Intelligent prediction management system for hypotension of hemodialysis patient in dialysis

Abstract

The invention discloses an intelligent prediction management system for hypotension in hemodialysis patient dialysis, which belongs to the field of hemodialysis monitoring and comprises an acquisition processing module, a dialysis monitoring module, a risk prediction module, a filling prediction module, a dynamic regulation and control module, a strategy execution module, a medical record analysis module and an information push module. According to the method, the perception capability of early symptoms of hypotension can be enhanced, the prediction model is ensured to have high individual adaptability and generalization capability, misinformation and missed information can be reduced, the medical efficiency is improved, sudden hypotension is reduced, and the stability and practicability of the system are improved; intelligent control can be carried out according to the current physiological state, dialysis flexibility is improved, dialysis multi-factor coupling characteristics can be comprehensively considered, the method is closer to real clinical conditions, decision making depending on medical staff experience is avoided, and intervention recommendation robustness and effect are improved.

Description

Technical Field

[0001] The present invention relates to the field of hemodialysis monitoring, and in particular to an intelligent prediction and management system for hemodialysis patients' intradialytic hypotension. Background Art

[0002] Hemodialysis is one of the most commonly used renal replacement therapies for patients with end-stage renal disease. It effectively removes metabolic waste and excess water from the body and maintains electrolyte and acid-base balance. However, during dialysis, approximately 20%–30% of patients experience recurrent intradialytic hypotension. This complication not only significantly impacts the continuity and safety of dialysis treatment but can also induce multiple organ damage, such as myocardial ischemia, cerebral ischemia, and decreased retinal perfusion, and even lead to an increased risk of acute cardiovascular and cerebrovascular events and death. Currently, most clinical dialysis monitoring systems still rely primarily on a single indicator for risk warning and lack dynamic adaptability to individual differences. Intervention strategies, such as adjusting ultrafiltration rate, dialysate sodium concentration, fluid temperature control, and intravenous saline infusion, often rely on the experience and judgment of medical staff, are subject to lags and uncertainties, and are difficult to achieve precise and forward-looking management.

[0003] After searching, Chinese patent number CN110152087A discloses a monitoring method for the hemodialysis process. Although this invention can effectively reduce the probability of hypotension in patients during hemodialysis, its ability to perceive early signs of hypotension is poor, and the prediction model does not have high individual adaptability and generalization capabilities, resulting in many false positives and missed positives, which reduces medical efficiency. In addition, the existing intelligent prediction and management system for hypotension during hemodialysis patients cannot be intelligently controlled according to the current physiological state, and is too dependent on the experience of medical staff for decision-making, which reduces the robustness and effectiveness of intervention recommendations. Therefore, we propose an intelligent prediction and management system for hypotension during hemodialysis patients. Summary of the Invention

[0004] The purpose of the present invention is to solve the defects in the prior art and to propose an intelligent prediction and management system for hypotension during dialysis in hemodialysis patients.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] An intelligent prediction and management system for hemodialysis patients' hypotension during dialysis, including an acquisition and processing module, a dialysis monitoring module, a risk prediction module, a filling prediction module, a dynamic control module, a strategy execution module, a medical record analysis module, and an information push module;

[0007] The acquisition and processing module is used to monitor the patient's physiological parameters in real time and process the monitored physiological data;

[0008] The dialysis monitoring module is used to monitor various dialysis parameters during the dialysis process in real time;

[0009] The risk prediction module constructs a patient-specific risk map based on real-time monitored patient physiological data and dialysis parameters, predicts the probability of intradialytic hypotension, and classifies the risk level;

[0010] The filling prediction module predicts the patient's plasma refill rate based on the patient's physiological data monitored in real time;

[0011] The dynamic control module is used to dynamically adjust the ultrafiltration rate based on the predicted individual patient condition and the patient's plasma refill rate;

[0012] The strategy execution module formulates and implements personalized intervention strategies based on risk classification;

[0013] The medical record parsing module is used to parse electronic medical records and extract information about various symptoms of patients;

[0014] The information push module hospital information system or mobile application pushes low blood pressure risk warnings to medical staff and patients and provides corresponding intervention strategies.

[0015] As a further solution of the present invention, the specific steps of the acquisition and processing module for processing the monitored physiological parameters are as follows:

[0016] S1.1: The acquisition and processing module collects patient physiological data collected in real time by external sensors, including blood pressure, blood volume, skin temperature, and heart rate. It synchronizes the time of each physiological data using dynamic time warping. It then uses a bandpass filter to remove high-frequency noise and low-frequency drift, and then uses a sliding mean filter to average the physiological data.

[0017] S1.2: Divide each type of physiological data into multiple groups according to the set time window, then calculate the mean μ and standard deviation σ of each type of physiological data in each time window, and then calculate the Z-score value of each type of physiological data in each time interval;

[0018] S1.3: If a Z-score value is higher than ±3σ, the corresponding physiological data is judged as an outlier and marked as a missing value. The missing value information in each type of physiological data is counted and filled by linear interpolation. Then, the Min-Max normalization method is used to normalize each type of physiological data to the interval [0, 1].

[0019] As a further embodiment of the present invention, the risk prediction module constructs a patient-specific risk map, and the specific steps for predicting the probability of occurrence of intradialytic hypotension are as follows:

[0020] S2.1: Build a patient-specific physiological signal graph using the various physiological data monitored by the acquisition and processing module and the dialysis parameters monitored by the dialysis monitoring module as nodes and the interactions between the various physiological data as edges. Calculate the dynamic correlations between the calculated physiological parameters and dialysis parameters, and construct the corresponding edge weight matrix.

[0021] S2.2: Signal propagation is performed on the established patient-specific physiological signal graph through the GCN network. The various data contained in each node in the patient-specific physiological signal graph are repeatedly fused through the multi-layer GCN network to obtain the final fused patient physiological status characteristics;

[0022] S2.3: Input the generated physiological state features of each patient into a gated recurrent unit. Based on the three gating mechanisms of the gated recurrent unit, namely the update gate, reset gate, and candidate hidden state, the hidden state of each patient's physiological state feature at each time step is calculated in sequence until the hidden state at the final time step is obtained as the representation of the patient's current physiological state.

[0023] S2.4: The patient's current physiological state is represented by the Sigmoid function of the fully connected layer and mapped to the range [0, 1] to generate a scalar probability value for the corresponding patient. If the scalar probability value is ≥ 0.7, a high-risk warning is triggered. If the scalar probability value is 0.3 ≤ 0.7, a moderate risk is indicated and adjustment of dialysis parameters is recommended. If the scalar probability value is < 0.3, it is judged as low risk and dialysis proceeds normally.

[0024] As a further embodiment of the present invention, the specific steps of the filling prediction module for predicting the patient's plasma refill rate are as follows:

[0025] S3.1: Record the baseline blood volume at the start of dialysis. Measure the patient's blood volume every minute using the bioelectrical impedance method. Calculate the relative change in blood volume during dialysis in real time based on the baseline blood volume and the patient's blood volume at that moment.

[0026] S3.2: Calculate the rate of blood volume decrease between two adjacent moments. Then, calculate the patient's fluid loss rate per minute based on the patient's ultrafiltration volume recorded in real time. Then, calculate the plasma refill volume for the corresponding time period based on the obtained blood volume decrease rate, fluid loss rate, and baseline blood volume at the start of dialysis.

[0027] S3.3: Collect plasma filling volume data for different time periods, establish filling datasets for the corresponding time periods, and set a set of N×N sliding windows. Then, move the sliding windows sequentially from the beginning of each filling dataset.

[0028] S3.4: After each movement, smooth the plasma filling volume data within the current sliding window to obtain the plasma refill rate at the current time, and compare the plasma refill rate at the current time with the patient's ultrafiltration rate to obtain the patient's vascular compensation rate index;

[0029] S3.5: If the vascular compensation rate index is lower than 0.5, it indicates that the patient may be at risk of hypotension and an alert or adjustment of the ultrafiltration rate should be issued. Otherwise, it indicates that the plasma refilling capacity is good.

[0030] As a further solution of the present invention, the specific steps of the dynamic control module dynamically adjusting the ultrafiltration rate are as follows:

[0031] S4.1: Extract the patient's historical dialysis data, including age, cardiovascular disease history, historical hypotension incidence, dialysis adequacy index, and plasma refill rate. Calculate the patient's maximum tolerated ultrafiltration rate and minimum tolerated ultrafiltration rate using a regression model based on the collected historical dialysis data.

[0032] S4.2: Based on the calculated maximum tolerated ultrafiltration rate and minimum tolerated ultrafiltration rate, initialize the ultrafiltration rate according to the patient's current state. Establish a state space based on the patient's physiological state and dialysis parameters, including the current systolic blood pressure, plasma refill rate, current ultrafiltration rate, and dialysate sodium concentration. Set the ultrafiltration rate adjustment step size to establish the corresponding action space.

[0033] S4.3: With the goal of reducing the risk of hypotension and maintaining dialysis adequacy, construct a corresponding reward function, randomly select actions in the action space, then calculate the long-term benefits of each action in the action space under the current patient state based on the reward function, and select the action with the highest current long-term benefit to adjust the ultrafiltration rate;

[0034] S4.4: Recalculate the patient's status based on the preset time interval and update the long-term benefits of each action. Reselect the action with the highest long-term benefit to adjust the ultrafiltration rate. If the systolic blood pressure decreases beyond the preset threshold within the preset time interval, immediately reduce the ultrafiltration rate.

[0035] As a further solution of the present invention, the specific steps of the policy execution module are as follows:

[0036] S5.1: Collect the patient's current risk level, hemodynamic parameters, and dialysis settings to establish a corresponding joint state set for the patient. Then, collect the interventions for adjusting sodium concentration, pre-injecting normal saline, and reducing ultrafiltration rate to establish corresponding action sets. The patient's current joint state set is then used as the root node.

[0037] S5.2: Starting from the root node, simulate the new patient state generated after executing each intervention action in the current patient state and use it as a new child node to construct the corresponding exploration tree. Then, starting from the root node, calculate the average reward of executing each action under the current state and the UCB value of each child node;

[0038] S5.3: Select the child node with the highest average reward and UCB value as the best child node. Repeat the child node selection process until a leaf node is selected that has not yet explored all actions. Based on the state of this leaf node, select an unsimulated action and generate a new patient state. This action is then added to the exploration tree as a new child node.

[0039] S5.4: Starting from the current new child node, use a random strategy to simulate future intervention operations until the patient completes dialysis. Based on the simulation results, calculate the total simulated utility corresponding to the current path. Then, trace the simulated total utility back to all nodes on the path and update the statistics of each node.

[0040] S5.5: Repeat the selection, expansion, simulation, and backtracking until the preset number of iterations is reached. Then, stop the iteration, traverse the final exploration tree, and select the child node action with the highest average utility from the root node. This action is used as the final recommended intervention, and the optimal path and corresponding intervention strategy are output.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. The present invention constructs an individualized physiological signal graph by fusing the patient's real-time physiological data with dialysis parameters, calculates the dynamic correlation between each parameter to form an edge weight matrix, extracts the fused physiological state features through multi-layer GCN, and inputs them into GRU for time modeling, outputs the current physiological state probability value, and divides the patient into three risk levels: high, medium and low based on the physiological state probability value. The blood volume changes are monitored every minute based on the bioimpedance method, and the real-time plasma refill rate is calculated by combining the baseline value and the fluid loss rate. After smoothing analysis using a sliding window, the corresponding vascular compensation rate index is generated. If the compensation rate is lower than 0.5, it indicates a potential risk of hypotension and triggers intervention. It can enhance the ability to perceive early signs of hypotension, ensure that the prediction model has high individual adaptability and generalization ability, help reduce false alarms and missed alarms, improve medical efficiency, reduce sudden hypotension, and improve system stability and practicality.

[0043] 2. The present invention extracts the patient's historical dialysis data and uses a regression model to calculate the individual's maximum and minimum tolerated ultrafiltration rates, and sets the initial ultrafiltration rate accordingly. It then constructs a state space containing information on blood pressure and plasma refill rate, as well as an action space centered on ultrafiltration rate adjustment. The ultrafiltration rate is continuously adjusted during dynamic dialysis to balance the risk of hypotension and dialysis adequacy. If a hypotension trend occurs, a state-action exploration tree is constructed to simulate the impact of each intervention operation on the future state, and the simulated total utility of each path is calculated. Ultimately, the intervention plan and path with the optimal average utility are selected to assist medical staff in making personalized treatment decisions. It can perform intelligent control based on the current physiological state, improve dialysis flexibility, and comprehensively consider the multi-factor coupling characteristics of dialysis, which is closer to real clinical conditions, avoids relying on the experience of medical staff in decision-making, and improves the robustness and effectiveness of intervention recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0045] Figure 1 This is a system block diagram of an intelligent prediction and management system for intradialytic hypotension in hemodialysis patients proposed by the present invention. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0047] Example 1

[0048] Reference Figure 1 , an intelligent prediction and management system for hypotension during dialysis in hemodialysis patients, including an acquisition and processing module, a dialysis monitoring module, a risk prediction module, a filling prediction module, a dynamic control module, a strategy execution module, a medical record analysis module and an information push module.

[0049] The acquisition and processing module is used to monitor the patient's physiological parameters in real time and process the monitored physiological data;

[0050] The dialysis monitoring module is used to monitor various dialysis parameters during the dialysis process in real time.

[0051] Specifically, the acquisition and processing module collects the patient's physiological data collected in real time by external sensors, including blood pressure, blood volume, skin temperature and heart rate, and synchronizes the time of each physiological data through the dynamic time warping method. Then, the high-frequency noise and low-frequency drift are removed through a bandpass filter, and the sliding mean filtering method is used for each physiological data. According to the set time window, each type of physiological data is divided into multiple groups of data, and then the mean μ and standard deviation σ of each type of physiological data in each time window are calculated respectively. Then, the Z-score value of each type of physiological data in each time interval is calculated. If there is a Z-score value higher than ±3σ, the corresponding physiological data is judged to be an abnormal value and marked as a missing value. The missing value information in each type of physiological data is counted, and the missing values are filled by linear interpolation. Then, the Min-Max normalization method is used to unify each type of physiological data to the [0, 1] interval.

[0052] The risk prediction module constructs a patient-specific risk map based on real-time monitored patient physiological data and dialysis parameters, predicts the probability of hypotension during dialysis, and divides the risk level.

[0053] Specifically, various types of physiological data of patients monitored by the acquisition and processing module and dialysis parameters monitored by the dialysis monitoring module are taken as nodes, and the mutual influence relationship between various types of physiological data is taken as edges. A patient-specific physiological signal graph is established, and the dynamic correlation between the calculated physiological parameters and the dialysis parameters is calculated. The corresponding edge weight matrix is constructed, and the signal propagation of the established patient-specific physiological signal graph is carried out through the GCN network. The various types of data contained in each node in the patient-specific physiological signal graph are repeatedly fused through the multi-layer GCN network to obtain the final fused patient physiological state characteristics, and the generated patient physiological state characteristics are input into the gate In the recurrent unit, the hidden state of each patient's physiological state feature at each time step is calculated in sequence according to the three gating mechanisms of the gated recurrent unit: the update gate, the reset gate, and the candidate hidden state, until the hidden state at the final time step is obtained. This hidden state is used as the representation of the patient's current physiological state. The Sigmoid function of the fully connected layer is used to map the patient's current physiological state representation to the range [0, 1] to generate a scalar probability value for the corresponding patient. If the scalar probability value is ≥0.7, a high-risk warning is triggered. If 0.3≤scalar probability value<0.7, a moderate risk is indicated, and adjustment of dialysis parameters is recommended. If the scalar probability value is <0.3, it is judged as low risk and dialysis proceeds normally.

[0054] The filling prediction module predicts the patient's plasma refill rate based on real-time monitored patient physiological data.

[0055] Specifically, the baseline blood volume at the start of dialysis was recorded, and the patient's blood volume at the current moment was detected every 1 minute by the bioelectrical impedance method. The relative change of blood volume during dialysis was calculated in real time based on the baseline blood volume and the patient's blood volume at the current moment. The blood volume decrease rate at every two adjacent moments was calculated, and then the patient's body fluid loss rate per minute was calculated based on the patient's ultrafiltration volume recorded in real time. Then, the plasma refill volume in the corresponding time period was calculated based on the corresponding patient's blood volume decrease rate, body fluid loss rate and the baseline blood volume at the start of dialysis. The plasma refill volume of each plasma refill volume in different time periods was collected. For plasma refill data, a filling data set for the corresponding time period has been established, and a set of sliding windows of size N×N has been set. The sliding windows are then moved sequentially from the head of each filling data set. After each movement, the plasma filling volume data in the current sliding window are smoothed to obtain the plasma refill rate at the current time, and the plasma refill rate at the current time is compared with the patient's ultrafiltration rate to obtain the patient's vascular compensation rate index. If the vascular compensation rate index is lower than 0.5, it indicates that the patient may be at risk of hypotension and should be warned or the ultrafiltration rate should be adjusted. Otherwise, it indicates that the plasma refill capacity is good.

[0056] Example 2

[0057] Reference Figure 1 , an intelligent prediction and management system for hypotension during dialysis in hemodialysis patients, including an acquisition and processing module, a dialysis monitoring module, a risk prediction module, a filling prediction module, a dynamic control module, a strategy execution module, a medical record analysis module and an information push module.

[0058] The dynamic control module is used to dynamically adjust the ultrafiltration rate based on the predicted individual patient conditions and the patient's plasma refill rate.

[0059] Specifically, the historical dialysis data of the corresponding patient are extracted, including age, cardiovascular history, historical hypotension incidence, dialysis adequacy index and plasma refill rate. According to the collected historical dialysis data, the patient's maximum tolerated ultrafiltration rate and minimum tolerated ultrafiltration rate are calculated through a regression model. Based on the calculated maximum tolerated ultrafiltration rate and minimum tolerated ultrafiltration rate of the patient, the ultrafiltration rate is initialized according to the patient's current state, and according to the patient's physiological state and dialysis parameters, a state space is established, which includes the current systolic blood pressure, plasma refill rate, current ultrafiltration rate and dialysate sodium concentration, and the ultrafiltration rate adjustment step is set. Long, in order to establish the corresponding action space, with the goal of reducing the risk of hypotension and maintaining dialysis adequacy, construct the corresponding reward function, randomly select actions in the action space, and then calculate the long-term benefits of each action in the action space under the current patient state based on the reward function, and select the action with the highest current long-term benefit to adjust the ultrafiltration rate. According to the preset time interval, the patient state is recalculated, and the long-term benefits of each action are updated. The action with the highest long-term benefit is reselected to adjust the ultrafiltration rate. If the systolic blood pressure downward trend exceeds the preset threshold within the preset time interval, the ultrafiltration rate is immediately reduced.

[0060] The strategy execution module formulates and implements personalized intervention strategies based on risk classification.

[0061] Specifically, the patient's current risk level, hemodynamic parameters, and dialysis settings are collected to establish a joint state set corresponding to the patient. Then, the intervention operations of adjusting sodium concentration, pre-injecting normal saline, and reducing ultrafiltration rate are collected to establish a corresponding action set. Then, the patient's current joint state set is used as the root node. Starting from the root node, the new patient state generated after the current patient state is simulated after each intervention action is performed is simulated and used as a new child node to construct the corresponding exploration tree. Then, starting from the root node, the average reward of each action performed under the current state and the UCB value of each child node are calculated. The child node with the highest average reward and UCB value is selected as the best child node. The child node selection is repeated until the child node that has not yet been selected is selected. After exploring the leaf nodes of all actions, an unsimulated action is selected based on the state of the leaf node, and a new patient state is generated. At the same time, it is added to the exploration tree as a new child node. Starting from the current new child node, a random strategy is used to simulate future intervention operations until the patient finishes dialysis. Based on the simulation results, the total simulated utility corresponding to the current path is calculated, and the total utility obtained by simulation is traced back to all nodes on the path, and the statistical information of each node is updated. The selection, expansion, simulation and backtracking are repeated until the preset number of iterations is reached. The iteration is stopped, the final exploration tree is traversed, and the child node action with the highest average utility is selected from the root node. It is used as the final recommended intervention, and the optimal path and corresponding intervention strategy are output at the same time.

[0062] The medical record parsing module is used to parse electronic medical records and extract information about the patient's various symptoms; the information push module uses the hospital information system or mobile application to push low blood pressure risk warnings to medical staff and patients and provide corresponding intervention strategies.

Claims

1. An intelligent prediction and management system for hemodialysis patients with hypotension during dialysis, characterized in that: It includes collection and processing module, dialysis monitoring module, risk prediction module, filling prediction module, dynamic control module, strategy execution module, medical record analysis module and information push module; The acquisition and processing module is used to monitor the patient's physiological parameters in real time and process the monitored physiological data; The dialysis monitoring module is used to monitor various dialysis parameters during the dialysis process in real time; The risk prediction module constructs a patient-specific risk map based on real-time monitored patient physiological data and dialysis parameters, predicts the probability of intradialytic hypotension, and classifies the risk level; The filling prediction module predicts the patient's plasma refill rate based on the patient's physiological data monitored in real time; The dynamic control module is used to dynamically adjust the ultrafiltration rate based on the predicted individual patient condition and the patient's plasma refill rate; The strategy execution module formulates and implements personalized intervention strategies based on risk classification; The medical record parsing module is used to parse electronic medical records and extract information about various symptoms of patients; The information push module hospital information system or mobile application pushes low blood pressure risk warnings to medical staff and patients and provides corresponding intervention strategies.

2. The intelligent prediction and management system for intradialysis hypotension in hemodialysis patients according to claim 1, characterized in that: The specific steps of the acquisition and processing module to process the monitored physiological parameters are as follows: S1.1: The acquisition and processing module collects patient physiological data collected in real time by external sensors, including blood pressure, blood volume, skin temperature, and heart rate. It synchronizes the time of each physiological data using dynamic time warping. It then uses a bandpass filter to remove high-frequency noise and low-frequency drift, and then uses a sliding mean filter to average the physiological data. S1.2: Divide each type of physiological data into multiple groups according to the set time window, then calculate the mean μ and standard deviation σ of each type of physiological data in each time window, and then calculate the Z-score value of each type of physiological data in each time interval; S1.3: If a Z-score value is higher than ±3σ, the corresponding physiological data is judged as an outlier and marked as a missing value. The missing value information in each type of physiological data is counted and filled by linear interpolation. Then, the Min-Max normalization method is used to normalize each type of physiological data to the interval [0, 1].

3. The intelligent prediction and management system for intradialytic hypotension in hemodialysis patients according to claim 2, characterized in that: The risk prediction module constructs a patient-specific risk map and predicts the probability of occurrence of intradialytic hypotension in the following specific steps: S2.1: Build a patient-specific physiological signal graph using the various physiological data monitored by the acquisition and processing module and the dialysis parameters monitored by the dialysis monitoring module as nodes and the interactions between the various physiological data as edges. Calculate the dynamic correlations between the calculated physiological parameters and dialysis parameters, and construct the corresponding edge weight matrix. S2.2: Signal propagation is performed on the established patient-specific physiological signal graph through the GCN network. The various data contained in each node in the patient-specific physiological signal graph are repeatedly fused through the multi-layer GCN network to obtain the final fused patient physiological status characteristics; S2.3: Input the generated physiological state features of each patient into a gated recurrent unit. Based on the three gating mechanisms of the gated recurrent unit, namely the update gate, reset gate, and candidate hidden state, the hidden state of each patient's physiological state feature at each time step is calculated in sequence until the hidden state at the final time step is obtained as the representation of the patient's current physiological state. S2.4: The patient's current physiological state is represented by the Sigmoid function of the fully connected layer and mapped to the range [0, 1] to generate a scalar probability value for the corresponding patient. If the scalar probability value is ≥ 0.7, a high-risk warning is triggered. If the scalar probability value is 0.3 ≤ 0.7, a moderate risk is indicated and adjustment of dialysis parameters is recommended. If the scalar probability value is < 0.3, it is judged as low risk and dialysis proceeds normally.

4. The intelligent prediction and management system for intradialysis hypotension in hemodialysis patients according to claim 3, characterized in that: The specific steps of the filling prediction module for predicting the patient's plasma refill rate are as follows: S3.1: Record the baseline blood volume at the start of dialysis. Measure the patient's blood volume every minute using the bioelectrical impedance method. Calculate the relative change in blood volume during dialysis in real time based on the baseline blood volume and the patient's blood volume at that moment. S3.2: Calculate the rate of blood volume decrease between two adjacent moments. Then, calculate the patient's fluid loss rate per minute based on the patient's ultrafiltration volume recorded in real time. Then, calculate the plasma refill volume for the corresponding time period based on the obtained blood volume decrease rate, fluid loss rate, and baseline blood volume at the start of dialysis. S3.3: Collect plasma filling volume data for different time periods, establish filling datasets for the corresponding time periods, and set a set of N×N sliding windows. Then, move the sliding windows sequentially from the beginning of each filling dataset. S3.4: After each movement, smooth the plasma filling volume data within the current sliding window to obtain the plasma refill rate at the current time, and compare the plasma refill rate at the current time with the patient's ultrafiltration rate to obtain the patient's vascular compensation rate index; S3.5: If the vascular compensation rate index is lower than 0.5, it indicates that the patient may be at risk of hypotension and an alert or adjustment of the ultrafiltration rate should be issued. Otherwise, it indicates that the plasma refilling capacity is good.

5. The intelligent prediction and management system for intradialysis hypotension in hemodialysis patients according to claim 4, characterized in that: The specific steps of the dynamic control module dynamically adjusting the ultrafiltration rate are as follows: S4.1: Extract the patient's historical dialysis data, including age, cardiovascular disease history, historical hypotension incidence, dialysis adequacy index, and plasma refill rate. Calculate the patient's maximum tolerated ultrafiltration rate and minimum tolerated ultrafiltration rate using a regression model based on the collected historical dialysis data. S4.2: Based on the calculated maximum tolerated ultrafiltration rate and minimum tolerated ultrafiltration rate, initialize the ultrafiltration rate according to the patient's current state. Establish a state space based on the patient's physiological state and dialysis parameters, including the current systolic blood pressure, plasma refill rate, current ultrafiltration rate, and dialysate sodium concentration. Set the ultrafiltration rate adjustment step size to establish the corresponding action space. S4.3: With the goal of reducing the risk of hypotension and maintaining dialysis adequacy, construct a corresponding reward function, randomly select actions in the action space, then calculate the long-term benefits of each action in the action space under the current patient state based on the reward function, and select the action with the highest current long-term benefit to adjust the ultrafiltration rate; S4.4: Recalculate the patient's status based on the preset time interval and update the long-term benefits of each action. Reselect the action with the highest long-term benefit to adjust the ultrafiltration rate. If the systolic blood pressure decreases beyond the preset threshold within the preset time interval, immediately reduce the ultrafiltration rate.

6. The intelligent prediction and management system for intradialysis hypotension in hemodialysis patients according to claim 5, characterized in that: The specific steps of the policy execution module are as follows: S5.1: Collect the patient's current risk level, hemodynamic parameters, and dialysis settings to establish a corresponding joint state set for the patient. Then, collect the interventions for adjusting sodium concentration, pre-injecting normal saline, and reducing ultrafiltration rate to establish corresponding action sets. The patient's current joint state set is then used as the root node. S5.2: Starting from the root node, simulate the new patient state generated after executing each intervention action in the current patient state and use it as a new child node to construct the corresponding exploration tree. Then, starting from the root node, calculate the average reward of executing each action under the current state and the UCB value of each child node; S5.3: Select the child node with the highest average reward and UCB value as the best child node. Repeat the child node selection process until a leaf node is selected that has not yet explored all actions. Based on the state of this leaf node, select an unsimulated action and generate a new patient state. This action is then added to the exploration tree as a new child node. S5.4: Starting from the current new child node, use a random strategy to simulate future intervention operations until the patient completes dialysis. Based on the simulation results, calculate the total simulated utility corresponding to the current path. Then, trace the simulated total utility back to all nodes on the path and update the statistics of each node. S5.5: Repeat the selection, expansion, simulation, and backtracking until the preset number of iterations is reached. Then, stop the iteration, traverse the final exploration tree, and select the child node action with the highest average utility from the root node. This action is used as the final recommended intervention, and the optimal path and corresponding intervention strategy are output.

Images