A digital twin driven personalized diagnosis and treatment system and method for psoriasis

Abstract

The application discloses a digital twin driven psoriasis personalized diagnosis and treatment system and method, and relates to the technical field of digital medical treatment and artificial intelligence. The method comprises the following steps: collecting multi-modal cross-scale medical data of a patient and preprocessing the multi-modal cross-scale medical data to generate a feature data set; constructing a multi-dimensional digital twin model containing geometric, physical, behavior and rule models based on the feature data set; obtaining a personalized multi-scale digital twin body through variable transmission, constraint correction and order reduction; inputting a pharmacological model to perform excitation simulation, dynamically deducing a single drug efficacy decay inflection point, and calculating a joint drug prediction effect by using a modified Bliss model; combining newly added monitoring data, using extended Kalman filtering and Bayesian inference to rollingly update and calibrate model parameters, and outputting a personalized treatment scheme. The application realizes accurate prediction of long-term efficacy of psoriasis and efficient evaluation of synergistic effect of joint drug use.

Description

Technical Field

[0001] This invention relates to the fields of digital healthcare and artificial intelligence, and in particular to a digital twin-driven personalized diagnosis and treatment system and method for psoriasis. Background Technology

[0002] Currently, the diagnosis and treatment of psoriasis mainly rely on clinical experience and standard treatment guidelines. For moderate to severe psoriasis, biologics such as IL-17A inhibitors have become important treatment methods. In basic research, animal models or in vitro cell models are typically used to study pathogenesis and screen drugs. Although some computer-aided technologies have been used for medical image analysis or pharmacokinetic / pharmacodynamic modeling, these models are mostly general and non-personalized, unable to integrate multi-dimensional, cross-scale vital sign data of patients. Meanwhile, although digital twin technology has been applied in industry, in the clinical application of complex chronic diseases like psoriasis, there is still no dynamic evolutionary personalized system that can integrate molecular, cellular, tissue, and organ information. The existing psoriasis research models, clinical treatment protocols, and computer-aided analysis technologies have the following shortcomings:

[0003] First, the research model is distorted: traditional animal or cell models cannot fully simulate the complex pathophysiological process of human psoriasis and have limitations in reflecting individual differences and the long-term effects of drugs.

[0004] Secondly, the treatment plan lacks foresight: clinical treatment is mostly "trial and error", making it difficult to accurately predict the long-term response of a specific patient to biologics before treatment, and it is impossible to predict and explain the phenomenon of efficacy decay.

[0005] Third, the data is not being fully utilized: the massive amount of multimodal data generated by clinical diagnosis and treatment has not been effectively integrated and deeply mined, making it impossible to form a complete view that dynamically reflects the evolution of the disease.

[0006] Fourth, there is a lack of combined drug simulation platforms: for novel therapies such as the combination of biological agents and traditional Chinese medicine, there is a lack of effective platforms to simulate synergistic mechanisms and predict efficacy, which hinders the development of personalized precision treatment. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a digital twin-driven personalized diagnosis and treatment method and system for psoriasis, so as to solve the technical problems in the prior art such as distorted psoriasis research models, lack of predictability in treatment plans, ineffective utilization of heterogeneous medical data, and lack of an effective simulation platform for combined biological agents and traditional Chinese medicine therapies.

[0008] To address the aforementioned technical problems, a first aspect of the present invention provides a digital twin-driven personalized diagnosis and treatment method for psoriasis, comprising the following steps:

[0009] Step S1: Multimodal data acquisition and preprocessing: Collect multimodal cross-scale medical data of psoriasis patients, and standardize, clean and denoise the multimodal cross-scale medical data and extract features to generate personalized feature datasets;

[0010] Step S2, Multidimensional Digital Twin Model Construction and Order Reduction: Based on the personalized feature dataset, a multidimensional digital twin model including a geometric model, a physical model, a behavioral model, and a rule model is constructed; the underlying data interaction method of variable transfer and constraint correction is used to couple the models, and the intrinsic orthogonal decomposition method is used to reduce the order of the models to construct a personalized multi-scale digital twin.

[0011] Step S3, Virtual Treatment Simulation and Efficacy Prediction: Input the pharmacological model into the personalized multi-scale digital twin for stimulation simulation; wherein, for single drug stimulation simulation, the efficacy decay process is dynamically deduced based on the preset triggering mechanism to output the long-term efficacy prediction curve and decay inflection point; for combined drug stimulation simulation, the final predicted effect of combined drug is calculated by introducing a modified Bliss independence model with synergistic interaction terms.

[0012] Step S4, Model Rolling Update and Solution Output: Obtain new monitoring data of the psoriasis patients, use the extended Kalman filter framework in combination with the new monitoring data to perform rolling updates and calibrations on the state vector and model parameters of the personalized multi-scale digital twin, and output personalized treatment recommendation solutions based on the long-term efficacy prediction curve and the final prediction effect.

[0013] Further, in step S2, constructing a physical model specifically includes: based on the skin thermodynamic data in the personalized feature dataset, constructing a skin heat conduction model using the Pennes bioheat transfer equation, and calculating and outputting the local temperature field of the skin tissue. The Pennes bioheat transfer equation is:

[0014] ;

[0015] in, c These represent the density, specific heat capacity, and thermal conductivity of skin tissue, respectively. Tissue temperature that varies with space and time; It is a time variable; Blood perfusion rate; , These are blood density, specific heat capacity, and arterial blood temperature, respectively. It is the metabolic heat production rate per unit volume of tissue; It is a divergence operator. It is a gradient operator;

[0016] Based on the mechanical property data in the personalized feature dataset, a skin biomechanical model is constructed using the Mooney-Rivlin hyperelastic model, and the strain energy density of the skin tissue is calculated and output. The strain energy density function of the Mooney-Rivlin hyperelastic model is:

[0017] ;

[0018] in, It is the strain energy per unit volume; and The material constants are obtained by fitting experimental data. and These are the first and second isocompressive strain invariants, respectively. It is the bulk modulus of the material. It is the determinant of the deformable gradient tensor.

[0019] Furthermore, in step S2, constructing the behavioral model specifically includes:

[0020] Based on the immunological and physiological biochemical data in the personalized feature dataset, a set of dynamic equations describing the inflammatory signal transduction axis is constructed.

[0021] By combining cell proliferation rate parameters regulated by local inflammatory factor concentrations, apoptosis rate parameters regulated by pro-apoptotic factor concentrations, and maximum cell carrying density, the ordinary differential equation describing the net growth of the cell population is solved, and the state evolution data of the output behavioral model in the time dimension is calculated. The ordinary differential equation is as follows:

[0022] ;

[0023] in, For cell number or density, For time variables, This is a parameter for cell proliferation rate. This is a parameter for the rate of apoptosis. Maximum cell carrying density;

[0024] In step S2, a low-level data interaction method using variable passing is employed to couple the various models. Specifically, this includes: using the local temperature field output by the physical model as the input parameter of the kinetic equation in the behavioral model, and determining the temperature-dependent reaction rate constant in the kinetic equation using the Arrhenius equation, which is:

[0025] ;

[0026] in, For local temperature field The changing reaction rate constant It is a pre-exponential factor for biochemical reactions. The activation energy of a biochemical reaction. Let be the ideal gas constant. The local temperature field is derived from the physical model.

[0027] Further, in step S2, constructing a rule model and executing hybrid decision-making specifically includes: structuring psoriasis medical knowledge to construct a psoriasis medical knowledge graph stored in a graph database in the form of triples; establishing a hybrid decision-making mechanism between the rule model and the data-based machine learning model; when the rule model and the machine learning model have a prediction conflict regarding non-safety taboo rules, obtaining the confidence score output by the machine learning model, and determining the evidence level weight of the rule model based on the evidence level in the psoriasis medical knowledge graph; calculating dynamic weights based on the confidence score, the evidence level weights, and the historical decision success rate, and performing a weighted fusion calculation on the prediction results of the rule model and the machine learning model based on the dynamic weights to obtain the final decision. The weighted fusion calculation formula is:

[0028] ;

[0029] in, For the final decision result, The prediction result of the rule model. The prediction result of the machine learning model. and Dynamic weights;

[0030] In step S2, a constraint-corrected underlying data interaction method is used to couple the various models, specifically including:

[0031] When the simulation results of the behavioral model exceed the physiological range defined in the psoriasis medical knowledge graph, a penalty term is applied to the optimization objective function of the behavioral model, and the parameters of the behavioral model are updated based on the gradient descent algorithm and the penalty term. The update formula is as follows:

[0032] ;

[0033] in, and These are the behavior model parameters before and after the update. For learning rate, The gradient of the loss function. The gradient of the penalty function; This is the penalty coefficient.

[0034] Further, step S3, which simulates the effect decay process for a single drug-induced stimulus and determines the decay inflection point, specifically includes: calculating the immunogenicity accumulation index inside the personalized multi-scale digital twin; determining the concentration of anti-drug antibody production when the immunogenicity accumulation index exceeds a preset anti-drug antibody production threshold; calculating the dynamically increasing drug clearance rate based on the production concentration and deriving the effective blood drug concentration decrease curve over time; calculating the long-term efficacy prediction curve based on the decrease curve and obtaining the first and second derivatives of the long-term efficacy prediction curve with respect to time; determining the current time point as the decay inflection point when the first derivative changes from negative to positive and the second derivative reaches a local positive maximum, and simultaneously satisfies the conditions that the clinical response improvement rate falls below a preset indicator threshold and the relative increase in drug clearance rate exceeds a critical value.

[0035] Furthermore, in step S3, for the combined drug stimulation simulation, the final predicted effect of the combined drug is calculated by introducing a modified Bliss independence model with synergistic interaction terms, specifically including:

[0036] The effective concentration vector of the second drug in the combination therapy regimen is obtained, and matrix multiplication is performed using the component-target effect strength matrix (characterizing the intensity of the component's effect on the target) and the target-pathway influence matrix (characterizing the weight of the target's influence on the pathway) to obtain the comprehensive regulatory effect vector of the second drug. The matrix multiplication calculation formula is as follows:

[0037] ;

[0038] in, This represents the overall regulatory effect vector of the second drug. This represents the effective concentration vector of the second drug. This is the component-target interaction intensity matrix. Target-pathway influence matrix;

[0039] Determine the first independent effect of the first drug, and the second independent effect of the second drug determined based on the comprehensive regulatory effect vector;

[0040] Extract the target of the first drug and the pathway of the second drug, and determine the synergistic interaction term that characterizes the synergistic relationship between the target and the pathway based on a preset rule function;

[0041] Combining the first independent effect, the second independent effect, and the synergistic interaction term, the final predicted effect of the combined medication is calculated and output using the modified Bliss independence model. The modified Bliss independence model formula is as follows:

[0042] ;

[0043] in, To predict the final effect of combination therapy, The first independent effect of the first drug. This is the second independent effect of the second drug. The collaborative interaction item is used to characterize the collaborative relationship.

[0044] Furthermore, in step S3, before outputting the long-term efficacy prediction curve, a process of optimizing the efficacy prediction model using a domain adaptive algorithm is also included, specifically including:

[0045] The simulation data generated by the personalized multi-scale digital twin is determined as the source domain data, and real clinical data of psoriasis is obtained as the target domain data.

[0046] A recurrent generative adversarial network (RGAN) comprising a generator and a discriminator is constructed. The source domain data and the target domain data are input into the RGAN, and training is performed using a total loss function including adversarial loss and cycle consistency loss to align the feature distributions of the source and target domains, and transfer data is generated. A machine learning model is trained using the transfer data and the real clinical data to construct a combined drug efficacy prediction model. The expression for the total loss function is:

[0047] ;

[0048] in, To generate the total loss function of the adversarial network in a loop, For source domain data, For target domain data, and For the two generators in the network, and For the two corresponding discriminators, For an adversarial loss function with gradient penalty, Let the cycle consistency loss function be... is the weighting coefficient for the cycle consistency loss.

[0049] Furthermore, step S4 involves rolling updates and calibrations of the state vector and model parameters of the personalized multi-scale digital twin. Specifically, this includes: performing a first-order Taylor expansion on the nonlinear system and using an extended Kalman filter framework to process the high-dimensional nonlinear evolution characteristics of the personalized multi-scale digital twin for parameter estimation; when there is a prediction deviation between the actual monitoring data and the model prediction data, applying Bayes' theorem to update the probability distribution of the model parameters and integrating the calibrated new parameters back into the multi-dimensional digital twin model to compensate for the prediction error.

[0050] To address the aforementioned technical problems, a second aspect of the present invention provides a digital twin-driven personalized psoriasis diagnosis and treatment system for performing a digital twin-driven personalized psoriasis diagnosis and treatment method as described in the first aspect, the system comprising:

[0051] The data acquisition module is used to collect multimodal, cross-scale medical data of psoriasis patients and perform standardization, cleaning and noise reduction, and feature extraction to generate personalized feature datasets.

[0052] The digital twin modeling module is communicatively connected to the data acquisition module and is used to construct a personalized multi-scale digital twin based on the personalized feature dataset; wherein, the digital twin modeling module integrates a variable transfer algorithm and a constraint correction algorithm for multi-dimensional model coupling, as well as an intrinsic orthogonal decomposition algorithm for model order reduction;

[0053] The simulation and prediction module is communicatively connected to the digital twin modeling module. It is used to input pharmacological models into the personalized multi-scale digital twin for stimulation simulation, and to perform efficacy attenuation extrapolation, attenuation inflection point determination and synergistic effect calculation of combined drug use, so as to output long-term efficacy prediction curve and final prediction effect.

[0054] The model update and scheme output module is connected to the simulation and prediction module. It is used to dynamically visualize the internal evolution state of the personalized multi-scale digital twin, and to input new monitoring data based on the extended Kalman filter framework to perform rolling updates and calibrations of model parameters, and output personalized treatment recommendations.

[0055] Furthermore, the system is deployed using a cloud-edge-device collaborative architecture; wherein: the device side includes a hardware layer for collecting the multimodal, cross-scale medical data; the edge side includes an edge computing server deployed locally in the hospital for real-time preprocessing, standardization, and preliminary feature extraction of the data collected on the device side; the cloud side includes a high-performance computing cluster for running the personalized multi-scale digital twin, executing the stimulus simulation, and training the combined drug efficacy prediction model; the hardware layer transmits skin image data via the DICOM standard protocol and transmits clinical information and laboratory results data via the HL7 standard protocol to achieve heterogeneous data interoperability with the edge computing server and the high-performance computing cluster.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] (1) Achieved highly personalized and precise diagnosis and treatment: Unlike all existing general models, this invention constructs a digital clone that corresponds one-to-one with a specific psoriasis patient by fusing multimodal cross-scale data and multidimensional models, which can accurately reflect individual differences of patients, thereby achieving truly personalized treatment plan design and efficacy prediction.

[0058] (2) Possesses strong efficacy prediction and mechanism insight capabilities: Through virtual drug stimulation simulation and dynamic deduction, this invention can prospectively predict the long-term efficacy and decay trend of biological agents such as IL-17A inhibitors, and can perform simulation explanations from the intrinsic mechanisms such as anti-drug antibody generation, breaking through the limitations of existing clinical methods and research models that cannot predict long-term effects.

[0059] (3) Accelerates the research and evaluation of innovative combination therapies: For complex therapies such as the combination of biological agents and traditional Chinese medicine, this invention provides an efficient and low-cost virtual laboratory by modifying the Bliss independence model and synergistic interaction terms, which can quickly screen combination schemes with synergistic potential and effectively guide the design of personalized clinical research.

[0060] (4) Enhanced the application value of multimodal medical data: This invention proposes a complete methodology based on digital twins, which effectively integrates fragmented multimodal data such as images, physiology and biochemistry in clinical practice, and transforms static medical data into dynamic knowledge models that can predict the future evolution of the disease, greatly enhancing the high-level utilization value of heterogeneous medical data. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This is an overall flowchart of the digital twin-driven personalized diagnosis and treatment method for psoriasis in this embodiment of the invention;

[0063] Figure 2 This is a block diagram of the module structure of the digital twin diagnosis and treatment system in an embodiment of the present invention;

[0064] Figure 3 This is a logic diagram of the multidimensional model fusion and generative adversarial network algorithm in this embodiment of the invention;

[0065] Figure 4 This is a schematic diagram illustrating the dynamic evolution of the drug stimulation and efficacy decay mechanism in an embodiment of the present invention;

[0066] Figure 5 The curves and data graphs are used to verify the efficacy prediction of therapeutic effects in the embodiments of the present invention. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are protected by this application.

[0068] It should be noted that the terms "comprising," "including," and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application, are intended to cover non-exclusive inclusion. For example, a process, method, terminal, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices. In the claims, specification, and accompanying drawings of this application, relational terms such as "first" and "second" are used merely to distinguish one entity / operation / object from another entity / operation / object, and do not necessarily require or imply any such immediate relationship or order between these entities / operations / objects. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0069] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0070] Currently, the diagnosis and treatment of psoriasis mainly rely on clinical experience and standard treatment guidelines. For moderate to severe psoriasis, biologics such as IL-17A inhibitors have become important treatment methods. In basic research, animal models or in vitro cell models are typically used to study pathogenesis and screen drugs. Although some computer-aided technologies have been used for medical image analysis or pharmacokinetic / pharmacodynamic modeling, these models are mostly general and non-personalized, unable to integrate multi-dimensional, cross-scale vital sign data of patients. Meanwhile, although digital twin technology has been applied in industry, in the clinical application of complex chronic diseases like psoriasis, there is still no dynamic evolutionary personalized system that can integrate molecular, cellular, tissue, and organ information. The existing psoriasis research models, clinical treatment protocols, and computer-aided analysis technologies have the following shortcomings:

[0071] First, the research model is distorted: traditional animal or cell models cannot fully simulate the complex pathophysiological process of human psoriasis and have limitations in reflecting individual differences and the long-term effects of drugs.

[0072] Secondly, the treatment plan lacks foresight: clinical treatment is mostly "trial and error", making it difficult to accurately predict the long-term response of a specific patient to biologics before treatment, and it is impossible to predict and explain the phenomenon of efficacy decay.

[0073] Third, the data is not being fully utilized: the massive amount of multimodal data generated by clinical diagnosis and treatment has not been effectively integrated and deeply mined, making it impossible to form a complete view that dynamically reflects the evolution of the disease.

[0074] Fourth, there is a lack of combined drug simulation platforms: for novel therapies such as the combination of biological agents and traditional Chinese medicine, there is a lack of effective platforms to simulate synergistic mechanisms and predict efficacy, which hinders the development of personalized precision treatment.

[0075] In view of this, embodiments of this application provide a digital twin-driven personalized diagnosis and treatment method and system for psoriasis. On the one hand, by collecting multimodal cross-scale data, a personalized multi-scale digital twin containing multi-dimensional model coupling is constructed, transforming static data into a complete view that dynamically reflects the evolution of the disease. On the other hand, by inputting pharmacological models for virtual stimulus simulation, the long-term efficacy of single drugs and their attenuation inflection points can be prospectively and dynamically extrapolated, and the synergistic effect of combination drugs can be evaluated using a modified Bliss model. Furthermore, by extending the Kalman filter framework to access newly added monitoring data for rolling updates, truly achieving accurate recommendations for personalized diagnosis and treatment plans.

[0076] To illustrate the technical solution of this application, specific embodiments are described below in conjunction with the accompanying drawings.

[0077] Please refer to Figure 1 This application provides an overall flowchart of a digital twin-driven personalized diagnosis and treatment method for psoriasis.

[0078] Specifically, the aforementioned digital twin-driven personalized treatment method for psoriasis may include the following steps S1 to S4.

[0079] Step S1: Multimodal data acquisition and preprocessing: Collect multimodal cross-scale medical data of psoriasis patients, and standardize, clean and denoise the multimodal cross-scale medical data and extract features to generate personalized feature datasets.

[0080] The system acquires patients' skin imaging data, clinical information, and laboratory biochemical indicators in real time or periodically through edge hardware devices. By standardizing the data to eliminate differences in data format between different devices and hospitals, and removing artifacts and noise, it extracts key geometric, thermodynamic, and immunological features to form a personalized feature dataset specific to the individual, providing a data foundation for the subsequent construction of a high-fidelity digital twin.

[0081] Step S2, Multidimensional Digital Twin Model Construction and Order Reduction: Based on the personalized feature dataset, a multidimensional digital twin model including a geometric model, a physical model, a behavioral model, and a rule model is constructed; the underlying data interaction method of variable transfer and constraint correction is used to couple the models, and the intrinsic orthogonal decomposition method is used to reduce the order of the models to construct a personalized multi-scale digital twin.

[0082] In this step, the specific model building and underlying data interaction coupling process is as follows: Figure 3 As shown in stage 1:

[0083] First, a physical model is constructed. Based on the skin thermodynamic data in the personalized feature dataset, a skin heat conduction model is built using the Pennes bioheat transfer equation to calculate and output the local temperature field of the skin tissue. The Pennes bioheat transfer equation is as follows:

[0084] ;

[0085] in, c These represent the density, specific heat capacity, and thermal conductivity of skin tissue, respectively. Tissue temperature that varies with space and time; It is a time variable; Blood perfusion rate; , These are blood density, specific heat capacity, and arterial blood temperature, respectively. It is the metabolic heat production rate per unit volume of tissue; It is a divergence operator. It is a gradient operator. Through this equation, it is possible to accurately simulate the increased microcirculation blood flow and abnormal temperature distribution characteristics in psoriatic lesions caused by local inflammation.

[0086] Simultaneously, based on the mechanical property data in the personalized feature dataset, a skin biomechanical model is constructed using the Mooney-Rivlin hyperelastic model to calculate the strain energy density of the skin tissue. The strain energy density function expression of the Mooney-Rivlin hyperelastic model is as follows:

[0087] ;

[0088] in, It is the strain energy per unit volume; and These are material constants obtained by fitting experimental data, reflecting the shear behavior of the material; and These are the first and second isochoric strain invariants, respectively, which are calculated from the deformation gradient tensor; It is the bulk modulus of a material, describing its compressibility; It is the determinant of the deformation gradient tensor, representing volume changes. This model aims to digitally characterize the changes in the mechanical properties of skin lesions caused by epidermal thickening and hyperkeratosis in psoriasis patients.

[0089] Secondly, a behavioral model is constructed. Combining cell proliferation rate parameters regulated by local inflammatory factor concentrations, cell apoptosis rate parameters regulated by pro-apoptotic factor concentrations, and maximum cell carrying density, the ordinary differential equation describing the net growth of the cell population is solved. The state evolution data of the behavioral model over time is then calculated and output. The ordinary differential equation is as follows:

[0090] ;

[0091] in, This refers to the number or density of cells. It is a time variable; This is a parameter representing the cell proliferation rate, which is a dynamic function regulated by the concentration of local inflammatory factors. The apoptosis rate parameter is affected by the concentration of pro-apoptotic factors. The maximum cell carrying density characterizes the contact inhibition effect. This equation precisely describes the process of excessive proliferation and abnormal differentiation of keratinocytes unique to psoriasis.

[0092] In this process, the physical model and the behavioral model use a low-level data interaction method of variable passing: the local temperature field output by the physical model is used as the input parameter of the kinetic equation in the behavioral model, and the reaction rate constant that changes with temperature in the kinetic equation is determined by the Arrhenius equation.

[0093] ;

[0094] in, For local temperature field The changing reaction rate constant; It is a pre-exponential factor for biochemical reactions; It is the activation energy of a biochemical reaction; It is the ideal gas constant; This represents the local temperature transferred from the physical model. This transfer mechanism directly maps macroscopic thermodynamic changes into physical alterations in microscopic cellular metabolic rates.

[0095] Finally, a rule-based model is constructed and a hybrid decision-making process is implemented. A psoriasis medical knowledge graph, stored in a graph database as triples, is built; a hybrid decision-making mechanism is established between the rule-based model and a data-driven machine learning model. When prediction conflicts occur, dynamic weights are calculated for weighted fusion, and the weighted fusion calculation formula is as follows:

[0096] ;

[0097] in, For the final decision result; The prediction result of the rule model; The prediction result of the machine learning model; and The weights are dynamic, and their magnitude is determined by the confidence level of the machine learning model, the evidence level of the rule, and the success rate of historical decisions.

[0098] Regarding the coupling between the rule model and the behavior model, a constraint correction approach is adopted: when the simulation results of the behavior model exceed the physiological range defined by the medical knowledge graph, a penalty term is applied to the optimization objective function of the behavior model, and the parameters are updated using the gradient descent algorithm.

[0099] ;

[0100] in, and These are the behavior model parameters before and after the update; The learning rate; The gradient of the loss function; This is a penalty function derived from the rule-based model; its value increases sharply when the parameters cross the knowledge boundary. This is the penalty coefficient.

[0101] Step S3, Virtual Treatment Simulation and Efficacy Prediction: Input the pharmacological model into the personalized multi-scale digital twin for stimulation simulation; wherein, for single-drug stimulation simulation, the efficacy decay process is dynamically deduced based on a preset triggering mechanism to output the long-term efficacy prediction curve and decay inflection point; for combined drug stimulation simulation, the final predicted effect of combined drug is calculated by introducing a modified Bliss independence model with synergistic interaction terms.

[0102] Specifically, please refer to Figure 4 For a single drug (such as an IL-17A inhibitor), the immunogenicity accumulation index within the twins is calculated. When it exceeds the anti-drug antibody production threshold, the concentration of anti-drug antibodies is determined, and the dynamic increase in drug clearance is calculated accordingly, thereby extrapolating the effective blood drug concentration decline curve. When the first derivative of the long-term efficacy prediction curve turns from negative to positive, and the second derivative reaches a local positive maximum, it is determined as the efficacy decay inflection point.

[0103] For combination therapy (such as biological agents like IL-17A inhibitors combined with active ingredients from traditional Chinese medicines like Xiaozheng Decoction), the comprehensive regulatory effect vector of the second drug is first calculated using matrix multiplication:

[0104] ;

[0105] in, This represents the overall regulatory effect vector of the second drug; is the effective concentration vector of the second drug; is the component-target interaction strength matrix characterizing the intensity of the component's effect on the target. The target-pathway influence matrix is ​​used to characterize the weight of the target's influence on the pathway.

[0106] Subsequently, the final predicted effects were summed using a modified Bliss independence model:

[0107] ;

[0108] in, The final predicted effect of combined drug use; This is the first independent effect of the first drug; The second independent effect of the second drug (based on) Sure); The collaborative interaction item is used to characterize the collaborative relationship.

[0109] Before outputting the final curve, it is also like Figure 3 As shown in stages 2 and 3, a recurrent generative adversarial network containing a generator and a discriminator is used for domain adaptive optimization to align the source domain data of the Siamese simulation. Compared with real clinical target domain data Its total loss function is:

[0110] ;

[0111] in, The total loss function for generating the adversarial network is generated iteratively; and These are two generators in the network; and These are the two corresponding discriminators; For an adversarial loss function with gradient penalty; Let this be the cycle consistency loss function; is the weighting coefficient for the cycle consistency loss.

[0112] Step S4, Model Rolling Update and Solution Output: Obtain new monitoring data of the psoriasis patients, use the extended Kalman filter framework in combination with the new monitoring data to perform rolling updates and calibrations on the state vector and model parameters of the personalized multi-scale digital twin, and output personalized treatment recommendation solutions based on the long-term efficacy prediction curve and the final prediction effect.

[0113] Since the human body is a high-dimensional nonlinear system, this embodiment performs a first-order Taylor expansion on the nonlinear system and uses an extended Kalman filter framework combined with detection constraints to handle its evolutionary characteristics. When there is a deviation between the actual monitoring data and the model prediction data, Bayes' theorem is applied to update the probability distribution of the model parameters, and the calibrated new parameters are integrated back into the twin to compensate for the error. This ensures that the treatment recommendation plan output by the system always closely follows the patient's real-time vital signs and changes.

[0114] To verify the accuracy of the diagnostic and treatment plan and efficacy prediction model described in this invention, such as Figure 5 As shown in the figure, this embodiment extracts real clinical follow-up data (PASI score) from the historical validation set and compares it with the virtual prediction results of the digital twin. The comparison graph of the PASI score prediction at 24 weeks and the real follow-up shows that this system accurately reproduces the efficacy decline trend. Furthermore, in the PASI 75 response prediction, the area under the ROC curve (AUC) of the model of this invention reached 0.92, significantly better than the 0.75 of the traditional statistical model. In addition, the MSE convergence curve of the machine learning model indicates that the model has good generalization ability, and the overall predictive efficacy (RMSE, MAE, R²) all meet the clinically usable standards.

[0115] Please refer to Figure 2 This application provides a module structure diagram of a digital twin-driven personalized psoriasis diagnosis and treatment system. Specifically, the digital twin-driven personalized psoriasis diagnosis and treatment system for performing the above methods may include:

[0116] The data acquisition module is used to collect multimodal, cross-scale medical data of psoriasis patients and perform standardization, cleaning and noise reduction, and feature extraction to generate personalized feature datasets.

[0117] A digital twin modeling module, which is communicatively connected to the data acquisition module, is used to construct a personalized multi-scale digital twin based on the personalized feature dataset; wherein, the digital twin modeling module integrates a variable transfer algorithm and a constraint correction algorithm for multi-dimensional model coupling, as well as an intrinsic orthogonal decomposition algorithm for model order reduction;

[0118] The simulation and prediction module is communicatively connected to the digital twin modeling module. It is used to input a pharmacological model into the personalized multi-scale digital twin for stimulation simulation, and to perform efficacy attenuation extrapolation, attenuation inflection point determination and synergistic effect calculation of combined drug use, so as to output long-term efficacy prediction curve and final prediction effect.

[0119] The model update and solution output module is communicatively connected to the simulation and prediction module. It is used to dynamically visualize the internal evolution state of the personalized multi-scale digital twin, and to input newly added monitoring data based on the extended Kalman filter framework with detection constraints to perform rolling updates and calibrations of model parameters, and output personalized treatment recommendations.

[0120] In a preferred implementation, the system employs a cloud-edge-device collaborative architecture. The device side includes a hardware layer for acquiring multimodal data; the edge side includes edge computing servers deployed locally in the hospital, which communicate with the hardware layer for real-time preprocessing; the cloud side includes a high-performance computing cluster, which communicates with the edge computing servers to execute high-performance digital twin operation and stimulus simulation. The hardware layer is configured to transmit skin image data via the DICOM standard protocol and clinical information and laboratory results data via the HL7 standard protocol, completely breaking down interoperability barriers between heterogeneous medical data.

[0121] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A digital twin-driven personalized diagnosis and treatment method for psoriasis, characterized in that, Includes the following steps: Step S1: Multimodal data acquisition and preprocessing: Collect multimodal cross-scale medical data of psoriasis patients, and standardize, clean and denoise the multimodal cross-scale medical data and extract features to generate personalized feature datasets; Step S2, Multidimensional Digital Twin Model Construction and Order Reduction: Based on the personalized feature dataset, a multidimensional digital twin model including a geometric model, a physical model, a behavioral model, and a rule model is constructed; the underlying data interaction method of variable transfer and constraint correction is used to couple the models, and the intrinsic orthogonal decomposition method is used to reduce the order of the models to construct a personalized multi-scale digital twin. Step S3, Virtual Treatment Simulation and Efficacy Prediction: Input the pharmacological model into the personalized multi-scale digital twin for stimulation simulation; wherein, for single drug stimulation simulation, the efficacy decay process is dynamically deduced based on the preset triggering mechanism to output the long-term efficacy prediction curve and decay inflection point; for combined drug stimulation simulation, the final predicted effect of combined drug is calculated by introducing a modified Bliss independence model with synergistic interaction terms. Step S4, Model Rolling Update and Solution Output: Obtain new monitoring data of the psoriasis patients, use the extended Kalman filter framework in combination with the new monitoring data to perform rolling updates and calibrations on the state vector and model parameters of the personalized multi-scale digital twin, and output personalized treatment recommendation solutions based on the long-term efficacy prediction curve and the final prediction effect.

2. The digital twin-driven personalized diagnosis and treatment method for psoriasis according to claim 1, characterized in that, The construction of the physical model in step S2 specifically includes: Based on the skin thermodynamic data in the personalized feature dataset, a skin heat conduction model is constructed using the Pennes bioheat transfer equation to calculate and output the local temperature field of the skin tissue. The Pennes bioheat transfer equation is as follows: ； in, c These represent the density, specific heat capacity, and thermal conductivity of skin tissue, respectively. Tissue temperature that varies with space and time; It is a time variable; Blood perfusion rate; , These are blood density, specific heat capacity, and arterial blood temperature, respectively. It is the metabolic heat production rate per unit volume of tissue; It is a divergence operator. It is a gradient operator; Based on the mechanical property data in the personalized feature dataset, a skin biomechanical model is constructed using the Mooney-Rivlin hyperelastic model, and the strain energy density of the skin tissue is calculated and output. The strain energy density function of the Mooney-Rivlin hyperelastic model is: ； in, It is the strain energy per unit volume; and The material constants are obtained by fitting experimental data. and These are the first and second isocompressive strain invariants, respectively. It is the bulk modulus of the material. It is the determinant of the deformable gradient tensor.

3. The digital twin-driven personalized diagnosis and treatment method for psoriasis according to claim 2, characterized in that, The construction of the behavioral model in step S2 specifically includes: Based on the immunological and physiological biochemical data in the personalized feature dataset, a set of dynamic equations describing the inflammatory signal transduction axis is constructed. By combining cell proliferation rate parameters regulated by local inflammatory factor concentrations, apoptosis rate parameters regulated by pro-apoptotic factor concentrations, and maximum cell carrying density, the ordinary differential equation describing the net growth of the cell population is solved, and the state evolution data of the output behavioral model in the time dimension is calculated. The ordinary differential equation is as follows: ； in, For cell number or density, For time variables, This is a parameter for cell proliferation rate. This is a parameter for the rate of apoptosis. Maximum cell carrying density; In step S2, a low-level data interaction method using variable passing is employed to couple the various models. Specifically, this includes: using the local temperature field output by the physical model as the input parameter of the kinetic equation in the behavioral model, and determining the temperature-dependent reaction rate constant in the kinetic equation using the Arrhenius equation, which is: ； in, For local temperature field The changing reaction rate constant, It is a pre-exponential factor for biochemical reactions. The activation energy of a biochemical reaction. Let be the ideal gas constant. The local temperature field is derived from the physical model.

4. The digital twin-driven personalized diagnosis and treatment method for psoriasis according to claim 3, characterized in that, Step S2, which involves constructing a rule model and executing hybrid decision-making, specifically includes: Structured processing of medical knowledge related to psoriasis is performed to construct a psoriasis medical knowledge graph stored in a graph database in the form of triples. A hybrid decision-making mechanism is established between the rule model and the data-based machine learning model. When the rule model and the machine learning model have a prediction conflict regarding non-safety contraindication rules, the confidence score output by the machine learning model is obtained, and the evidence level weight of the rule model is determined according to the evidence level in the psoriasis medical knowledge graph. Dynamic weights are calculated based on the confidence score, the evidence level weights, and the historical decision success rate. The prediction results of the rule model and the machine learning model are then weighted and fused according to the dynamic weights to obtain the final decision. The weighted fusion calculation formula is as follows: ； in, For the final decision result, The prediction result of the rule model. The prediction result of the machine learning model. and Dynamic weights; In step S2, a constraint-corrected underlying data interaction method is used to couple the various models. Specifically, this includes: when the simulation results of the behavioral model exceed the physiological range defined in the psoriasis medical knowledge graph, a penalty term is applied to the optimization objective function of the behavioral model, and the parameters of the behavioral model are updated based on the gradient descent algorithm and the penalty term. The update formula is as follows: ； in, and These are the behavior model parameters before and after the update. For learning rate, The gradient of the loss function. The gradient of the penalty function; This is the penalty coefficient.

5. The digital twin-driven personalized diagnosis and treatment method for psoriasis according to claim 1, characterized in that, Step S3, which simulates the effect of a single drug and dynamically extrapolates the attenuation process to determine the attenuation inflection point, specifically includes: Calculate the cumulative immunogenicity index inside the personalized multiscale digital twin; When the immunogenicity accumulation index exceeds the preset anti-drug antibody production threshold, the concentration of anti-drug antibody production is determined. The dynamically increasing drug clearance rate is calculated based on the generated concentration, and the effective blood drug concentration decrease curve over time is derived. The long-term efficacy prediction curve is calculated based on the descent curve, and the first and second derivatives of the long-term efficacy prediction curve with respect to time are obtained. When the first derivative changes from negative to positive and the second derivative reaches a local positive maximum, and the clinical response improvement rate falls below a preset threshold and the relative increase in drug clearance exceeds a critical value, the current time point is determined as the decay inflection point.

6. The digital twin-driven personalized diagnosis and treatment method for psoriasis according to claim 5, characterized in that, In step S3, for the simulation of combined drug use incentives, the final predicted effect of combined drug use is calculated by introducing a modified Bliss independence model with synergistic interaction terms. Specifically, this includes: The effective concentration vector of the second drug in the combination therapy regimen is obtained, and matrix multiplication is performed using the component-target effect strength matrix (characterizing the intensity of the component's effect on the target) and the target-pathway influence matrix (characterizing the weight of the target's influence on the pathway) to obtain the comprehensive regulatory effect vector of the second drug. The matrix multiplication calculation formula is as follows: ； in, This represents the overall regulatory effect vector of the second drug. This represents the effective concentration vector of the second drug. This is the component-target interaction intensity matrix. Target-pathway influence matrix; Determine the first independent effect of the first drug, and the second independent effect of the second drug determined based on the comprehensive regulatory effect vector; Extract the target of the first drug and the pathway of the second drug, and determine the synergistic interaction term that characterizes the synergistic relationship between the target and the pathway based on a preset rule function; Combining the first independent effect, the second independent effect, and the synergistic interaction term, the final predicted effect of the combined medication is calculated and output using the modified Bliss independence model. The modified Bliss independence model formula is as follows: ； in, To predict the final effect of combination therapy, The first independent effect of the first drug. This is the second independent effect of the second drug. The collaborative interaction item is used to characterize the collaborative relationship.

7. The digital twin-driven personalized diagnosis and treatment method for psoriasis according to claim 1, characterized in that, Before outputting the long-term efficacy prediction curve, step S3 further includes optimizing the efficacy prediction model using a domain adaptive algorithm, specifically including: The simulation data generated by the personalized multi-scale digital twin is determined as the source domain data, and real clinical data of psoriasis is obtained as the target domain data. A recurrent generative adversarial network (RGAN) comprising a generator and a discriminator is constructed. The source domain data and the target domain data are input into the RGAN, and training is performed using a total loss function including adversarial loss and cycle consistency loss to align the feature distributions of the source and target domains, and transfer data is generated. A machine learning model is trained using the transfer data and the real clinical data to construct a combined drug efficacy prediction model. The expression for the total loss function is: ； in, To generate the total loss function of the adversarial network in a loop, For source domain data, For target domain data, and For the two generators in the network, and For the two corresponding discriminators, For an adversarial loss function with gradient penalty, Let the cycle consistency loss function be... is the weighting coefficient for the cycle consistency loss.

8. The digital twin-driven personalized diagnosis and treatment method for psoriasis according to claim 1, characterized in that, Step S4 involves rolling updates and calibrations of the state vector and model parameters of the personalized multi-scale digital twin, specifically including: By performing a first-order Taylor expansion on the nonlinear system, the extended Kalman filter framework is used to process the high-dimensional nonlinear evolution characteristics of personalized multi-scale digital twins for parameter estimation. When the actual monitoring data deviates from the model prediction data, Bayes' theorem is applied to update the probability distribution of the model parameters, and the calibrated new parameters are integrated back into the multidimensional digital twin model to compensate for the prediction error.

9. A digital twin-driven personalized psoriasis diagnosis and treatment system, used to execute the digital twin-driven personalized psoriasis diagnosis and treatment method as described in any one of claims 1 to 8, characterized in that, The system includes: The data acquisition module is used to collect multimodal, cross-scale medical data of psoriasis patients and perform standardization, cleaning and noise reduction, and feature extraction to generate personalized feature datasets. A digital twin modeling module, which is communicatively connected to the data acquisition module, is used to construct a personalized multi-scale digital twin based on the personalized feature dataset; wherein, the digital twin modeling module integrates a variable transfer algorithm and a constraint correction algorithm for multi-dimensional model coupling, as well as an intrinsic orthogonal decomposition algorithm for model order reduction; The simulation and prediction module is communicatively connected to the digital twin modeling module. It is used to input a pharmacological model into the personalized multi-scale digital twin for stimulation simulation, and to perform efficacy attenuation extrapolation, attenuation inflection point determination and synergistic effect calculation of combined drug use, so as to output long-term efficacy prediction curve and final prediction effect. The model update and scheme output module is communicatively connected to the simulation and prediction module. It is used to input new monitoring data based on the extended Kalman filter framework to perform rolling updates and calibrations of model parameters, and output personalized treatment recommendations.

10. The digital twin-driven personalized psoriasis diagnosis and treatment system according to claim 9, characterized in that, The system is deployed using a cloud-edge-device collaborative architecture, wherein: The endpoint includes a hardware device layer for acquiring the multimodal, multiscale medical data; The edge side includes an edge computing server deployed locally in the hospital. The edge computing server is communicatively connected to the hardware device layer and is used to perform real-time preprocessing, standardization, and preliminary feature extraction on the data collected from the edge side. The cloud side includes a high-performance computing cluster, which is communicatively connected to the edge computing server and is used to run the personalized multi-scale digital twin, perform the stimulus simulation, and train the combined drug efficacy prediction model. The hardware device layer is configured to transmit skin image data via the DICOM standard protocol and transmit clinical information and laboratory results data to the edge computing server via the HL7 standard protocol.

Images