An AI Agent-Based Intelligent Hemodialysis Prescription Optimization System and Method

Abstract

This invention discloses an intelligent hemodialysis prescription optimization system and method based on an AI agent. The system includes a sensing module, an inference module, an action module, and a feedback module, forming a closed-loop control architecture. The inference module embeds a dual-chamber sodium kinetic model (diffusion equilibrium concentration C_eq=[Na]d / σ (σ=0.97 is the Gibbs-Donnan factor)) to calculate sodium clearance in real time and predict the end-of-dialysis state. Under the premise of meeting three constraints—ultrafiltration rate safety, serum sodium range, and hemodynamic stability—the system optimizes the combination of dialysate sodium concentration, ultrafiltration volume, and dialysis time to achieve the target sodium clearance value. During dialysis, the system can dynamically adjust prescription parameters, and after dialysis, it continuously calibrates model parameters through feedback learning. This invention is the first to combine a sodium kinetic model with an AI agent architecture, achieving an upgrade from "open-loop experience-based prescriptions" to "closed-loop precision prescriptions."

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, medical information systems, and blood purification technology, specifically to an intelligent hemodialysis prescription optimization system and method based on AIAgent. Background Technology

[0002] 1. Problems with existing dialysis prescription formulation Currently, dialysis prescriptions (dialysis fluid sodium concentration, ultrafiltration volume, and dialysis time) are set by doctors based on experience before dialysis and are usually not adjusted during the dialysis process. This "open-loop" control method cannot cope with dynamic changes during dialysis (such as blood pressure fluctuations, deviations in actual ultrafiltration volume, etc.) and cannot guarantee that the sodium clearance volume reaches the target value for each dialysis session.

[0003] 2. Limitations of existing dialysis machine "sodium curves" Some dialysis machines offer a "sodium profiling" function, which allows the sodium concentration in the dialysate to change according to a preset curve (such as linear decrease, stepwise decrease, etc.) during dialysis. However, these curves are fixed patterns, not based on the individual patient's sodium kinetic parameters, nor are they adjusted according to real-time data, thus failing to achieve truly individualized and precise control.

[0004] 3. Limitations of existing dialysis systems and technologies Existing dialysis information systems only have data recording and statistical functions, lacking real-time reasoning and optimization capabilities. Currently, no system combines sodium kinetics mathematical models with an AI agent architecture to achieve closed-loop intelligent control of dialysis prescriptions.

[0005] Existing technologies have also attempted to automatically adjust dialysis parameters, but all have fundamental limitations: (1) Methods based on fixed rule tables (if-then) can only respond to preset discrete scenarios and cannot handle continuously changing multi-parameter spaces, and conflicts may occur between rules; (2) Methods based on fixed PID controllers only perform feedback adjustment for a single control variable (such as ultrafiltration rate) and cannot simultaneously optimize the combination of multiple parameters such as dialysate sodium concentration, ultrafiltration volume, and dialysis time; (3) Methods based on single-objective optimization only pursue the optimization of a certain indicator (such as achieving the target ultrafiltration volume) and cannot balance multiple constraints such as sodium clearance target, hemodynamic safety, and serum sodium range. All of the above methods lack the forward-looking reasoning ability based on physical models and the adaptive ability to learn from historical data.

[0006] 4. Advantages of AI Agent Architecture An AI Agent is an intelligent agent architecture with autonomous perception, reasoning, and action capabilities. Unlike traditional rule engines or simple feedback control, an Agent can: - Integrate multi-source heterogeneous data (physiological parameters, equipment parameters, historical data); - Prospective reasoning based on mathematical models (predicting the state at the end of dialysis); - Optimize decision-making under multiple constraints (simultaneously satisfying sodium removal targets, ultrafiltration rate safety, and serum sodium range); - Learn from historical data to continuously improve prediction accuracy. Summary of the Invention

[0007] The technical problem to be solved by this invention is to provide an intelligent hemodialysis prescription optimization system based on an AI Agent architecture, which embeds a sodium dynamics mathematical model into a closed-loop control architecture of perception-reasoning-action-feedback, so as to realize real-time optimization and individualized control of dialysis prescriptions, and upgrade from "open-loop experience prescriptions" to "closed-loop precision prescriptions".

[0008] Unlike dialysis parameter adjustment methods based on fixed rules or single-objective control, the AI ​​Agent of this invention simultaneously evaluates sodium clearance targets, hemodynamic safety constraints, and individualized historical response patterns within each decision cycle, and dynamically generates action strategies through an inference module. This multi-constraint, model-driven, and adaptive decision-making mechanism is something that rule tables, PID controllers, and single-objective optimizers cannot achieve.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides an intelligent hemodialysis prescription optimization system based on AI Agent, comprising a perception module, an inference module, an action module, and a feedback module, which are connected in sequence to form a closed-loop control architecture.

[0010] I. Perception Module The sensing module is used to collect the following data in real time: (1) Patient physiological parameters: including predialysis weight, dry weight, continuous blood pressure monitoring data, and serum sodium concentration (measured before dialysis or monitored online); (2) Dialysis machine operating parameters: including actual ultrafiltration rate, dialysate sodium concentration set value, blood flow rate, and dialysate flow rate; (3) Dialyzer parameters: including sodium dialyzer clearance rate K_Na, which can be estimated from dialyzer specifications and operating parameters; (4) Historical data: including the patient’s previous dialysis records, weight gain pattern between dialysis sessions, and blood pressure change trend.

[0011] II. Reasoning Module The reasoning module is connected to the perception module and includes three sub-modules: (1) Sodium kinetics calculation engine: Based on the dual-chamber sodium kinetics model (C_eq = [Na]d / σ, where σ=0.97 is the Gibbs-Donnan factor, which reflects the influence of plasma proteins on mobile sodium concentration; perturbation ICF buffer model; solvent-free drag constraint of aquaporins), the current cumulative sodium clearance, the predicted serum sodium concentration at the end of dialysis and the total sodium clearance are calculated in real time; (2) Safety constraint evaluator: Real-time assessment of three safety constraints, including ultrafiltration rate <10 ml / h / kg, serum sodium after dialysis within the range of 135-145 mmol / L, and blood pressure drop during dialysis <20 mmHg; (3) Prescription optimizer: Under the premise of meeting safety constraints, with the target value of sodium removal being Na_target = (W_pre - W_dry) × [Na]s / 0.93 as the optimization objective, the combination of three adjustable parameters is optimized, namely dialysate sodium concentration [Na]d, ultrafiltration volume V_UF (or ultrafiltration rate), and dialysis time t.

[0012] The core coupled differential equations of the dual-chamber sodium kinetic model built into the inference module are as follows: ECF sodium mass balance equation: Ve×dCe / dt = -K_Na×(Ce - C_eq) - j_ic, where Ve is the extracellular fluid volume, Ce is the extracellular fluid sodium concentration, K_Na is the sodium dialyzer clearance rate, C_eq = [Na]d / σ (σ=0.97) is the diffusion equilibrium concentration, and j_ic is the intracellular and extracellular sodium exchange flux; ICF sodium mass balance equation: d(Vi×Ci) / dt = j_ic, where Vi is the intracellular fluid volume and Ci is the intracellular fluid sodium concentration; the intracellular and extracellular sodium exchange adopts a perturbation reference model: j_ic = K_ic×[(Ce-Ci)-(Ce0-Ci0)], where K_ic is the cell membrane sodium exchange coefficient, and Ce0 and Ci0 are the initial concentrations before dialysis.

[0013] III. Action Module The action module is connected to the reasoning module and is used to perform the following operations: (1) Before dialysis: Generate optimized initial prescription recommendations, including recommended dialysate sodium concentration, ultrafiltration volume, dialysis time, and dialysate sodium concentration control curves throughout the dialysis process; (2) During dialysis: The sodium balance calculation is updated every 5-15 minutes based on real-time data. When the predicted sodium clearance deviates from the target, the sodium concentration of the dialysate is dynamically adjusted. When the blood pressure drops beyond the threshold, the ultrafiltration rate is reduced and the dialysis time is extended accordingly. (3) Early warning: When the predicted parameters exceed the safe range, an early warning is issued to medical staff; (4) Record: Record the complete sodium balance data for each dialysis session.

[0014] IV. Feedback Module The feedback module is connected to the action module and the reasoning module to enable the system's self-learning and continuous optimization. (1) Compare the dialysis results (actual serum sodium changes, actual sodium clearance) with the model predictions; (2) Automatically calibrate model parameters (including K_Na, K_ic, Lp, etc.) to gradually adapt the model to individual patient characteristics by minimizing prediction errors; (3) Establish individualized sodium kinetic parameter profiles for patients to improve the accuracy of subsequent predictions. Attached Figure Description

[0015] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying 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.

[0016] Figure 1 This is a system architecture diagram of the AI ​​Agent of the present invention, showing a closed-loop control architecture of perception → reasoning → action → feedback.

[0017] Figure 2 This is a diagram of the internal structure of the inference module of the present invention, showing the collaborative relationship between the sodium dynamics engine, the safety constraint evaluator, and the prescription optimizer.

[0018] Figure 3 This is a schematic diagram illustrating the dynamic control of the Agent during the dialysis process of the present invention, showing the adjustment process of the sodium concentration of the dialysate over time.

[0019] Figure 4 This is a flowchart of the feedback learning process of the present invention, showing the comparison between model predictions and actual results, as well as the parameter calibration process.

[0020] Figure 5 This is an integrated architecture diagram of the system of the present invention and the dialysis machine / information system. Detailed Implementation

[0021] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the specific embodiments. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this patent, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this patent.

[0022] Example 1: System Overall Architecture like Figure 1 As shown, the present invention provides an intelligent hemodialysis prescription optimization system based on AI Agent, including a perception module 100, an inference module 200, an action module 300 and a feedback module 400, which are connected in sequence to form a closed-loop control architecture.

[0023] The input terminal of the sensing module 100 is connected to the patient monitoring equipment, the dialysis machine control system, and the hospital information system for real-time acquisition of multi-source data. The output terminal of the sensing module 100 is connected to the input terminal of the inference module 200, transmitting the acquired data to the inference module 200 for processing.

[0024] The output of the inference module 200 is connected to the input of the action module 300, transmitting the optimized prescription parameters to the action module 300. The output of the action module 300 is connected to the dialysis machine control system to perform prescription adjustment operations. Simultaneously, the output of the action module 300 is also connected to the input of the feedback module 400, transmitting the actual execution results to the feedback module 400.

[0025] The output of the feedback module 400 is connected to the inference module 200 to compare the actual dialysis results with the model predictions and to automatically calibrate the model parameters in the inference module 200.

[0026] Example 2: Specific Structure of the Sensing Module like Figure 1 As shown, the sensing module includes the following data acquisition units: The patient physiological parameter acquisition unit 101 is used to collect predialysis weight, dry weight, continuous blood pressure monitoring data, and serum sodium concentration. Serum sodium concentration can be obtained through predialysis blood tests or in real-time through online blood volume monitoring equipment.

[0027] The dialysis machine operating parameter acquisition unit 102 is used to read operating parameters such as actual ultrafiltration rate, dialysate sodium concentration set value, blood flow rate, and dialysate flow rate from the dialysis machine control system.

[0028] The dialyzer parameter acquisition unit 103 is used to obtain the sodium clearance rate (K_Na) of the dialyzer currently in use. This parameter can be estimated by combining the specifications in the dialyzer manual with the current blood flow rate and dialysate flow rate.

[0029] The historical data interface unit 104 is used to access the patient's past dialysis records from the hospital information system, including prescription parameters for each dialysis session, actual ultrafiltration volume, blood pressure change curves, and weight gain patterns between dialysis sessions.

[0030] Example 3: Specific Structure of the Inference Module like Figure 2As shown, the inference module includes a sodium kinetics calculation engine 201, a safety constraint evaluator 202, and a prescription optimizer 203, which are connected in sequence.

[0031] The input terminal of the sodium kinetics calculation engine 201 is connected to the sensing module 100, receiving real-time acquired patient physiological parameters and dialysis machine operating parameters. The sodium kinetics calculation engine 201 has a pre-built dual-chamber sodium kinetics model with the equation: C_eq = [Na]d / σ (where σ=0.97), employing a perturbation ICF buffer model without solvent drag constraints. The sodium kinetics calculation engine 201 calculates the current cumulative sodium clearance in real time based on the input parameters and predicts the serum sodium concentration and total sodium clearance at the end of dialysis. The calculation results are transmitted to the safety constraint evaluator 202.

[0032] The safety constraint evaluator 202 receives prediction results from the sodium kinetics calculation engine 201 and real-time blood pressure monitoring data from the sensing module 100. The safety constraint evaluator 202 has three preset safety thresholds: ultrafiltration rate <10 ml / h / kg, post-dialysis serum sodium within the range of 135-145 mmol / L, and blood pressure reduction during dialysis <20 mmHg. The evaluator determines in real time whether the current state and prediction results meet the above safety constraints and transmits the evaluation results to the prescription optimizer 203.

[0033] The prescription optimizer 203 receives the prediction results from the sodium kinetics calculation engine 201 and the evaluation results from the safety constraint evaluator 202. The prescription optimizer 203 has a pre-defined optimization objective function: sodium removal reaches the target value Na_target = ΔW × [Na]s / 0.93. Under the premise of meeting safety constraints, the prescription optimizer 203 performs combined optimization on three adjustable parameters—dialysis fluid sodium concentration [Na]d, ultrafiltration volume V_UF (or ultrafiltration rate), and dialysis time t—to generate the optimal combination of prescription parameters. The optimization results are transmitted to the action module 300.

[0034] Example 4: Specific Structure of the Action Module like Figure 1 and Figure 3 As shown, the action module includes a predialysis prescription generation unit 301, a dynamic control unit during dialysis 302, an early warning unit 303, and a recording unit 304.

[0035] The input terminal of the predialysis prescription generation unit 301 is connected to the inference module 200, receiving the optimized initial prescription parameters. This unit displays the recommended combination of dialysate sodium concentration, ultrafiltration volume, and dialysis time on the user interface for medical staff to confirm or adjust.

[0036] The input terminal of the dynamic control unit 302 during dialysis is connected to the inference module 200, continuously receiving real-time optimization results during dialysis. When the predicted sodium removal rate deviates from the target value, the control unit automatically adjusts the dialysate sodium concentration setpoint and sends the adjustment command to the dialysis machine control system. Figure 3 As shown, the sodium concentration in the dialysate exhibits a dynamic change curve with dialysis time. When the blood pressure monitoring data exceeds the drop threshold, the control unit automatically reduces the ultrafiltration rate and correspondingly extends the dialysis time to ensure that the total ultrafiltration volume reaches the target value.

[0037] The input terminal of the early warning unit 303 is connected to the inference module 200. When the predicted parameters exceed the safe range, the early warning unit will display an early warning prompt on the user interface and remind medical staff to pay attention through sound or light signals.

[0038] The input terminal of the recording unit 304 is connected to each unit inside the action module 300 to record complete sodium balance data for each dialysis session, including prescription parameters, real-time adjustment records, and final dialysis results, and transmits the recorded data to the feedback module 400.

[0039] Example 5: Specific Structure of the Feedback Module like Figure 1 and Figure 4 As shown, the feedback module includes a prediction-actual comparison unit 401, a parameter calibration unit 402, and an individualized profile unit 403.

[0040] The input of the prediction-actual comparison unit 401 is connected to the recording unit 304, receiving actual dialysis result data and simultaneously acquiring model prediction data before dialysis from the inference module 200. The comparison unit compares the actual serum sodium changes and actual sodium clearance with the model prediction values ​​and calculates the prediction error. The comparison results are transmitted to the parameter calibration unit 402.

[0041] The parameter calibration unit 402 receives the comparison results and automatically calibrates key parameters in the sodium kinetic model based on the prediction error, including sodium dialyzer clearance rate K_Na, intracellular sodium transfer coefficient K_ic, and permeability coefficient Lp. The calibration algorithm uses gradient descent or Kalman filtering to make the model prediction gradually approximate the patient's actual physiological response. The calibrated parameters are then transmitted to the individualized profile unit 403.

[0042] The individualized profile unit 403 receives the calibrated parameters and establishes an independent sodium kinetic parameter profile for each patient. This profile is stored in a database. During the patient's next dialysis session, the inference module 200 automatically retrieves the individualized parameters from this profile for prediction and optimization, thereby improving the prediction accuracy and prescription optimization effect of subsequent dialysis.

[0043] Example 6: Pre-dialysis prescription generation Patient's basic parameters: predialysis weight 67 kg, dry weight 65 kg, predialysis serum sodium [Na]s = 140 mmol / L. The system automatically calculates the target sodium clearance: Na_target = (67-65) × 140 / 0.93 = 301 mmol. The inference module searches for the optimal prescription combination under safety constraints based on a two-chamber sodium kinetic model. The system recommends: dialysate sodium concentration [Na]d = 138 mmol / L, ultrafiltration volume 2.0 L, dialysis time 4 hours. The predicted sodium clearance is 326 mmol, R_Na = 1.08, indicating adequate dialysis.

[0044] Example 7: Dynamic Adjustment During Dialysis like Figure 3 As shown, the system updates the sodium balance calculation every 10 minutes after dialysis begins. At the 90th minute of dialysis, the patient's blood pressure dropped from 140 / 80 to 115 / 70 mmHg (a decrease of 25 mmHg, exceeding the 20 mmHg threshold). The action module automatically performed: (1) increasing the dialysate sodium concentration from 138 to 142 mmol / L to reduce diffuse sodium excretion and stabilize blood pressure; (2) decreasing the ultrafiltration rate from 8.3 ml / h / kg to 6.0 ml / h / kg; (3) extending the dialysis time from 4 hours to 4.5 hours to maintain the total ultrafiltration volume. At the 150th minute of dialysis, after the blood pressure stabilized, the system gradually reduced the dialysate sodium concentration back to 136 mmol / L, increasing diffuse sodium excretion for the remaining time to make up for the previous sodium clearance deficit.

[0045] Example 8: Feedback Learning and Parameter Calibration like Figure 4 As shown, after dialysis, the system compared the predicted and actual values: the predicted serum sodium level after dialysis was 138.2 mmol / L, while the actual measured value was 137.5 mmol / L, a deviation of 0.7 mmol / L. The feedback module analyzed the source of the deviation and increased the sodium dialyzer clearance rate K_Na from 0.025 to 0.027 L / min, improving the prediction accuracy for the next dialysis session. After 5-10 iterative calibrations using dialysis, the model prediction error converged to within ±0.3 mmol / L.

[0046] Example 9: System Workflow The following is combined with Figures 1 to 5 The complete workflow of the system of the present invention is described below: First, before the patient begins dialysis, the perception module 100 collects basic data such as the patient's pre-dialysis weight, dry weight, and serum sodium concentration, and retrieves the patient's previous dialysis records through the historical data interface unit 104. Based on this data and the patient's individualized sodium kinetic parameter profile, the reasoning module 200 predicts the sodium clearance requirement for this dialysis session through the sodium kinetic calculation engine 201, assesses the safety boundaries through the safety constraint evaluator 202, and generates an optimized initial prescription through the prescription optimizer 203. The pre-dialysis prescription generation unit 301 of the action module 300 displays the initial prescription on the user interface for confirmation by medical staff.

[0047] After dialysis begins, the sensing module 100 continuously collects real-time blood pressure monitoring data and dialysis machine operating parameters. The inference module 200 updates the status and makes predictions every 5 minutes. When the predicted sodium clearance deviates from the target value by more than 5%, the prescription optimizer 203 recalculates the optimization parameters. The dynamic control unit 302 in the action module 300 adjusts the dialysate sodium concentration setpoint based on the optimization results. When blood pressure drops by more than 20 mmHg, the control unit automatically reduces the ultrafiltration rate and calculates the required extended dialysis time to ensure that the total ultrafiltration volume reaches the target value.

[0048] During dialysis, if the reasoning module 200 predicts that any parameter is about to exceed the safe range (such as predicting that serum sodium will be lower than 135 mmol / L after dialysis), the early warning unit 303 will immediately issue an early warning to the medical staff, indicating that manual intervention is required.

[0049] After dialysis, the recording unit 304 of the action module 300 saves the complete data of this dialysis session. The prediction-actual comparison unit 401 of the feedback module 400 compares the actual results with the model predictions, the parameter calibration unit 402 calibrates the model parameters based on the comparison results, and the individualized profile unit 403 updates the patient's sodium kinetic parameter profile for prescription optimization for the next dialysis session.

[0050] Example 10: System Integration Method like Figure 5 As shown, the system of the present invention can be integrated into existing dialysis machine control systems or hospital dialysis information management systems via software.

[0051] In the first integration method, the system of the present invention operates as a functional module of the dialysis machine control system, directly adjusting parameters such as sodium concentration and ultrafiltration rate of the dialysate through the control interface of the dialysis machine, and realizing human-computer interaction through the display screen of the dialysis machine.

[0052] In the second integration method, the system of this invention runs as an independent server within the hospital's intranet, exchanging data with the dialysis machine control system and the hospital information system via standard communication protocols (such as HL7 and FHIR). Medical staff can access the system interface via mobile terminals or workstations.

[0053] As can be seen from the above embodiments, the present invention achieves the following beneficial effects: (1) For the first time, the sodium dynamics mathematical model was embedded into the AI ​​Agent architecture to realize closed-loop intelligent control of dialysis prescriptions, upgrading from "open-loop experience prescriptions" to "closed-loop precision prescriptions", significantly improving the individualization level of dialysis treatment.

[0054] (2) The four-module architecture of perception-reasoning-action-feedback enables the system to have autonomous optimization capabilities, dynamically adjust prescription parameters based on real-time data, effectively cope with various changes in the dialysis process, and improve the adaptability and safety of treatment.

[0055] (3) Through a multi-constraint optimization mechanism, the sodium clearance effect is maximized under the premise of ensuring safety, while satisfying the three constraints of ultrafiltration rate safety, serum sodium range and hemodynamic stability, thereby reducing the risk of dialysis complications.

[0056] (4) The feedback learning mechanism enables the system to continuously improve. The results of each dialysis are used to calibrate the model parameters, and the prediction accuracy increases with the number of uses, thus realizing true personalized precision medicine.

[0057] (5) The system can be integrated with existing dialysis machines and dialysis information systems without replacing dialysis equipment. It can be made intelligent through software upgrades, and has good compatibility and promotion value.

[0058] (6) The modules are closely connected and the data flows smoothly. While realizing multiple functions, the system ensures the clarity and scalability of the overall architecture, making it easy to deploy and use in actual clinical environments.

[0059] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0060] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An intelligent hemodialysis prescription optimization system based on AI Agent, characterized in that, include: The sensing module is used to collect patient physiological parameters, dialysis machine operating parameters, dialyzer parameters, and historical dialysis data in real time. The patient physiological parameters include predialysis weight, dry weight, blood pressure, and serum sodium concentration. The dialysis machine operating parameters include actual ultrafiltration rate, dialysate sodium concentration setpoint, blood flow rate, and dialysate flow rate. The inference module includes a sodium kinetics calculation engine, a safety constraint evaluator, and a prescription optimizer. The sodium kinetics calculation engine is based on a two-chamber sodium kinetics model and uses the diffusion equilibrium concentration formula C_eq = [Na]d / σ (where σ = 0.97 is the Gibbs-Donnan factor, reflecting the influence of plasma proteins on mobile sodium concentration) to calculate cumulative sodium clearance in real time and predict serum sodium concentration and total sodium clearance at the end of dialysis. The safety constraint evaluator assesses ultrafiltration rate safety, serum sodium range, and hemodynamic stability in real time. The prescription optimizer, under the premise of meeting safety constraints, optimizes the combination of dialysate sodium concentration, ultrafiltration volume, and dialysis time to achieve the target sodium clearance value Na_target = (W_pre - W_dry) × [Na]s / 0.

93. The action module is used to generate optimized initial prescription suggestions before dialysis, dynamically adjust dialysate sodium concentration and ultrafiltration rate during dialysis, and issue warnings to medical staff when predicted parameters exceed safe ranges. The feedback module is used to compare dialysis results with model predictions, automatically calibrate model parameters, and establish individualized sodium kinetic parameter profiles for patients.

2. The system as described in claim 1, characterized in that, The dual-chamber sodium dynamics model in the sodium dynamics calculation engine divides the patient's body fluid into extracellular fluid chambers and intracellular fluid chambers, establishing a coupled set of differential equations: Ve × dCe / dt = -K_Na × (Ce - C_eq) - j_ic d(Vi×Ci) / dt = j_ic Among them, cell membrane aquaporins transport only water and do not carry sodium ions, and water transmembrane transport does not produce solvent-drafted sodium transport.

3. The system as described in claim 1, characterized in that, The safety constraints assessed in real time by the safety constraint evaluator include: ultrafiltration rate not exceeding 10 ml / h / kg, serum sodium after dialysis within the range of 135-145 mmol / L, and blood pressure reduction during dialysis not exceeding 20 mmHg.

4. The system as described in claim 1, characterized in that, The dynamic adjustment strategy of the action module during dialysis includes: when the predicted sodium clearance at the end of dialysis is lower than the target value, reducing the sodium concentration of the dialysate to increase diffuse sodium excretion; when the blood pressure drops above a preset threshold, increasing the sodium concentration of the dialysate and reducing the ultrafiltration rate; and when the ultrafiltration rate approaches a safe threshold, automatically extending the dialysis time to maintain the total ultrafiltration volume unchanged.

5. The system as described in claim 1, characterized in that, The parameter calibration method of the feedback module is as follows: compare the model-predicted serum sodium after dialysis with the actual measured value, and automatically adjust the sodium dialyzer clearance rate K_Na, intracellular and extracellular sodium transport coefficient K_ic, and permeability coefficient Lp by minimizing the prediction error, so that the model gradually adapts to the individual sodium dynamics characteristics of the patient.

6. The system as described in claim 1, characterized in that, The system is integrated into the dialysis information system or dialysis machine. It automatically obtains ultrafiltration volume and dialysate sodium concentration parameters from the dialysis machine through a standard communication interface, obtains patient weight and serum sodium data from the electronic medical record, and automatically generates a sodium balance assessment report at the end of dialysis.

7. A method for optimizing intelligent hemodialysis prescriptions based on AI Agents, characterized in that, include: Predialysis phase: Collect the patient's predialysis weight and serum sodium concentration, calculate the target sodium clearance Na_target = (W_pre - W_dry) × [Na]s / 0.93, and generate an initial prescription suggestion based on a two-chamber sodium kinetic model; During dialysis: Sodium balance calculation is updated based on real-time data at preset time intervals. When the predicted sodium clearance at the end of dialysis deviates from the target, the sodium concentration of the dialysate is dynamically adjusted. When the ultrafiltration rate approaches the safety threshold or the blood pressure drops beyond the preset value, the ultrafiltration rate and dialysis time are automatically adjusted. Post-dialysis stage: Calculate the actual sodium clearance and sodium clearance ratio R_Na of this dialysis, feed the results back to the model parameter calibration module, and generate a dialysis report including sodium balance analysis.

8. The method as described in claim 7, characterized in that, The real-time data update frequency during the dialysis phase is once every 5-15 minutes. Each time an update is performed, the dual-chamber sodium kinetic model is re-run to predict the amount of sodium removed during the remaining dialysis time.

9. The method as described in claim 7, characterized in that, The initial prescription recommendation includes a dialysate sodium concentration control curve [Na]d(t) for the entire dialysis process. The design principle of the control curve is: a higher dialysate sodium concentration is used in the first half of dialysis to maintain hemodynamic stability, and the dialysate sodium concentration is gradually reduced in the second half of dialysis to increase diffuse sodium excretion and ensure that the total sodium clearance reaches the target.

Images