Dynamic pet-ct pharmacokinetic modeling method combined with mechanism and data

Abstract

The application relates to a dynamic PET-CT pharmacokinetics fusion modeling method combining mechanism and data, and belongs to the technical field of pharmacokinetic modeling, wherein the method comprises the following steps: acquiring original image data of historical cases, and constructing a training data set according to the original image data; constructing a dynamic pharmacokinetic model based on a mechanism model and a data-driven network, and training the dynamic pharmacokinetic model according to the training data set; collecting actual image data of a current subject for a period of time; inputting the actual image data into the trained dynamic pharmacokinetic model to obtain a prediction result. The application fuses a mechanism model and a data-driven network, can accurately predict a complete pharmacokinetic process only by using image data for a short period of time in the early stage, solves problems that a traditional model depends on long-time scanning and is easily disturbed by motion artifacts, greatly improves parameter analysis efficiency and accuracy, and is suitable for clinical rapid diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pharmacokinetic modeling, in particular to a dynamic PET-CT pharmacokinetic fusion modeling method combining mechanism and data. BACKGROUND

[0002] Dynamic PET-CT imaging, as a core technology in the field of molecular imaging, has irreplaceable value in pharmacokinetic and pharmacodynamic research. This imaging technology can track the distribution, metabolism and clearance of drugs in real time at the living body level by using radioactive tracers, and can realize the analysis of pharmacokinetic parameters by combining quantitative modeling. This technology breaks through the limitations of traditional in vitro experiments on static time points, and provides key imaging evidence for revealing drug action mechanisms, optimizing drug administration schemes and exploring individualized treatment.

[0003] In the prior art, the patent for invention with publication number CN105868537B, "PET-CT dynamic medical image intelligent quantitative analysis system and analysis method", proposes an intelligent quantitative analysis system. The system realizes the optimization of dynamic model parameters by using a hybrid intelligent optimization algorithm such as artificial immune network and gradient descent nesting, and generates accurate regions of interest (ROIs) based on the clustering of time-radioactivity curves of dynamic characteristics distribution. This method greatly improves the efficiency and accuracy of parameter estimation by optimizing model parameters and accurately locking the region of interest.

[0004] However, the above method still needs to rely on complete time-radioactivity curves as data support before the construction of the model. Currently, the generation of time-radioactivity curves requires continuous dynamic PET-CT scanning for about an hour to capture the drug dynamic process. During the long scanning process, children, the elderly or critically ill subjects have poor tolerance, and it is difficult for them to fix their body positions, which easily leads to motion artifacts in the collected imaging, affecting the quality of the imaging data and reducing the accuracy of subsequent analysis of pharmacokinetic parameters. SUMMARY

[0005] In order to improve the problem that the prior art relies on long continuous dynamic scanning, which easily leads to motion artifacts interfering with the quality of the imaging, and then reduces the accuracy of pharmacokinetic parameter analysis, the present application provides a dynamic PET-CT pharmacokinetic fusion modeling method combining mechanism and data.

[0006] In a first aspect, the present application provides a dynamic PET-CT pharmacokinetic fusion modeling method combining mechanism and data, which adopts the following technical solution:

[0007] A dynamic PET-CT pharmacokinetic fusion modeling method combining mechanism and data, comprising:

[0008] Acquire all original imaging data from historical cases and construct a training dataset based on all the original imaging data;

[0009] Constructing a dynamic pharmacokinetic model based on the mechanism model and the deep learning network, and completing the training of the dynamic pharmacokinetic model according to the training data set;

[0010] Collect actual imaging data of the current subject in the early period;

[0011] Inputting the actual image data into the trained dynamic pharmacokinetic model to obtain a prediction result, wherein the prediction result is a predicted tissue time-activity curve;

[0012] The constructing of a training data set based on all the original image data includes:

[0013] Based on the raw image data between all dynamic frames in each historical case, a plasma input function and a standard tissue time-activity curve corresponding to each case are obtained; the plasma input function is used to reflect the concentration dynamics of the radioactive tracer in the plasma of subjects with different historical cases, and the standard tissue time-activity curve is used to reflect the concentration dynamics of the radioactive tracer in the region of interest of subjects with different historical cases;

[0014] The original imaging data of each historical case is divided into two parts, namely, early imaging data and full imaging data;

[0015] A training dataset was constructed based on the early imaging data, full imaging data, plasma input function, and standard tissue time-activity curve of each historical case.

[0016] By adopting the above technical solutions, raw imaging data from historical cases was obtained to construct a training dataset, providing a rich sample base for model training, enabling the model to learn the pharmacokinetic characteristics and patterns of different individuals. A physical framework was constructed based on mechanistic models (such as the two-compartment model), ensuring that the model has clear biophysical significance and interpretability, and can reflect basic processes such as tracer transport and metabolism in the body. At the same time, deep learning networks (parameter adjustment network and residual prediction network) were introduced to capture the time-varying nature of individual physiological states and to compensate for complex physiological mechanisms not covered by the mechanistic model, thereby significantly improving the model's adaptability to individual differences and dynamic prediction accuracy.

[0017] In practical applications, this application only needs to collect actual imaging data of the initial short period (such as 10 minutes) after the current subject is injected with a radioactive tracer, and the complete pharmacokinetic process (such as 60 minutes) can be predicted through the trained dynamic pharmacokinetic model. This greatly shortens the clinical scanning time, reduces the discomfort and motion artifacts caused by long-term scanning to the subject, and improves data quality.

[0018] In a specific embodiment, constructing a training dataset based on the original image data includes:

[0019] Performing spatial alignment on the original image data between each dynamic frame in each historical case to obtain spatially aligned registered image data;

[0020] extracting a signal of a target area in the registered image data as a plasma input function;

[0021] Extract the original tissue time-activity curve within the preset region of interest;

[0022] The original tissue time-activity curve was normalized to obtain the standard tissue time-activity curve.

[0023] In a specific embodiment, the dynamic PET-CT pharmacokinetic model includes a mechanism model and several deep learning networks, wherein the deep learning network includes a parameter adjustment network and a residual prediction network; the mechanism model is the basic physical framework of the dynamic PET-CT pharmacokinetic model;

[0024] The parameter adjustment network is used to obtain a time-activity curve of the tissue to be verified based on the plasma input function corresponding to the actual image data and the standard tissue time-activity curve;

[0025] The residual prediction network is used to obtain the predicted tissue time-activity curve based on the plasma input function corresponding to the actual image data and the standard tissue time-activity curve.

[0026] In a specific embodiment, the parameter adjustment network is specifically used to obtain time-varying parameters based on the plasma input function corresponding to the actual image data and the standard tissue time-activity curve, and to modify the population basis parameters of the two-compartment differential equation in the mechanism model based on the time-varying parameters; the parameter adjustment network is also used to drive the mechanism model to solve the two-compartment differential equation to obtain the tissue time-activity curve to be verified;

[0027] The residual prediction network is specifically used to obtain a residual correction term based on the plasma input function corresponding to the actual image data and the standard tissue time-activity curve, and to perform nonlinear compensation on the tissue time-activity curve to be verified based on the residual correction term to obtain the predicted tissue time-activity curve.

[0028] In a specific embodiment, the early imaging data is used to reflect the actual drug distribution dynamics of the subject at the beginning stage after the radioactive tracer is injected; the full imaging data is used to reflect the actual drug distribution dynamics of the subject from the beginning stage to the end stage after the radioactive tracer is injected; and the predicted tissue time-activity curve is used to reflect the drug concentration dynamics obtained by prediction in the region of interest from the beginning stage to the end stage.

[0029] In a specific embodiment, the constructing of a dynamic pharmacokinetic model based on a mechanism model and a deep learning network, and completing the training of the dynamic pharmacokinetic model according to the training data set includes:

[0030] designing a parameter adjustment network, wherein the parameter adjustment network is used to accurately reflect the actual physiological process;

[0031] Designing a residual prediction network, wherein the residual prediction network is used to compensate for the systematic deviation of the mechanism model;

[0032] Based on the early imaging data, the full imaging data, the plasma input function and the standard tissue time-activity curve, the mechanism model, the parameter adjustment network and the residual prediction network are fused and trained to obtain a dynamic pharmacokinetic model.

[0033] In a specific embodiment, the fusion training of the mechanism model, the parameter adjustment network, and the residual prediction network based on the early imaging data, the full imaging data, the plasma input function, and the standard tissue time-activity curve to obtain a dynamic pharmacokinetic model includes:

[0034] Determining the initial population basic parameters of the mechanism model based on the standard tissue time-activity curve corresponding to the full-course imaging data to complete pre-training;

[0035] Fixing the population basic parameters, and training the parameter adjustment network according to the standard tissue time-activity curve corresponding to the full-course imaging data to obtain time-varying parameter correction terms;

[0036] Fixing the population basic parameters and the trained parameter adjustment network, and training the parameter adjustment network according to a standard tissue time-activity curve corresponding to the full-course imaging data to obtain a residual correction term;

[0037] The entire architecture is fine-tuned end-to-end to collaboratively optimize the mechanism model, the parameter adjustment network, and the residual prediction network to obtain a trained dynamic pharmacokinetic model.

[0038] In a specific embodiment, in the step of determining the population basis parameters of the mechanism model, a set of initial population basis parameters is determined by nonlinear least squares fitting.

[0039] In a specific embodiment, the dynamic pharmacokinetic model unifies the data of the predicted tissue time-activity curve and the predicted observed concentration output by the mechanistic model through an observation function, and the observation function is as follows:

[0040] ;

[0041] in, represents the predicted observed concentration output by the mechanistic model; It represents the vascular fraction, which reflects the volume ratio of blood vessels in the tissue and is a dimensionless parameter. ; It represents the concentration contribution of the plasma tracer that is not taken up in tissue blood vessels to the PET signal, and is used to reflect the weight of the intravascular tracer in the PET signal; Represents the volume proportion of the nonvascular part of the tissue; is the independent variable, representing time; is the model parameter vector; It represents the concentration of radiotracer that is unbound in the tissue, i.e., freely diffusing in the extracellular fluid of the tissue; represents the plasma input function.

[0042] In a second aspect, the present application provides a smart terminal, which adopts the following technical solution:

[0043] An intelligent terminal, characterized in that it includes a memory and a processor, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement a dynamic PET-CT pharmacokinetic fusion modeling method combining mechanism and data as described in the first aspect.

[0044] In a third aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution:

[0045] A computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement a dynamic PET-CT pharmacokinetic fusion modeling method combining mechanism and data as described in the first aspect.

[0046] In summary, this application includes at least one of the following beneficial technical effects:

[0047] 1. Compared to traditional models that rely on long-term continuous dynamic scanning, this application only requires a relatively short initial imaging data acquisition period to reconstruct the complete pharmacokinetic process, significantly shortening the scanning time, reducing the difficulty of positioning children, elderly, or critically ill subjects, and reducing the interference of motion artifacts on data quality, thereby effectively improving the efficiency and accuracy of overall parameter analysis;

[0048] 2. This application introduces time-varying parameters through a parameter adjustment network to capture the time-varying nature of individual physiological states, breaking through the limitations of the traditional model's "fixed parameters";

[0049] 3. This application uses a residual prediction network to compensate for complex factors such as nonlinear metabolism and spatial heterogeneity noise that are not covered by the mechanistic model, making the prediction results output by the model closer to the actual physiological signals and breaking through the limitations of the "simplified assumptions" of traditional models;

[0050] 4. This application uses the differential equations of the mechanism model as the physical backbone and the neural network as the dynamic compensation mechanism. The two complement each other through end-to-end training, and realize the establishment of a modeling architecture that is both interpretable and flexible. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A flow chart of a dynamic PET-CT pharmacokinetic fusion modeling method combining mechanism and data provided in one embodiment of the present application is shown.

[0052] Figure 2 A diagram showing an output process of the prediction result in step S400 according to an embodiment of the present application is shown.

[0053] Figure 3 This is a structural diagram of an electronic device disclosed in another embodiment of the present application.

[0054] Description of reference numerals: 500, electronic device; 501, processor; 502, communication bus; 503, user interface; 504, network interface; 505, memory. DETAILED DESCRIPTION

[0055] In order for those skilled in the art to better understand the technical solutions in the specification, the technical solutions in the specification will be clearly and completely described below in conjunction with the accompanying drawings in the specification embodiments. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments.

[0056] In the description of the embodiments of the present application, the words such as "for example" or "for instance" are used to represent an example, illustration or description. Any embodiment or design scheme described as "for example" or "for instance" in the embodiments of the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the words such as "for example" or "for instance" are intended to present the relevant concept in a specific manner.

[0057] In the description of the embodiments of the present application, the term "a plurality of" means two or more. For example, a plurality of systems means two or more systems, and a plurality of screen terminals means two or more screen terminals. In addition, the terms "first" and "second" are used only for the purpose of description, and should not be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more of the features. The terms "include", "contain", "have" and their variants mean "include but are not limited to", unless otherwise specifically emphasized.

[0058] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments.

[0059] The existing pharmacokinetic modeling method mainly adopts the compartment model. The compartment model is a mathematical model used to describe the distribution, transport and metabolism process of substances (such as drugs, hormones, toxins, etc.) in a biological body or system. The core idea is to abstract the complex biological body or system as a combination of several "compartments", and to establish a dynamic equation to simulate the actual process by analyzing the transport rate of substances between the compartments.

[0060] However, the traditional compartment model has significant limitations:

[0061] The mechanism of this type of model is often based on the assumption of simplifying the actual physiological conditions, that is, pre-set fixed parameters (such as constant drug transport rate, metabolism constant, etc.) or default parameter changes conforming to linear kinetics are usually used, which is essentially different from the complex biological system;

[0062] On the one hand, the assumption of fixed parameters makes it difficult to capture the diverse physiological dynamics of each subject. In real physiological environments, parameters such as drug transport rate (e.g., transmembrane transporter activity) and metabolic enzyme concentration (e.g., cytochrome P450 expression) exhibit significant time-varying and individual heterogeneity. For example, high expression of vascular endothelial growth factor (VEGF) in tumor tissue can lead to a dynamic increase in drug permeability (a transport parameter in compartmental models) over time, and the static parameters of traditional models cannot capture such changes.

[0063] On the other hand, the default of linear dynamics will lead to the neglect of spatial heterogeneity and nonlinear processes; due to the significant differences in the physiological microenvironment of different regions of the organism (such as the liver, kidney, and tumor), drug metabolism may exhibit nonlinear characteristics (such as enzyme saturation effects), but the compartmental model usually regards the entire tissue as a homogeneous space and describes the transport process with linear differential equations, resulting in the model being unable to accurately reflect the actual pharmacokinetic behavior (such as the difference in drug distribution at the edge and center of the tumor).

[0064] More importantly, the fixed-parameter model is highly dependent on data acquisition, and its subsequent parameter estimation process is highly dependent on the complete tissue time-activity curve (such as 60 minutes of continuous dynamic PET-CT scan data), and is extremely sensitive to the sampling scheme: high-frequency sampling is required in the early stage to capture the dynamics of the blood flow phase, and extended acquisition is required in the later stage to accurately fit the terminal trend of metabolism; however, in actual clinical scenarios, due to equipment resolution limitations, counting statistical noise, motion artifacts of subjects (especially children, elderly or critically ill subjects) and individual physiological differences, the tissue time-activity curve obtained after a long time of acquisition and analysis often has a low signal-to-noise ratio, causing the fixed parameter estimation results to fall into the trap of "high variance-low stability", seriously weakening the efficiency and accuracy of the above-mentioned traditional compartment model in pharmacokinetic parameter analysis.

[0065] Based on the above, the actual contradiction between the simplified assumptions of the traditional compartment model and physiological reality, as well as the low efficiency of PET-CT scanning equipment, have caused it to face bottlenecks in complex pharmacokinetic analysis. It is urgent to introduce new modeling methods to improve the ability to characterize real biological motion processes; therefore, one embodiment of the present application discloses a dynamic PET-CT pharmacokinetic fusion modeling method that combines mechanism and data.

[0066] Reference Figure 1 A dynamic PET-CT pharmacokinetic fusion modeling method that combines mechanism and data includes:

[0067] S100, obtaining all original imaging data in historical cases and constructing a training dataset based on all original imaging data;

[0068] Among them, historical cases refer to previous PET-CT examination records of subjects, usually including complete dynamic PET-CT scan data from the injection of radioactive tracers to the end (covering the early rapid distribution stage and the late metabolic clearance stage), basic information of the subjects (such as age, weight, medical history), radioactive dose, scanning parameters (such as time resolution) and region of interest (ROI) annotations (such as tumor lesions, organ boundaries); original image data refers to four-dimensional dynamic image signals collected by PET-CT equipment and including the time dimension. The original image data is used to reflect the distribution intensity of the radioactive tracer in the subject's body at different time points; specifically, constructing a training dataset based on all original image data in S100 includes the following steps:

[0069] S110, performing spatial alignment on the original image data between each dynamic frame to obtain spatially aligned registered image data;

[0070] The original image data of this embodiment is Indicates that the registered image data is Specifically, the relationship between the original image data and the registered image data can be expressed as follows: Where T is the spatial transformation matrix calculated by the inter-frame registration algorithm (such as rigid transformation, affine transformation or non-rigid deformation model), which is used to correct the displacement or deformation error caused by factors such as breathing and body movement during the dynamic PET-CT scanning process of the subject; are spatial coordinates and time coordinates, t corresponds to each time frame in the dynamic scan, Corresponding to the specific location of tissues or organs in the body, x and y usually correspond to the two-dimensional coordinates of the image, indicating the rows and columns of the scanned slice, and are used to locate the pixel position within a certain layer. z corresponds to the depth coordinate of the image, indicating the stacking order of different scan slices, and is used to locate the voxel position in three-dimensional space. / Representation Space The change in radioactive signal intensity at time t.

[0071] This embodiment ensures that images at different time points are aligned in the spatial dimension through this registration process, thereby eliminating the interference of motion artifacts on subsequent pharmacokinetic parameter analysis and laying the foundation for accurate extraction of tissue time-activity curves.

[0072] S120, obtaining a plasma input function corresponding to each case based on the original image data of all dynamic frames in each historical case;

[0073] Considering that the heart is the core of blood circulation, the signal in the heart region can directly reflect the concentration dynamics of the radioactive tracer in the plasma. Therefore, the heart region is selected as the target region, and the plasma input function is used to reflect the concentration dynamics of the radioactive tracer in the plasma of subjects with different historical cases. Specifically, S120 mainly extracts the signal of the target region in the registered image data as the plasma input function. The plasma input function is as follows:

[0074] ;

[0075] in, is the plasma input function; is the number of voxels corresponding to the heart region, which is mainly obtained by segmenting the heart region and counting the total number of voxels; i is the voxel index of the region, corresponding to the spatial coordinates of each voxel in the heart region .

[0076] It should be noted that in medical image processing, the calculation of the number of voxels is usually based on the following steps:

[0077] The specific area, including the heart area in step S120 and the region of interest in the subsequent step S130, first needs to be divided from the registered image data The corresponding region is segmented out from the image; after spatial alignment, the same spatial region in each registered image data corresponds to the same tissue / organ region. Therefore, in this embodiment, common region segmentation methods include but are not limited to: manual outlining (experienced physicians manually circle the heart boundary on the image), threshold segmentation (based on the difference in radioactivity intensity between the heart tissue and the surrounding tissue, setting a threshold to extract the target region), template matching (using a predefined heart template, such as an ellipse model, for fitting), deep learning segmentation (using a network such as U-Net to automatically identify the heart region), etc.; the segmentation result is usually represented as a binary mask :

[0078] ;

[0079] After obtaining the mask of the heart region, the number of voxels can be calculated using the following formula:

[0080] , that is, count all voxels with value 1 in the mask.

[0081] S130, extracting the original tissue time-activity curve in the preset region of interest;

[0082] The region of interest refers to a target tissue region specified by the user based on diagnostic or research needs, such as a tumor lesion, liver / kidney, or other organs, and is used to reflect the dynamic process of the concentration of the radioactive tracer in the specific region changing over time. The principle of the region of interest segmentation method is the same as that of the heart region segmentation in step S120, which is a prior art and will not be repeated here. Specifically, the formula for extracting the original tissue time-activity curve is as follows:

[0083] ;

[0084] in, is the number of voxels corresponding to the region of interest, which is mainly obtained by image segmentation of the region of interest and counting the total number of voxels; i is the voxel index of the region, corresponding to the spatial coordinates of each voxel in the region of interest .

[0085] S140, normalizing the original tissue time-activity curve to obtain a standard tissue time-activity curve;

[0086] Normalization is mainly used to improve the convergence and cross-individual comparability of the model. The standard tissue time-activity curve is used to reflect the concentration dynamics of the radiotracer in the region of interest of subjects with different historical cases. Specifically, the normalization formula of the standard tissue time-activity curve is as follows:

[0087] ;

[0088] in, is the standard tissue time-activity curve; D is the dose of injected radiotracer, and W is the weight of the subject.

[0089] S150, dividing the original imaging data of each historical case into two parts, obtaining early imaging data and full imaging data;

[0090] The raw image data in step S100 is a continuous long-term dynamic sequence. The division into two parts in this embodiment mainly refers to intercepting different stage data of the same dynamic sequence (i.e., raw image data) according to the time range; the early image data is used to reflect the actual drug distribution dynamics of the subject at the beginning stage after the injection of the radioactive tracer; the full image data is used to reflect the actual drug distribution dynamics of the subject from the beginning stage to the end stage after the injection of the radioactive tracer; in this embodiment, the beginning stage is 0-10 minutes and the end stage is 0-60 minutes as an example. The standard tissue time-activity curve corresponding to the early image data can be recorded as The standard tissue time-activity curve corresponding to the full imaging data can be recorded as ,in exist Time and Completely overlap; Take 10 minutes, Take 60 minutes.

[0091] In this embodiment, the early imaging data is acquired with high temporal resolution. For example, within 0-10 minutes after the subject is injected with the radioactive tracer, one image frame is acquired every 10-30 seconds to capture the process of rapid diffusion of the tracer into the tissue with the blood flow, such as the initial uptake of 18F-FDG by the myocardium; the full imaging data covers the early distribution, mid-term metabolic conversion, and late clearance and excretion stages. During this complete pharmacokinetic motion process, the temporal resolution of the imaging data decreases as the scanning progresses. For example, within the subsequent 10-60 minutes, the acquisition frequency is reduced to one image frame every 5-10 minutes to characterize the metabolism and clearance process of the tracer. The full imaging data covers the early imaging data and is used to provide kinetic details at the end of metabolism (such as the retention of the tracer by the tumor or the clearance trend of the tracer by the liver and kidneys).

[0092] S160: constructing a training dataset based on the early imaging data, full imaging data, plasma input function, and standard tissue time-activity curve of each historical case;

[0093] The training data set is represented as follows:

[0094] ;

[0095] in, represents the training data set; j represents the jth case, and N represents the number of subjects, i.e., the number of cases; represents the standard tissue time-activity curve corresponding to the early imaging data of the jth case; represents the standard tissue time-activity curve corresponding to the full imaging data of the jth case; represents the plasma input function corresponding to the jth case.

[0096] S200 builds a dynamic pharmacokinetic model based on the mechanism model and deep learning network, and completes the training of the dynamic pharmacokinetic model based on the training data set;

[0097] The dynamic pharmacokinetic model, as the physical foundation of the entire fusion architecture, directly reflects the transport and metabolism of the tracer in the subject. The construction of the dynamic pharmacokinetic model uses mechanistic models with clear biophysical significance, such as the two-compartment model, as its physical framework. Combined with the dynamic compensation capabilities of deep learning networks, this model forms a hybrid architecture of "mechanistic constraints + deep learning." Specifically:

[0098] In terms of using the mechanism model as the basic physical framework, this embodiment mainly takes the two-compartment model as an example. The differential equation of the dynamic pharmacokinetic model of this embodiment can be expressed as:

[0099] ;

[0100] in, It represents the concentration of radiotracer that is unbound in the tissue, i.e., freely diffusing in the extracellular fluid of the tissue; It is the concentration of the tracer that binds to or becomes a metabolite in the tissue, that is, binds to the tissue target or enters the metabolic pathway; is the transport rate of the tracer into the tissue, is the rate of transport of the tracer out of the tissue, is the phosphorylation rate of the tracer, is the dephosphorylation rate of the tracer; is the plasma input function.

[0101] Because the differential equation output of the mechanistic model is a continuous-time theoretical concentration, while PET imaging data is a radioactive count signal at discrete time points, and actual measurements include complex factors such as residual tracer in blood vessels and tissue-specific uptake differences, to ensure data consistency, this embodiment primarily introduces the vascular fraction to construct an observation function to separate the tracer signals in blood vessels and tissues, thereby ensuring consistency between the final predicted output of the mechanistic model and the actual PET signal (including plasma residual and tissue uptake). Specifically, the observation function of this embodiment is expressed as follows:

[0102] ;

[0103] in, The predicted observed concentration of the mechanistic model is the radioactivity concentration observed in the imaging data predicted by the mechanistic model, in Bq / mL. It represents the vascular fraction, which reflects the volume ratio of blood vessels in the tissue. It is a dimensionless parameter with a value range of , Indicates that the tissue is completely avascular (containing only extracellular fluid and cells). Indicates that the tissue is entirely vascular (an extreme case, usually absent); Represents the concentration contribution of plasma tracer not taken up by tissue blood vessels to the PET signal, and the unit is Consistent, essentially a portion of the concentration, used to reflect the weight of the intravascular tracer in the PET signal; represents the volume ratio of the non-vascular part of the tissue (i.e., the effective volume of the extracellular fluid that can take up the tracer); t is the independent variable, representing time; is the model parameter vector, including When t changes, Output the predicted observed concentration corresponding to different time points. When changes, The output corresponds to the predicted observed concentration under different parameter settings.

[0104] Therefore, this embodiment uses the blood vessel score As the core link between the "theoretical pharmacokinetic process" and the "measured imaging data", it realizes the physical mapping between the output of the mechanism model (tissue and plasma concentrations) and the actual PET signal. This design not only retains the biological interpretability of the two-compartment model, but also achieves the goal of the two-compartment model through the parameter Flexible adjustment of the contribution of blood vessel and tissue signals is achieved, laying the foundation for the modeling of "mechanism-data fusion" in this embodiment.

[0105] It should be noted that in the above observation function, the pharmacokinetic parameters The estimation of can be transformed into a nonlinear least squares fitting problem and the observed concentration is predicted by minimizing Compared with actual PET observation data The error between ; Thus, this embodiment can continuously adjust the parameters , so that the predicted observed concentration As close as possible to the actual PET observation data at all time points t This optimization effectively ensures that the dynamic pharmacokinetic model of this embodiment can fit the actual dynamic imaging data while having good biological interpretability, and can be used for clinical tasks such as drug uptake analysis, metabolic flux estimation, and efficacy prediction of tumor lesions.

[0106] Furthermore, because traditional mechanistic models (such as the two-compartment model) use fixed parameters and simplified assumptions, they are unable to capture the time-varying nature of individual physiological states (such as fluctuations in metabolic enzyme activity) and complex physiological mechanisms (such as nonlinear metabolic pathways and spatially heterogeneous noise). Therefore, to improve the model's adaptability to individual differences and dynamic prediction accuracy, this embodiment primarily implements dynamic compensation for the mechanistic model by designing a dual neural network architecture consisting of a parameter adjustment network and a residual prediction network. The specific steps include:

[0107] S210, designing a parameter adjustment network; wherein the parameter adjustment network is used to break through the limitation of the traditional model "fixed parameters", that is, the limitation of being unable to accurately capture the time-varying nature of individual physiological states. In this embodiment, the time-varying parameters are introduced. To more accurately reflect the actual physiological process, specifically, The calculation formula is as follows: ;

[0108] in, Characterize the basic parameter values ​​at the population level, characterize the pharmacokinetic parameters under the "average physiological state", such as the average transport rate of healthy people , can be estimated using all the original imaging data of historical cases in S100; represents the model parameter vector corresponding to t; Used to capture individual specificity and time-varying characteristics to adjust the value of the model parameter vector corresponding to t, is a time-varying parameter, a four-dimensional vector, where each element corresponds to an adjustment factor of a basic parameter; specifically:

[0109] ;

[0110] in, Corresponding to the four core rate constants in the two-compartment model; if , indicating that the subject's tracer uptake rate is higher than the group average. , indicating that the binding rate of the tracer to this subject decreases over time (e.g., receptor saturation).

[0111] Parameter adjustment network The tissue time-activity curve in the first 10 minutes of PET-CT and plasma input function As input, predict the time-varying parameters :

[0112] ;

[0113] in, Tune network parameters for parameters; express The standard tissue time-activity curve is used to reflect the early distribution dynamics of the tracer in the target tissue; express The plasma input function between the two groups is used to reflect the individual's hemodynamic characteristics (such as cardiac output, tracer injection rate, etc.); for The time-varying parameters between , for The number of time points.

[0114] It should be noted that in this embodiment, the parameter adjustment network integrates convolutional layers, bidirectional GRU and attention mechanisms, which can extract time series features from an early period of PET data, map them to four pharmacokinetic parameter adjustment factors on the complete time series through a multi-layer perceptron, and apply parameter constraints to ensure physiological rationality, thereby realizing individualized time-varying parameter prediction to enhance the flexibility of the mechanism model.

[0115] Specifically, the design of the parameter adjustment network is as follows:

[0116] The convolutional layer in this example uses two layers of 1D convolution (kernel size = 3, number of channels = 32→64), equipped with BatchNorm and ReLU activation, to extract local features in the time series, such as the slope and fluctuation frequency of the early drug uptake peak, and convert the original signal into a high-dimensional feature vector. The hidden dimension of the bidirectional GRU is 128, which is used to capture the long-term dependencies of the time series data, such as the impact of trends within the previous 10 minutes on subsequent metabolism. The bidirectional mechanism of the bidirectional GRU simultaneously considers information from past and future time points. That is, even if the input is only the previous period (10 minutes), the GRU can infer subsequent trends (the next 50 minutes) through sequence modeling. An 8-head attention mechanism is used to calculate the attention weights between time points, which is used to focus on key time points (such as the peak time after tracer injection and the distribution equilibrium time) to suppress the interference of noise on parameter prediction.

[0117] In the application of multi-layer perceptron, the structure of Flatten layer → fully connected layer (256 neurons, ReLU activation) → Dropout (0.2) is mainly used to map the features extracted by convolution-GRU to the parameter adjustment space and generate the four-dimensional adjustment factors corresponding to the time points of the entire time course. Regarding parameter constraints, this embodiment mainly scales the correction term to a reasonable range by setting α and β based on physiological priors, such as etc. and ensure non-negative, , (If metabolism is irreversible, ).

[0118] In summary, the architecture design in the parameter adjustment network is as follows in actual operation:

[0119] Input layer:

[0120] - # Time-activity curve of the first 10 minutes

[0121] - #Plasma input function for the first 10 minutes

[0122] Feature extraction module:

[0123] 1. Temporal feature encoding layer:

[0124] -1D convolutional layer (kernel size = 3, stride = 1, number of channels = 32)

[0125] -BatchNormalization

[0126] -ReLU activation

[0127] -1D convolutional layer (kernel size = 3, stride = 1, number of channels = 64)

[0128] -BatchNormalization

[0129] -ReLU activation

[0130] 2. Timing dynamic capture module:

[0131] - Bidirectional GRU (hidden layer dimension = 128)

[0132] -Self-attention mechanism (8-head attention), focusing on key time points

[0133] 3. Feature integration layer:

[0134] -Flatten

[0135] -Dropout(0.2)

[0136] -Dense(256,activation='relu')

[0137] -Dropout(0.2)

[0138] Parameter generation module:

[0139] 4. Time-varying parameter prediction:

[0140] -

[0141] - #Predict the changes in the four parameters on the complete time series

[0142] 5. Parameter constraint layer:

[0143] -Tanh activation function (scaled to [-1,1])

[0144] -Linear transformation (scaled to a reasonable range): ;

[0145] in , Scaling parameters set based on physiological prior knowledge

[0146] Output layer:

[0147] - #Four time-varying parameter adjustment factors.

[0148] S220, design a residual prediction network;

[0149] Among them, the residual prediction network is used to break through the limitations of the traditional model's "simplified assumptions", that is, it cannot fully reflect the limitations of complex physiological processes such as nonlinear metabolism, spatial heterogeneity, measurement noise, and unmodeled pathways. This embodiment introduces a residual correction term to compensate for the systematic deviation of the mechanism model. The residual prediction network and the parameter adjustment network in S210 have the same input data, and the prediction covers the entire process. The residual correction term is calculated as follows:

[0150] ;

[0151] in, Predict network parameters for the residual; is the residual correction term, and its dimension is , for The number of time points.

[0152] It should be noted that in this embodiment, the residual prediction network combines multi-scale convolution, UNet and bidirectional LSTM architecture to simultaneously capture local details and global trends from PET data of a previous period of time, extrapolate short sequences to complete time series through the autoregressive mechanism, and apply smoothing constraints to generate continuous residual correction terms, thereby effectively compensating for the complex dynamic processes that the mechanism model cannot describe.

[0153] Specifically, the multi-scale convolution of this embodiment mainly adopts parallel 1D convolution branches. The three branches use 1D convolution with kernel sizes of 3, 5, and 7 respectively to capture local features at different time scales: a small kernel of size 3 is used to extract short-term fluctuations, such as high-frequency changes in the early rapid distribution stage, and a large kernel of size 5 or 7 is used to extract long-term trends, such as low-frequency features of late metabolic clearance. Finally, the outputs of the three branches are spliced ​​to form a feature map containing multi-scale information. In this way, this embodiment can avoid the limitations of a single kernel size and enhance the network's ability to capture complex dynamic patterns.

[0154] The UNet structure used is as follows:

[0155] The encoding path is a 3-layer downsampling Conv1D with the number of channels being 64→128→256, with a step size of 2. Each downsampling expands the receptive field and extracts global context features (such as the trend of the entire early time series). The decoding path is a 3-layer upsampling Conv1D with the number of channels being 256→128→64. The time dimension resolution is restored through transposed convolution, and the shallow details and deep semantics are fused with the skip connection (SkipConnection) of the encoding path. In this way, this embodiment can achieve "compression-decompression" of time series data through the symmetric architecture of UNet, while retaining local details such as the precise position of early peaks while extracting global trends. The hidden dimension of the bidirectional LSTM is 128, which is used to process the fused feature sequence and capture bidirectional time series dependencies, such as the impact of future time points on current predictions. The autoregressive prediction mechanism mainly uses Teacher-Forcing in the training stage, taking the true residual as the input for subsequent predictions to alleviate the gradient disappearance, and recursively generates subsequent sequences based on the predicted residual terms in the inference stage to achieve the goal of "from arrive Extrapolation of; Among them, in the initialization stage, the hidden state of LSTM is composed of The feature representation between the two is initialized to ensure the consistency of the extrapolation process with the known data.

[0156] In the application of smoothing constraints, this embodiment uses 1D convolution with a kernel size of 5 to smooth the prediction residuals, suppress high-frequency noise, and The residuals of known data points are subjected to stronger smoothness constraints, assuming that physiological processes change gradually over time, thus avoiding abrupt predictions.

[0157] In summary, the architecture design in the residual prediction network is as follows in actual operation:

[0158] Input layer:

[0159] - # Time-activity curve of the first 10 minutes

[0160] - #Plasma input function for the first 10 minutes

[0161] Feature extraction module:

[0162] 1. Multi-scale feature extraction:

[0163] -Three parallel 1D convolution branches (kernel sizes 3, 5, and 7 respectively)

[0164] -Each branch: Conv1D→BatchNorm→ReLU→MaxPooling

[0165] -Feature connection layer: Concatenate

[0166] 2. Context encoder:

[0167] - 1D-UNet structure:

[0168] - Encoding path: 3 layers of down-sampling Conv1D (increasing number of channels: 64 → 128 → 256)

[0169] - Decoding path: 3 layers of up-sampling Conv1D (decreasing number of channels: 256 → 128 → 64)

[0170] - Skip connections: between corresponding layers of encoding-decoding pair

[0171] 3. Global-local feature fusion:

[0172] - Global feature: GlobalAveragePooling → Dense(128)

[0173] - Local feature: UNet output

[0174] - Fusion layer: Concatenate → Conv1D

[0175] Temporal prediction module:

[0176] 4. Time extrapolation layer:

[0177] - Bidirectional LSTM (128 units)

[0178] - Autoregressive prediction mechanism:

[0179] - Train using Teacher-Forcing

[0180] - Initialize hidden state using feature representation of previous 10 minutes

[0181] - Predictor network: Dense(1) → linear activation

[0182] 5. Smoothing constraint layer:

[0183] - Temporal smoothing filter: Conv1D (kernel size=5, stride=1)

[0184] - Decreasing smoothing weight mechanism: stronger smoothing constraint for positions further away from known data points

[0185] Output layer:

[0186] - # Complete 60-minute residual term.

[0187] After completing the architecture design of the parameter adjustment network in step S210 and the residual prediction network in step S220, these two deep learning modules need to be integrated end-to-end with the mechanistic model to form a complete dynamic pharmacokinetic model. At this point, the model architecture already has the dual-driving capabilities of "mechanistic framework + data compensation", but it is necessary to optimize the parameters of each component through a staged training process to ensure:

[0188] The basic parameters of the mechanistic model are biophysically reasonable;

[0189] The parameters of the deep learning network can accurately capture dynamic features not covered by the mechanism model;

[0190] When the two work together, they meet the balance between physiological constraints and prediction accuracy;

[0191] Therefore, S200 of this embodiment further includes the following steps:

[0192] S230, integrating the training mechanism model and the deep learning network to obtain a dynamic pharmacokinetic model;

[0193] In order to ensure that the fusion training starts from a meaningful physical solution space, it is necessary to provide biophysically reasonable initial values ​​for the subsequent neural network. Therefore, S230 includes:

[0194] S231, determining the population basic parameters of the mechanism model based on the standard tissue time-activity curve corresponding to the full-process imaging data to complete pre-training;

[0195] Among them, the population basic parameters are used to establish the initial dynamic framework of physiological laws; specifically, the calculation formula of the population basic parameters is as follows:

[0196] ;

[0197] in, is the basic parameter of the group, which is the same as that in S210 Corresponding; N represents the number of subjects; Characterizes the basic parameter value at the population level; i represents the i-th case; represents the complete time series of the i-th case A collection of time points; The time-activity curve of the tissue to be verified output by the mechanism model of the i-th case is obtained by solving the differential equation of the two-compartment model and depends on the parameter ; Represents the standard tissue time-activity curve corresponding to the full imaging data in the i-th case.

[0198] It should be noted that in step S231, this embodiment mainly determines a set of population basic parameters by nonlinear least squares fitting. , so that the time-activity curve of the tissue to be verified output by the mechanism model Standard tissue time-activity curve corresponding to the actual measurement The error between them is minimal to ensure that the model complies with physiological laws such as conservation of matter and transport direction.

[0199] Pre-trained The physiological rationality at the group level has been met. However, in order to avoid the neural network from destroying the basic physical constraints during optimization, this embodiment mainly completes the training by training the parameter adjustment network and the residual prediction network in stages:

[0200] S232, fixing the population basic parameters, and adjusting the network according to the training parameters of the training data set to obtain a time-varying parameter correction term;

[0201] Specifically, the calculation formula of the time-varying parameter correction term is as follows:

[0202] ;

[0203] in, Represents the optimized optimal weight parameter set of the parameter adjustment network, including neural network parameters such as convolutional layer, GRU layer, and attention mechanism; represents the time-varying parameter correction term of the i-th case, which is a four-dimensional vector; The time-activity curve of the tissue to be verified, which is output by the mechanism model after the introduction of time-varying parameters, is obtained by solving the differential equation of the two-compartment model; is a preset regularization hyperparameter used to control the amplitude of the time-varying parameter correction term to prevent was over-adjusted to keep the corrected time-varying parameters within a physiologically plausible range.

[0204] It should be noted that, in step S232, if the tumor tissue of a subject is rich in blood vessels ( Higher), the parameter adjustment network may learn , indicating that its tracer uptake rate is higher than the group average, thus Dynamically adjust the transport rate parameters of tracers into tissues to accommodate individual differences.

[0205] After S232, the mechanistic model and parameter adjustment network have solved the problems of "parameter fixation" and "individual differences", while the residual network focuses on dealing with the limitations of the model structure itself (such as the enzyme saturation effect that is not modeled in the two-compartment model). Therefore, to avoid optimization confusion caused by multi-module coupling, S230 also includes:

[0206] S233, fixing the population basic parameters and the trained parameter adjustment network, and training the parameter adjustment network according to the training data set to obtain a residual correction term;

[0207] Specifically, the calculation formula of the residual correction term is as follows:

[0208] ;

[0209] in, Represents the optimized optimal network weight parameters of the residual prediction network; Indicates that the i-th case is The residual correction term at the moment; It is a preset residual smoothness regularization hyperparameter used to promote the continuous and smooth change of residuals over time, which is consistent with the continuous change characteristics of physiological processes; It is a preset residual amplitude regularization hyperparameter, which is used to control the absolute value of the residual correction term to ensure that the residual is a fine-tuning of the mechanism model rather than a dominant one; represents the difference between the residual of the i-th case at the next moment and the residual at the current moment, that is, the sum of squared differences of the residuals at adjacent time points, Used to penalize mutations.

[0210] It should be noted that in step S233, if the measured PET signal fluctuates abnormally in the later stage (possibly caused by PET device noise), the residual prediction network will learn the corresponding residual correction term, but the amplitude constraint will limit its overreaction, and the smoothness constraint will ensure that the fluctuation does not violate physiological continuity.

[0211] S234, performing end-to-end fine-tuning on the entire architecture, collaboratively optimizing the mechanism model, the parameter adjustment network, and the residual prediction network, to obtain a trained dynamic pharmacokinetic model;

[0212] Specifically, the calculation formula for collaborative optimization is as follows:

[0213] ;

[0214] in, is the total loss function, which is used to balance prediction accuracy, physical constraints and model complexity;

[0215] The specific calculation formula of the total loss function is as follows:

[0216] ;

[0217] in, is an adjustable weight factor of the physical constraint loss, which is used to adjust the strength of the physiological plausibility constraint; is an adjustable weight factor for regularization loss, used to control model complexity; The prediction loss is used to measure the fitting error between the model prediction value and the measured data; It is the physical constraint loss, which is used to ensure that the model parameters conform to physiological laws; It is a regularization loss used to prevent overfitting and enhance the generalization ability of the model;

[0218] Specifically, the prediction loss, physical constraint loss, and regularization loss are expressed as:

[0219] ;

[0220] Among them, M is the upper limit of the binding site ratio, that is, the upper limit of the rate ratio, usually M≥1; The time-activity curve of the tissue to be verified is output by the dynamic pharmacokinetic model, i.e., the fusion model; It represents the ratio of phosphorylation rate to dephosphorylation rate, reflecting metabolic reversibility; is the adjustable weight factor for the rate ratio constraint; Used to force the parameter to be non-negative. , then the forward loss is , otherwise 0.

[0221] S300, collecting actual imaging data of the current subject in the early period;

[0222] The early period refers to the beginning stage after the subject is injected with the radioactive tracer. This embodiment takes the first 10 minutes as an example. The actual image data is the four-dimensional dynamic image signal including the time dimension currently acquired by the PET-CT device, which is essentially equivalent to the early image data in S100. The acquisition method thereof will not be repeated here.

[0223] S400: Input the actual image data into the trained dynamic pharmacokinetic model to obtain a prediction result.

[0224] The prediction result is a predicted tissue time-activity curve, which is used to reflect the dynamics of drug concentration in the preset region of interest from the beginning to the end. Figure 2 When the actual image data is input into the trained dynamic pharmacokinetic model, the dynamic pharmacokinetic model can perform the following steps:

[0225] ① Based on the same principle as steps S100-S200 above, extract the standard tissue time-activity curve of the region of interest in the actual image data and plasma input function ;

[0226] The standard tissue time-activity curve is used to reflect the early tracer distribution dynamics in the region of interest (i.e., a specific tissue / organ), such as uptake rate and peak time. The plasma input function is used to reflect the hemodynamic characteristics of different subjects, such as cardiac output and tracer injection rate, through changes in tracer concentration in the heart region. The plasma input function can assist the dynamic pharmacokinetic model in capturing the impact of inter-individual blood flow differences on pharmacokinetics. Both the standard tissue time-activity curve and the plasma input function serve as core input features for the subsequent parameter adjustment network and residual prediction network. In this embodiment, if the plasma input function of an individual is difficult to accurately obtain, an input function model based on population statistics can also be used for estimation.

[0227] ②Adjust the network through parameters Complete the following steps:

[0228] ② / 1 According to the network parameters trained in step S232 , standard tissue time-activity curve and plasma input function, predict the time-varying parameter correction term:

[0229] .

[0230] ② / 2 Update the population basic parameters of the mechanism model according to the time-varying parameter correction term to obtain the basic parameters corresponding to the current subject: .

[0231] ② / 3 Solve the mechanism model based on the time-varying parameters and basic parameters to obtain the time-activity curve of the tissue to be verified based on the actual imaging data of the subject in the early period:

[0232] ,in, Represents the process of solving ordinary differential equations.

[0233] ③Through residual prediction network Complete the following steps:

[0234] ③ / 1 According to the network parameters trained in step S233 , standard tissue time-activity curve and plasma input function, predict the residual correction term:

[0235] .

[0236] ③ / 2 Superimpose the time-activity curve of the tissue to be verified output by the mechanism model and the residual correction term to obtain the fused prediction result, that is, the predicted tissue time-activity curve:

[0237] .

[0238] Based on the same inventive concept, the application further discloses a smart terminal, which comprises a memory and a processor, the memory stores at least one instruction, at least one program, a code set or an instruction set, the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to realize the dynamic PET-CT pharmacokinetic fusion modeling method of combining mechanism and data provided by the above method embodiment.

[0239] Based on the same inventive concept, the application further discloses a computer readable storage medium, the storage medium stores at least one instruction, at least one program, a code set or an instruction set, at least one instruction, at least one program, code set or instruction set can be loaded and executed by the processor to realize the dynamic PET-CT pharmacokinetic fusion modeling method of combining mechanism and data provided by the above method embodiment.

[0240] In addition, in order to guarantee the accuracy of the dynamic pharmacokinetic model as much as possible, the embodiment can also periodically check the model through evaluation and verification, wherein the quantitative evaluation can select the MAE and R2 of the curve prediction accuracy, the physiological reasonableness test can select the non-negative parameter constraint and the rate ratio constraint, and the qualitative evaluation can use the sensitivity, specificity and AUC; the above evaluation means is the prior art, and will not be described here.

[0241] Referring to Figure 3 The application further discloses an electronic device, which can include at least one processor 501, at least one network interface 504, a user interface 503, a memory 505, and at least one communication bus 502.

[0242] The communication bus 502 is used to realize the connection and communication between the components.

[0243] The user interface 503 can include a display screen (Display) and a camera (Camera), and the optional user interface 503 can further include a standard wired interface and a wireless interface.

[0244] The network interface 504 can optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0245] The processor 501 may include one or more processing cores. Using various interfaces and circuits, the processor 501 connects to various components within the server. It executes instructions, programs, code sets, or instruction sets stored in the memory 505, as well as accesses data stored in the memory 505, to perform various server functions and process data. Optionally, the processor 501 may be implemented using at least one of the following hardware forms: a digital signal processing (DSP), a field-programmable gate array (FPGA), or a programmable logic array (PLA). The processor 501 may integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. The CPU primarily processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing content displayed on the display; and the modem handles wireless communications. It is understood that the modem may also be implemented as a separate chip, rather than integrated into the processor 501.

[0246] Memory 505 may include random access memory (RAM) or read-only memory (ROM). Optionally, memory 505 may include non-transitory computer-readable storage medium. Memory 505 may be used to store instructions, programs, code, code sets, or instruction sets. Memory 505 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, sound playback function, image playback function, etc.), instructions for implementing the aforementioned method embodiments, and the data storage area may store data related to the aforementioned method embodiments. Memory 505 may also optionally be at least one storage device located remotely from the aforementioned processor 501.

[0247] The memory 505 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an application program for a dynamic PET-CT pharmacokinetic fusion modeling method combining mechanism and data.

[0248] exist Figure 3In the electronic device 500 shown, the user interface 503 is mainly used to provide an input interface for the user and obtain the data input by the user; and the processor 501 can be used to call the application program stored in the memory 505 for a dynamic PET-CT pharmacokinetic fusion modeling method that combines mechanism and data. When executed by one or more processors 501, the electronic device 500 executes one or more methods in the above-mentioned embodiments. It should be noted that for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should know that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for this application.

[0249] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0250] In the several embodiments provided in this application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are merely schematic, such as the division of units, which is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interface, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0251] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0252] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0253] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable memory. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a memory and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The aforementioned memory includes: a U disk, a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0254] The above are only exemplary embodiments of the present disclosure, and cannot limit the scope of the present disclosure. That is, any equivalent changes and modifications made in accordance with the teachings of the present disclosure are still within the scope of the present disclosure. Other embodiments of the present disclosure will be readily apparent to those skilled in the art upon considering the specification and practicing the true principles of the present disclosure.

[0255] The present application is intended to cover any variations, uses or adaptive changes of the present disclosure that follow the general principles of the present disclosure and include common knowledge or conventional technical means in the technical field not described in the present disclosure. The specification and examples are only considered as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A dynamic PET-CT pharmacokinetic fusion modeling method combining mechanism and data, characterized by: include: Acquire all original imaging data from historical cases and construct a training dataset based on all the original imaging data; Constructing a dynamic pharmacokinetic model based on the mechanism model and the deep learning network, and completing the training of the dynamic pharmacokinetic model according to the training data set; Collect actual imaging data of the current subject in the early period; Inputting the actual image data into the trained dynamic pharmacokinetic model to obtain a prediction result, wherein the prediction result is a predicted tissue time-activity curve; The constructing of a training data set based on all the original image data includes: Based on the raw image data between all dynamic frames in each historical case, a plasma input function and a standard tissue time-activity curve corresponding to each case are obtained; the plasma input function is used to reflect the concentration dynamics of the radioactive tracer in the plasma of subjects with different historical cases, and the standard tissue time-activity curve is used to reflect the concentration dynamics of the radioactive tracer in the region of interest of subjects with different historical cases; The original imaging data of each historical case is divided into two parts, namely, early imaging data and full imaging data; A training dataset was constructed based on the early imaging data, full imaging data, plasma input function, and standard tissue time-activity curve of each historical case. In the construction of the dynamic pharmacokinetic model: The dynamic PET-CT pharmacokinetic model includes a mechanism model and several deep learning networks, wherein the deep learning networks include a parameter adjustment network and a residual prediction network. The dynamic PET-CT pharmacokinetic model directly uses the differential equation of the mechanism model as a mathematical framework; The parameter adjustment network is used to obtain a time-activity curve of the tissue to be verified based on the plasma input function corresponding to the actual image data and the standard tissue time-activity curve, and the time-activity curve of the tissue to be verified is used to accurately reflect the actual physiological process; The residual prediction network is used to obtain the predicted tissue time-activity curve based on the plasma input function corresponding to the actual image data and the standard tissue time-activity curve. The predicted tissue time-activity curve is used to compensate for the systematic deviation of the mechanism model.

2. The dynamic PET-CT pharmacokinetic fusion modeling method according to claim 1, characterized in that: The plasma input function and the standard tissue time-activity curve corresponding to each case are obtained based on the original image data of all dynamic frames in each historical case, including: Performing spatial alignment on the original image data between each dynamic frame in each historical case to obtain spatially aligned registered image data; extracting a signal of a target area in the registered image data as a plasma input function; Extract the original tissue time-activity curve within the preset region of interest; The original tissue time-activity curve was normalized to obtain the standard tissue time-activity curve.

3. The dynamic PET-CT pharmacokinetic fusion modeling method according to claim 1, characterized in that: The differential equation includes a plurality of population basis parameters, each of which corresponds to an initial pharmacokinetic parameter, and a set of the population basis parameters is used to establish an initial kinetic framework of physiological laws; the parameter adjustment network is specifically used to obtain time-varying parameters based on the plasma input function corresponding to the actual imaging data and the standard tissue time-activity curve, and to modify the population basis parameters based on the time-varying parameters; the parameter adjustment network is also used to drive the mechanism model to solve the differential equation to obtain the tissue time-activity curve to be verified; The residual prediction network is specifically used to obtain a residual correction term based on the plasma input function corresponding to the actual image data and the standard tissue time-activity curve, and to perform nonlinear compensation on the tissue time-activity curve to be verified based on the residual correction term to obtain the predicted tissue time-activity curve.

4. The dynamic PET-CT pharmacokinetic fusion modeling method according to claim 1, characterized in that: The early imaging data is used to reflect the actual drug distribution dynamics of the subject at the beginning stage after the radioactive tracer is injected; the full imaging data is used to reflect the actual drug distribution dynamics of the subject from the beginning stage to the end stage after the radioactive tracer is injected; the predicted tissue time-activity curve is used to reflect the drug concentration dynamics obtained by prediction in the region of interest from the beginning stage to the end stage.

5. The dynamic PET-CT pharmacokinetic fusion modeling method according to claim 4, characterized in that: The step of constructing a dynamic pharmacokinetic model based on a mechanism model and a deep learning network, and completing the training of the dynamic pharmacokinetic model according to the training data set includes: designing a parameter adjustment network, wherein the parameter adjustment network is used to accurately reflect the actual physiological process; Designing a residual prediction network, wherein the residual prediction network is used to compensate for the systematic deviation of the mechanism model; Based on the early imaging data, the full imaging data, the plasma input function and the standard tissue time-activity curve, the mechanism model, the parameter adjustment network and the residual prediction network are fused and trained to obtain a dynamic pharmacokinetic model.

6. The dynamic PET-CT pharmacokinetic fusion modeling method according to claim 5, characterized in that: The step of fusing and training the mechanism model, the parameter adjustment network, and the residual prediction network based on the early imaging data, the full imaging data, the plasma input function, and the standard tissue time-activity curve to obtain a dynamic pharmacokinetic model includes: Determining the initial population basic parameters of the mechanism model based on the standard tissue time-activity curve corresponding to the full-course imaging data to complete pre-training; Fixing the population basic parameters, and training the parameter adjustment network according to the standard tissue time-activity curve corresponding to the full-course imaging data to obtain time-varying parameter correction terms; Fixing the population basic parameters and the trained parameter adjustment network, and training the parameter adjustment network according to a standard tissue time-activity curve corresponding to the full-course imaging data to obtain a residual correction term; The entire architecture is fine-tuned end-to-end to collaboratively optimize the mechanism model, the parameter adjustment network, and the residual prediction network to obtain a trained dynamic pharmacokinetic model.

7. The dynamic PET-CT pharmacokinetic fusion modeling method according to claim 6, characterized in that: In the step of determining the initial population basic parameters of the mechanism model, a set of initial population basic parameters is determined by nonlinear least squares fitting.

8. The dynamic PET-CT pharmacokinetic fusion modeling method according to claim 1, characterized in that: The dynamic pharmacokinetic model unifies the predicted tissue time-activity curve with the predicted observed concentration output by the mechanism model through an observation function, and the observation function is as follows: ； in, represents the predicted observed concentration output by the mechanistic model; It represents the vascular fraction, which reflects the volume ratio of blood vessels in the tissue and is a dimensionless parameter. ; It represents the concentration contribution of the plasma tracer that is not taken up in tissue blood vessels to the PET signal, and is used to reflect the weight of the intravascular tracer in the PET signal; Represents the volume proportion of the nonvascular part of the tissue; is the independent variable, representing time; is the model parameter vector; It represents the concentration of radiotracer that is unbound in the tissue, i.e., freely diffusing in the extracellular fluid of the tissue; represents the plasma input function.

9. A computer-readable storage medium, characterized in that The readable storage medium stores at least one instruction, at least one program, code set, or instruction set, which is loaded and executed by a processor to implement a dynamic PET-CT pharmacokinetic fusion modeling method combining mechanism and data as described in any one of claims 1 to 8.

Images