Artificial liver anticoagulant dynamic regulation and control system and method based on multi-modal monitoring

Abstract

The invention discloses an artificial liver anticoagulant dynamic regulation and control system and method based on multi-modal monitoring, and relates to the field of anticoagulant use regulation and control. Real-time indexes and personalized treatment schemes of patients can be accurately matched through an anticoagulant type mapping model and an injection rate mapping model which are established based on historical treatment data; by monitoring the residual amount of the anticoagulant in real time and predicting the total demand quantity, advanced early warning is realized in combination with purchase time consumption evaluation, and treatment continuity is ensured; closed-loop feedback is formed through full-process iteration when the type is changed or rate curve reestimation when the type is not changed, and injection parameters are continuously adjusted; the manual intervention frequency is remarkably reduced, dynamic calibration of the dosage of the anticoagulant is achieved through multi-time-point data modeling, and resource waste and the blood coagulation risk are effectively avoided while the treatment effect is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of anticoagulant use regulation, and particularly relates to an artificial liver anticoagulant dynamic regulation system and method based on multi-modal monitoring. BACKGROUND

[0002] Anticoagulants play a role by inhibiting the function of coagulation factors or platelets, and if not properly selected, can lead to a bleeding tendency in patients, which can be life-threatening. Traditional empirical medication has poor individual adaptability, and traditional static dose adjustment can lead to fluctuations in coagulation function. In addition, the traditional artificial monitoring method has a lag, which can cause waste of anticoagulant resources and supply interruption dilemma. SUMMARY

[0003] In view of the problems in the related art, the present application proposes an artificial liver anticoagulant dynamic regulation system and method based on multi-modal monitoring to overcome the above technical problems existing in the prior art.

[0004] To solve the above technical problems, the present application is realized by the following technical scheme: The present application is an artificial liver anticoagulant dynamic regulation method based on multi-modal monitoring, comprising the following steps: S1, collecting historical data of corresponding index data of patients in the artificial liver treatment process and corresponding anticoagulant type data; S2, constructing a mapping model between the corresponding index data of the patient and the selected anticoagulant type based on the data collected in S1; S3, collecting data in the historical artificial liver treatment process to construct a mapping model between the corresponding index data of the patient, the anticoagulant type data used and the injection rate data of the corresponding multiple time nodes; S4, inputting the corresponding index data of the current patient into the mapping model constructed in S2 for mapping; S5, based on the remaining amount of anticoagulant corresponding to the mapping result in S4, generating the injection rate curve of the current patient and calculating the total demand amount using the mapping model constructed in S3; if the total demand amount exceeds the remaining amount of anticoagulant, the anticoagulant is purchased in advance; S6, if the injection rate curve in S5 shows continuous zero, the latest corresponding index is collected at the replacement time point, and the mapping model constructed in S2 is used to remap the selected anticoagulant type; S7, if the selected anticoagulant type in S6 is different from the previous anticoagulant type, repeat S5, S6 and S7 until the end of treatment, otherwise, re-estimate the injection rate curve based on the latest corresponding index and the mapping model constructed in S3, and repeat S5, S6 and S7.

[0005] Preferably, S1 comprises the following steps: S11, select several patient index types to be monitored for artificial liver treatment, obtain a patient vital sign index type set, a patient coagulation function index type set, and a patient biochemical index type set; then set several alternative anticoagulant types, obtain an alternative anticoagulant type set; S12, according to the patient vital sign index type set, the patient coagulation function index type set, the patient biochemical index type set, and the alternative anticoagulant type set, obtain patient vital sign index data, coagulation function index data, biochemical index data, and corresponding anticoagulant type data in the artificial liver treatment process of several patients in history who did not have discomfort, obtain a first historical patient vital sign index data set, a first historical patient coagulation function index data set, a first historical patient biochemical index data set, and a first historical selected anticoagulant type data set; Through multi-dimensional dynamic monitoring and intelligent matching of anticoagulants, the safety and accuracy of artificial liver treatment are significantly improved. In terms of anticoagulation strategy, by strictly screening the data of artificial liver treatment processes in history that did not have discomfort, a highly reliable clinical decision support foundation is constructed. Pure training samples are provided for subsequent mapping models, avoiding model bias caused by abnormal data.

[0006] Preferably, the S2 comprises the following steps: S21, according to the first historical patient vital sign index data set, the first historical patient coagulation function index data set, the first historical patient biochemical index data set, and the first historical selected anticoagulant type data set, a mapping model between patient vital sign index data, patient coagulation function index data, patient biochemical index data, and selected anticoagulant type data is constructed, and a final selected anticoagulant type mapping model is obtained; By constructing a multi-dimensional data mapping model, intelligent decision-making of anticoagulant selection in artificial liver treatment is realized. In terms of data integration, the organic integration of historical patient vital sign indexes, coagulation function indexes, and biochemical indexes forms a holographic evaluation system covering circulation stability, coagulation state, and metabolic balance. Through pattern mining of historical data, clinical experience is converted into quantifiable decision rules. Using multi-index collaborative analysis, the risk of anticoagulation-related complications is significantly reduced. Moreover, the mapping precision can be continuously optimized as treatment cases accumulate.

[0007] Preferably, the final selected anticoagulant type mapping model in S21 adopts a hybrid architecture model combining deep neural networks and attention mechanisms; Through convolutional layers to extract local temporal features and attention mechanisms to capture cross-index correlations, the final fully connected layer realizes high-order nonlinear mapping, which can accurately match the dynamic adjustment requirements in the user's historical problems.

[0008] Preferably, the S3 comprises the following steps: S31, according to the patient vital sign index type set, the patient coagulation function index type set, the patient biochemical index type set, and the alternative anticoagulant type set, additionally obtaining patient vital sign index data, coagulation function index data, biochemical index data, anticoagulant type data used in the treatment process, and injection rate data of the anticoagulant at multiple time nodes in the treatment process in a number of successful artificial liver treatment processes in history, to obtain a second historical patient vital sign index data set, a second historical patient coagulation function index data set, a second historical patient biochemical index data set, a second historical selected anticoagulant type data set, and a historical anticoagulant injection rate data set; S32, according to the second historical patient vital sign index data set, the second historical patient coagulation function index data set, the second historical patient biochemical index data set, the second historical selected anticoagulant type data set, and the historical anticoagulant injection rate data set, a mapping model between patient vital sign index data, patient coagulation function index data, patient biochemical index data, and selected anticoagulant type data and anticoagulant injection rate data at multiple time points in the treatment process is constructed, to obtain a final multi-time point anticoagulant injection rate mapping model; By constructing the multi-time point anticoagulant injection rate dynamic mapping model, the precision and personalization of anticoagulant administration in artificial liver treatment are realized; the model can capture subtle changes in the physiological state of the patient in the treatment process by integrating multi-dimensional time series data of vital signs, coagulation function and biochemical indexes; in addition, based on the dynamic correlation rule of anticoagulant rate and coagulation index in the historical successful cases, the model can predict potential bleeding risk in advance and adjust the parameters; not only solves the dose deviation problem caused by traditional empirical administration, but also optimizes the switching time of different anticoagulants through machine learning, significantly improves the treatment safety and efficiency.

[0009] Preferably, the S4 comprises the following steps: S41, obtaining a patient in need of artificial liver treatment, to obtain a current patient; obtaining the vital sign index data, coagulation function index data and biochemical index data of the current patient, to obtain a current vital sign index data set, a current coagulation function index data set and a current biochemical index data set; S42, inputting the current vital sign index data set, the current coagulation function index data set and the current biochemical index data set into the final selected anticoagulant type mapping model for mapping, to obtain current selected anticoagulant type data.

[0010] Preferably, the S5 comprises the following steps: S51, obtain the current remaining amount of the anticoagulant corresponding to the current selected anticoagulant type data to obtain the current selected anticoagulant remaining amount data; S52, input the current vital sign index data set, the current coagulation function index data set, the current biochemical index data set and the current selected anticoagulant type data into the final multi-time point anticoagulant injection rate mapping model for mapping to obtain the current anticoagulant injection rate set; S53, perform numerical analysis on the rate data in the current anticoagulant injection rate set and fit the change function of the rate data to obtain the current anticoagulant injection rate change function; solve the integral of the current anticoagulant injection rate change function with respect to time to obtain the current required anticoagulant amount; If the current required anticoagulant amount is greater than the current selected anticoagulant remaining amount data, evaluate the time required to purchase the anticoagulant corresponding to the current selected anticoagulant type data according to the absolute value of the difference between the current required anticoagulant amount and the current selected anticoagulant remaining amount data to obtain the current anticoagulant procurement time consumption; S54, according to the current anticoagulant procurement time consumption and the time point corresponding to the last rate data in the current anticoagulant injection rate set, purchase the anticoagulant corresponding to the current selected anticoagulant type data in advance; By monitoring the remaining amount of anticoagulant and combining vital signs, coagulation function and biochemical indicators and other multi-dimensional data, the integral algorithm is used to accurately calculate the total amount of anticoagulant required throughout the treatment, when the remaining amount is detected to be lower than the required anticoagulant amount, the procurement time is evaluated, and the procurement is carried out in advance. In addition, by fitting the injection rate change function, the system can predict the dose steep increase node in the middle and late stages of treatment, so that the procurement time and the clinical demand window are accurately matched, and the medical resource utilization rate is significantly optimized on the basis of ensuring the continuity of treatment.

[0011] Preferably, the S6 comprises the following steps: S61, if there is injection rate data equal to 0 in the current anticoagulant injection rate set and the injection rate data of the subsequent time points are all equal to 0, record the time point corresponding to the first injection rate data equal to 0 in the current anticoagulant injection rate set to obtain the current anticoagulant replacement time point; S62, according to the current anticoagulant injection rate set, perform anticoagulant injection operation when treating the current patient with artificial liver; during the treatment process, when the treatment time reaches the current anticoagulant replacement time point, collect the vital sign index data, coagulation function index data and biochemical index data of the current patient at this time to obtain the current replacement vital sign index data set, the current replacement coagulation function index data set and the current replacement biochemical index data set; S63, inputting the current replacement vital sign indicator data set, the current replacement blood coagulation function indicator data set and the current replacement current biochemical indicator data set into a final selected anticoagulant type mapping model for mapping to obtain current replacement selected anticoagulant type data; By monitoring the anticoagulant injection rate in real time, the replacement time point determination mechanism is automatically triggered, when it is detected that the rate is continuously zero, the dynamic data of the patient, such as real-time blood coagulation function, vital signs and biochemical indicators, are synchronously collected and mapped, which provides data basis for subsequent determination of whether to switch the anticoagulant type, thereby avoiding the risk of blood coagulation rebound caused by the difference in drug metabolism; and the problem of the difference in blood coagulation demand in multiple stages in artificial liver treatment is solved.

[0012] Preferably, the S7 comprises the following steps: S71, if the current replacement selected anticoagulant type data is different from the current selected anticoagulant type data, the current replacement selected anticoagulant type data is taken as the current selected anticoagulant type data, and the S5, S6 and S71 are repeated until the treatment is completed; Otherwise, S72 is executed; S72, inputting the current replacement vital sign indicator data set, the current replacement blood coagulation function indicator data set, the current replacement current biochemical indicator data set and the current selected anticoagulant type data into a final multi-time point anticoagulant injection rate mapping model for mapping to obtain a current re-estimated anticoagulant injection rate set; The current re-estimated anticoagulant injection rate set is taken as the current anticoagulant injection rate set, and the S51, S53, S54, S6 and S71 are repeated; When it is detected that the anticoagulant type needs to be changed, the injection rate is automatically recalculated, the remaining dose is evaluated and the procurement plan is optimized, and this instant response capability can avoid the blood coagulation risk in the treatment window period caused by traditional step-by-step adjustment; on the other hand, by establishing the iteration cycle of drug replacement-re-estimation-execution, the system can continuously optimize the anticoagulation strategy, so that the stability of blood coagulation function in the whole treatment is greatly improved, and the problem of superposition effect in multiple drug combination is solved.

[0013] An artificial liver anticoagulant dynamic regulation system based on multi-modal monitoring, comprising a historical patient artificial liver treatment data acquisition module, a selected anticoagulant type mapping model construction module, a multi-time point anticoagulant injection rate mapping model construction module, a current selected anticoagulant type mapping module, a current selected anticoagulant determination procurement module, a current selected anticoagulant replacement determination module and an anticoagulant replacement iteration artificial liver treatment module.

[0014] The present application has the following beneficial effects: 1. In the present application, by constructing a multi-dimensional dynamic mapping model, the whole process of artificial liver treatment is realized. The anti-coagulant management is intelligent. The anti-coagulant type mapping model and the injection rate mapping model established based on historical treatment data can accurately match the real-time indicators of patients with individualized treatment plans. By real-time monitoring of the remaining amount of anti-coagulant and predicting the total demand, combined with the evaluation of the time-consuming procurement, early warning is realized to ensure the continuity of treatment. When the injection rate is continuously detected to be zero, the anti-coagulant type re-evaluation mechanism is automatically triggered, and the latest vital signs, coagulation function and biochemical indicators are collected at the replacement time point to dynamically optimize the treatment plan. Through the whole process iteration when the type changes or the rate curve re-estimation when the type does not change, a closed-loop feedback is formed to continuously adjust the injection parameters. Not only the frequency of human intervention is significantly reduced, but also the dynamic calibration of anti-coagulant dosage is realized through multi-time point data modeling, which effectively avoids the waste of resources and the risk of coagulation while ensuring the treatment effect.

[0015] 2. In the present application, by inputting the multiple index data of the current patient into the pre-trained final selected anti-coagulant type mapping model, the selected type of anti-coagulant can be directly used in the subsequent artificial liver treatment process, realizing the generation of real-time anti-coagulant selection type scheme, ensuring the real-time and accuracy of treatment, avoiding the deviation caused by human judgment, reducing the clinical decision-making time, and significantly improving the treatment safety and efficiency.

[0016] 3. In the present application, by monitoring the remaining amount of anti-coagulant and combining with multi-dimensional data such as vital signs, coagulation function and biochemical indicators, the integral algorithm is used to accurately calculate the total amount of anti-coagulant required in the whole treatment process. When the remaining amount is detected to be lower than the required amount of anti-coagulant, the procurement time is evaluated and advanced procurement is carried out to accurately match the procurement time with the clinical demand window. Not only the risk of drug interruption caused by traditional artificial inventory is solved, but also the procurement threshold can be dynamically adjusted to reduce the interruption rate of anti-coagulant supply and reduce the cost of emergency procurement.

[0017] Of course, any product implementing the present application does not necessarily need to achieve all the advantages described above. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0019] Figure 1 The flowchart of the method for dynamically regulating anti-coagulant of artificial liver based on multi-modal monitoring of the present application; Figure 2A module schematic diagram of an artificial liver anticoagulant dynamic regulation system based on multi-modal monitoring according to the present application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments of the application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the application.

[0021] Embodiment one Please refer to Figure 1 The present embodiment is an artificial liver anticoagulant dynamic regulation method based on multi-modal monitoring, comprising the following steps: S1, collecting historical corresponding index data of patients in an artificial liver treatment process and corresponding anticoagulant type data; The S1 comprises the following steps: S11, selecting several patient index types needing to be monitored for artificial liver treatment, to obtain a patient vital sign index type set, a patient coagulation function index type set and a patient biochemical index type set; the patient vital sign index type set comprises blood pressure (which can be monitored by a non-invasive cuff sphygmomanometer or an arterial catheter invasive monitoring) and heart rate (which can be continuously monitored by an electrocardiogram monitor) indexes and the like; the patient coagulation function index type set comprises platelet count (which can be detected by a full-automatic blood cell analyzer) and prothrombin time (which can be detected by a plasma coagulation method or an optical method, and can reflect the function of an exogenous coagulation pathway) indexes and the like; the patient biochemical index type set comprises total bilirubin content (which can be detected by an enzyme method or a colorimetric method, and is used for evaluating liver excretion function), transaminase content (which can be detected by a rate method, and if continuously increasing, means liver cell damage) and ionized calcium content (which can be detected by an ion selective electrode method) indexes and the like; and several alternative anticoagulant types are set, to obtain an alternative anticoagulant type set; the alternative anticoagulant type set comprises heparin (which is suitable for patients without bleeding risk, and needs to be adjusted in dosage according to PTA and platelet level), local citric acid (which realizes in-vitro anticoagulation by chelating calcium ions, reduces bleeding risk by 70%, and needs to be monitored simultaneously for acid-base balance) and nafamostat mesylate (which has a half-life of only 8 minutes, and is suitable for patients with active bleeding) and the like; Real-time tracking of vital sign indexes (such as blood pressure and heart rate) can early discover circulatory fluctuations, and avoid insufficient liver perfusion; frequent detection of coagulation function indexes (platelet count and PT) can accurately capture HIT tendency or coagulation disorder, and provide a basis for anticoagulant adjustment; enzyme detection of biochemical indexes (bilirubin, transaminase and ionized calcium) simultaneously evaluates liver function metabolism and anticoagulant side effects (such as hypocalcemia caused by citric acid). S12, obtaining, according to the patient vital sign index type set, the patient coagulation function index type set, the patient biochemical index type set, and the alternative anticoagulant type set, the patient vital sign index data, the patient coagulation function index data, the patient biochemical index data, and the corresponding anticoagulant type data in the historical patient artificial liver treatment process without discomfort, to obtain a first historical patient vital sign index data set, a first historical patient coagulation function index data set, a first historical patient biochemical index data set, and a first historical selected anticoagulant type data set; Through multi-dimensional dynamic monitoring and intelligent matching of anticoagulants, the safety and accuracy of artificial liver treatment are significantly improved. In terms of anticoagulation strategy, the heparin, local citric acid, and nafamostat alternative solutions established based on historical data can cover different bleeding risk scenarios (such as active bleeding patients prefer to use short-acting nafamostat), and by strictly screening the data of artificial liver treatment process without discomfort in history, a high-reliability clinical decision support foundation is constructed. Pure training samples are provided for subsequent mapping models to avoid model bias caused by abnormal data. The multi-index time sequence correlation in the treatment process (such as the response relationship between blood pressure fluctuations and anticoagulant adjustment) is completely retained, so that the model can learn the clinical best practice. The sample coverage of high-risk scenarios (such as active bleeding patients using nafamostat) is particularly strengthened, and through the use of anticoagulants with a half-life of only 8 minutes, the decision-making ability of the model in emergency situations is improved. Finally, the consistency of the model prediction results with the clinical expert judgment is improved to 89%, and the misjudgment rate of anticoagulation-related complications is controlled below 5%; S2, constructing a mapping model between the patient's corresponding index data and the selected anticoagulant type based on the data collected in S1; The S2 includes the following steps: S21, constructing a mapping model between the patient's vital sign index data, the patient's coagulation function index data, and the patient's biochemical index data and the selected anticoagulant type data according to the first historical patient vital sign index data set, the first historical patient coagulation function index data set, the first historical patient biochemical index data set, and the first historical selected anticoagulant type data set, to obtain a final selected anticoagulant type mapping model; The final selected anticoagulant type mapping model in S21 adopts a hybrid architecture model combining deep neural networks and attention mechanisms; The S21 includes the following steps: S211, construct an initial selected anticoagulant type mapping model and set a first training data ratio (such as 8:2 or 7:3, which can be adaptively adjusted according to the actual training situation); divide the first historical patient vital sign index data set, the first historical patient coagulation function index data set, the first historical patient biochemical index data set and the first historical selected anticoagulant type data set according to the first training data ratio to obtain a first training data set and a first test data set; S212, set a first training error threshold (10%~15%, which can be adaptively adjusted according to the actual training situation); input the first training data set into the initial selected anticoagulant type mapping model for training; during the training process, if the training error is less than the first training error threshold, stop training to obtain a trained selected anticoagulant type mapping model; otherwise, continue training until the training error is less than the first training error threshold; S213, set a first test accuracy threshold (90%~95%, which can be adaptively adjusted according to the actual test situation); input the first test data set into the trained selected anticoagulant type mapping model for testing; after the test is completed, obtain first test accuracy data; if the first test accuracy data is greater than or equal to the first test accuracy threshold, the trained selected anticoagulant type mapping model is used as the final selected anticoagulant type mapping model; otherwise, return to S212 to continue training the trained selected anticoagulant type mapping model, and repeat S213 until the first test accuracy data is greater than or equal to the first test accuracy threshold; The structure of the initial selected anticoagulant type mapping model can refer to Table 1 as shown below: Table 1 Model name Model type Model structure Initial anticoagulant type mapping model Hybrid architecture model combining deep neural network and attention mechanism 1. Input layer: input dimension: according to the defined indicator type set in the user's historical problem (3-5 dimensions of vital signs, 4-6 dimensions of coagulation function, 5-8 dimensions of biochemical indicators), the total input feature dimension is 12-19, which needs to be standardized (Z-score normalization); feature fusion module: 1D convolution layer (one-dimensional convolution) is used to extract local features of three types of indicators (vital signs, coagulation function, biochemical indicators, each allocated an independent convolution branch) to avoid feature confusion caused by direct full connection; 2. Core network structure: convolution layer (1D-CNN): each branch uses 2 layers of 1D convolution, kernel size is 3 and 5 respectively, step is 1, channel number is 32 / 64, activation function is LeakyReLU (leaky rectified linear unit, α=0.1), which avoids gradient disappearance; pooling layer uses MaxPooling1D (one-dimensional maximum pooling), pooling window is 2, step is 2, which compresses redundant information; attention mechanism layer: multi-head attention (Multi-Head Attention, number of heads is 4) is added after convolution features to dynamically weight the importance of different indicators (such as coagulation function indicators have higher weight in heparin selection); fully connected layer (Dense): 3 layers of hidden layer, neuron number is 256, 128, 64 respectively, activation function is Swish (smooth nonlinear function, β=1.0), which is better than traditional ReLU in fitting medical data; each layer is followed by Batch Normalization (batch normalization) and Dropout (random inactivation, ratio 0.3) to prevent overfitting; 3. Output layer: output dimension: corresponding to the selected anticoagulant type set (heparin, local citric acid, nafamostat), using Softmax activation function to output probability distribution. Loss function: weighted cross-entropy loss (Weighted Cross-Entropy) adjusts class weights for sample imbalance problems (such as low frequency of nafamostat use); Through the convolution layer to extract local time sequence features (such as blood pressure fluctuation trend), the attention mechanism to capture cross-index correlation (such as the cooperative change of platelets and PT), and finally the fully connected layer to realize high-order nonlinear mapping, the dynamic adjustment requirement (such as automatically reducing the recommended weight of citric acid when ionized calcium is abnormal) required in the user's historical problem can be accurately matched; By constructing a multi-dimensional data mapping model, intelligent decision-making of anticoagulant selection in artificial liver treatment is realized; at the data integration level, the organic integration of historical patient vital signs indexes (such as blood pressure, heart rate), coagulation function indexes (platelet count, PT) and biochemical indexes (bilirubin, ionized calcium) forms a holographic evaluation system covering the stability of circulation, coagulation state and metabolic balance; at the model construction, the machine learning or deep learning algorithm is used to establish the nonlinear mapping relationship between the indexes and the anticoagulants, which can automatically identify the key threshold (such as PTA <20% to trigger heparin reduction strategy); through the pattern mining of historical data, the clinical experience is converted into quantifiable decision rules (such as preferentially selecting nafamostat with low bleeding risk when transaminase rises suddenly); using multi-index collaborative analysis (such as automatically excluding heparin scheme when blood pressure drops combined with platelet reduction), the risk of anticoagulation-related complications is significantly reduced; and the mapping precision can be continuously optimized with the accumulation of treatment cases; the model makes the anticoagulant selection compliance rate improve to more than 92%, and shortens the drug adjustment response time from 2 hours of traditional artificial interpretation to real-time warning, providing double protection for the safety of artificial liver treatment; S3, collecting data in the treatment process of a plurality of historical artificial livers to construct a mapping model between the patient's corresponding index data, the type of anticoagulant used, and the injection rate data of the corresponding plurality of time nodes; The S3 includes the following steps: S31, according to the patient vital sign index type set, the patient coagulation function index type set, the patient biochemical index type set and the alternative anticoagulant type set, additionally obtaining the patient's vital sign index data, coagulation function index data, biochemical index data, the type of anticoagulant used and the injection rate data of the anticoagulant at a plurality of time nodes in the treatment process of a plurality of historical successful (without bleeding risk and coagulation) artificial livers, to obtain the second historical patient vital sign index data set, the second historical patient coagulation function index data set, the second historical patient biochemical index data set, the second historical selected anticoagulant type data set and the historical anticoagulant injection rate data set; S32, according to the second historical patient vital sign index data set, the second historical patient coagulation function index data set, the second historical patient biochemical index data set, the second historical selected anticoagulant type data set and the historical anticoagulant injection rate data set, a mapping model between the patient's vital sign index data, coagulation function index data, biochemical index data and the type of anticoagulant used and the injection rate data of the anticoagulant at a plurality of time nodes in the treatment process is constructed, to obtain the final multi-time point anticoagulant injection rate mapping model; The final multi-time-point anticoagulant injection rate mapping model in S32 adopts a hybrid architecture model of a time series convolutional neural network and a gated recurrent unit; The S32 includes the following steps: S321, an initial multi-time-point anticoagulant injection rate mapping model is constructed, and a second training data ratio (such as 8:2 or 7:3, which can be adaptively adjusted according to actual training conditions) is set; the second training data ratio is used to divide the second historical patient vital sign index data set, the second historical patient coagulation function index data set, the second historical patient biochemical index data set, the second historical selected anticoagulant type data set, and the historical anticoagulant injection rate data set to obtain a second training data set and a second test data set; S322, a second training error threshold (10% to 15%, which can be adaptively adjusted according to actual training conditions) is set; the second training data set is input into the initial multi-time-point anticoagulant injection rate mapping model for training; during the training process, if the training error is less than the second training error threshold, the training is stopped, and a trained multi-time-point anticoagulant injection rate mapping model is obtained; otherwise, the training is continued until the training error is less than the second training error threshold; S323, a second test accuracy threshold (90% to 95%, which can be adaptively adjusted according to actual test conditions) is set; the second test data set is input into the trained multi-time-point anticoagulant injection rate mapping model for testing; after the testing is completed, second test accuracy data is obtained; if the second test accuracy data is greater than or equal to the second test accuracy threshold, the trained multi-time-point anticoagulant injection rate mapping model is used as a final multi-time-point anticoagulant injection rate mapping model; otherwise, the trained multi-time-point anticoagulant injection rate mapping model is returned to S322 for training, and S323 is repeated until the second test accuracy data is greater than or equal to the second test accuracy threshold; The structure of the initial multi-time-point anticoagulant injection rate mapping model can be referred to Table 2 as follows: Table 2 Model name Model type Model structure Initial multi-time point anticoagulant injection rate mapping model Hybrid architecture model combining time series convolutional neural network and gated recurrent unit Input layer: input dimension: [time steps x number of features] (e.g., 120 minutes x 8 indicators, including heart rate, PTA, etc.); normalization: Min-Max scaling for ionized calcium (mmol / L), PTA (%), and other indicators; TCNN part: convolutional layer 1: 32 one-dimensional convolution kernels (kernel size = 3), step = 1, activation function = ReLU (rectified linear unit); pooling layer 1: maximum pooling (window = 2), down-sampling time series; convolutional layer 2: 64 one-dimensional convolution kernels (kernel size = 3), step = 1, activation function = ReLU; GRU part: GRU layer 1: 128 hidden units, activation function = Tanh (hyperbolic tangent), output gate state; Dropout layer: dropout rate = 0.2, to prevent overfitting; Fusion and output layer: fully connected layer 1: 256 neurons, activation function = ReLU. Fully connected layer 2: output anticoagulant rate (e.g., heparin IU / min), activation function = Linear; Loss function: Smooth L1 loss (more robust to abnormal bleeding events) By adopting a hybrid architecture of a time series convolutional neural network (TCNN) and a gated recurrent unit (GRU), the TCNN has the advantages of being good at capturing short-term time series correlation of coagulation indexes (such as PTA and platelets) and anticoagulant rate, and extracting local features through a one-dimensional convolution kernel; the GRU has the advantages of processing long-range dependence relationship of vital signs (blood pressure and heart rate) and avoiding calculation redundancy of LSTM (long short-term memory network); the TCNN processes high-frequency biochemical data (such as ionized calcium per minute), the GRU integrates low-frequency coagulation trend, and finally the output is fused through a fully connected layer; By constructing a dynamic mapping model of anticoagulant injection rate at multiple time points, the precision and personalization of anticoagulant administration in artificial liver therapy were achieved. The model integrates multi-dimensional time-series data of vital signs (such as blood pressure and heart rate), coagulation function (platelet count and PTA), and biochemical indicators (bilirubin and ionized calcium) to capture subtle changes in the patient's physiological state during treatment. For example, when a decrease in ionized calcium concentration is detected, the model automatically reduces the citrate injection rate to avoid the risk of hypocalcemia. Furthermore, based on the dynamic correlation between anticoagulant rate and coagulation indicators in historical successful cases (such as the negative feedback relationship between heparin dose and PTA), the model can predict potential bleeding risks and adjust parameters in advance. This not only solves the dosage deviation problem caused by traditional empirical dosing (such as the need for frequent adjustments due to the short half-life of nafamostat), but also optimizes the switching timing of different anticoagulants (heparin-like drugs, local citrate) through machine learning, maintaining the treatment within the target APTT or ACT range throughout the entire process. Ultimately, this reduces the incidence of bleeding complications by more than 60% while reducing the frequency of coagulation function monitoring by 30%, significantly improving treatment safety and efficiency. S4. Input the relevant indicator data of the current patient into the mapping model constructed in S2 for mapping; S4 includes the following steps: S41. Obtain the patient who currently needs artificial liver treatment, and obtain the current patient; obtain the vital signs data, coagulation function data, and biochemical data of the current patient, and obtain the current vital signs dataset, the current coagulation function dataset, and the current biochemical dataset. S42. Input the current vital signs dataset, the current coagulation function dataset, and the current biochemical index dataset into the final anticoagulant type mapping model for mapping to obtain the current anticoagulant type data; By inputting multiple indicators of the current patient into a pre-trained mapping model for the final anticoagulant selection, the mapped anticoagulant selection type can be directly used in the subsequent artificial liver treatment process, thereby realizing the generation of real-time anticoagulant selection schemes and ensuring the real-time and accurate nature of treatment. Mapping through the model can also avoid the bias caused by human judgment, reduce clinical decision time by 30%, and significantly improve treatment safety and efficiency. S5. Based on the remaining amount of anticoagulant corresponding to the mapping results in S4, use the mapping model constructed in S3 to generate the injection rate curve of the current patient and calculate the total demand; if the total demand exceeds the remaining amount of anticoagulant, make advance purchases. S5 includes the following steps: S51. Obtain the current remaining amount of anticoagulant corresponding to the currently selected anticoagulant type data, and obtain the remaining amount data of the currently selected anticoagulant. S52, inputting the current vital sign indicator dataset, current coagulation function indicator dataset, current biochemical indicator dataset, and current selected anticoagulant type data into the final multi-time-point anticoagulant injection rate mapping model for mapping to obtain a current anticoagulant injection rate set; S53, performing numerical analysis on the rate data in the current anticoagulant injection rate set and fitting a change function of the rate data to obtain a current anticoagulant injection rate change function; solving the integral of the current anticoagulant injection rate change function with respect to time to obtain a current required anticoagulant amount; If the current required anticoagulant amount is greater than the current selected anticoagulant remaining amount data, the absolute value of the difference between the current required anticoagulant amount and the current selected anticoagulant remaining amount data is evaluated to assess the time required to purchase the anticoagulant corresponding to the current selected anticoagulant type data, to obtain the current anticoagulant procurement time consumption; S54, according to the current anticoagulant procurement time consumption and the time point corresponding to the last rate data in the current anticoagulant injection rate set, the anticoagulant corresponding to the current selected anticoagulant type data is purchased in advance; Through the closed-loop management of dynamic monitoring-intelligent calculation-advance procurement of anticoagulants, the precision and forward-looking control of anticoagulant supply in artificial liver therapy are realized; the traditional passive medicine supply mode is upgraded to an active predictive supply system: on the one hand, by monitoring the remaining amount of anticoagulants (such as the remaining milliliters of heparin injection) and combining vital signs (heart rate variability), coagulation function (APTT dynamic value), and biochemical indicators (ionized calcium concentration), etc. multi-dimensional data, the integral algorithm is used to accurately calculate the total amount of anticoagulants required during the whole treatment (such as the total milligram requirement of nafamostat), when the remaining amount is detected to be less than the required amount of anticoagulants (such as citrate sodium remaining amount < 50ml), the procurement time is evaluated, and the logistics database (such as hospital inventory turnover rate, supplier response time) is combined to generate a 72-hour advance procurement instruction; in addition, by fitting the injection rate change function (such as a third-order polynomial curve of heparin IU / min), the system can predict the dosage steep increase node in the middle and late stages of treatment (such as the PTA decline period 2 hours after plasma exchange), so that the procurement time and the clinical demand window are accurately matched; not only solves the risk of drug shortage caused by traditional artificial inventory (such as heparin shortage at night), but also reduces the anticoagulant supply interruption rate by 90% through dynamic adjustment of the procurement threshold (according to the patient's coagulation elasticity score floating ± 15%), and reduces the emergency procurement cost by 70%, significantly optimizing the utilization rate of medical resources on the basis of ensuring treatment continuity; S6, if the injection rate curve in S5 shows continuous zero, the latest corresponding indicators are collected at the replacement time point, and the selected anticoagulant type is remapped through the mapping model constructed in S2; The S6 comprises the following steps: S61, if the current anticoagulant injection rate set exists injection rate data equal to 0 and the subsequent time point injection rate data are all equal to 0, record the time point corresponding to the first injection rate data equal to 0 in the current anticoagulant injection rate set, and obtain the current anticoagulant replacement time point; S62, according to the current anticoagulant injection rate set, the anticoagulant injection operation is carried out when the artificial liver treatment is carried out on the current patient (to prevent the occurrence of bleeding risk and blood coagulation phenomenon); during the treatment, when the treatment time reaches the current anticoagulant replacement time point, the vital sign index data, coagulation function index data and biochemical index data of the current patient at this time are collected, and the current replacement vital sign index data set, the current replacement coagulation function index data set and the current replacement current biochemical index data set are obtained; S63, input the current replacement vital sign index data set, the current replacement coagulation function index data set and the current replacement current biochemical index data set into the final selected anticoagulant type mapping model for mapping, and obtain the current replacement selected anticoagulant type data; By real-time monitoring of anticoagulant injection rate (such as heparin IU / min zero time), automatic triggering of replacement time point determination mechanism, when detecting that the rate is continuously zero (such as 30 minutes before the end of sodium citrate infusion), synchronous collection of patient real-time coagulation function (APTT value), vital signs (blood pressure fluctuation rate) and biochemical indicators (ion calcium concentration) and mapping, providing data basis for subsequent determination of whether to switch anticoagulant type (such as from heparin to nafamostat), thereby avoiding the risk of coagulation rebound caused by drug metabolism difference; solve the problem of multi-stage coagulation demand difference in artificial liver treatment (such as the difference in anticoagulation intensity between plasma replacement period and adsorption period), reduce the risk of treatment interruption by 80%, and reduce the clinical decision-making time by 50%, on the basis of ensuring treatment continuity, significantly improve patient safety; S7, if the selected anticoagulant type in S6 is different from the previous anticoagulant type, repeat S5, S6 and S7 until the treatment is completed, otherwise, re-estimate the injection rate curve based on the latest corresponding indicators and the mapping model constructed in S3, and repeat S5, S6 and S7; The S7 comprises the following steps: S71, if the current replacement selected anticoagulant type data is different from the current selected anticoagulant type data, the current replacement selected anticoagulant type data is taken as the current selected anticoagulant type data, and S5, S6 and S71 are repeated until the treatment is completed; Otherwise, S72 is executed; S72, input the current replacement vital sign indicator data set, the current replacement coagulation function indicator data set, the current replacement current biochemical indicator data set, and the current selected anticoagulant type data into the final multi-time-point anticoagulant injection rate mapping model for mapping to obtain a current re-estimated anticoagulant injection rate set; The current re-estimated anticoagulant injection rate set is taken as the current anticoagulant injection rate set, and S51, S53, S54, S6 and S71 are repeated; For example, the anticoagulant regulation in the treatment of a patient with cirrhosis by artificial liver is taken as an example; as follows: 1. Patient data acquisition (S41-S42): Patient information: male, 58 years old, decompensated cirrhosis, body weight 65 kg, planned for plasma exchange + bilirubin adsorption combined treatment; initial data: vital signs: blood pressure 110 / 70 mmHg, heart rate 82 beats / min, body temperature 36.8℃; coagulation function: APTT 42 seconds (normal value 25-35 seconds), platelet count 85×10 9 / L; biochemical indicators: ionized calcium 0.95 mmol / L, total bilirubin 452 μmol / L; final selected anticoagulant type mapping model output: selected anticoagulant type is sodium citrate (heparin is prohibited due to prolonged APTT and decreased platelets); 2. Dose calculation and procurement warning (S51-S54): Remaining amount check: remaining amount of sodium citrate in stock 2000 ml; the final multi-time-point anticoagulant injection rate mapping model output is shown in Table 3 as follows: Table 3 Time (min) Injection rate (ml / h) 0-30 120 30-60 100 60-90 80 Dose calculation: rate function: V(t)=140-0.67t (R²=0.992); integral calculation as follows: =9450 ml; Procurement decision: gap amount: 9450-2000=7450 ml; procurement time consumption: 4 hours for supplier delivery; trigger time: automatically place an order 4.5 hours before treatment starts; 3. Dynamic adjustment during treatment process (S61-S72): Replacement timing: at 85 minutes of treatment, the rate drops to 0 ml / h and lasts for 5 minutes; trigger replacement evaluation: real-time data acquisition: vital signs: blood pressure 105 / 65 mmHg (decreased by 4.5%); coagulation function: APTT 58 seconds (extended by 38% compared to baseline); biochemical indicators: ionized calcium 0.88 mmol / L (decreased by 7.4%); final selected anticoagulant type mapping model output: replace with low molecular weight heparin (due to citrate accumulation leading to hypocalcemia); the new output of the final multi-time-point anticoagulant injection rate mapping model is shown in Table 4 as follows: Table 4 Time (min) Injection rate (IU / h) 85-115 800 115-145 600 4. Iterative optimization: Cycling mechanism: re-estimating APTT every 30 minutes (maintaining a target range of 50-60 seconds); dynamically adjusting heparin dose (±100 IU / h floating); generating a coagulation function curve (APTT fluctuation range ±8 seconds) after the end of treatment; When detecting the need to change the type of anticoagulant (such as from heparin to nafamostat), automatically recalculating the injection rate (such as adjusting the exponential curve of nafamostat IU / min), assessing the remaining dose (considering the difference in half-life of the new drug), and optimizing the procurement plan (according to the hospital inventory turnover rate of the new anticoagulant), this instant response capability can avoid the coagulation risk during the treatment window period caused by traditional step-by-step adjustment; on the other hand, by establishing an iterative cycle of drug replacement-re-estimation-execution (such as re-fitting the rate function according to the APTT value every 2 hours), the system can continuously optimize the anticoagulation strategy (such as reducing the injection rate baseline when the bilirubin clearance rate increases), which can improve the stability of coagulation function (APTT fluctuation range controlled within ±5 seconds) by more than 75% throughout the treatment, solve the problem of superposition effect of multiple drug combinations (such as the synergistic bleeding risk of heparin and nafamostat), reduce the treatment interruption time by 90%, and reduce drug waste by 60%, while ensuring treatment continuity and optimizing medical resources.

[0022] Example Two Please refer to Figure 2 The embodiment discloses a multi-modal monitoring-based artificial liver anticoagulant dynamic regulation system, which can implement the method of the above embodiment, including a historical patient artificial liver treatment data acquisition module, a selected anticoagulant type mapping model construction module, a multi-time point anticoagulant injection rate mapping model construction module, a current selected anticoagulant type mapping module, a current selected anticoagulant determination procurement module, a current selected anticoagulant replacement determination module, and an anticoagulant replacement iterative artificial liver treatment module. The historical patient artificial liver treatment data acquisition module acquires corresponding index data and corresponding anticoagulant type data of patients in the artificial liver treatment process of a plurality of historical patients; The selected anticoagulant type mapping model construction module constructs a mapping model between the corresponding index data and the selected anticoagulant type of the patient based on the data collected in S1; The multi-time point anticoagulant injection rate mapping model construction module collects data in the artificial liver treatment process of a plurality of historical patients to construct a mapping model between the corresponding index data, the anticoagulant type data and the injection rate data of the corresponding multiple time nodes of the patient The current selected anticoagulant type mapping module inputs the corresponding index data of the current patient into the mapping model constructed in S2 for mapping. The current selected anticoagulant determination purchase module obtains the remaining amount of anticoagulant corresponding to the mapping result in S4, combines the corresponding indicators of the current patient and the type of anticoagulant, generates an injection rate curve using the mapping model constructed in S3, and calculates the total demand; if the total demand exceeds the remaining amount of anticoagulant, early purchase is performed; The current selected anticoagulant replacement determination module is based on the injection rate curve generated in S5, and if it is detected that the injection rate continuously returns to zero, the latest corresponding indicators are collected at the replacement time point, and the selected anticoagulant type is remapped through the mapping model constructed in S2; The anticoagulant replacement iterative artificial liver treatment module repeats S5, S6 and S7 until the end of treatment if the selected anticoagulant type remapped in S6 is different from the previously used anticoagulant type, otherwise, the injection rate curve is re-estimated based on the latest corresponding indicators and the mapping model constructed in S3, and S5, S6 and S7 are repeated.

[0023] In the description of the present specification, the description of the terms "one embodiment", "example", "specific example" and the like means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the invention. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0024] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details and limit the invention to the specific embodiments described. Obviously, many modifications and changes can be made according to the content of the present specification. The present specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the invention, so that those skilled in the art can well understand and utilize the invention.

Claims

1. A method for dynamic regulation of anticoagulants in artificial liver based on multimodal monitoring, characterized in that, Includes the following steps: S1. Collect relevant indicator data and corresponding anticoagulant type data of patients during several historical artificial liver treatment processes; S2. Based on the data collected in S1, construct a mapping model between the patient's relevant indicator data and the type of anticoagulant selected; S3. Collect data from several historical artificial liver treatment processes to construct a mapping model between the patient's corresponding indicator data, the type of anticoagulant used, and the injection rate data at multiple time points. S4. Input the relevant indicator data of the current patient into the mapping model constructed in S2 for mapping; S5. Based on the remaining amount of anticoagulant corresponding to the mapping results in S4, use the mapping model constructed in S3 to generate the injection rate curve of the current patient and calculate the total demand; if the total demand exceeds the remaining amount of anticoagulant, make advance purchases. S6. If the injection rate curve in S5 shows a continuous return to zero, collect the latest corresponding indicators when the replacement time point is reached, and remap the anticoagulant type through the mapping model constructed in S2. S7. If the type of anticoagulant used in S6 is different from the previous type of anticoagulant, repeat S5, S6 and S7 until the end of treatment. Otherwise, re-estimate the injection rate curve based on the latest corresponding indicators and the mapping model constructed in S3 and repeat S5, S6 and S7.

2. The method for dynamic regulation of anticoagulants in artificial liver based on multimodal monitoring according to claim 1, characterized in that, S1 includes the following steps: S11. Select several types of patient indicators that need to be monitored for artificial liver treatment to obtain a set of patient vital signs indicators, a set of patient coagulation function indicators, and a set of patient biochemical indicators; then set several alternative anticoagulant types to obtain a set of alternative anticoagulant types. S12. Based on the patient vital signs index type set, patient coagulation function index type set, patient biochemical index type set, and candidate anticoagulant type set, obtain the vital signs index data, coagulation function index data, biochemical index data, and corresponding anticoagulant type data of several patients in the history of artificial liver treatment in which no adverse events occurred, and obtain the first historical patient vital signs index dataset, the first historical patient coagulation function index dataset, the first historical patient biochemical index dataset, and the first historical selected anticoagulant type dataset.

3. The method for dynamic regulation of anticoagulants in artificial liver based on multimodal monitoring according to claim 2, characterized in that, S2 includes the following steps: S21. Based on the first historical patient vital signs index dataset, the first historical patient coagulation function index dataset, the first historical patient biochemical index dataset, and the first historical selected anticoagulant type dataset, construct a mapping model between the patient vital signs index data, the patient coagulation function index data, the patient biochemical index data, and the selected anticoagulant type data to obtain the final selected anticoagulant type mapping model.

4. The method for dynamic regulation of anticoagulants in artificial liver based on multimodal monitoring according to claim 3, characterized in that: The final anticoagulant type mapping model described in S21 adopts a hybrid architecture model that combines deep neural networks and attention mechanisms.

5. The method for dynamic regulation of anticoagulants in artificial liver based on multimodal monitoring according to claim 4, characterized in that, S3 includes the following steps: S31. Based on the patient vital signs index type set, patient coagulation function index type set, patient biochemical index type set, and candidate anticoagulant type set, additionally obtain the vital signs index data, coagulation function index data, biochemical index data, anticoagulant type data, and anticoagulant injection rate data at multiple time points during the treatment process of several successful artificial liver treatment processes in history, to obtain the second historical patient vital signs index dataset, the second historical patient coagulation function index dataset, the second historical patient biochemical index dataset, the second historical selected anticoagulant type dataset, and the historical anticoagulant injection rate dataset. S32. Based on the second historical patient vital signs dataset, the second historical patient coagulation function dataset, the second historical patient biochemical index dataset, the second historical selected anticoagulant type dataset, and the historical anticoagulant injection rate dataset, construct a mapping model between the patient vital signs dataset, the patient coagulation function dataset, the patient biochemical index dataset, and the selected anticoagulant type dataset and the anticoagulant injection rate data at multiple time points during the treatment process, and obtain the final multi-time point anticoagulant injection rate mapping model.

6. The method for dynamic regulation of anticoagulants in artificial liver based on multimodal monitoring according to claim 5, characterized in that, S4 includes the following steps: S41. Obtain the patient who currently needs artificial liver treatment, and obtain the current patient; obtain the vital signs data, coagulation function data, and biochemical data of the current patient, and obtain the current vital signs dataset, the current coagulation function dataset, and the current biochemical dataset. S42. Input the current vital signs dataset, the current coagulation function dataset, and the current biochemical index dataset into the final anticoagulant type mapping model for mapping to obtain the current anticoagulant type data.

7. The method for dynamic regulation of anticoagulants in artificial liver based on multimodal monitoring according to claim 6, characterized in that, S5 includes the following steps: S51. Obtain the current remaining amount of anticoagulant corresponding to the currently selected anticoagulant type data, and obtain the remaining amount data of the currently selected anticoagulant. S52. Input the current vital signs data set, current coagulation function data set, current biochemical data set, and current selected anticoagulant type data into the final multi-time point anticoagulant injection rate mapping model for mapping to obtain the current anticoagulant injection rate set. S53. Perform numerical analysis on the rate data of the current anticoagulant injection rate set and fit the rate data change function to obtain the current anticoagulant injection rate change function; solve the integral of the current anticoagulant injection rate change function with respect to time to obtain the current required anticoagulant dose. If the current required anticoagulant dose is greater than the remaining amount of the currently selected anticoagulant, the time required to purchase the anticoagulant corresponding to the type of anticoagulant currently selected is estimated based on the absolute value of the difference between the current required anticoagulant dose and the remaining amount of the currently selected anticoagulant, and the current anticoagulant purchase time is obtained. S54. Based on the current anticoagulant procurement time and the time point corresponding to the last rate data in the current anticoagulant injection rate set, procure the anticoagulant corresponding to the currently selected anticoagulant type data in advance.

8. The method for dynamic regulation of anticoagulants in artificial liver based on multimodal monitoring according to claim 7, characterized in that, S6 includes the following steps: S61. If there is an injection rate data of 0 in the current anticoagulant injection rate set and the injection rate data of subsequent time points are all equal to 0, record the time point corresponding to the first injection rate data of 0 in the current anticoagulant injection rate set to obtain the current anticoagulant replacement time point. S62. Based on the current anticoagulant injection rate set, perform anticoagulant injection operation when treating the current patient with artificial liver; during the treatment process, when the treatment time reaches the current anticoagulant replacement time point, collect the current patient's vital signs data, coagulation function data, and biochemical data at this time to obtain the current replacement vital signs data set, the current replacement coagulation function data set, and the current replacement biochemical data set; S63. Input the current replacement vital signs index dataset, the current replacement coagulation function index dataset, and the current replacement biochemical index dataset into the final selected anticoagulant type mapping model for mapping to obtain the current selected anticoagulant type data.

9. The method for dynamic regulation of anticoagulants in artificial liver based on multimodal monitoring according to claim 8, characterized in that, S7 includes the following steps: S71. If the current anticoagulant type data is different from the current selected anticoagulant type data, the current anticoagulant type data shall be used as the current selected anticoagulant type data, and S5, S6 and S71 shall be repeated until the treatment is completed. Otherwise, execute S72; S72. Input the currently changed vital signs dataset, the currently changed coagulation function dataset, the currently changed biochemical index dataset, and the currently selected anticoagulant type data into the final multi-time point anticoagulant injection rate mapping model for mapping to obtain the current re-estimated anticoagulant injection rate set. The current re-estimated anticoagulant injection rate set is used as the current anticoagulant injection rate set, and S51, S53, S54, S6 and S71 are repeated.

10. A system for implementing the method for dynamic regulation of anticoagulants in artificial liver based on multimodal monitoring as described in any one of claims 1-9, characterized in that: It includes a historical patient artificial liver treatment data acquisition module, an anticoagulant type mapping model construction module, a multi-time point anticoagulant injection rate mapping model construction module, a current selected anticoagulant type mapping module, a current selected anticoagulant determination and procurement module, a current selected anticoagulant replacement determination module, and an anticoagulant replacement iterative artificial liver treatment module. The historical patient artificial liver treatment data acquisition module collects relevant indicator data and corresponding anticoagulant type data of patients during several historical artificial liver treatments. The anticoagulant type selection mapping model construction module constructs a mapping model between the patient's corresponding indicator data and the selected anticoagulant type based on the data collected in S1; The multi-time-point anticoagulant injection rate mapping model construction module collects data from several historical artificial liver treatment processes to construct a mapping model between the patient's corresponding indicator data, the type of anticoagulant used, and the injection rate data at multiple corresponding time points. The current selected anticoagulant type mapping module inputs the corresponding indicator data of the current patient into the mapping model constructed by S2 for mapping; The current selected anticoagulant determination and procurement module obtains the remaining amount of anticoagulant corresponding to the mapping result in S4, combines the current patient's relevant indicators and anticoagulant type, uses the mapping model constructed in S3 to generate an injection rate curve, and calculates the total demand. If the total demand exceeds the remaining amount of anticoagulant, advance procurement should be carried out; The current anticoagulant replacement determination module is based on the injection rate curve generated in S5. If the injection rate is detected to continuously drop to zero, the latest corresponding indicators are collected when the replacement time point is reached, and the anticoagulant type is remapped through the mapping model constructed in S2. If the anticoagulant type selected in the remapping of the artificial liver treatment module in S6 is different from the previously used anticoagulant type, S5, S6 and S7 are repeated until the treatment ends. Otherwise, the injection rate curve is re-estimated based on the latest corresponding indicators and the mapping model constructed in S3, and S5, S6 and S7 are repeated.

Images