit between the model predictions and the measured TAC.

[0058] (9)

[0059] Among them, RMSE i Let represent the root mean square error corresponding to the i-th antibody parameter, and ε be a very small positive number to prevent the denominator from being zero.

[0060] Based on fitness results, the artificial immune network generates new antibodies through a clonal expansion mechanism. The number of clones is directly proportional to antibody affinity, which can be expressed as:

[0061] (10)

[0062] Where, β c N is the cloning rate constant. Ab This refers to the antibody population size. This mechanism ensures that high-affinity individuals have more opportunities to replicate, thereby accelerating the search for optimal regions.

[0063] Subsequently, mutation operations were performed on the cloned antibodies. The magnitude of the mutation was negatively correlated with antibody affinity, meaning that individuals with high affinity showed less mutation and individuals with low affinity showed more mutation.

[0064] (11)

[0065] Where η is the variability rate, and N(0,1) is the standard normal random perturbation. The normalized affinity has a value range of [0,1].

[0066] In the process of antibody population evolution, inhibition and selection mechanisms are introduced to maintain population diversity. When the similarity between two antibodies exceeds a set threshold δ... s At this time, antibodies with lower fitness are suppressed or eliminated. Similarity can be calculated using Euclidean distance or cosine distance:

[0067] (12)

[0068] If Sim(Ab) i Ab j )>δ s If the condition is met, then the suppression operation will be performed, retaining those with higher fitness.

[0069] After multiple rounds of cloning, mutation, and suppression iterations, the population structure of the artificial immune network gradually converges, forming a set of candidate parameters with high affinity and high diversity. The individuals in this candidate set represent the optimal or suboptimal dynamic parameters obtained from the global search at the current stage, providing initial input for the next step of adaptive particle swarm optimization.

[0070] Step 4: Using the high-affinity candidate parameters output by the artificial immune network as the initial particle swarm, an adaptive particle swarm optimization algorithm is employed for local refinement optimization. This algorithm achieves a balance between exploratory and convergent processes during parameter search by dynamically adjusting particle velocity and learning factor, further improving the accuracy and stability of dynamic parameter estimation.

[0071] In each iteration, the particle's position and velocity are updated according to the following formula:

[0072] (13) (14)

[0073] Where ω(t) is the inertia weight, which adopts a linear decay form:

[0074] (15)

[0075] To ensure the algorithm possesses global search capabilities in its initial stages and achieves local fine-tuning in later stages, c1 and c2 are adaptive learning factors that are dynamically adjusted with iterations to maintain the balance of the population search. An early stopping criterion is implemented during optimization; the iteration terminates prematurely when the change in the global optimum is less than a threshold for several consecutive generations, preventing overfitting and ineffective searches. Through this local optimization process, high-precision parameter estimation can be achieved based on the global candidate solutions provided by the artificial immune network, significantly improving the model's fitting effect and physiological rationality.

[0076] Step 5: Verify and evaluate the obtained parameter results through a hybrid optimization experiment with multiple groups of different random seeds. After completing the hybrid optimization of artificial immune network and adaptive particle swarm optimization, calculate the mean, standard deviation, and confidence interval of the parameter results obtained from each group of optimizations to evaluate the stability and consistency of parameter estimation.

[0077] Through the above verification and evaluation steps, the estimated parameters can be guaranteed to have high robustness and physiological rationality. The results can effectively assist in the quantitative differentiation between hepatocellular carcinoma and normal tissues, providing a basis for clinical diagnosis and treatment evaluation.

[0078] The specific parameter estimation results and experimental evaluation indicators of this method are shown in Table 1:

[0079] Table 1 Experimental results of the present invention

[0080] To verify the effectiveness of the parameter estimation method described in this invention, statistical analysis was performed on the estimation results of hepatocellular carcinoma tissue and normal liver tissue. A paired-samples t-test was used to compare the significance of the parameter means of the two independent samples. The test output was a P-value; a P-value < 0.05 indicated a significant difference between the two groups of parameters.

[0081] The results showed that all the kinetic parameters obtained in this invention were significantly different between the two types of tissues, indicating that the optimized method can effectively distinguish between hepatocellular carcinoma tissue and normal liver tissue. From a physiological mechanism perspective, 18 F–FDG–6P can be converted to F–FDG–6P via dephosphorylation. 18 F–FDG metabolism is more active in normal liver tissue, therefore the k4 value in normal tissue is usually higher than that in hepatocellular carcinoma tissue. Regarding other parameters, K1, k3, and v... b The value of f is generally lower in normal liver tissue than in hepatocellular carcinoma tissue, reflecting the enhanced glucose uptake and metabolic activity in tumor tissue. Furthermore, because the liver has dual blood supply from both the hepatic artery and portal vein, normal liver tissue primarily receives blood from the portal vein, while hepatocellular carcinoma tissue mainly relies on the hepatic artery. Therefore, in the parameter estimation results, f... aIt is significantly higher in hepatocellular carcinoma tissue than in normal liver tissue, which can effectively characterize the differences in hemodynamics between the two tissues.

[0082] All parameter estimation results in this invention accurately reflect physiological characteristics while maintaining low RMSE, and significantly distinguish hepatocellular carcinoma tissue from normal liver tissue. This method provides reliable data support for the assessment, grading, and prognostic analysis of hepatocellular tumor characteristics and after resection.

[0083] The specific embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for estimating dynamic parameters based on artificial immune networks and adaptive particle swarm optimization, characterized in that, The specific steps of the method are as follows: Step 1: Based on the dual-input multi-compartment dynamics model, the tissue metabolic process is mathematically modeled, and combined with real medical imaging time-radioactivity curve data, an optimization objective function with the residual sum of squares as the core is constructed. Step 2: Set initial parameter values ​​and physiologically reasonable boundaries to initialize the antibody population of the artificial immune network; Step 3: Use an artificial immune algorithm to perform a global search for kinetic parameters. Generate a set of high-affinity candidate parameters through clonal amplification, mutation evolution and suppression screening mechanisms to obtain the optimal candidate solution in the global scope. Step 4: The candidate parameters output by the artificial immune algorithm are used as the initial particle swarm. An improved adaptive particle swarm optimization algorithm is used for local search. A balanced search is achieved by using linearly decaying inertial weights and dynamically adjusted learning factors. The early stopping mechanism is combined to improve the convergence speed and stability, and obtain more accurate parameter estimation results. Step 5: Through a multi-random seed repeated mixing optimization process, calculate the mean, standard deviation, and global optimal validation index of the parameters; then calculate the goodness of fit of RMSE, AIC, and BIC, and use independent samples t-test to analyze the significance of differences between normal tissues and tumor tissues, outputting a visual fitting curve and statistical results table to achieve the quantification and verifiability of parameter estimation.

2. The PET / CT dynamic parameter estimation method based on artificial immune algorithm and adaptive particle swarm optimization according to claim 1, characterized in that: In Step 1, the objective function is defined by minimizing the root mean square error (RMSE) between the TAC value predicted by the computational model and the measured TAC value.

3. The PET / CT dynamic parameter estimation method based on artificial immune algorithm and adaptive particle swarm optimization according to claim 1, characterized in that: In Step 2, the antibody population of the artificial immune network is initialized, and patient data including time series and arterial and venous radioactivity concentration curves are loaded. Based on the physiological characteristics of the tissue dynamics model, the parameter boundary range is set, and the midpoint of the boundary interval is used as the unified initial parameter; the parameter range is dynamically constrained through the boundary correction function to ensure the rationality of the search space. Simultaneously, hyperparameters of the artificial immune algorithm are set, including cloning rate, mutation rate, inhibition threshold, similarity threshold, and maximum number of generations, to control the diversity and convergence speed of the antibody population, thereby providing an initial basis for parameter optimization for subsequent global search.

4. The dynamic parameter estimation method based on artificial immune network and adaptive particle swarm optimization according to claim 1, characterized in that: In Step 3, an artificial immune algorithm is used to perform a global search. A set of high-affinity candidate parameters is generated through clonal amplification, mutation update and inhibition screening mechanisms. The affinity evaluation function is used to select and evolve individual antibodies, thereby realizing the global exploration of a multi-peak complex target space.

5. The dynamic parameter estimation method based on artificial immune network and adaptive particle swarm optimization according to claim 1, characterized in that: In Step 4, the candidate parameters output by the artificial immune algorithm are used as the initial particle swarm, and an adaptive particle swarm optimization algorithm is used for local fine-grained search. This algorithm achieves a balance between exploration and utilization by linearly decaying inertial weights and dynamically adjusting learning factors c1 and c2, and improves the convergence speed and stability of the algorithm by combining early stopping criteria, thereby obtaining high-precision tissue dynamics parameter estimation results.

6. The dynamic parameter estimation method based on artificial immune network and adaptive particle swarm optimization according to claim 1, characterized in that: In Step 4, the adaptive particle swarm optimization algorithm achieves a balance between global exploration and local utilization by linearly decaying inertia weights and dynamically adjusting learning factors c1 and c2. The inertia weights gradually decrease linearly from a large initial value during iteration to enhance the algorithm's early global search capability and improve local search accuracy in the later convergence phase. The learning factors c1 and c2 correspond to individual particle experience and swarm experience, respectively, and their values ​​are dynamically adjusted with iteration to strengthen individual exploration in the early search phase and enhance swarm aggregation in the convergence phase, thereby achieving adaptive adjustment of search direction and speed.

7. The dynamic parameter estimation method based on artificial immune network and adaptive particle swarm optimization according to claim 1, characterized in that: In Step 5, the mixed optimization process is repeatedly executed with multiple random seeds to calculate the mean, standard deviation, and improvement rate of the parameters in each optimization result, so as to verify the stability and global optimality of the parameter estimation results. Furthermore, the goodness-of-fit indices, including RMSE, AIC, and BIC, are calculated, and the differences in parameters between normal tissue and tumor tissue are statistically analyzed using paired-samples t-test. Finally, the fitting curve and statistical results table are output to realize the visualization and quantitative verification of the kinetic parameter estimation.

Images