AI fusion enables crrt intelligent anticoagulation decision system
By constructing a hybrid AI model that combines CNN and LSTM networks, the problem of insufficient data fusion in traditional CRRT anticoagulation decision-making was solved, realizing intelligent and precise CRRT treatment and improving the treatment effect and safety of critically ill patients.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JIAXING NO 1 HOSPITAL
- Filing Date
- 2026-01-07
- Publication Date
- 2026-06-05
AI Technical Summary
Traditional CRRT anticoagulation decisions rely on experience-based judgment, cannot deeply integrate multi-dimensional data, cannot predict coagulation trends, resulting in response lag and errors, and cannot form intelligent closed-loop control.
A hybrid AI model was constructed, combining CNN and LSTM networks, and integrating patient signs and CRRT instrument parameters to achieve dynamic modeling and generate adaptive anticoagulation parameter trajectories.
It enables multi-dimensional, dynamic, and precise modeling of CRRT anticoagulation decisions, improving the continuity and safety of treatment and reducing human delays and errors.
Smart Images
Figure CN122158053A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent control technology for medical devices, specifically to an AI-enabled intelligent anticoagulation decision-making system for CRRT. Background Technology
[0002] Continuous renal replacement therapy (CRRT) is a crucial supportive treatment for critically ill patients. Its core objective is to stabilize the patient's internal environment through continuous blood purification. Anticoagulation management is a critical aspect of CRRT, directly determining the success or failure of the treatment. An ideal anticoagulation regimen requires a precise balance between preventing coagulation during extracorporeal circulation and avoiding bleeding. However, the conditions of critically ill patients are complex and variable, and their coagulation status often fluctuates dynamically due to multiple factors such as primary disease, infection, and surgery. This makes the formulation and adjustment of anticoagulation regimens extremely challenging. In the traditional model, CRRT anticoagulation decisions heavily rely on the experience and judgment of the medical team, manually adjusting parameters by intermittently checking coagulation indicators. This discrete and lagging management model is ill-suited to the complex and rapidly changing physiological state of the human body.
[0003] In existing technologies, attempts to make CRRT intelligent are mostly focused on threshold alarms for single parameters or simple logic control based on fixed rules. For example, an alarm is issued when the transmembrane pressure (TMP) exceeds a preset threshold, or an initial anticoagulant dose is calculated based on the patient's weight. However, these methods have significant drawbacks. They lack the ability to deeply integrate and correlate multi-dimensional and heterogeneous data, and their decision-making is based on one-sided criteria. They cannot capture and learn the trend of parameter changes over time and its deep mapping relationship with treatment results. For example, they cannot predict that the slow rise in TMP will eventually lead to clotting of the filter after several hours. Therefore, they cannot achieve prospective intervention. Moreover, they are mostly open-loop assistance, meaning that manual execution is still required after suggestions are provided. There are human delays and errors in response speed and execution accuracy, and they have failed to form an intelligent closed loop of perception, decision-making, and execution.
[0004] Therefore, there is an urgent clinical need for an intelligent system that can simulate clinical thinking. This system should be able to comprehensively utilize real-time and historical data to deeply understand the dynamic interaction between the patient's individualized pathophysiological state and the CRRT treatment process, and be able to autonomously and accurately execute dynamic adjustments to individualized anticoagulation regimens to ensure the continuity and safety of treatment. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an AI-enabled intelligent anticoagulation decision-making system for CRRT. This system achieves multi-dimensional, dynamic, and precise modeling of the complex problem of CRRT anticoagulation decision-making by constructing a hybrid AI model that integrates spatial and temporal feature extraction capabilities. It utilizes a convolutional neural network (CNN) to mine deep spatial correlation features between static parameters such as patient weight, coagulation indicators, and platelet count. Simultaneously, it uses a long short-term memory network (LSTM) to process high-frequency time series formed by machine parameters such as transmembrane pressure and filter pressure, learning the dynamic laws of filter performance decay and subtle trends of precoagulation precursors. Through end-to-end training, this hybrid model fuses and maps these two types of features at a high level, ultimately forming a dynamic digital complex that comprehensively reflects the patient's real-time coagulation risk, filter status, and treatment efficacy. This allows the system to predict the probability distribution of treatment outcomes over a future period based on the current system state, generating not a static, fixed plan, but an optimal parameter trajectory that adaptively adjusts as the system state evolves.
[0006] To solve the above-mentioned technical problems, this invention provides the following technical solution: an AI-integrated CRRT intelligent anticoagulation decision-making system, which includes: a data acquisition module, a data preprocessing module, an AI computing module, a decision output module, and a control interface module, wherein:
[0007] The data acquisition module is used to collect patient parameters and CRRT instrument parameters in real time. The patient parameters include age, weight, body temperature, height, heart rate, blood pressure, diagnostic information, coagulation function parameters, platelet count, and coagulation factors. The CRRT instrument parameters include filter pressure, arterial pressure, venous pressure, and transmembrane pressure.
[0008] The data preprocessing module is used to standardize, normalize, and extract features from the collected raw data to generate standardized data for use by the AI computing module.
[0009] Based on the standardized data, the AI computing module generates recommended treatment parameters through machine learning algorithms, including anticoagulation dose, blood flow, replacement fluid volume, and dehydration volume, and dynamically adjusts the parameters according to the treatment goals.
[0010] The decision output module is used to output the treatment recommendation parameters in a visual form, including screen display and sound prompts, to guide medical staff to make adjustments;
[0011] The control interface module is used to connect to the CRRT machine to automatically adjust the recommended treatment parameters, forming a closed-loop control system.
[0012] Furthermore, the data acquisition module includes:
[0013] The patient data unit is used to automatically obtain the patient's static parameters and dynamic test parameters from the electronic medical record. The static parameters include age, weight, height, and basic diagnosis, while the dynamic test parameters include real-time updated coagulation function parameters, platelet count, electrolyte level, and inflammatory markers.
[0014] The instrument data unit is used to communicate with the CRRT machine through a data interface to continuously acquire the filter pressure, arterial pressure, venous pressure, transmembrane pressure, current blood flow, replacement fluid flow rate and dehydration rate at a preset sampling frequency;
[0015] The data fusion unit is used to time-align and integrate information from the patient data unit and the instrument data unit.
[0016] Furthermore, the execution process of the data preprocessing module includes:
[0017] Standardization involves identifying and handling missing and outlier values. For continuous parameters, the sliding window mean method is used for imputation, and for outliers, the historical quantiles of patients of the same type are used for truncation.
[0018] Normalization transforms parameters of different dimensions and magnitudes to the same standard normal distribution.
[0019] For feature engineering, composite features are generated based on clinical knowledge, including the ratio of blood pressure to heart rate and the rate of change of transmembrane pressure over time.
[0020] Furthermore, the AI computing module has a pre-trained AI algorithm model built in, and the AI algorithm model construction process is as follows:
[0021] Model architecture: A CNN-LSTM hybrid network structure is adopted, in which the CNN layer is used to extract parametric spatial features, including the correlation between patient weight and anticoagulation dose, and the mapping relationship between PT value and bleeding risk, and the LSTM layer is used to extract temporal features, including the trend of TMP change over time and the dynamic fluctuation pattern of PT value.
[0022] Training data: Structured data of CRRT treatment cases include: patient baseline parameters, test parameters, operating parameters, treatment goals and treatment results;
[0023] Training process: With the goal of achieving treatment, the loss function is defined as the combination of cross-entropy loss and mean squared error between the predicted treatment result and the actual result. The network weights are iteratively adjusted using the training set data and the backpropagation algorithm.
[0024] Model Deployment and Iteration: The trained model is integrated into the AI computing module for real-time processing of input parameters. By learning the mapping relationship between patient parameters, machine parameters, and treatment results in historical data, the real-time input parameters are feature-encoded. The gradient is adjusted by calculating parameters in combination with preset treatment goals to generate optimal treatment recommendation parameters, including anticoagulation dose, blood flow rate, replacement fluid volume, and dehydration volume.
[0025] Furthermore, the anticoagulant dosage ,in, Basic anticoagulant dose and , For coagulation function correction factor and , The transmembrane pressure correction factor and , Treatment time correction factor and , represents weight, Indicates age, and These represent the upper limit of normal prothrombin time and the patient's current PT value, respectively. This indicates the safe threshold for transmembrane pressure. Indicates the current transmembrane pressure. Indicates the running time. Indicates the target treatment time.
[0026] Furthermore, the blood flow velocity ,in, Basic blood flow velocity and , The platelet count correction factor and , The filter pressure correction factor and , For weight, This indicates the lower limit of normal platelet count. This indicates the patient's current platelet count. Indicates the initial filter pressure. This indicates the current filter pressure.
[0027] Furthermore, the displacement fluid volume ,in, Basic replacement fluid volume and , The correction factor for serum creatinine and , It is the capacity state correction factor and , For weight, This indicates the upper limit of normal serum creatinine levels. This indicates the patient's current serum creatinine level. This indicates the liquid load index.
[0028] Furthermore, the amount of water removed Where TBW is the estimated total liquid volume and TBW , The urine volume correction factor and , For blood pressure correction factor and , For weight, Indicates the target treatment time. This indicates normal hourly urine output. This indicates the amount of urine per hour. This indicates the patient's baseline systolic blood pressure. This indicates the current systolic blood pressure.
[0029] Compared with existing technologies, this AI-integrated intelligent anticoagulation decision-making system for CRRT has the following beneficial effects:
[0030] I. This invention constructs a hybrid AI model that integrates spatial and temporal feature extraction capabilities, achieving multi-dimensional, dynamic, and precise modeling of the complex problem of CRRT anticoagulation decision-making. It utilizes a convolutional neural network (CNN) to mine deep spatial correlation features between static parameters such as patient weight, coagulation indicators, and platelet count. Simultaneously, it employs a long short-term memory network (LSTM) to process high-frequency time series formed by machine parameters such as transmembrane pressure and filter pressure, learning the dynamic laws of filter performance decay and subtle trends of precoagulation precursors. This hybrid model, through end-to-end training, fuses and maps these two types of features at a high level, ultimately forming a dynamic digital complex that comprehensively reflects the patient's real-time coagulation risk, filter status, and treatment efficacy. This allows the system to predict the probability distribution of treatment outcomes over a future period based on the current system state, thus generating not a static, fixed plan, but an optimal parameter trajectory that adaptively adjusts with the evolution of the system state.
[0031] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description
[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.
[0033] Figure 1 Operation flowchart for AI-enabled CRRT intelligent anticoagulation decision-making system;
[0034] Figure 2 A block diagram of the modules that enable AI-integrated intelligent anticoagulation decision-making system for CRRT. Detailed Implementation
[0035] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.
[0036] Example 1
[0037] This embodiment aims to explain in detail the working principle of the AI-enabled CRRT intelligent anticoagulation decision-making system, such as... Figure 2 As shown, the system includes a data acquisition module, a data preprocessing module, an AI computing module, a decision output module, and a control interface module. The data acquisition module acquires multi-dimensional physiological parameters of the patient and operating parameters of the CRRT instrument in real time. After standardization, normalization, and feature engineering by the data preprocessing module, the data is input to the AI computing module based on a CNN-LSTM hybrid network architecture. The AI computing module combines preset treatment goals and uses a trained algorithm model and specific calculation formulas to dynamically generate recommended treatment parameters such as anticoagulation dose, blood flow, replacement fluid volume, and dehydration volume. The decision output module then guides medical staff in operation in a visual form. At the same time, the control interface module links with the CRRT machine to achieve automatic parameter adjustment, forming a control system of acquisition, processing, calculation, decision-making, and execution. This solves the problems of traditional CRRT anticoagulation decision-making relying on experience and response lag, and improves treatment accuracy and safety.
[0038] (I) Working principle of the data acquisition module
[0039] The data acquisition module, as the system's data input source, is responsible for comprehensively acquiring CRRT treatment-related data. Its operation revolves around the collaborative efforts of the patient data unit, instrument data unit, and data fusion unit, specifically as follows:
[0040] The patient data unit automatically extracts static and dynamic test parameters from the hospital's electronic medical record system, ensuring convenient and timely data acquisition. Static parameters refer to the patient's basic information, which is relatively stable and changes with low frequency during the treatment period, including age, weight, height, and basic diagnosis. Dynamic test parameters refer to the patient's physiological indicators, which change in real time with the treatment process and require regular monitoring and updating. These parameters include real-time updated coagulation function parameters, platelet count, electrolyte levels, and inflammatory markers. These parameters directly reflect the patient's current coagulation status and the trend of disease progression. The patient data unit establishes a data interface with the hospital's electronic medical record system and automatically retrieves parameters at preset time intervals, avoiding errors and delays caused by data entry.
[0041] The instrument data unit is responsible for establishing real-time communication with the CRRT machine and continuously collecting key parameters during machine operation. This provides a basis for assessing the status of the treatment equipment and adjusting treatment parameters. The unit connects to the CRRT machine through a standardized data interface and acquires parameters such as filter pressure, arterial pressure, venous pressure, transmembrane pressure (TMP), current blood flow, replacement fluid flow rate, and dehydration rate at a preset sampling frequency. Among these parameters, filter pressure reflects the degree of filter blockage, arterial and venous pressures assess the patency of the extracorporeal circulation pathway, and transmembrane pressure is a core indicator for judging filter performance. Current blood flow, replacement fluid flow rate, and dehydration rate are the core operating parameters of CRRT treatment, which are matched with the patient's physiological state to ensure treatment effectiveness and safety. During the acquisition process, the instrument data unit performs preliminary data verification. If abnormal values are found, they are marked as data to be processed and further processed by the data preprocessing module.
[0042] Because the parameters acquired by the patient data unit and the parameters acquired by the instrument data unit have different timestamps, the data fusion unit needs to perform time alignment and integration of the two types of data to form a unified dataset for use by subsequent modules. At the same time, the unit will add identification information to the integrated data, including patient ID, treatment start time, current treatment stage, etc., to facilitate the subsequent AI computing module to trace the data source, analyze the parameter change patterns at different treatment stages, and finally generate a structured time-series dataset, which is then transmitted to the data preprocessing module.
[0043] (II) Working principle of the data preprocessing module
[0044] The raw data acquired by the data acquisition module contains issues such as missing values, outliers, and inconsistent units, making it unsuitable for direct input into the AI calculation module. The data preprocessing module needs to transform the raw data into standardized data that meets the input requirements of the AI model through standardization, normalization, and feature engineering. Specifically:
[0045] The core objective of standardization is to identify and process missing and outlier values in the raw data to ensure data integrity and reliability. For continuous parameters such as transmembrane pressure, PT value, and platelet count, missing value processing uses a sliding window mean method, calculating the mean of data points before and after a time point within the window as the imputation value. For outlier processing, historical data quantiles from similar patients are used for truncation. Historical data from similar patients with similar baseline diagnoses, ages, and weights are selected, and the 95th and 5th percentiles for the corresponding parameters in this data set are calculated. Values in the current data that exceed the 95th percentile or fall below the 5th percentile are truncated to the 95th percentile. For outliers caused by sudden changes in a patient's condition, such as the 5th percentile, the original data is retained in conjunction with clinical judgment to avoid truncation that masks the true changes in the patient's condition. In this case, the data preprocessing module will mark the outliers as clinically relevant anomalies and synchronize them to the decision output module to remind medical staff to pay attention to the patient's condition. For normalization, parameters of different dimensions and magnitudes are transformed to the same standard normal distribution to eliminate differences in dimensions and magnitudes and ensure that the AI model allocates weights to each parameter reasonably. For feature engineering, a single parameter in the original data is difficult to fully reflect the correlation between the patient's physiological state and the effect of CRRT treatment. Feature engineering generates composite features based on clinical knowledge to improve the AI model's ability to capture key information.
[0046] (III) Working principle of AI computing module
[0047] The AI computing module, based on the standardized data output by the data preprocessing module, dynamically generates treatment recommendation parameters through a trained CNN-LSTM hybrid network model, specifically:
[0048] AI Algorithm Model Construction and Training: A CNN-LSTM hybrid network structure is adopted, combining the advantages of both networks to achieve multi-dimensional feature extraction. The CNN layer, or Convolutional Neural Network layer, is mainly used to extract the spatial features of parameters, i.e., the correlation between different parameters, such as the correlation between patient weight and anticoagulation dose, and the mapping relationship between PT value and bleeding risk. The CNN layer slides through the standardized dataset using convolution kernels, performing convolution operations on the features of different parameters to extract local correlation features, and uses pooling layers to reduce feature dimensionality while retaining key information. The LSTM layer, or Long Short-Term Memory network layer, is used to extract the temporal features of parameters, i.e., the pattern of parameter changes over time, such as the trend of TMP changes over time and the dynamic fluctuation pattern of PT value. The LSTM layer uses a gating mechanism to memorize long-term temporal information, avoiding short-term memory loss, capturing the long-term trend of parameter changes, and providing a basis for predicting future treatment outcomes. The training data comes from structured data of CRRT treatment cases in hospitals, covering patient baseline parameters, test parameters, operational parameters, and treatment... Treatment goals and outcomes are considered, with achieving the treatment goal as the optimization objective. The loss function is defined as the combination of cross-entropy loss and mean squared error between the predicted and actual treatment outcomes. Cross-entropy loss evaluates the model's prediction accuracy for classification tasks, while mean squared error evaluates the model's prediction error for continuous parameters. These two are combined through weighting coefficients to form the total loss function. During training, training data is input into the model, and the network weights are iteratively adjusted using the backpropagation algorithm. The total loss function value is calculated after each iteration. If the loss function value decreases, the current weights are retained; if the loss function value does not decrease after multiple iterations, training is stopped to avoid overfitting. Simultaneously, validation data is used to periodically evaluate model performance, and hyperparameters are adjusted based on the evaluation results to obtain the optimal training model. The AI calculation module inputs the standardized data output from the data preprocessing module into the trained model. Combined with the preset treatment goals, the model calculates and generates four core recommended treatment parameters: anticoagulation dose, blood flow rate, replacement fluid volume, and dehydration volume. The specific calculation process is as follows:
[0049] Anticoagulation dose calculation: Anticoagulation dose is a key parameter for balancing coagulation and bleeding risks during CRRT treatment. It is calculated based on the product of the baseline anticoagulation dose and correction factors for coagulation function, transmembrane pressure, and treatment time. The formula is as follows: ,in, Basic anticoagulant dose and , For coagulation function correction factor and , The transmembrane pressure correction factor and , Treatment time correction factor and , represents weight, Indicates age, and These represent the upper limit of normal prothrombin time and the patient's current PT value, respectively. This indicates the safe threshold for transmembrane pressure. Indicates the current transmembrane pressure. Indicates the running time. Indicates the target treatment time.
[0050] Blood flow calculation: Blood flow affects the clearance efficiency and circulatory stability of CRRT treatment. It is calculated based on the product of baseline blood flow velocity, platelet count correction factor, and filter pressure correction factor. The formula is as follows: ,in, Basic blood flow velocity and , The platelet count correction factor and , The filter pressure correction factor and , For weight, This indicates the lower limit of normal platelet count. This indicates the patient's current platelet count. Indicates the initial filter pressure. This indicates the current filter pressure.
[0051] Replacement fluid volume calculation: Replacement fluid volume affects solute clearance and patient volume balance during CRRT treatment. It is calculated based on the product of the baseline replacement fluid volume and the serum creatinine correction factor and volume status correction factor. The formula is as follows: ,in, Basic replacement fluid volume and , The correction factor for serum creatinine and , It is the capacity state correction factor and , For weight, This indicates the upper limit of normal serum creatinine levels. This indicates the patient's current serum creatinine level. This indicates the liquid load index.
[0052] Dehydration calculation: Dehydration is used to regulate the patient's volume balance. It is calculated based on the product of the total fluid volume estimate and the urine volume correction factor and blood pressure correction factor, using the following formula: Where TBW is the estimated total liquid volume and TBW , The urine volume correction factor and , For blood pressure correction factor and , For weight, Indicates the target treatment time. This indicates normal hourly urine output. This indicates the amount of urine per hour. This indicates the patient's baseline systolic blood pressure. This indicates the current systolic blood pressure.
[0053] After generating the above four treatment recommendation parameters, the AI computing module will dynamically adjust the model's prediction of future treatment outcomes. For example, if the model predicts an increased risk of filter coagulation, the anticoagulant dose will be appropriately increased and the blood flow rate will be reduced; if the model predicts an increased risk of bleeding, the anticoagulant dose will be appropriately reduced. Finally, the optimal treatment recommendation parameters will be determined and transmitted to the decision output module.
[0054] (iv) Working principle of the decision output module
[0055] The core function of the decision output module is to output the treatment recommendation parameters generated by the AI computing module in an intuitive and easy-to-understand visual form, providing clear operational guidance for medical staff, while conveying patient condition and treatment risk warnings. The decision output module outputs information by combining screen display and sound prompts. After outputting the information, the decision output module allows medical staff to manually adjust the treatment recommendation parameters. If medical staff believe that the parameters do not meet the patient's actual situation, they can enter the adjusted parameters on the interface and fill in the reason for the adjustment. The system will record the manual adjustment record and transmit the adjusted parameters to the control interface module.
[0056] (V) Working principle of the control interface module
[0057] The control interface module acts as a bridge between the system and the CRRT machine, enabling automatic adjustment of recommended treatment parameters to form a closed-loop control system. The interface communication module uses a standardized medical device communication protocol to establish communication with the CRRT machine, ensuring data transmission security and compatibility. During communication, the system authenticates the CRRT machine; upon successful authentication, an encrypted communication channel is established to prevent data tampering or leakage. Communication between the control interface module and the CRRT machine includes data transmission and reception. The communication frequency matches the sampling frequency of the instrument's data unit to ensure real-time parameter adjustment. The control interface module receives treatment recommendations transmitted from the decision output module. After the CRRT machine adjusts according to the received parameters, it feeds back the actual operating parameters to the control interface module in real time. The control interface module compares the actual operating parameters with the recommended treatment parameters and calculates the deviation value. If the deviation value is within the allowable range, the adjustment is considered normal. If the deviation value exceeds the allowable range, the control interface module analyzes the cause of the deviation and feeds the deviation information back to the AI calculation module. The AI calculation module recalculates the recommended treatment parameters and prompts medical staff through the decision output module. At the same time, the system records the time, recommended value, actual value, deviation value, and correction measures for each parameter adjustment, forming an adjustment log to facilitate subsequent analysis of the treatment process and optimization of the treatment plan by medical staff.
[0058] This embodiment elaborates on the working principles of the data acquisition module, data preprocessing module, AI calculation module, decision output module, and control interface module of the AI-integrated CRRT intelligent anticoagulation decision-making system. It fully presents the closed-loop control process of the system from data acquisition to parameter execution. Through the collaborative work of multiple modules, the entire system effectively solves the problems of traditional CRRT anticoagulation decision-making, such as reliance on experience, response lag, and one-sided parameter adjustment. It realizes the intelligent, individualized, and precise treatment of CRRT, providing reliable technical support for the treatment of critically ill patients and helping to improve treatment efficacy and patient safety.
[0059] Example 2
[0060] Based on Example 1, this example provides the specific steps of the AI-empowered CRRT intelligent anticoagulation decision-making system in making CRRT intelligent anticoagulation decisions, such as... Figure 1 As shown, the specific steps are as follows:
[0061] (1) Real-time data acquisition
[0062] Automatically obtain patient static parameters from electronic medical records.
[0063] Real-time collection of dynamic parameters of patients.
[0064] Instrument parameters are continuously read via the CRRT machine interface.
[0065] Align and integrate patient parameters with instrument parameters over time.
[0066] (2) Data preprocessing
[0067] Handling missing values: For continuous parameters, imputation is performed using the sliding window mean.
[0068] Handling outliers: Truncate outliers using quantiles from historical data of similar patients.
[0069] Normalization: Convert all parameters to a standard normal distribution format.
[0070] Generate composite features: Calculate clinically derived indicators.
[0071] (3) AI intelligent computing
[0072] Spatial features are extracted using CNN layers.
[0073] Temporal features are extracted using LSTM layers.
[0074] By integrating spatial and temporal characteristics and combining them with preset treatment goals, dynamic recommended parameters are generated: anticoagulation dose, blood flow, replacement fluid volume, and dehydration volume.
[0075] (4) Decision output and prompts
[0076] Recommended parameters are dynamically visualized on the display screen.
[0077] Trigger an audible alarm for critical anomalies.
[0078] Provide medical staff with instructions on adjusting procedures.
[0079] (5) Automatic closed-loop control
[0080] Recommended parameters are automatically sent to the CRRT machine via the control interface.
[0081] Adjust the instrument's execution parameters in real time, collect machine feedback data, and re-enter the data into the system to start a new round of decision-making cycle.
[0082] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.
Claims
1. An AI-integrated intelligent anticoagulation decision-making system for CRRT, characterized in that, The system includes: a data acquisition module, a data preprocessing module, an AI computing module, a decision output module, and a control interface module, wherein: The data acquisition module is used to collect patient parameters and CRRT instrument parameters in real time. The patient parameters include age, weight, body temperature, height, heart rate, blood pressure, diagnostic information, coagulation function parameters, platelet count, and coagulation factors. The CRRT instrument parameters include filter pressure, arterial pressure, venous pressure, and transmembrane pressure. The data preprocessing module is used to standardize, normalize, and extract features from the collected raw data to generate standardized data for use by the AI computing module. Based on the standardized data, the AI computing module generates recommended treatment parameters through machine learning algorithms, including anticoagulation dose, blood flow, replacement fluid volume, and dehydration volume, and dynamically adjusts the parameters according to the treatment goals. The decision output module is used to output the treatment recommendation parameters in a visual form, including screen display and sound prompts, to guide medical staff to make adjustments; The control interface module is used to connect to the CRRT machine to automatically adjust the recommended treatment parameters, forming a closed-loop control system.
2. The AI-integrated CRRT intelligent anticoagulation decision-making system according to claim 1, characterized in that, The data acquisition module includes: The patient data unit is used to automatically obtain the patient's static parameters and dynamic test parameters from the electronic medical record. The static parameters include age, weight, height, and basic diagnosis, while the dynamic test parameters include real-time updated coagulation function parameters, platelet count, electrolyte level, and inflammatory markers. The instrument data unit is used to communicate with the CRRT machine through a data interface to continuously acquire the filter pressure, arterial pressure, venous pressure, transmembrane pressure, current blood flow, replacement fluid flow rate and dehydration rate at a preset sampling frequency; The data fusion unit is used to time-align and integrate information from the patient data unit and the instrument data unit.
3. The AI-integrated CRRT intelligent anticoagulation decision-making system according to claim 1, characterized in that, The execution process of the data preprocessing module includes: Standardization involves identifying and handling missing and outlier values. For continuous parameters, the sliding window mean method is used for imputation, and for outliers, the historical quantiles of patients of the same type are used for truncation. Normalization transforms parameters of different dimensions and magnitudes to the same standard normal distribution. For feature engineering, composite features are generated based on clinical knowledge, including the ratio of blood pressure to heart rate and the rate of change of transmembrane pressure over time.
4. The AI-integrated CRRT intelligent anticoagulation decision-making system according to claim 1, characterized in that, The AI computing module has a built-in trained AI algorithm model, and the AI algorithm model construction process is as follows: Model architecture: A CNN-LSTM hybrid network structure is adopted, in which the CNN layer is used to extract parametric spatial features, including the correlation between patient weight and anticoagulation dose, and the mapping relationship between PT value and bleeding risk, and the LSTM layer is used to extract temporal features, including the trend of TMP change over time and the dynamic fluctuation pattern of PT value. Training data: Structured data of CRRT treatment cases include: patient baseline parameters, test parameters, operating parameters, treatment goals and treatment results; Training process: With the goal of achieving treatment, the loss function is defined as the combination of cross-entropy loss and mean squared error between the predicted treatment result and the actual result. The network weights are iteratively adjusted using the training set data and the backpropagation algorithm. Model Deployment and Iteration: The trained model is integrated into the AI computing module for real-time processing of input parameters. By learning the mapping relationship between patient parameters, machine parameters, and treatment results in historical data, the real-time input parameters are feature-encoded. The gradient is adjusted by calculating parameters in combination with preset treatment goals to generate optimal treatment recommendation parameters, including anticoagulation dose, blood flow rate, replacement fluid volume, and dehydration volume.
5. The AI-integrated CRRT intelligent anticoagulation decision-making system according to claim 4, characterized in that, The anticoagulant dosage ,in, Basic anticoagulant dose and , For coagulation function correction factor and , The transmembrane pressure correction factor and , Treatment time correction factor and W represents weight, and Age represents age. and These represent the upper limit of normal prothrombin time and the patient's current PT value, respectively. This represents the safe threshold for transmembrane pressure, and TMP represents the current transmembrane pressure. Indicates the running time. Indicates the target treatment time.
6. The AI-integrated CRRT intelligent anticoagulation decision-making system according to claim 4, characterized in that, The blood flow velocity ,in, Basic blood flow velocity and , The platelet count correction factor and , The filter pressure correction factor and , For weight, This indicates the lower limit of normal platelet count. This indicates the patient's current platelet count. Indicates the initial filter pressure. This indicates the current filter pressure.
7. The AI-integrated CRRT intelligent anticoagulation decision-making system according to claim 4, characterized in that, The displacement fluid volume ,in, Basic replacement fluid volume and , The correction factor for serum creatinine and , It is the capacity state correction factor and , For weight, This indicates the upper limit of normal serum creatinine levels. This indicates the patient's current serum creatinine level. This indicates the liquid load index.
8. The AI-integrated CRRT intelligent anticoagulation decision-making system according to claim 4, characterized in that, The amount of water removed Where TBW is the estimated total liquid volume and TBW , The urine volume correction factor and , For blood pressure correction factor and , For weight, Indicates the target treatment time. This indicates normal hourly urine output. This indicates the amount of urine per hour. This indicates the patient's baseline systolic blood pressure. This indicates the current systolic blood pressure.