Artificial intelligence-based crrt treatment parameter processing system

Abstract

The application relates to the technical field of medical equipment, and discloses a CRRT treatment parameter processing system based on artificial intelligence, which comprises a data acquisition unit, a data preprocessing and feature extraction unit, a physiological model and prediction unit, an intelligent decision and instruction generation unit, a treatment control and execution unit, and an information interaction and monitoring unit; the method comprises the following steps: collecting CRRT parameters in real time, preprocessing and extracting characteristic data such as urea reduction rate, and constructing an individual physiological model to predict a trend; then, through multi-level AI decision, a regulation instruction is generated by containing rules and reinforcement learning to drive equipment execution, and information is displayed to medical staff and intervention instructions are received. The application realizes individual real-time optimization and risk early warning by regulating and controlling CRRT parameters through an AI closed loop; in combination with precise data processing, multi-level intelligent decision and man-machine interaction, the treatment accuracy, safety and efficiency are improved, and manual intervention and testing are reduced.

Description

Technical Field

[0001] This invention relates to the field of medical device technology, specifically to an artificial intelligence-based CRRT treatment parameter processing system. Background Technology

[0002] Continuous renal replacement therapy (CRRT) is a vital life support technology in modern medicine for treating critically ill patients. Through extracorporeal blood purification, it continuously removes excess water and harmful solutes from the patient's body, maintaining internal homeostasis. With the deepening development of artificial intelligence (AI) technology in the medical field, its potential to improve the precision, safety, and efficiency of complex treatment processes is receiving increasing attention. AI technology, especially its autonomous learning and adaptability, combined with the traditional algorithmic system's pursuit of efficiency and determinism, provides new ideas for building more intelligent medical devices.

[0003] In the prior art related to this invention, CRRT devices have seen continuous advancements at the hardware level, such as in precise fluid control. Meanwhile, monitoring technologies, such as online clearance monitoring (OCM), have been used to assess the clearance of specific small molecule toxins, providing some reference for clinical decision-making. However, these technologies primarily focus on single or limited dimensions of treatment evaluation.

[0004] Despite the progress made in existing CRRT technology, significant challenges remain in clinical practice. The effectiveness of CRRT treatment is highly dependent on the accurate understanding and timely response to the patient's dynamically changing physiological state. Currently, monitoring of key biochemical indicators relies mainly on intermittent laboratory tests, which struggles to capture real-time changes in the patient's internal environment, often leading to delays in adjusting treatment parameters. More importantly, existing CRRT systems generally lack intelligent and automated control capabilities based on individual patient physiological feedback, failing to automatically optimize treatment parameters, such as adjusting dialysate / replacement fluid composition or ultrafiltration rate, based on monitored dynamic equilibrium changes. This limitation not only increases the burden on healthcare professionals but also restricts further improvements in the individualization, refinement, and proactive risk management of CRRT treatment, making it difficult to adequately address the complex and rapidly changing conditions of critically ill patients.

[0005] Therefore, this invention proposes an artificial intelligence-based CRRT treatment parameter processing system to address the shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based CRRT treatment parameter processing system, which solves the problem that in continuous renal replacement therapy, the adjustment of treatment parameters mainly relies on human experience and intermittent monitoring data, making it difficult to achieve intelligent, individualized, and automatic optimization and adjustment based on the patient's real-time dynamic physiological state.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an artificial intelligence-based CRRT treatment parameter processing system, the system comprising the following modules:

[0008] The data acquisition unit is used to acquire patient physiological and treatment parameters from the CRRT circuit in real time and generate raw digital signals.

[0009] The data preprocessing and feature extraction unit is used to preprocess and feature engineer the original digital signal to generate structured feature data containing urea reduction rate, creatinine clearance rate and urea clearance index.

[0010] The physiological model and prediction unit are used to construct and dynamically calibrate an individualized physiological model based on the patient's baseline information and the structured feature data, and to use the individualized physiological model to predict the trend of physiological parameter changes and generate physiological parameter change trend prediction information.

[0011] The intelligent decision-making and instruction generation unit is used to analyze the treatment status and generate treatment parameter adjustment instructions and decision basis based on the structured feature data and the physiological parameter change trend prediction information, using a multi-level artificial intelligence decision-making logic that includes rule-based fast response logic and reinforcement learning-based treatment optimization logic.

[0012] The treatment control and execution unit is used to receive the adjustment command, convert it into a control signal, and drive the device actuator to adjust the parameters.

[0013] The information interaction and monitoring unit is used to graphically display treatment-related information, decision-making basis, and alarm information to medical staff, and to receive intervention instructions from medical staff that affect the generation of subsequent adjustment instructions.

[0014] Preferably, the data acquisition unit includes an electrolyte sensor and a metabolite sensor, which are deployed at the filter inlet, filter outlet, and filtrate outlet of the CRRT circuit; the electrolyte sensor is used to monitor the electrolyte concentration in the fluid at the corresponding deployment location, and the metabolite sensor is used to monitor the metabolite concentration in the fluid at the corresponding deployment location.

[0015] Preferably, when the data preprocessing and feature extraction unit performs the feature engineering operation to generate the structured feature data, it specifically performs the calculation of the urea reduction rate; to calculate the urea reduction rate, the data preprocessing and feature extraction unit obtains the urea concentration of the blood at the inlet of the CRRT treatment loop filter, which is collected in real time by the data acquisition unit, and denoted as C. Urea,pre And the urea concentration in the blood at the outlet of the CRRT treatment circuit filter, denoted as C. Urea,postBased on the obtained urea concentration, the following formula is used to calculate the urea decrease rate:

[0016]

[0017] In the formula: URR represents the urea reduction rate; C Urea,pre This indicates the urea concentration in the blood at the inlet of the filter in the CRRT treatment circuit; C Urea,post This indicates the urea concentration in the blood at the outlet of the filter in the CRRT treatment circuit.

[0018] Preferably, when performing the feature engineering operation to generate the structured feature data, the data preprocessing and feature extraction unit also specifically performs the calculation of creatinine clearance rate; to calculate the creatinine clearance rate, the data preprocessing and feature extraction unit obtains the creatinine concentration in the CRRT treatment circuit filtrate collected in real time by the data acquisition unit, denoted as C. Crea,uf Obtain the current ultrafiltration rate of the CRRT device, denoted as Q. UF The system acquires the creatinine concentration of the blood at the inlet of the CRRT treatment circuit filter, which is collected in real time by the data acquisition unit and denoted as C. Crea,pre Based on the obtained creatinine concentration and the ultrafiltration rate, the creatinine clearance rate is calculated using the following formula:

[0019]

[0020] In the formula: CrCl represents the creatinine clearance rate; C Crea,uf Q represents the creatinine concentration in the filtrate of the CRRT treatment circuit; UF This indicates the ultrafiltration rate of the CRRT device; C Crea,pre This indicates the creatinine concentration of the blood at the inlet of the filter in the CRRT treatment circuit.

[0021] Preferably, when performing the dynamic calibration operation of the individualized physiological model, the physiological model and prediction unit uses the extended Kalman filter algorithm or the recursive least squares algorithm, and adjusts specific model parameters within the individualized physiological model online and continuously based on the structured feature data generated by the data preprocessing and feature extraction unit; the specific model parameters include parameters that characterize the effective distribution volume of the target solute in the patient's body, and parameters that characterize the effective clearance capacity of the CRRT filter for the specific solute.

[0022] Preferably, the rule-based rapid response logic in the multi-level artificial intelligence decision-making logic adopted by the intelligent decision-making and instruction generation unit includes a pre-set rule set composed of multiple IF-THEN structure rules; this rule set compares and judges the current values ​​of the patient's physiological parameters collected in real time by the data acquisition unit and the rate of change of these physiological parameter values ​​with various preset safety thresholds in the system; when the comparison and judgment result indicates that the patient has an acute physiological imbalance, the rule-based rapid response logic automatically generates treatment parameter adjustment instructions for immediate intervention.

[0023] Preferably, the reinforcement learning-based treatment optimization logic in the multi-level artificial intelligence decision-making logic adopted by the intelligent decision-making and instruction generation unit is used to dynamically optimize the ultrafiltration rate during CRRT treatment. The reinforcement learning-based treatment optimization logic makes decisions based on a state space, which includes the urea reduction rate, creatinine clearance rate, and cumulative fluid balance in the structured feature data generated by the data preprocessing and feature extraction unit, as well as the predicted values ​​of specific physiological parameters in the physiological parameter change trend prediction information generated by the physiological model and prediction unit. The reinforcement learning-based treatment optimization logic uses a reward function to guide the optimization of the ultrafiltration rate. The reward function is used to comprehensively evaluate the solute clearance efficiency, the achievement of fluid balance goals, and the maintenance of patient physiological stability in CRRT treatment.

[0024] Preferably, before finally determining the specific treatment parameter adjustment instruction to be output, the intelligent decision-making and instruction generation unit also performs a decision pre-simulation operation. The decision pre-simulation operation uses the individualized physiological model provided by the physiological model and prediction unit as a simulation platform to simulate the dynamic change trajectory of the parameter adjustment instruction on various physiological parameters of the patient within a preset time for the candidate treatment parameter adjustment instructions. The intelligent decision-making and instruction generation unit then performs a risk-benefit assessment on each candidate treatment parameter adjustment instruction based on the simulated dynamic change trajectory, and confirms or modifies the finally generated treatment parameter adjustment instruction based on the assessment results.

[0025] Preferably, the alarm information displayed by the information interaction and monitoring unit includes:

[0026] A parameter anomaly alarm is triggered when the patient's physiological parameter values ​​collected by the data acquisition unit exceed the preset safety range or the preset critical value range.

[0027] A technical fault alarm is triggered when a sensor or other component of the system within the data acquisition unit malfunctions.

[0028] A predictive risk alarm is triggered when the intelligent decision-making and instruction generation unit predicts, based on analysis, that the patient is about to experience a potential physiological risk.

[0029] The parameter anomaly alarm, the technical fault alarm, and the predictive risk alarm all include alarm category information, parameter identifiers related to the alarm, parameter value information when the alarm is triggered, and system-suggested countermeasures.

[0030] This invention also provides an artificial intelligence-based method for processing CRRT treatment parameters, the method comprising:

[0031] S1. The patient's physiological and treatment parameters are collected in real time from the CRRT circuit through the data acquisition unit to generate raw digital signals;

[0032] S2. The data preprocessing and feature extraction unit performs preprocessing and feature engineering operations on the original digital signal to generate structured feature data, which includes urea reduction rate, creatinine clearance rate and urea clearance index.

[0033] S3. Based on the patient's baseline information and the generated structured feature data, construct and dynamically calibrate the patient's individualized physiological model using the physiological model and prediction unit, and use the individualized physiological model to predict the trend of the patient's physiological parameters in the future to generate physiological parameter change trend prediction information.

[0034] S4. By combining the structured feature data and the physiological parameter change trend prediction information through the intelligent decision-making and instruction generation unit, a multi-level artificial intelligence decision-making logic including rule-based fast response logic and reinforcement learning-based treatment optimization logic is adopted to analyze the current treatment status and generate CRRT treatment parameter adjustment instructions and decision basis information corresponding to the adjustment instructions.

[0035] S5. The treatment control and execution unit receives the adjustment command of the CRRT treatment parameters and converts the adjustment command into a control signal for the corresponding actuator in the CRRT device, so as to drive the actuator to perform the adjustment of the CRRT treatment parameters;

[0036] S6. The information interaction and monitoring unit graphically displays treatment-related information, decision-making basis information for the adjustment instructions, and necessary alarm information to medical staff, and receives intervention instructions input by medical staff, which are used to influence the adjustment instructions of subsequent CRRT treatment parameters.

[0037] This invention provides an artificial intelligence-based CRRT treatment parameter processing system. It has the following beneficial effects:

[0038] 1. This invention integrates artificial intelligence algorithms for closed-loop management and intelligent control of CRRT treatment parameters, achieving a high degree of individualization and real-time optimization of treatment plans, and improving the response speed and accuracy to the dynamically changing physiological needs of critically ill patients. Compared with the existing technology that relies on the experience of medical staff, periodic testing, and manual parameter adjustment, this invention overcomes the shortcomings of its often lagging adjustments and inability to fully adapt to individual patient differences and rapid disease evolution, reducing unnecessary human intervention through AI optimization.

[0039] 2. This invention utilizes a dynamically calibrated individualized physiological model and prediction unit, which can proactively infer the future trends of key physiological indicators in patients based on real-time monitoring data, thereby achieving early warning of potential treatment risks (such as hemodynamic fluctuations). Compared with existing technologies that mainly rely on current indicators for alarms and lack the ability to predict trend deterioration, this invention solves the problem of passive risk prevention and difficulty in actively avoiding adverse events in CRRT treatment.

[0040] 3. This invention constructs a data preprocessing and feature extraction module, which can efficiently extract core efficacy indicators such as urea reduction rate and creatinine clearance rate from continuously monitored raw signals. Combined with multi-level artificial intelligence decision-making logic, it achieves more precise and dynamic CRRT treatment guidance. This invention improves upon existing methods that insufficiently utilize multi-source data and rely primarily on a small number of discrete biochemical indicators for judgment, addressing their limitations in comprehensively assessing treatment effectiveness and rapidly adjusting treatment strategies, while reducing the need for frequent blood tests.

[0041] 4. This invention includes an intelligent decision-making and instruction generation unit with rapid response rules, reinforcement learning optimization, and decision pre-playing mechanisms, supplemented by a comprehensive alarm system, which greatly enhances the safety and efficiency of CRRT treatment. Compared to traditional CRRT systems with relatively simple alarm logic and a lack of intelligent decision support for complex situations, this invention effectively solves the problems of insufficient response to acute physiological imbalances, reliance on human experience for treatment parameter optimization, and inadequate potential risk assessment, thereby improving the overall safety and operational efficiency of treatment.

[0042] 5. This invention utilizes a user-friendly information interaction and monitoring unit that clearly displays the treatment process and the rationale behind AI decision-making. It also provides medical staff with a convenient interface for supervision and intervention, realizing a new paradigm of intelligent treatment through human-machine collaboration. Compared to the potential problems of AI decision-making "black boxes" or poor human-machine interaction in existing technologies, this invention promotes medical staff's understanding and trust in the intelligent system, ensures the effective integration of clinical experience and AI intelligence, optimizes the allocation of human resources, and guarantees medical staff's ultimate control over the treatment process. Attached Figure Description

[0043] Figure 1 This is a system architecture diagram of the present invention;

[0044] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0045] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Please see Figure 1 This invention provides an artificial intelligence-based CRRT treatment parameter processing system, comprising the following steps:

[0047] The data acquisition unit is used to acquire patient physiological and treatment parameters from the CRRT circuit in real time and generate raw digital signals.

[0048] This data acquisition unit is used to continuously and in real-time acquire patient-related physiological parameters and treatment parameters related to the operation of the CRRT device from the continuous renal replacement therapy (CRRT) circuit. This acquired information is then converted into raw digital signals by the unit, providing a basis for subsequent data processing and analysis.

[0049] Specifically, the data acquisition unit is equipped with various types of sensors to achieve accurate monitoring of key parameters. These sensors include electrolyte sensors for monitoring the concentration of specific electrolytes in the fluid, and metabolite sensors for monitoring the concentration of specific metabolites in the fluid.

[0050] In one possible implementation, to comprehensively capture changes in solute exchange and patient fluid composition during CRRT, the electrolyte sensor and the metabolite sensor are positioned at multiple key monitoring points in the CRRT treatment circuit. These monitoring points are preferably located at the filter inlet of the CRRT circuit, where the composition of the blood before entering the filter can be monitored; at the filter outlet, where the composition of the blood after processing by the filter can be monitored; and at the filtrate outlet, where the composition of waste fluid removed from the patient can be monitored. This helps to comprehensively understand the filter's operating status and solute removal.

[0051] The electrolyte sensors, each deployed at its respective location, are responsible for real-time monitoring of the concentration of one or more electrolytes in the fluid flowing through that location. These electrolytes, such as sodium ions, potassium ions, calcium ions, and chloride ions, have a direct impact on maintaining the stability of the patient's internal environment due to their dynamic changes in concentration.

[0052] Similarly, the metabolite sensor, at its deployment location, is responsible for real-time monitoring of the concentration of one or more metabolic waste products in the fluid flowing through that location. These metabolites, such as urea and creatinine, are key indicators for assessing renal function and the effectiveness of CRRT treatment. Real-time tracking of these metabolite concentrations helps in evaluating the immediate effects of treatment.

[0053] Alternatively, in addition to the specially deployed electrolyte and metabolite sensors mentioned above, the data acquisition unit can also integrate and receive data from sensors built into the CRRT device itself. This data includes blood pathway pressures reported by the CRRT device, such as arterial pressure, venous pressure, and transmembrane pressure.

[0054] In some embodiments, the data acquisition unit can also connect to and integrate data from external patient monitoring devices. These external monitoring devices can continuously monitor the patient's vital signs, such as systemic arterial blood pressure, heart rate, and body temperature; by integrating multi-source data, a more comprehensive and three-dimensional assessment of the patient's overall physiological state can be formed.

[0055] Therefore, the range of physiological parameters collected by this data acquisition unit is extensive, including urea concentration, creatinine concentration, concentration of various electrolytes, blood pressure, heart rate, body temperature, and various pressure monitoring values ​​in the CRRT circuit.

[0056] Meanwhile, the treatment parameters collected by this data acquisition unit mainly reflect the actual operating status of the CRRT equipment, including the blood pump flow rate, dialysate pump flow rate, replacement fluid pump flow rate, ultrafiltration pump flow rate, and infusion rate of anticoagulants, such as citrate or heparin, which are set or actually running on the CRRT equipment. These parameters are directly related to treatment intensity and solute clearance efficiency.

[0057] Typically, various sensors installed in CRRT circuits or integrated into related equipment convert detected physical quantities, such as pressure, or chemical quantities, such as the concentration of a specific substance, into corresponding analog electrical signals. In order for these signals to be recognized and processed by subsequent digital processing units, the data acquisition unit or its closely integrated modules contain necessary signal conditioning and conversion circuits.

[0058] Specifically, analog electrical signals undergo a series of processing steps. For example, signal amplification circuits enhance potentially weak sensor signals to improve the signal-to-noise ratio and ensure measurement accuracy. Signal filtering circuits, such as low-pass or band-pass filters, remove environmental noise and high-frequency interference components from the signal, purifying it into a valid signal. Subsequently, the conditioned analog signal is fed into an analog-to-digital converter, which converts the continuously changing analog signal into a discrete digital signal sequence at a preset sampling frequency and a certain digital precision.

[0059] To ensure data integrity and traceability for subsequent analysis, each group or individual digital signal sample is assigned a precise timestamp. This timestamp records the exact moment the data was acquired, which is crucial for subsequent time series analysis, dynamic trend assessment, and correlation analysis between different parameters.

[0060] The data acquisition unit outputs one or more raw digital signal streams with timestamps. These signal streams constitute a real-time, unprocessed raw record of the patient's physiological state and the CRRT treatment process, and are stably transmitted to the data preprocessing and feature extraction unit in the system.

[0061] The data preprocessing and feature extraction unit is used to preprocess and feature engineer the original digital signal to generate structured feature data containing urea reduction rate, creatinine clearance rate and urea clearance index.

[0062] In this implementation, the data preprocessing and feature extraction unit receives the raw digital signals generated by the aforementioned data acquisition unit and performs a series of preprocessing and feature engineering operations on these signals. The aim is to transform the raw, potentially noisy or irregular data into structured feature data that has significant value for subsequent intelligent analysis and decision-making. This structured feature data includes key evaluation indicators such as urea reduction rate, creatinine clearance rate, and urea clearance index.

[0063] In the data preprocessing stage, this unit focuses on improving the quality and usability of the data.

[0064] In general, raw digital signals may contain noise from various sources, such as fluctuations introduced by sensor characteristics, electromagnetic interference, or patient physiological activities. To remove these noise components, improve the signal-to-noise ratio, and thus ensure the accuracy of subsequent analysis, this unit employs digital filtering techniques. Alternatively, median filtering, Gaussian filtering, or adaptive filtering algorithms can be selected based on the signal characteristics and noise type.

[0065] Considering that during long-term continuous monitoring, some data points may be missing due to temporary sensor failure, signal transmission interruption, or other unexpected situations, this unit performs missing data processing to ensure the integrity of the data sequence and avoid affecting the continuity and reliability of the analysis results due to missing data. Specifically, mathematical interpolation algorithms, such as linear interpolation or spline interpolation, can be used to estimate and fill missing values ​​based on the valid data points before and after the missing points. In some embodiments, predictive models based on historical data can also be used to fill more complex missing patterns.

[0066] Furthermore, since data may originate from different types of sensors or devices, their time bases and dimensions may differ. To ensure temporal consistency and numerical comparability of multi-source data, this unit performs data calibration and synchronization. Using timestamp information provided by the data acquisition unit, data points from different data sources are precisely time-aligned. Simultaneously, measurements from different sensors are calibrated as needed to eliminate systematic errors and standardize data dimensions.

[0067] After data preprocessing, this unit further performs feature engineering operations. The purpose of feature engineering is to extract or calculate characteristic indicators from the preprocessed clean data that can more directly and profoundly reflect the treatment effect of CRRT and the evolution of the patient's physiological state.

[0068] In this embodiment, when the data preprocessing and feature extraction unit performs the feature engineering operation to generate the structured feature data, it specifically calculates the urea reduction rate. To calculate the urea reduction rate, the data preprocessing and feature extraction unit obtains the real-time urea concentration of the blood at the inlet of the CRRT treatment loop filter from the data acquisition unit; this concentration is denoted as C. Urea,pre Furthermore, the urea concentration in the blood at the outlet of the CRRT treatment circuit filter is obtained, and this concentration is denoted as C. Urea,post Based on the obtained urea concentrations of blood at the filter inlet and outlet, the following formula is used to calculate the urea reduction rate:

[0069] In the formula: URR represents the urea reduction rate, which directly reflects the urea removal efficiency of CRRT treatment during a single filter flush; C Urea,pre This indicates the urea concentration in the blood at the inlet of the filter in the CRRT treatment circuit; C Urea,post This indicates the urea concentration in the blood at the outlet of the filter in the CRRT treatment circuit.

[0070] In one possible implementation, the data preprocessing and feature extraction unit, while performing the feature engineering operation, also specifically calculates the creatinine clearance rate. Creatinine is another important metabolite for evaluating renal function and dialysis effectiveness. To calculate the creatinine clearance rate, the data preprocessing and feature extraction unit obtains the real-time creatinine concentration in the CRRT treatment circuit filtrate from the data acquisition unit, denoted as C0. Crea,uf Simultaneously, this unit obtains the current actual ultrafiltration rate of the CRRT device from the CRRT device or through a data acquisition unit, and this rate is denoted as Q. UF Furthermore, this unit also obtains the real-time creatinine concentration of the blood at the inlet of the CRRT treatment circuit filter from the data acquisition unit, and this concentration is denoted as C. Crea,preBased on the obtained creatinine concentration in the filtrate, the ultrafiltration rate, and the creatinine concentration in the blood at the filter inlet, the creatinine clearance rate is calculated using the following formula:

[0071] In the formula: CrCl represents the creatinine clearance rate, which reflects the equivalent plasma volume of creatinine removed from the blood per unit time; C Crea,uf Q represents the creatinine concentration in the filtrate of the CRRT treatment circuit; UF Indicates the current ultrafiltration rate of the CRRT device; C Crea,pre This indicates the creatinine concentration of the blood at the inlet of the filter in the CRRT treatment circuit.

[0072] As an option, this unit also calculates the urea clearance index, a commonly used one being Kt / V. To calculate Kt / V, the total urea clearance K from a single CRRT treatment or over a specific time period must first be determined. The K value can be calculated by analyzing the difference in urea concentration in the blood at the filter inlet and outlet, combined with blood flow and effective treatment time; alternatively, in some cases, it can be estimated using parameters provided by the CRRT device itself and specific mathematical models. Secondly, the effective treatment duration t needs to be determined, typically the actual time the current CRRT treatment has been performed. Finally, the urea distribution volume V in the patient's body needs to be estimated. The V value is usually initially estimated based on the patient's baseline information such as weight, height, and sex, using generally accepted empirical formulas in the biomedical field (e.g., the Watson formula or the Hume-Weyers formula to estimate the total fluid volume, assuming uniform urea distribution within this fluid volume), and may be further individualized during subsequent physiological model calibration. The urea clearance index Kt / V is then obtained by multiplying K by t and then dividing by V.

[0073] In addition to the core indicators mentioned above, this data preprocessing and feature extraction unit can also extract other clinically significant dynamic features. For example, it can calculate the rate of change of key physiological parameters (such as blood pressure and heart rate) within a specific time window, which helps to determine the trend and urgency of changes in the patient's condition. It can also calculate the mean, median, standard deviation, or coefficient of variation of these parameters within a specific time window to reflect the overall level and stability of the parameters. Furthermore, it can track and calculate the patient's cumulative fluid balance since the start of CRRT treatment, which is crucial for guiding volume management.

[0074] All feature data obtained through preprocessing and feature engineering operations, including but not limited to calculated urea reduction rate, creatinine clearance rate, urea clearance index, and other extracted dynamic features, are organized into a structured data format. This structured data is typically represented as a series of time-stamped feature vectors, where each feature vector represents the patient's overall state at a specific moment. This structured feature data is then transmitted to subsequent physiological models and prediction units, as well as intelligent decision-making and instruction generation units, serving as key input information for model calibration, trend prediction, and decision-making.

[0075] The physiological model and prediction unit are used to construct and dynamically calibrate an individualized physiological model based on the patient's baseline information and the structured feature data, and to use the individualized physiological model to predict the trend of physiological parameter changes and generate physiological parameter change trend prediction information.

[0076] In this embodiment, the core function of the physiological model and prediction unit is to construct and continuously and dynamically calibrate an individualized physiological model that reflects the physiological characteristics of a specific patient, based on the patient's individualized baseline information and the real-time structured feature data generated by the aforementioned data preprocessing and feature extraction unit. Subsequently, this calibrated individualized physiological model is used to predict the changing trends of various key physiological parameters of the patient over a future period, thereby generating predictive information on physiological parameter change trends that has forward-looking guiding significance for clinical decision-making.

[0077] Specifically, a patient's baseline information is the initial basis for constructing an individualized physiological model. This information typically includes the patient's age, sex, medical history, laboratory test results at admission, such as initial blood urea nitrogen levels, serum creatinine levels, and other relevant clinical static or slowly changing data.

[0078] In one possible implementation, the construction of individualized physiological models can employ mathematical models based on physiological mechanisms. For example, compartmentalized models, such as single-compartment urea kinetic models or more complex two-compartment urea kinetic models, can be used to quantitatively describe the distribution, endogenous production, metabolism, and dynamic processes of substance exchange and clearance of target solutes (e.g., urea, creatinine) across different physiological compartments in the patient's body. The initial parameters of such models, such as the initial apparent volume of distribution of the target solute in the body and the rate of endogenous metabolic production, can be initially set based on the patient's baseline information, combined with existing clinical empirical formulas or population statistics.

[0079] Crucially, in order for the physiological model to accurately reflect the dynamic changes in the patient's physiological state and individualized response to treatment intervention during CRRT treatment, the physiological model and prediction unit perform dynamic calibration of the individualized physiological model as CRRT treatment continues.

[0080] In this embodiment, when performing the dynamic calibration operation of the individualized physiological model, the unit employs advanced estimation algorithms, such as the extended Kalman filter algorithm or the recursive least squares algorithm. These algorithms can effectively handle nonlinear systems and noisy measurement data.

[0081] This dynamic calibration process is based on structured feature data generated in real time by the data preprocessing and feature extraction unit. This input data includes, for example, continuously calculated urea reduction rate, creatinine clearance rate, and concentrations of specific solutes in the blood monitored in real time by sensors.

[0082] Driven by these online, continuous observational data, the extended Kalman filter algorithm or the recursive least squares algorithm can adjust one or more pre-set specific model parameters within the individualized physiological model online and continuously.

[0083] The adjustment of specific model parameters aims to minimize the difference between the model output and actual observational data, thereby improving the model's fit and predictive accuracy. These dynamically adjusted specific model parameters include: First, parameters characterizing the effective volume of distribution of the target solute in a particular patient. Due to individual differences, pathological conditions, and changes in fluid balance, the actual volume of distribution of the solute may differ from the initial value estimated based on standard formulas. Dynamic calibration can obtain an effective volume of distribution closer to the patient's actual condition. Second, parameters characterizing the effective clearance capacity of the current CRRT filter for a specific solute. The clearance efficiency of the filter may be affected by various factors, such as filtration time, filter characteristic degradation, and fluctuations in blood flow. Dynamic calibration allows for real-time evaluation and updating of parameters in the model representing the filter's actual clearance rate or mass transfer coefficient for a specific solute.

[0084] Through this online, continuous, and dynamic calibration mechanism, the physiological model can continuously "learn" and adapt to the individual specificity and time-varying nature of patients, evolving from a general model framework into a highly individualized dynamic model that can accurately reflect the current physiological state of the patient.

[0085] After the individualized physiological model has undergone sufficient dynamic calibration and can fit current and recent patient data well, the physiological model and prediction unit use this model to make prospective predictions.

[0086] Specifically, this unit combines the current treatment parameter settings of the CRRT device (such as blood flow, dialysate / replacement fluid flow rate, ultrafiltration rate, etc.) with the patient's latest physiological state (represented by the calibrated internal state variables of the model) to predict the possible changes of the patient's key physiological parameters over a preset period of time (such as the next 1 hour, 2 hours, or a longer time span set according to clinical needs) by numerically solving or simulating the model equations.

[0087] These predicted key physiological parameters may include trends in blood urea and creatinine concentrations, fluctuations in key electrolyte levels (such as potassium and sodium ions), and the evolution of the patient's overall fluid balance.

[0088] The physiological model and prediction unit output information predicting the trends of physiological parameter changes. This information is usually not just a single sequence of predicted values, but may also include descriptions of the confidence intervals and probability distributions of the predicted results, or assessments of the probability of reaching key thresholds (such as the hypotension threshold and the hyperkalemia threshold). This predictive information, with its forward-looking and uncertain quantitative characteristics, provides crucial decision support for the subsequent intelligent decision-making and instruction generation units to formulate more accurate and safer treatment strategies.

[0089] The intelligent decision-making and instruction generation unit is used to analyze the treatment status and generate treatment parameter adjustment instructions and decision basis based on the structured feature data and the physiological parameter change trend prediction information, using a multi-level artificial intelligence decision-making logic that includes rule-based fast response logic and reinforcement learning-based treatment optimization logic.

[0090] In this embodiment, the intelligent decision-making and instruction generation unit integrates and analyzes real-time structured feature data from the aforementioned data preprocessing and feature extraction unit, as well as prospective physiological parameter change trend prediction information from the aforementioned physiological model and prediction unit. Based on this comprehensive information, the unit employs an advanced multi-level artificial intelligence decision-making logic to deeply analyze and evaluate the patient's current CRRT treatment status, and autonomously generates specific and executable CRRT treatment parameter adjustment instructions, while simultaneously providing decision-making basis information to support these instructions.

[0091] Specifically, the input information received by the intelligent decision-making and instruction generation unit mainly includes two aspects: First, structured feature data generated by the data preprocessing and feature extraction unit, such as real-time updated urea reduction rate (URR), creatinine clearance rate (CrCl), urea clearance index (e.g., Kt / V), and the patient's cumulative fluid balance. This data objectively reflects the patient's current physiological state and the immediate effect of CRRT treatment. Second, predictive information on the changing trends of physiological parameters generated by the physiological model and prediction unit, such as predictions of the dynamic changes in key indicators such as blood urea concentration, blood pressure, and electrolyte levels in the patient over the next few hours. This information provides predictability for decision-making.

[0092] In this embodiment, the multi-level artificial intelligence decision-making logic employed by the intelligent decision-making and instruction generation unit combines different levels of artificial intelligence technology to adapt to the complex and ever-changing needs of clinical scenarios. This multi-level logic includes a rule-based rapid response logic.

[0093] This rule-based rapid response logic embeds a pre-defined, clinically validated and optimized rule set. This rule set consists of multiple "IF-THEN" structured condition-action rules, which encode a wealth of mature clinical knowledge and established treatment guidelines. For example, a rule might be defined as: "IF The patient's current arterial blood pressure is below the pre-defined individualized safety lower limit AND the rate of blood pressure change shows a sustained rapid downward trend THEN The system recommends immediately and significantly reducing the ultrafiltration rate, and simultaneously issues a high-priority alert to healthcare personnel, suggesting a review of the patient's blood volume status and vasoactive drug usage."

[0094] When the rule set is running, it will compare and make logical judgments in real time with the current accurate values ​​of various physiological parameters of the patient after being collected in real time by the data acquisition unit and processed by the data preprocessing and feature extraction unit, as well as the parameter change rate calculated by analyzing historical data of these parameters, and the various safety thresholds (such as upper and lower limits of blood pressure, upper and lower limits of heart rate, critical values ​​of specific electrolyte concentrations, etc.) preset by the system for the specific patient.

[0095] When the result of this comparison clearly indicates that the patient has an acute physiological imbalance, such as, but not limited to, severe hypotension, life-threatening hyperkalemia or hypokalemia, or acute acid-base imbalance, the rule-based rapid response logic will be activated first and can automatically and quickly generate CRRT treatment parameter adjustment instructions for immediate intervention and stabilization of the patient's vital signs without complex calculations.

[0096] The multi-level artificial intelligence decision-making logic also includes a reinforcement learning-based treatment optimization logic. This logic is mainly used to dynamically and continuously optimize one or more key adjustable treatment parameters during CRRT treatment, while ensuring patient safety. A typical optimization goal is the setting of the ultrafiltration rate, but it can also be extended to the adjustment of blood flow, dialysate flow rate, or replacement fluid flow rate and composition. The aim is to more effectively achieve preset long-term treatment goals, such as ideal solute clearance efficiency, precise fluid balance management, and maintaining the stability of the patient's overall physiological state. The reinforcement learning-based treatment optimization logic makes decisions based on a predefined state space. This state space consists of a series of feature variables that can comprehensively describe the current treatment situation, specifically including the urea decrease rate, creatinine clearance rate, and current cumulative fluid balance from the structured feature data generated by the data preprocessing and feature extraction unit; in addition, it also includes predicted values ​​of specific physiological parameters from the physiological parameter change trend prediction information generated by the physiological model and prediction unit, such as predicted values ​​of urea concentration, blood pressure, or key electrolyte levels for the next hour or several hours.

[0097] The agent's action space defines the set of treatment parameter adjustments that the logic can take, such as increasing or decreasing the ultrafiltration rate by a specific amount, or choosing to keep the current parameters unchanged. To guide and evaluate the quality of the parameter adjustments made by the agent, this reinforcement learning-based treatment optimization logic utilizes a reward function. This reward function is a mathematical expression used to comprehensively evaluate the short-term and long-term effects of each treatment parameter adjustment decision; the reward function can be expressed as: R total =w eff ·f eff (Efficacy)+w bal ·f bal (Balance)+w stab ·f stab (Stability)-P risk (Risk);

[0098] In the formula: R total f represents the total reward value of the treatment decision; eff (Efficacy) represents a performance evaluation function related to solute clearance efficiency, and its inputs include indicators such as urea reduction rate and creatinine clearance rate; bal (Balance) represents an evaluation function related to the achievement of the liquid balance target, and its input is, for example, the cumulative deviation between the liquid balance and the target balance; f stab (Stability) represents an evaluation function related to the maintenance of a patient's physiological stability, and its inputs are fluctuations or predicted values ​​of parameters such as blood pressure and heart rate; P risk(Risk) indicates a penalty associated with a potential risk or adverse event, such as a predicted risk of hypotension or electrolyte imbalance; eff w bal w stab These are weighting coefficients for efficacy, balance, and stability, which can be adjusted according to clinical goals.

[0099] Specifically, the reward function comprehensively considers the following aspects: whether the solute clearance efficiency of CRRT treatment reaches or approaches the clinical target (e.g., whether URR or CrCl is within the expected range); the degree to which fluid balance targets are achieved (e.g., the magnitude of the deviation between the actual cumulative ultrafiltration volume and the target ultrafiltration volume set by the healthcare professional); and the maintenance of the patient's physiological stability (e.g., whether parameter adjustments have triggered or exacerbated large fluctuations in blood pressure, abnormal changes in heart rate, or electrolyte disturbances). In one implementation, a deep Q-network, the Actor-Critic method (such as A2C or A3C), or other advanced reinforcement learning algorithms can be used to train the decision model (i.e., the policy network) of this optimization logic. The agent learns through extensive interactive learning (trial and error) with a simulated CRRT treatment environment (which can be constructed based on the aforementioned physiological model and the individualized physiological model provided by the prediction unit), or, under strict safety supervision and ethical permission, through offline learning or online fine-tuning using partially screened and desensitized real clinical data, thereby gradually learning and forming a strategy capable of making near-optimal treatment parameter adjustment decisions under various complex conditions.

[0100] Furthermore, to further enhance the reliability and security of decision-making, the intelligent decision-making and instruction generation unit performs a crucial decision-making pre-simulation operation before finally determining and outputting a specific CRRT treatment parameter adjustment instruction. The core idea of ​​this decision-making pre-simulation operation is to utilize a dynamically calibrated individualized physiological model, provided by the physiological model and prediction unit, which can highly simulate the current individual patient's physiological characteristics, as a virtual "digital twin" patient or simulation platform.

[0101] In response to the various candidate treatment parameter adjustment instructions initially generated by rule-based logic or reinforcement learning-based logic, this unit will simulate on this digital twin platform how the patient's various key physiological parameters (e.g., blood urea concentration, blood pressure, heart rate, fluid balance, etc.) will dynamically evolve after executing these candidate instructions within a preset future time period (e.g., several hours or a treatment shift), i.e., generate simulated dynamic change trajectories of these parameters.

[0102] The intelligent decision-making and instruction generation unit performs a comprehensive risk-benefit assessment of each candidate treatment parameter adjustment instruction based on the simulated dynamic change trajectories of various physiological parameters. This assessment comprehensively considers the expected therapeutic benefits of the instruction (e.g., the extent of improvement in solute clearance efficiency and the speed of improvement in fluid balance) and its potential risks (e.g., increased probability of hypotension and severity of electrolyte disturbances).

[0103] Based on the results of this risk-benefit assessment, the system will make a final decision: if a candidate instruction is assessed as having significant benefits and manageable risks, it will be confirmed; if the assessment shows that the risks are high or the benefits are not obvious, the candidate instruction may be modified as necessary (e.g., appropriately reducing the magnitude of the adjustment, adjusting the timing of execution, or adding auxiliary monitoring measures), or in extreme cases, the instruction may be rejected and other alternative solutions may be tried.

[0104] This mechanism of decision-making simulation and risk-benefit assessment can identify and mitigate potential adverse consequences before actually applying adjustment instructions to patients, thereby significantly improving the robustness of system decision-making and the safety of clinical application.

[0105] The treatment control and execution unit is used to receive the adjustment command, convert it into a control signal, and drive the device actuator to adjust the parameters.

[0106] In this embodiment, the treatment control and execution unit acts as a bridge between intelligent decision-making and physical devices. Its main responsibility is to receive treatment parameter adjustment instructions issued by the intelligent decision-making and instruction generation unit, and accurately convert these high-level instructions into low-level control signals that the CRRT device actuators can recognize and execute, thereby driving these actuators to complete the precise adjustment of treatment parameters.

[0107] Specifically, upon receiving an adjustment instruction, the unit first parses and verifies its validity. These instructions specify the treatment parameters to be adjusted (such as the blood pump flow rate Q). blood,target Ultrafiltration pump rate Q uf,target (etc.) and their target settings.

[0108] Subsequently, the unit performs a crucial instruction conversion operation, converting the target parameter values ​​into specific physical control signals. This conversion process can be summarized as follows:

[0109] In the formula: S control This represents the specific control signals generated, such as the voltage and current of the drive motor, the duty cycle of the pulse width modulation (PWM) signal, or digital commands for a specific communication protocol; P targetThis represents the target parameter setting value in the instruction, such as the target flow rate or target infusion rate; C actuator These represent characteristic parameters or calibration data of the corresponding actuator, such as the flow-voltage characteristic curve of a pump. This represents a transformation function or algorithm that maps high-level parameter targets to low-level physical control signals based on the characteristics of the actuator.

[0110] These generated control signals are stably applied to the various physical actuators of the CRRT device, which mainly include: blood pumps, dialysate pumps, replacement fluid pumps, ultrafiltration pumps, and anticoagulant infusion pumps. Upon receiving the control signals, the actuators adjust their operating status.

[0111] In one possible implementation, to ensure the accuracy and responsiveness of parameter adjustments, the treatment control and execution unit can integrate feedback control logic. By monitoring the actual operating status of the actuator (e.g., real-time monitoring of actual flow rate via a flow sensor) and comparing it with the target setpoint, closed-loop control is formed to correct deviations in real time.

[0112] In addition, the unit includes necessary safety monitoring and fault handling mechanisms. This includes checking the rationality of instructions, monitoring the working status of the actuators, and triggering preset safety measures, such as switching the device to a safe mode and alerting medical personnel, when communication errors, actuator malfunctions, or parameter adjustments fail to meet expectations are detected.

[0113] The information interaction and monitoring unit is used to graphically display treatment-related information, decision-making basis, and alarm information to medical staff, and to receive intervention instructions from medical staff that affect the generation of subsequent adjustment instructions;

[0114] In this embodiment, this unit serves as the core interactive interface between the system and medical staff. Its main responsibilities include: displaying CRRT treatment-related information, the decision-making basis of the intelligent system, and various alarm information to medical staff in a graphical manner in real time; and receiving intervention instructions from medical staff, which will affect the generation of subsequent treatment parameter adjustment instructions.

[0115] Specifically, the information interaction and monitoring unit displays the following key information through a graphical user interface (GUI):

[0116] Current values ​​and trend graphs of real-time physiological parameters (such as urea concentration) collected by the data acquisition unit.

[0117] The current values ​​and changes of structured feature data (such as urea decrease rate URR, creatinine clearance rate CrCl, and cumulative fluid balance) calculated by the data preprocessing and feature extraction unit.

[0118] The prediction information on the future trend of key physiological parameters (such as predicted blood pressure and blood potassium concentration) generated by the physiological model and prediction unit may include confidence intervals.

[0119] The treatment parameter adjustment suggestions and detailed decision-making basis proposed by the intelligent decision-making and instruction generation unit.

[0120] Regarding alarm information display, this unit provides comprehensive alarm functions, including the following alarm information:

[0121] Parameter abnormality alarm: Triggered when the patient's physiological parameter values ​​collected by the data acquisition unit exceed the preset safe range or critical value range.

[0122] Technical fault alarm: Triggered when a sensor in the data acquisition unit or other components of the system malfunctions or experiences a communication failure.

[0123] Predictive risk alarm: Triggered when the intelligent decision-making and instruction generation unit predicts, based on analysis, that a patient is about to experience a potential physiological risk (such as hypotension or severe electrolyte imbalance).

[0124] The aforementioned parameter anomaly alarms, technical fault alarms, and predictive risk alarms all include: alarm category information, the identifier of the parameter or component related to the alarm, the parameter value or fault description information when the alarm is triggered, and the preliminary response measures suggested by the system based on preset logic or model analysis. Alarm information is usually presented in a conspicuous manner (such as color changes or sound prompts) and can be graded according to the urgency level.

[0125] Regarding receiving intervention instructions from healthcare professionals, the unit allows healthcare professionals to perform the following operations via an interactive interface (such as a touchscreen):

[0126] The system can confirm, modify, or reject treatment parameter adjustment suggestions proposed by the intelligent decision-making and instruction generation unit. It can also manually set various treatment parameters for the CRRT device. Alarm messages issued by the system can be acknowledged, silenced, or marked as processed. External clinical information that the system fails to automatically collect (such as the latest medications or external examination results) can be used by the system to update the model or adjust decisions.

[0127] The intervention instructions from medical staff will be recorded by the system and will directly affect the work of the subsequent intelligent decision-making and instruction generation units. For example, manually set parameters will serve as new constraints or goals, thereby achieving human-machine collaborative treatment management.

[0128] Please see Figure 2 The present invention also provides an artificial intelligence-based method for processing CRRT treatment parameters, the method comprising:

[0129] S1. The patient's physiological and treatment parameters are collected in real time from the CRRT circuit through the data acquisition unit to generate raw digital signals;

[0130] S2. The data preprocessing and feature extraction unit performs preprocessing and feature engineering operations on the original digital signal to generate structured feature data, which includes urea reduction rate, creatinine clearance rate and urea clearance index.

[0131] S3. Based on the patient's baseline information and the generated structured feature data, construct and dynamically calibrate the patient's individualized physiological model using the physiological model and prediction unit, and use the individualized physiological model to predict the trend of the patient's physiological parameters in the future to generate physiological parameter change trend prediction information.

[0132] S4. By combining the structured feature data and the physiological parameter change trend prediction information through the intelligent decision-making and instruction generation unit, a multi-level artificial intelligence decision-making logic including rule-based fast response logic and reinforcement learning-based treatment optimization logic is adopted to analyze the current treatment status and generate CRRT treatment parameter adjustment instructions and decision basis information corresponding to the adjustment instructions.

[0133] S5. The treatment control and execution unit receives the adjustment command of the CRRT treatment parameters and converts the adjustment command into a control signal for the corresponding actuator in the CRRT device, so as to drive the actuator to perform the adjustment of the CRRT treatment parameters;

[0134] S6. The information interaction and monitoring unit graphically displays treatment-related information, decision-making basis information for the adjustment instructions, and necessary alarm information to medical staff, and receives intervention instructions input by medical staff, which are used to influence the adjustment instructions of subsequent CRRT treatment parameters.

[0135] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based CRRT treatment parameter processing system, characterized in that, The system includes the following modules: The data acquisition unit is used to acquire patient physiological and treatment parameters from the CRRT circuit in real time and generate raw digital signals. The data preprocessing and feature extraction unit is used to preprocess and feature engineer the original digital signal to generate structured feature data containing urea reduction rate, creatinine clearance rate and urea clearance index. The physiological model and prediction unit are used to construct and dynamically calibrate an individualized physiological model based on the patient's baseline information and the structured feature data, and to use the individualized physiological model to predict the trend of physiological parameter changes and generate physiological parameter change trend prediction information. The intelligent decision-making and instruction generation unit is used to analyze the treatment status and generate treatment parameter adjustment instructions and decision basis based on the structured feature data and the physiological parameter change trend prediction information, using a multi-level artificial intelligence decision-making logic that includes rule-based fast response logic and reinforcement learning-based treatment optimization logic. The treatment control and execution unit is used to receive the adjustment command, convert it into a control signal, and drive the device actuator to adjust the parameters. The information interaction and monitoring unit is used to graphically display treatment-related information, decision-making basis, and alarm information to medical staff, and to receive intervention instructions from medical staff that affect the generation of subsequent adjustment instructions.

2. The CRRT treatment parameter processing system based on artificial intelligence according to claim 1, characterized in that, The data acquisition unit includes an electrolyte sensor and a metabolite sensor, which are deployed at the filter inlet, filter outlet, and filtrate outlet of the CRRT circuit. The electrolyte sensor is used to monitor the electrolyte concentration in the fluid at the corresponding deployment location, and the metabolite sensor is used to monitor the metabolite concentration in the fluid at the corresponding deployment location.

3. The CRRT treatment parameter processing system based on artificial intelligence according to claim 1, characterized in that, When performing the feature engineering operation to generate the structured feature data, the data preprocessing and feature extraction unit specifically calculates the urea reduction rate. To calculate the urea reduction rate, the data preprocessing and feature extraction unit obtains the urea concentration in the blood at the inlet of the CRRT treatment loop filter, which is collected in real time by the data acquisition unit and denoted as C. Urea,pre And the urea concentration in the blood at the outlet of the CRRT treatment circuit filter, denoted as C. Urea,post Based on the obtained urea concentration, the following formula is used to calculate the urea decrease rate: In the formula: URR represents the urea reduction rate; C Urea,pre This indicates the urea concentration in the blood at the inlet of the filter in the CRRT treatment circuit; C Urea,post This indicates the urea concentration in the blood at the outlet of the filter in the CRRT treatment circuit.

4. The CRRT treatment parameter processing system based on artificial intelligence according to claim 3, characterized in that, When performing the feature engineering operation to generate the structured feature data, the data preprocessing and feature extraction unit also specifically calculates the creatinine clearance rate. To calculate the creatinine clearance rate, the data preprocessing and feature extraction unit obtains the creatinine concentration in the CRRT treatment circuit filtrate, which is collected in real time by the data acquisition unit and denoted as C. Crea,uf Obtain the current ultrafiltration rate of the CRRT device, denoted as Q. UF The system acquires the creatinine concentration of the blood at the inlet of the CRRT treatment circuit filter, which is collected in real time by the data acquisition unit and denoted as C. Crea,pre Based on the obtained creatinine concentration and the ultrafiltration rate, the creatinine clearance rate is calculated using the following formula: In the formula: CrCl represents the creatinine clearance rate; C Crea,uf Q represents the creatinine concentration in the filtrate of the CRRT treatment circuit; UF This indicates the ultrafiltration rate of the CRRT device; C Crea,pre This indicates the creatinine concentration of the blood at the inlet of the filter in the CRRT treatment circuit.

5. The CRRT treatment parameter processing system based on artificial intelligence according to claim 1, characterized in that, When performing dynamic calibration of the individualized physiological model, the physiological model and prediction unit use an extended Kalman filter algorithm or a recursive least squares algorithm, and adjust specific model parameters within the individualized physiological model online and continuously based on the structured feature data generated by the data preprocessing and feature extraction unit. The specific model parameters include parameters that characterize the effective distribution volume of the target solute in the patient's body, and parameters that characterize the effective clearance capacity of the CRRT filter for the specific solute.

6. The CRRT treatment parameter processing system based on artificial intelligence according to claim 1, characterized in that, The rule-based fast response logic in the multi-level artificial intelligence decision-making logic adopted by the intelligent decision-making and instruction generation unit includes a pre-set rule set composed of multiple IF-THEN structure rules; the rule set compares and judges the current values ​​of the patient's physiological parameters collected in real time by the data acquisition unit and the rate of change of these physiological parameter values ​​with various preset safety thresholds in the system. When the comparison results indicate that the patient is in an acute physiological imbalance, the rule-based rapid response logic automatically generates treatment parameter adjustment instructions for immediate intervention.

7. The CRRT treatment parameter processing system based on artificial intelligence according to claim 1, characterized in that, The reinforcement learning-based treatment optimization logic employed by the intelligent decision-making and instruction generation unit in the multi-level artificial intelligence decision-making logic is used to dynamically optimize the ultrafiltration rate during CRRT treatment. This reinforcement learning-based treatment optimization logic makes decisions based on a state space, which includes the urea reduction rate, creatinine clearance rate, and cumulative fluid balance from the structured feature data generated by the data preprocessing and feature extraction unit, as well as the predicted values ​​of specific physiological parameters from the physiological parameter change trend prediction information generated by the physiological model and prediction unit. The reinforcement learning-based treatment optimization logic utilizes a reward function to guide the optimization of the ultrafiltration rate. This reward function comprehensively evaluates the solute clearance efficiency, the achievement of fluid balance goals, and the maintenance of patient physiological stability during CRRT treatment.

8. The CRRT treatment parameter processing system based on artificial intelligence according to claim 1, characterized in that, Before finally determining the specific treatment parameter adjustment instruction to be output, the intelligent decision-making and instruction generation unit also performs a decision-making pre-simulation operation. The decision-making pre-simulation operation uses the individualized physiological model provided by the physiological model and prediction unit as a simulation platform to simulate the dynamic change trajectory of the parameter adjustment instruction on various physiological parameters of the patient within a preset time for the candidate treatment parameter adjustment instructions. The intelligent decision-making and instruction generation unit then performs a risk-benefit assessment on each candidate treatment parameter adjustment instruction based on the simulated dynamic change trajectory, and confirms or modifies the finally generated treatment parameter adjustment instruction based on the assessment results.

9. The CRRT treatment parameter processing system based on artificial intelligence according to claim 1, characterized in that, The alarm information displayed by the information interaction and monitoring unit includes: A parameter anomaly alarm is triggered when the patient's physiological parameter values ​​collected by the data acquisition unit exceed the preset safety range or the preset critical value range. A technical fault alarm is triggered when a sensor or other component of the system within the data acquisition unit malfunctions. A predictive risk alarm is triggered when the intelligent decision-making and instruction generation unit predicts, based on analysis, that the patient is about to experience a potential physiological risk. The parameter anomaly alarm, the technical fault alarm, and the predictive risk alarm all include alarm category information, parameter identifiers related to the alarm, parameter value information when the alarm is triggered, and system-suggested countermeasures.

10. An artificial intelligence-based CRRT treatment parameter processing method, applied to the system described in any one of claims 1-9, characterized in that, The method includes: S1. The patient's physiological and treatment parameters are collected in real time from the CRRT circuit through the data acquisition unit to generate raw digital signals; S2. The data preprocessing and feature extraction unit performs preprocessing and feature engineering operations on the original digital signal to generate structured feature data, which includes urea reduction rate, creatinine clearance rate and urea clearance index. S3. Based on the patient's baseline information and the generated structured feature data, construct and dynamically calibrate the patient's individualized physiological model using the physiological model and prediction unit, and use the individualized physiological model to predict the trend of the patient's physiological parameters in the future to generate physiological parameter change trend prediction information. S4. By combining the structured feature data and the physiological parameter change trend prediction information through the intelligent decision-making and instruction generation unit, a multi-level artificial intelligence decision-making logic including rule-based fast response logic and reinforcement learning-based treatment optimization logic is adopted to analyze the current treatment status and generate CRRT treatment parameter adjustment instructions and decision basis information corresponding to the adjustment instructions. S5. The treatment control and execution unit receives the adjustment command of the CRRT treatment parameters and converts the adjustment command into a control signal for the corresponding actuator in the CRRT device, so as to drive the actuator to perform the adjustment of the CRRT treatment parameters; S6. The information interaction and monitoring unit graphically displays treatment-related information, decision-making basis information for the adjustment instructions, and necessary alarm information to medical staff, and receives intervention instructions input by medical staff, which are used to influence the adjustment instructions of subsequent CRRT treatment parameters.

Images