CRRT data omnibearing acquisition and optimization processing system based on artificial intelligence

Abstract

The invention relates to the field of medical artificial intelligence, and discloses a CRRT data omnibearing acquisition and optimization processing system based on artificial intelligence, and the system comprises an acquisition calibration module which is used for acquiring CRRT equipment parameters, physiological signals and laboratory inspection data, forming multi-source data, and carrying out the time alignment and space calibration of the multi-source data, and obtaining calibration data; the modeling estimation module is used for constructing a space-time dynamic model for substance transportation and biochemical reaction in the CRRT treatment process based on a random partial differential equation, estimating a state variable in combination with calibration data, and generating high-dimensional state data; and the fusion reasoning module is used for carrying out dimensionality reduction and adversarial denoising processing on the high-dimensional state data. Through multi-source acquisition and space-time calibration of CRRT equipment parameters, physiological signals and laboratory data, a treatment data basis is constructed, space-time dynamics modeling of substance transportation and biochemical reaction is performed based on a random partial differential equation, and dynamic description of a treatment process is realized.

Description

Technical Field

[0001] The present invention relates to the field of medical artificial intelligence technology, and specifically to an artificial intelligence-based CRRT data all-round acquisition and optimization processing system. Background Art

[0002] Continuous renal replacement therapy (CRRT), a crucial life-support method in critical care medicine, uses an extracorporeal circulation device to continuously remove metabolic waste and maintain water and electrolyte balance. It is widely used to treat patients with critical illnesses such as acute kidney injury and sepsis. The CRRT treatment process involves dynamic adjustments to multiple parameters, including blood flow, ultrafiltration rate, and replacement fluid composition. It also requires real-time adjustments to complex changes in the patient's physiological parameters (such as ionized calcium concentration and acid-base balance), placing extremely high demands on the accuracy and real-time performance of treatment parameters.

[0003] Traditional CRRT optimization processing systems often rely on statistical association models (such as regression analysis and machine learning classification algorithms) to train historical treatment data to establish the correlation between treatment parameters and physiological indicators. However, they can only reveal the correlation between variables and cannot analyze the causal mechanism. As a result, during the parameter adjustment process, it is impossible to distinguish between true causal effects and interference from confounding factors, causing unexpected treatment side effects. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides an artificial intelligence-based comprehensive CRRT data acquisition and optimization processing system to solve the problem that traditional CRRT optimization processing systems are unable to distinguish between true causal effects and interference from confounding factors during the parameter adjustment process, thereby causing unexpected treatment side effects.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: an artificial intelligence-based CRRT data comprehensive acquisition and optimization processing system, comprising: Acquisition and calibration module: used to collect CRRT equipment parameters, physiological signals and laboratory test data to form multi-source data, and perform time alignment and spatial calibration on them to obtain calibration data; Modeling and Estimation Module: This module is used to construct a spatiotemporal dynamics model of material transport and biochemical reactions during CRRT treatment based on stochastic partial differential equations, and to estimate state variables using calibration data to generate high-dimensional state data. Fusion reasoning module: used to perform dimensionality reduction and adversarial denoising on high-dimensional state data to generate low-dimensional feature vectors. Based on the low-dimensional feature vectors, it constructs a causal relationship map between treatment parameters and physiological indicators and calculates the intervention effect. Control strategy module: used to build a model predictive control framework based on the spatiotemporal dynamics model, generate a basic treatment plan that meets clinical constraints based on the intervention effect, and adjust its parameters individually through deep reinforcement learning; Instruction generation module: used to integrate basic treatment plans and personalized adjustment parameters, generate device control instructions, perform real-time control at the edge through the edge-cloud collaborative architecture, and perform iterative training of spatiotemporal dynamics models and deep reinforcement learning strategies in the cloud.

[0006] By adopting the above technical solutions, multi-source acquisition and spatiotemporal calibration of CRRT equipment parameters, physiological signals, and laboratory data are carried out to build a treatment data foundation. Based on the spatiotemporal dynamics modeling of material transport and biochemical reactions using stochastic partial differential equations, a dynamic characterization of the treatment process is achieved. Through dimensionality reduction, denoising, and causal relationship map construction, the intervention effect of treatment parameters on physiological indicators is quantified. Combined with model predictive control and deep reinforcement learning, a treatment plan with both clinical safety and individual adaptability is generated. Real-time control and model iteration are achieved through an edge-cloud collaborative architecture, thereby clarifying the true action path of treatment parameters and physiological indicators through the causal map, avoiding decision-making bias caused by confounding factors. With the help of stochastic partial differential equation modeling and deep reinforcement learning, the model's adaptability to individual differences and the real-time performance of parameter adjustment are simultaneously improved, achieving scientific analysis and precise intervention of causal relationships during CRRT treatment. This solves the problem that traditional CRRT optimization processing systems cannot distinguish between true causal effects and interference from confounding factors during parameter adjustment, causing unexpected treatment side effects.

[0007] Preferably, the acquisition and calibration module includes an acquisition unit, a communication unit and a calibration unit. The acquisition unit is used to acquire CRRT equipment parameters, physiological signals and laboratory test data through a three-level sensor network to form multi-source data. The communication unit is used to perform real-time interaction of various data through HL7 FHIR interface protocol conversion. The calibration unit is used to perform time alignment and spatial calibration on the multi-source data through hardware synchronous clock triggered sampling combined with fractional calculus interpolation algorithm to obtain calibration data.

[0008] Preferably, the time alignment adopts fractional calculus interpolation, and its formula is: ,in, is the optimization parameter and its value range is , is the gamma function, is the interpolation signal at the target time point, is the first-order derivative of the original signal, is the signal acquisition time window, Align the target time point.

[0009] Preferably, the modeling and estimation module includes a discretization unit, a parameter identification unit and a state estimation unit. The discretization unit is used to convert the material transport and biochemical reactions in the CRRT treatment process into a weak form of a three-dimensional stochastic partial differential equation. The parameter identification unit is used to optimize the parameters of the three-dimensional stochastic partial differential equation through variational Bayesian inference. The state estimation unit is used to solve the three-dimensional stochastic partial differential equation based on calibration data, estimate state variables, and generate high-dimensional state data.

[0010] Preferably, the stochastic partial differential equation is , where the state vector Including ion calcium concentration field, citrate concentration field and blood flow velocity field, diffusion matrix Is a diagonal matrix, the elements are the diffusion coefficients of each substance, the reaction terms Describe the kinetics of biochemical reactions. is the noise intensity matrix, is the Wiener process; The optimization is to optimize the parameters of the three-dimensional stochastic partial differential equation by maximizing the lower bound of evidence. The expression of the maximization lower bound of evidence is: ,in, is the variational distribution, is the prior distribution, is the observation data, is the variational distribution The expectation operator under is the Kullback-Leibler divergence.

[0011] Preferably, the fusion reasoning module includes a dimensionality reduction unit, a denoising unit and a causal modeling unit. The dimensionality reduction unit is used to perform manifold dimensionality reduction on high-dimensional state data through a local linear embedding algorithm. The denoising unit is used to perform denoising on the reduced high-dimensional state data based on a conditional generative adversarial network to generate a low-dimensional feature vector. The causal modeling unit is used to construct a causal relationship map between treatment parameters and physiological indicators based on the low-dimensional feature vector, and calculate the intervention effect through Do-Calculus.

[0012] Preferably, the locally linear embedding algorithm performs dimensionality reduction by solving the following optimization problem: ,in, is the low-dimensional manifold eigenvector, To locally reconstruct the weight matrix, by minimizing Sure, is the original high-dimensional data; The formula for calculating the intervention effect is ,in, is the treatment parameter, As physiological indicators, for The parent node set of for Medium External variables, For intervention hour The probability distribution of .

[0013] Preferably, the control strategy module includes a basic control unit and an intelligent optimization unit. The basic control unit is used to construct a model predictive control framework based on the spatiotemporal dynamics model, and generate a basic treatment plan that meets clinical constraints in combination with the intervention effect. The clinical constraints include blood flow range, ionized calcium concentration threshold and ultrafiltration rate upper limit. The cost function of the model predictive control framework is ,in, For the prediction time domain, To control the time domain, is the predicted state vector, is the reference trajectory, is the control variable, and are weight matrices of state and control variables respectively. The intelligent optimization unit is used to adjust the parameters of the basic treatment plan through deep reinforcement learning. The reward function of the deep reinforcement learning is ,in, To control the smoothing coefficient, is the complication avoidance bonus coefficient, is the indicator function, As the control variable of the basic treatment plan, is the reference state vector.

[0014] Preferably, the instruction generation module includes a weight fusion unit, a safety verification unit and a collaborative architecture unit. The weight fusion unit is used to dynamically adjust the weights of the basic treatment plan and the personalized adjustment parameters based on the treatment stage, perform fusion, and output parameter adjustment instructions. The safety verification unit is used to filter abnormal instructions from the parameter adjustment instructions based on causal constraints and clinical thresholds, and generate device control instructions. The collaborative architecture unit is used to perform real-time control according to the device control instructions at the edge through the edge-cloud collaborative architecture, and perform iterative training of spatiotemporal dynamics models and deep reinforcement learning strategies through federated learning in the cloud.

[0015] The artificial intelligence-based CRRT data comprehensive acquisition and optimization processing method is applied to the artificial intelligence-based CRRT data comprehensive acquisition and optimization processing system described above, and includes the following steps: Acquisition and calibration: Collect CRRT equipment parameters, physiological signals, and laboratory test data to form multi-source data, and perform time alignment and spatial calibration on them to obtain calibration data; Modeling and estimation: Based on stochastic partial differential equations, a spatiotemporal dynamics model of material transport and biochemical reactions during CRRT treatment is constructed. State variables are estimated using calibration data to generate high-dimensional state data. Fusion reasoning: Perform dimensionality reduction and adversarial denoising on high-dimensional state data to generate low-dimensional feature vectors. Based on these low-dimensional feature vectors, a causal relationship map between treatment parameters and physiological indicators is constructed to calculate the intervention effect. Control strategy: A model predictive control framework is constructed based on the spatiotemporal dynamics model. The intervention effect is combined to generate a basic treatment plan that meets clinical constraints, and deep reinforcement learning is used to adjust its parameters in a personalized manner. Instruction generation: Integrates basic treatment plans and personalized adjustment parameters to generate device control instructions, performs real-time control at the edge through an edge-cloud collaborative architecture, and performs iterative training of spatiotemporal dynamics models and deep reinforcement learning strategies in the cloud.

[0016] The present invention provides an artificial intelligence-based CRRT data all-round acquisition and optimization processing system, which has the following beneficial effects: 1. The present invention builds a treatment data foundation through multi-source acquisition and spatiotemporal calibration of CRRT equipment parameters, physiological signals and laboratory data, and realizes dynamic characterization of the treatment process by spatiotemporal dynamics modeling of material transport and biochemical reactions based on stochastic partial differential equations. Through dimensionality reduction, denoising and causal relationship map construction, the intervention effect of treatment parameters on physiological indicators is quantified. Combined with model predictive control and deep reinforcement learning, a treatment plan with both clinical safety and individual adaptability is generated. Real-time control and model iteration are achieved through an edge-cloud collaborative architecture, thereby realizing scientific analysis and precise intervention of causal relationships in the CRRT treatment process.

[0017] 2. The present invention introduces stochastic partial differential equations to mathematically describe the material transport and biochemical reaction processes such as ion calcium diffusion and citrate complexation in CRRT treatment. The numerical solution framework is constructed by discretizing the time and space domain using the finite element method, and variational Bayesian inference is used to optimize model parameters such as the diffusion coefficient and reaction rate, so that the equation parameters are accurately matched with the statistical characteristics of the patient's actual treatment data. In this way, the stochastic finite element method is combined with the Euler-Markov time discretization technology to solve the dynamic evolution distribution of state variables such as the ion calcium concentration field and the blood flow velocity field in real time, and generate high-dimensional state data that characterize the spatiotemporal characteristics of the treatment process. This better fits the dynamic changes of material distribution and physiological indicators in CRRT treatment, and realizes refined mathematical modeling of the treatment process.

[0018] 3. The present invention uses a local linear embedding algorithm to perform manifold dimensionality reduction on high-dimensional state data, retains the key state patterns of CRRT treatment, and then uses a conditional generative adversarial network to remove data noise and construct a high-fidelity low-dimensional feature vector. Based on low-dimensional features, the PC algorithm is used to construct a causal relationship map between treatment parameters and physiological indicators, clarify the direct causal paths and indirect influence mechanisms between variables, and quantify the intervention effect of treatment parameter adjustment on physiological indicators through Do-Calculus, so as to distinguish between true causal effects and interference from confounding factors, avoid parameter adjustment errors caused by false associations, provide a traceable physical mechanism basis for treatment decisions, and improve the safety and clinical interpretability of CRRT treatment.

[0019] 4. The present invention uses model predictive control based on spatiotemporal dynamics models and clinical constraints, and utilizes an optimized cost function to generate a basic treatment plan that meets safety specifications, ensuring that the treatment parameter adjustments are within the clinically acceptable range; deep reinforcement learning uses real-time state deviation, control smoothness, and complication avoidance as reward functions to perform personalized parameter adjustments on the basic plan, and learn the optimal treatment parameters corresponding to the patient's individual characteristics. This not only maintains the bottom line of treatment safety through model predictive control, but also uses deep reinforcement learning to explore the optimal solution under individual differences, achieving an organic unity of standardization and personalization of the treatment plan, and improving the adaptability and efficacy of CRRT treatment for different patients.

[0020] 5. The present invention dynamically adjusts the weights of the basic plan and personalized parameters according to the treatment stage, and generates compliant device control instructions after filtering abnormal instructions by combining causal constraints and clinical thresholds; the edge end relies on the HL7FHIR protocol interface to drive the CRRT equipment execution control in real time to ensure the millisecond-level real-time performance of the treatment response; the cloud end is based on federated learning technology to associate the real-time treatment data collected by the edge end with the spatiotemporal dynamics model and deep reinforcement learning strategy for training, and iteratively optimize the model parameters and control strategy, thereby meeting the strict real-time requirements of CRRT treatment and adapting to changes in patients' conditions and new clinical scenarios through continuous learning, thereby continuously improving the ability to respond to complex treatment environments in long-term use, and promoting the development of CRRT treatment towards intelligence and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is the system architecture diagram of the artificial intelligence-based CRRT data all-round acquisition and optimization processing system proposed in the present invention; Figure 2 This is a flow chart of the method for comprehensive acquisition and optimization processing of CRRT data based on artificial intelligence proposed in the present invention. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solution of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0023] Please see the attached Figure 1 The embodiment of the present invention provides an artificial intelligence-based CRRT data comprehensive acquisition and optimization processing system, including: Acquisition and calibration module: used to collect CRRT equipment parameters, physiological signals and laboratory test data to form multi-source data, and perform time alignment and spatial calibration on them to obtain calibration data; the acquisition and calibration module includes an acquisition unit, a communication unit and a calibration unit. The acquisition unit is used to collect CRRT equipment parameters, physiological signals and laboratory test data through a three-level sensor network to form multi-source data. The communication unit is used to perform real-time interaction of various data through HL7 FHIR interface protocol conversion. The calibration unit is used to perform time alignment and spatial calibration on multi-source data through hardware synchronous clock trigger sampling combined with fractional calculus interpolation algorithm to obtain calibration data.

[0024] Time alignment uses fractional calculus interpolation, and its formula is ,in, is the optimization parameter and its value range is , is the gamma function, is the interpolation signal at the target time point, is the first-order derivative of the original signal, is the signal acquisition time window, Align the target time point.

[0025] Specifically, the acquisition and calibration module is used to establish a link for the collection, transmission, and spatiotemporal calibration of multi-source CRRT data, providing standardized input for the modeling and estimation modules. The acquisition unit collects data in layers through a three-level sensor network: the first level connects to the CRRT equipment to collect operating parameters such as pressure, flow, and replacement fluid concentration; the second level connects to physiological monitoring equipment to collect physiological signals such as electrocardiogram, blood oxygen, and pulse rate; and the third level connects to the laboratory information system to obtain test data such as blood gas analysis and ion concentration. This cross-device data collection covers all dimensions of CRRT treatment information, forming a multi-source heterogeneous data set.

[0026] Based on the HL7 FHIR interface protocol, the communication unit performs protocol conversion on the multi-source data output by the acquisition unit, breaking down the barriers of private protocols of equipment manufacturers and converting heterogeneous data into standardized medical data formats. At the same time, relying on the real-time interaction mechanism, it preliminarily aligns the timestamps of multi-source data, providing basic data input for the spatiotemporal calibration of the calibration unit, and ensuring the real-time and compatibility of data transmission.

[0027] The calibration unit adopts a strategy combining hardware synchronous clock triggered sampling with fractional-order calculus interpolation: the hardware synchronous clock gives a unified time base to multi-source data, triggering each sensor to synchronously sample at a preset time, solving the problem of sampling asynchrony between devices; for the dynamically changing non-uniform sampling data during CRRT treatment, such as fluctuating blood flow velocity and instantaneous ion concentration, time alignment is performed using a fractional-order calculus interpolation algorithm, the formula of which is: ,in To optimize the parameters, take the value , adapted to the dynamic characteristics of CRRT data, is the gamma function, Fit the interpolated signal at the target time point, Characterizes the rate of change of the original signal, The system defines the signal acquisition time window. Simultaneously, based on a built-in spatial calibration model, it maps the spatially deployed differences in physiological signal sensors, such as surface electrodes and internal sensors, converting multi-source data into a unified spatial coordinate system. After calibration, the output calibration data has a unified sampling benchmark in the time dimension and coordinates are normalized in the spatial dimension. This provides a precise foundation for the modeling and estimation module to construct spatiotemporal dynamics models, effectively reducing model errors caused by data heterogeneity.

[0028] When the acquisition unit obtains the blood flow velocity signal of the CRRT device, the signal is distributed in a non-uniform time interval due to the device sampling mechanism and clinical fluctuations. The calibration unit uses the target time point triggered by the hardware synchronous clock As a reference, the first derivative of the original blood flow velocity signal As input, set the optimization parameters , which adapts to the dynamic change characteristics of blood flow velocity and signal acquisition time window , which covers recent key volatility data, and substitutes the fractional calculus interpolation formula , the target time is obtained by integration operation fitting Blood flow velocity interpolation signal at ; The interpolation signal realizes the time axis alignment of non-uniformly sampled blood flow velocity data with other physiological signals such as electrocardiogram and blood oxygen, provides time-synchronized standardized data for the modeling and estimation module to build a spatiotemporal dynamic model, and ensures the spatiotemporal consistency of the estimation of state variables such as ion concentration field and blood flow velocity field.

[0029] Modeling and estimation module: used to construct a spatiotemporal dynamic model of material transport and biochemical reactions in the CRRT treatment process based on stochastic partial differential equations, and estimate state variables in combination with calibration data to generate high-dimensional state data; the modeling and estimation module includes a discretization unit, a parameter identification unit and a state estimation unit. The discretization unit is used to convert the material transport and biochemical reactions in the CRRT treatment process into a weak form of a three-dimensional stochastic partial differential equation. The parameter identification unit is used to optimize the parameters of the three-dimensional stochastic partial differential equation through variational Bayesian inference. The state estimation unit is used to solve the three-dimensional stochastic partial differential equation based on calibration data, estimate the state variables, and generate high-dimensional state data.

[0030] The stochastic partial differential equation is , where the state vector Including ion calcium concentration field, citrate concentration field and blood flow velocity field, diffusion matrix Is a diagonal matrix, the elements are the diffusion coefficients of each substance, the reaction terms Describe the kinetics of biochemical reactions. is the noise intensity matrix, is the Wiener process; The optimization is to optimize the parameters of the three-dimensional stochastic partial differential equation by maximizing the evidence lower bound. The expression for maximizing the evidence lower bound is ,in, is the variational distribution, is the prior distribution, is the observation data, is the variational distribution The expectation operator under is the Kullback-Leibler divergence.

[0031] Specifically, the modeling and estimation module receives the calibration data output by the acquisition and calibration module, which includes multi-source data such as time-aligned and spatially normalized CRRT equipment parameters and physiological signals. Through the collaborative work of the discretization unit, parameter identification unit, and state estimation unit, it constructs a spatiotemporal dynamic model of the CRRT treatment process and estimates the state variables.

[0032] Among them, the discretization unit first converts the material transport processes such as ion calcium diffusion, citrate distribution, and the continuous processes of biochemical reactions such as calcium-citrate complexation in CRRT treatment into the weak form of three-dimensional stochastic partial differential equations; through the finite element discretization method, the spatial domain is divided into polyhedral grids, and the time domain is split into small steps, and the continuous partial differential equations are converted into discrete equations that can be numerically solved, laying the foundation for subsequent calculations. The parameter identification unit uses the discretized equation structure as a framework, and takes the observed values ​​such as ion calcium concentration and blood flow velocity in the calibration data as , introduce the variational Bayesian inference method; set the prior distribution , for example, assuming that the diffusion coefficient follows a normal distribution; and the variational distribution , to approximate the posterior distribution of the true parameter, construct the evidence lower bound , iterative optimization maximizes the lower bound and solves the diffusion matrix in the stochastic partial differential equation , is the diffusion coefficient of each substance, reaction term The dynamic parameters of the noise intensity matrix , so that the model parameters match the statistical characteristics of the calibration data and improve the accuracy of the model in describing the CRRT process. The state estimation unit is based on the stochastic partial differential equation optimized by the parameter identification unit, which is in the form of ,in It covers the ion calcium concentration field, citrate concentration field, and blood flow velocity field, and substitutes calibration data as initial conditions and boundary constraints; the stochastic finite element method combined with Euler-Markov time discretization is used to solve the numerical solution of the equation in the time and space domain, estimate the distribution and evolution law of each state variable, and generate high-dimensional state data; this data fully characterizes the material distribution and dynamic changes in CRRT treatment, provides accurate state input for dimensionality reduction, denoising and causal analysis of the fusion reasoning module, and supports the model foundation for subsequent treatment parameter optimization.

[0033] In this embodiment, the parameter identification unit receives the calibration data output by the acquisition and calibration module, including the observed values ​​of ionized calcium concentration, citrate concentration, blood flow velocity, etc. at different times, as , combined with the three-dimensional stochastic partial differential equation framework constructed by discretization units, the form is ,in Characterize the state vector of ionized calcium concentration field, citrate concentration field, and blood flow velocity field; set the prior distribution , for example, assuming the diffusion matrix The diagonal elements of follow normal distribution, and the response term The biochemical reaction rate obeys the Gamma distribution, constructing the variational distribution Approximate the posterior distribution of the parameter and substitute it into the lower bound formula of the evidence , maximized through iterative optimization , output the optimized diffusion matrix , is the diffusion coefficient of each substance, reaction term Kinetic parameters such as calcium-citrate complexation rate, noise intensity matrix , so that the parameters of the stochastic partial differential equation match the actual material transport and biochemical reaction characteristics of CRRT. The state estimation unit calls the optimized parameters output by the parameter identification unit, using the calibration data as initial conditions and boundary constraints, such as the ionized calcium concentration and blood flow velocity at the start of treatment in the initial conditions, and the filter inlet flow and replacement fluid concentration in the boundary constraints, and substitutes them into the stochastic partial differential equation , the stochastic finite element method is used to discretize the state vector in the time and space domain , the dynamic evolution distribution of ion calcium concentration field, citrate concentration field, and blood flow velocity field, that is, high-dimensional state data, is obtained, which realizes the spatiotemporal dynamics characterization of material transport and biochemical reactions during CRRT treatment, and provides an accurate state basis for the fusion reasoning module to analyze the relationship between treatment parameters and physiological indicators.

[0034] Fusion reasoning module: used to perform dimensionality reduction and adversarial denoising on high-dimensional state data, generate low-dimensional feature vectors, and construct a causal relationship map between treatment parameters and physiological indicators based on the low-dimensional feature vectors, and calculate the intervention effect; the fusion reasoning module includes a dimensionality reduction unit, a denoising unit and a causal modeling unit. The dimensionality reduction unit is used to perform manifold dimensionality reduction on high-dimensional state data through a local linear embedding algorithm. The denoising unit is used to perform denoising on the reduced high-dimensional state data based on a conditional generative adversarial network to generate low-dimensional feature vectors. The causal modeling unit is used to construct a causal relationship map between treatment parameters and physiological indicators based on the low-dimensional feature vectors, and calculate the intervention effect through Do-Calculus.

[0035] The locally linear embedding algorithm performs dimensionality reduction by solving the following optimization problem: ,in, is the low-dimensional manifold eigenvector, To locally reconstruct the weight matrix, by minimizing Sure, is the original high-dimensional data; The formula for calculating the intervention effect is ,in, is the treatment parameter, As physiological indicators, for The parent node set of for Medium External variables, For intervention hour The probability distribution of .

[0036] Specifically, the fusion reasoning module receives the high-dimensional state data output by the modeling and estimation module, which covers the spatiotemporal evolution information of the ion calcium concentration field, citrate concentration field, and blood flow velocity field. It is processed in sequence by the dimensionality reduction unit, the denoising unit, and the causal modeling unit to provide the control strategy module with low-dimensional feature vectors and the causal relationship between treatment parameters and physiological indicators.

[0037] Among them, the dimensionality reduction unit uses a local linear embedding algorithm for the nonlinear manifold structure of high-dimensional state data: each spatiotemporal sample in the high-dimensional state data is used as the original high-dimensional data. For example, at a certain moment and at a certain spatial grid point, the ion calcium concentration and blood flow velocity combination are calculated first. The local reconstruction weight matrix between samples is , which is achieved by minimizing Determine and characterize the local linear dependency between samples; then solve the low-dimensional manifold eigenvector based on the weight matrix , and its optimization goal is , mapping high-dimensional state data to low-dimensional space, preserving its intrinsic manifold structure, while reducing the data dimension, retaining the key change patterns of material transport and physiological state during CRRT treatment, simplifying the calculation for subsequent denoising and causal analysis.

[0038] The denoising unit receives the low-dimensional manifold feature vector output by the dimensionality reduction unit and constructs a conditional generative adversarial network: the noisy low-dimensional feature is used as the generator input, combined with the physiological constraint prior output by the modeling estimation module, such as the clinically reasonable range of ionized calcium concentration, the generator outputs denoised feature candidates; the discriminator uses the low-dimensional features of the stable treatment period in the acquisition and calibration module as the real sample. The low-dimensional feature filters the self-calibration data, distinguishes the generated features from the real features, and through iterative optimization of adversarial training, the generator output is close to the real low-dimensional feature, removing the noise introduced by dimensionality reduction and the measurement noise of the original data, generating a more accurate low-dimensional feature vector, and improving the reliability of subsequent causal modeling.

[0039] The causal modeling unit extracts the association between treatment parameters and physiological indicators based on the denoised low-dimensional feature vector. In one embodiment, the treatment parameters include blood flow and ultrafiltration rate, which are obtained from the CRRT equipment parameter mapping of the acquisition and calibration module. Physiological indicators include ionized calcium concentration and blood gas parameters, which are extracted from physiological signals and laboratory test data. The PC algorithm is used to analyze the conditional independence between variables and construct a causal relationship map, which clarifies the parent node and child node relationship between treatment parameters and physiological indicators, such as the direct effect of ultrafiltration rate on ionized calcium concentration and intermediate variables. Do-Calculus is introduced to convert treatment parameters into blood flow and ultrafiltration rate. As an intervention variable, such as adjusting ultrafiltration rate, to determine physiological indicators based on causal mapping The parent node set , such as ionized calcium concentration, substitute into the intervention effect formula , quantify the probability distribution of the impact of treatment parameter adjustment on physiological indicators, provide a causal basis for the control strategy module to optimize the treatment plan, avoid the misleading of statistical correlation, and improve the targetedness of treatment parameter adjustment.

[0040] In this embodiment, the dimensionality reduction unit receives the high-dimensional state data output by the modeling and estimation module, which includes the spatiotemporal sampling information of the ionized calcium concentration field, the citrate concentration field, and the blood flow velocity field. , first calculate the linear dependence of the local neighborhood of the sample and minimize Determine the local reconstruction weight matrix ; then As constraints, solve the optimization problem , output low-dimensional manifold feature vector , which retains the key correlation patterns between material transport and physiological status during CRRT treatment, achieves dimensionality reduction of high-dimensional data, and provides streamlined and fidelity input for the denoising unit.

[0041] The causal modeling unit receives the low-dimensional feature vector output by the denoising unit to extract the treatment parameters , such as ultrafiltration rate; physiological indicators , such as ion calcium concentration, the PC algorithm constructs a causal relationship map to determine The parent node set , including variables such as blood flow velocity and citrate concentration; As an intervention variable, screening Medium External variables , substitute into the formula , combined with the statistical distribution of low-dimensional feature vectors, derived from the modeling estimation module data , Computational Intervention hour The probability distribution of treatment parameters is used to quantify the causal effects of treatment parameters on physiological indicators, providing accurate causal basis for the control strategy module to optimize the treatment plan.

[0042] Control strategy module: used to build a model predictive control framework based on the spatiotemporal dynamics model, generate a basic treatment plan that meets clinical constraints in combination with the intervention effect, and adjust its parameters individually through deep reinforcement learning; the control strategy module includes a basic control unit and an intelligent optimization unit. The basic control unit is used to build a model predictive control framework based on the spatiotemporal dynamics model, generate a basic treatment plan that meets clinical constraints in combination with the intervention effect. Clinical constraints include blood flow range, ionized calcium concentration threshold, and ultrafiltration rate upper limit. The cost function of the model predictive control framework is ,in, For the prediction time domain, To control the time domain, is the predicted state vector, is the reference trajectory, is the control variable, and are the weight matrices of state and control variables respectively. The intelligent optimization unit is used to adjust the parameters of the basic treatment plan through deep reinforcement learning. The reward function of deep reinforcement learning is ,in, To control the smoothing coefficient, is the complication avoidance bonus coefficient, is the indicator function, As the control variable of the basic treatment plan, is the reference state vector.

[0043] Specifically, the control strategy module receives the spatiotemporal dynamics model output by the modeling and estimation module and the causal intervention effect of the fusion reasoning module, such as the probability distribution of the impact of ultrafiltration rate adjustment on ionized calcium concentration. The spatiotemporal dynamics model describes the state evolution laws of ionized calcium concentration field, blood flow velocity field, etc. during CRRT treatment. Relying on the coordinated operation of the basic control unit and the intelligent optimization unit, it generates a treatment plan that meets clinical constraints and adapts to individual differences.

[0044] Among them, the basic control unit takes the spatiotemporal dynamics model as the prediction core, takes the causal intervention effect of the fusion reasoning module as the constraint condition, integrates the clinical constraints extracted by the acquisition and calibration module, including blood flow range, ionized calcium concentration threshold, and ultrafiltration rate upper limit, and constructs a model prediction control framework: setting the prediction time domain and control time domain , used to match the dynamic adjustment rhythm of CRRT treatment to model the predicted state vector output by the estimation module , such as the predicted distribution of ionized calcium concentration field at future moments, clinical target reference trajectory , which derives and controls variables from the normal physiological indicators of the acquisition and calibration module , such as the adjustment of blood flow and ultrafiltration rate as input, by optimizing the cost function ,in, 、 It is the weight matrix of state deviation and control energy consumption, balances the treatment effect and equipment operation stability, solves the control sequence that meets clinical constraints, generates a basic treatment plan, and ensures the safety and basic effectiveness of the treatment.

[0045] The intelligent optimization unit receives the control variables of the basic treatment plan and the real-time state vector of the modeling and estimation module , which extracts key physiological indicators from high-dimensional state data and introduces deep reinforcement learning: With the reference state vector Deviation, and The difference is the penalty term, combined with the complication events monitored by the acquisition and calibration module, to drive the indicator function through physiological signal judgment. , construct the reward function ,in, Smoothing out fluctuations in control parameters, Incentivize the avoidance of complications; optimize the control strategy through iterative learning, and personalize the parameters of the basic treatment plan based on the individual characteristics of the patient, such as underlying diseases and metabolic rate, and the multi-source data implicit in the acquisition and calibration module to improve the adaptability of the treatment plan to the patient.

[0046] Instruction generation module: used to fuse the basic treatment plan and personalized adjustment parameters, generate device control instructions, perform real-time control at the edge through the edge-cloud collaborative architecture, and perform iterative training of spatiotemporal dynamics models and deep reinforcement learning strategies in the cloud. The instruction generation module includes a weight fusion unit, a safety verification unit, and a collaborative architecture unit. The weight fusion unit is used to dynamically adjust the weights of the basic treatment plan and personalized adjustment parameters based on the treatment stage, perform fusion, and output parameter adjustment instructions. The safety verification unit is used to filter abnormal instructions from the parameter adjustment instructions based on causal constraints and clinical thresholds, and then generate device control instructions. The collaborative architecture unit is used to perform real-time control according to the device control instructions at the edge through the edge-cloud collaborative architecture, and perform iterative training of the spatiotemporal dynamics model and deep reinforcement learning strategies in the cloud through federated learning.

[0047] Specifically, the instruction generation module receives the basic treatment plan and personalized adjustment parameters output by the control strategy module, as well as treatment phase information from the acquisition and calibration module, such as the start-up and stabilization phase indicators. After weight fusion, safety verification, and coordinated execution, it generates device control instructions and drives model iteration. The weight fusion unit dynamically adjusts the weights of the basic treatment plan and personalized parameters based on the treatment progress monitored by the acquisition and calibration module, taking into account the device's operating time and the physiological signal trend determination phase. During the startup phase, the basic treatment plan's safety constraints are prioritized, receiving a higher weight. During the stabilization phase, personalized adaptation is gradually strengthened, increasing the weight of personalized parameters. The adjusted parameters are integrated and output as parameter adjustment instructions, facilitating the safety verification phase. Upon receiving the parameter adjustment instructions, the safety verification unit invokes the causal relationship graph of the fusion reasoning module to extract the causal constraints from the "treatment parameter → physiological indicator" link, such as the path from a sudden change in ultrafiltration rate to a decrease in ionized calcium. Combined with the clinical threshold set by the acquisition and calibration module, such as ionized calcium ≥ 1.0 mmol / L, it filters out abnormal instructions that violate causal logic or thresholds. The unit then integrates compliance instructions and generates device control instructions to ensure that the instructions comply with treatment safety and clinical standards. After receiving device control instructions, the collaborative architecture unit adapts to the HL7 FHIR protocol at the edge through the device interface of the acquisition and calibration module, driving the CRRT equipment to perform real-time control, such as adjusting the ultrafiltration rate and blood flow; synchronously collects real-time treatment data at the edge, such as device parameters and physiological feedback, and uploads it to the cloud; based on federated learning, the cloud associates the real-time data with the spatiotemporal dynamics model of the modeling and estimation module and the deep reinforcement learning strategy of the control strategy module for training, iteratively optimizes the model and strategy, realizes the coordination of edge real-time control and cloud model evolution, and improves the system's adaptability to CRRT treatment.

[0048] The acquisition and calibration module collects equipment parameters, physiological signals and laboratory data through a three-level sensor network, and combines hardware synchronous clocks with fractional-order calculus interpolation algorithms to complete time alignment and spatial calibration of multi-source data, thereby providing a high-precision standardized data foundation for subsequent modeling; the modeling and estimation module performs spatiotemporal dynamics modeling of material transport and biochemical reactions in CRRT treatment based on stochastic partial differential equations, and estimates state variables such as ion calcium concentration field and blood flow velocity field in combination with calibration data to generate high-dimensional state data that characterizes the dynamic characteristics of the treatment process, thereby realizing a mathematical description of the CRRT treatment mechanism; the fusion reasoning module performs manifold dimensionality reduction and adversarial denoising on the high-dimensional state data, constructs a causal relationship map between treatment parameters and physiological indicators through the PC algorithm, and uses Do-Calculus to quantify the intervention effect, thereby solving the problem of CRRT treatment. The control strategy module combines the spatiotemporal dynamics model with the causal intervention effect to construct a model predictive control framework to generate a basic treatment plan that meets clinical constraints, and then optimizes personalized parameters through deep reinforcement learning to achieve the coordination of treatment plan safety and individual adaptability; the instruction generation module integrates the basic plan and personalized parameters, and generates equipment control instructions after verification of causal constraints and clinical thresholds. Real-time control is performed at the edge through the edge-cloud collaborative architecture, and federated learning is used in the cloud to iteratively optimize models and strategies, thereby achieving scientific analysis and precise intervention of causal relationships in CRRT treatment, and solving the problem that traditional CRRT optimization processing systems cannot distinguish between true causal effects and interference from confounding factors during parameter adjustment, causing unexpected treatment side effects.

[0049] Please see the attached Figure 2 The method for all-around acquisition and optimization processing of CRRT data based on artificial intelligence is applied to the above-mentioned all-around acquisition and optimization processing system of CRRT data based on artificial intelligence, and includes the following steps: Acquisition and calibration: Collect CRRT equipment parameters, physiological signals, and laboratory test data to form multi-source data, and perform time alignment and spatial calibration on them to obtain calibration data; Modeling and estimation: Based on stochastic partial differential equations, a spatiotemporal dynamics model of material transport and biochemical reactions during CRRT treatment is constructed. State variables are estimated using calibration data to generate high-dimensional state data. Fusion reasoning: Perform dimensionality reduction and adversarial denoising on high-dimensional state data to generate low-dimensional feature vectors. Based on these low-dimensional feature vectors, a causal relationship map between treatment parameters and physiological indicators is constructed to calculate the intervention effect. Control strategy: A model predictive control framework is constructed based on the spatiotemporal dynamics model. The intervention effect is combined to generate a basic treatment plan that meets clinical constraints, and deep reinforcement learning is used to adjust its parameters in a personalized manner. Instruction generation: Integrates basic treatment plans and personalized adjustment parameters to generate device control instructions, performs real-time control at the edge through an edge-cloud collaborative architecture, and performs iterative training of spatiotemporal dynamics models and deep reinforcement learning strategies in the cloud.

[0050] Specifically, a three-level sensor network collects CRRT equipment operating parameters, physiological monitoring signals, and laboratory test data in a hierarchical manner, generating multi-source heterogeneous data. The communication unit utilizes the HL7FHIR interface protocol to complete data conversion and preliminary timestamp alignment. The calibration unit triggers multi-device synchronous sampling using a hardware-synchronized clock. For non-uniform dynamic data such as blood flow velocity fluctuations, a fractional-order calculus interpolation algorithm is introduced to fit the interpolated signal at the target time point, achieving multi-source data time alignment. Furthermore, based on a spatial coordinate mapping model, differences in sensor deployment are corrected for, completing data spatial calibration. The output of this calibration data provides precise, spatiotemporally synchronized input for subsequent modeling, eliminating interference from data heterogeneity.

[0051] After receiving the collected and calibrated calibration data, the discretization unit converts the transport and biochemical reactions of substances such as ionized calcium diffusion and citrate complexation during CRRT treatment into a weak form of three-dimensional stochastic partial differential equations. Using the finite element method, the space-time domain is discretized to construct a numerical solution framework. The parameter identification unit introduces variational Bayesian inference, sets prior distributions for parameters such as the diffusion coefficient and reaction rate, constructs a variational distribution approximation posterior, and uses the evidence lower bound formula for iterative optimization to match the model parameters with the statistical characteristics of the calibration data. The state estimation unit substitutes the calibration data as initial and boundary conditions and uses the stochastic finite element method combined with Euler-Markov time discretization to solve the dynamic evolution distribution of state variables such as the ionized calcium concentration field and blood flow velocity field, generating high-dimensional state data to characterize the spatiotemporal dynamics of CRRT treatment.

[0052] For the high-dimensional state data estimated by modeling, the dimensionality reduction unit uses a local linear embedding algorithm to first calculate the local reconstruction weights between samples. Then, based on these weights, the high-dimensional data is mapped to a low-dimensional manifold space, preserving the key state patterns of the CRRT process. The denoising unit constructs a conditional generative adversarial network, which uses noisy low-dimensional features as input. Combined with the physiological constraint priors estimated by modeling, it removes data noise through adversarial training between the generator and the discriminator, outputting an accurate low-dimensional feature vector. The causal modeling unit extracts treatment parameters and physiological indicators from the low-dimensional features, constructs a causal graph using the PC algorithm, clarifies the parent node relationship of the variables, and then uses the Do-Calculus formula to quantify the intervention effect of treatment parameter adjustment on physiological indicators, providing a causal basis for control strategies and avoiding misleading statistical associations.

[0053] Relying on the spatiotemporal dynamics model estimated through modeling and the intervention effects derived through fusion reasoning, the basic control unit constructs a model predictive control framework, sets the prediction and control time domains, and uses the predicted state vector, clinical reference trajectory, and control variables as inputs to optimize the cost function and generate a basic treatment plan that meets clinical constraints. The intelligent optimization unit incorporates deep reinforcement learning, using the deviation between the real-time state and the reference state, and the difference between the control variables and the basic plan as penalty terms. Complication indicators are combined to construct a reward function, iteratively optimize the control strategy, and personalize the basic plan based on individual patient characteristics to improve treatment adaptability.

[0054] The basic scheme and personalized parameters of the control strategy are received, and the weight fusion unit dynamically adjusts the weights of the two according to the treatment stage of the collected calibration feedback, and outputs parameter adjustment instructions. The safety verification unit calls the causal graph of the fusion reasoning and the clinical threshold of the collected calibration to filter out abnormal instructions that violate the causal or threshold and generate compliant equipment control instructions. The collaborative architecture unit performs real-time control on the edge side based on the device interface of the collected calibration through edge-cloud collaboration, and synchronously collects real-time data and uploads it to the cloud; based on federated learning, the cloud associates the real-time data with the spatiotemporal dynamics model estimated by modeling and the reinforcement learning strategy of the control strategy for training, iteratively optimizes the model and strategy, realizes the closed loop of edge control and cloud evolution, and continuously improves the system's optimization capabilities for CRRT treatment.

[0055] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The CRRT data comprehensive acquisition and optimization processing system based on artificial intelligence is characterized by: include: Acquisition and calibration module: used to collect CRRT equipment parameters, physiological signals and laboratory test data to form multi-source data, and perform time alignment and spatial calibration on them to obtain calibration data; Modeling and Estimation Module: This module is used to construct a spatiotemporal dynamics model of material transport and biochemical reactions during CRRT treatment based on stochastic partial differential equations, and to estimate state variables using calibration data to generate high-dimensional state data. Fusion reasoning module: used to perform dimensionality reduction and adversarial denoising on high-dimensional state data to generate low-dimensional feature vectors. Based on the low-dimensional feature vectors, it constructs a causal relationship map between treatment parameters and physiological indicators and calculates the intervention effect. Control strategy module: used to build a model predictive control framework based on the spatiotemporal dynamics model, generate a basic treatment plan that meets clinical constraints based on the intervention effect, and adjust its parameters individually through deep reinforcement learning; Instruction generation module: used to integrate basic treatment plans and personalized adjustment parameters, generate device control instructions, perform real-time control at the edge through the edge-cloud collaborative architecture, and perform iterative training of spatiotemporal dynamics models and deep reinforcement learning strategies in the cloud.

2. The artificial intelligence-based CRRT data omnidirectional acquisition and optimization processing system according to claim 1, characterized in that: The acquisition and calibration module includes an acquisition unit, a communication unit, and a calibration unit. The acquisition unit is used to collect CRRT equipment parameters, physiological signals, and laboratory test data through a three-level sensor network to form multi-source data. The communication unit is used to perform real-time interaction of various data through HL7 FHIR interface protocol conversion. The calibration unit is used to perform time alignment and spatial calibration on multi-source data through hardware synchronous clock triggered sampling combined with fractional calculus interpolation algorithm to obtain calibration data.

3. The artificial intelligence-based CRRT data omnidirectional acquisition and optimization processing system according to claim 2, characterized in that: The time alignment adopts fractional calculus interpolation, and its formula is: ,in, is the optimization parameter and its value range is , is the gamma function, is the interpolation signal at the target time point, is the first-order derivative of the original signal, is the signal acquisition time window, Align the target time point.

4. The artificial intelligence-based CRRT data omnidirectional acquisition and optimization processing system according to claim 1 is characterized in that: The modeling and estimation module includes a discretization unit, a parameter identification unit and a state estimation unit. The discretization unit is used to convert the material transport and biochemical reactions in the CRRT treatment process into a weak form of a three-dimensional stochastic partial differential equation. The parameter identification unit is used to optimize the parameters of the three-dimensional stochastic partial differential equation through variational Bayesian inference. The state estimation unit is used to solve the three-dimensional stochastic partial differential equation based on calibration data, estimate state variables, and generate high-dimensional state data.

5. The artificial intelligence-based CRRT data omnidirectional acquisition and optimization processing system according to claim 4 is characterized in that: The stochastic partial differential equation is , where the state vector Including ion calcium concentration field, citrate concentration field and blood flow velocity field, diffusion matrix Is a diagonal matrix, the elements are the diffusion coefficients of each substance, the reaction terms Describe the kinetics of biochemical reactions. is the noise intensity matrix, is the Wiener process; The optimization is to optimize the parameters of the three-dimensional stochastic partial differential equation by maximizing the lower bound of evidence. The expression of the maximization lower bound of evidence is: ,in, is the variational distribution, is the prior distribution, is the observation data, is the variational distribution The expectation operator under is the Kullback-Leibler divergence.

6. The artificial intelligence-based CRRT data omnidirectional acquisition and optimization processing system according to claim 1, characterized in that: The fusion reasoning module includes a dimensionality reduction unit, a denoising unit and a causal modeling unit. The dimensionality reduction unit is used to perform manifold dimensionality reduction on high-dimensional state data through a local linear embedding algorithm. The denoising unit is used to perform denoising on the reduced high-dimensional state data based on a conditional generative adversarial network to generate a low-dimensional feature vector. The causal modeling unit is used to construct a causal relationship map between treatment parameters and physiological indicators based on the low-dimensional feature vector, and calculate the intervention effect through Do-Calculus.

7. The artificial intelligence-based CRRT data omnidirectional acquisition and optimization processing system according to claim 6, characterized in that: The locally linear embedding algorithm performs dimensionality reduction by solving the following optimization problem: ,in, is the low-dimensional manifold eigenvector, To locally reconstruct the weight matrix, by minimizing Sure, is the original high-dimensional data; The formula for calculating the intervention effect is ,in, is the treatment parameter, As physiological indicators, for The parent node set of for Medium External variables, For intervention hour The probability distribution of .

8. The artificial intelligence-based CRRT data omnidirectional acquisition and optimization processing system according to claim 1 is characterized by: The control strategy module includes a basic control unit and an intelligent optimization unit. The basic control unit is used to construct a model predictive control framework based on the spatiotemporal dynamics model, and generate a basic treatment plan that meets clinical constraints in combination with the intervention effect. The clinical constraints include blood flow range, ionized calcium concentration threshold and ultrafiltration rate upper limit. The cost function of the model predictive control framework is ,in, For the prediction time domain, To control the time domain, is the predicted state vector, is the reference trajectory, is the control variable, and are weight matrices of state and control variables respectively. The intelligent optimization unit is used to adjust the parameters of the basic treatment plan through deep reinforcement learning. The reward function of the deep reinforcement learning is ,in, To control the smoothing coefficient, is the complication avoidance bonus coefficient, is the indicator function, As the control variable of the basic treatment plan, is the reference state vector.

9. The artificial intelligence-based CRRT data omnidirectional acquisition and optimization processing system according to claim 1, characterized in that: The instruction generation module includes a weight fusion unit, a safety verification unit and a collaborative architecture unit. The weight fusion unit is used to dynamically adjust the weights of the basic treatment plan and personalized adjustment parameters based on the treatment stage, perform fusion, and output parameter adjustment instructions. The safety verification unit is used to filter abnormal instructions from the parameter adjustment instructions based on causal constraints and clinical thresholds, and generate device control instructions. The collaborative architecture unit is used to perform real-time control according to the device control instructions at the edge through an edge-cloud collaborative architecture, and perform iterative training of spatiotemporal dynamics models and deep reinforcement learning strategies through federated learning in the cloud.

10. An artificial intelligence-based method for all-around acquisition and optimization of CRRT data, characterized by: The artificial intelligence-based CRRT data all-round acquisition and optimization processing system as described in any one of claims 1 to 9 comprises the following steps: Acquisition and calibration: Collect CRRT equipment parameters, physiological signals, and laboratory test data to form multi-source data, and perform time alignment and spatial calibration on them to obtain calibration data; Modeling and estimation: Based on stochastic partial differential equations, a spatiotemporal dynamics model of material transport and biochemical reactions during CRRT treatment is constructed. State variables are estimated using calibration data to generate high-dimensional state data. Fusion reasoning: Perform dimensionality reduction and adversarial denoising on high-dimensional state data to generate low-dimensional feature vectors. Based on these low-dimensional feature vectors, a causal relationship map between treatment parameters and physiological indicators is constructed to calculate the intervention effect. Control strategy: A model predictive control framework is constructed based on the spatiotemporal dynamics model. The intervention effect is combined to generate a basic treatment plan that meets clinical constraints, and deep reinforcement learning is used to adjust its parameters in a personalized manner. Instruction generation: Integrates basic treatment plans and personalized adjustment parameters to generate device control instructions, performs real-time control at the edge through an edge-cloud collaborative architecture, and performs iterative training of spatiotemporal dynamics models and deep reinforcement learning strategies in the cloud.

Images