In-vitro pipeline and filter blood coagulation monitoring method and system

Abstract

The invention discloses an in-vitro pipeline and filter blood coagulation monitoring method and system, and belongs to the technical field of intelligent medical instruments, and the in-vitro pipeline and filter blood coagulation monitoring method comprises the following steps: S1, obtaining operation data and operation parameters of a corresponding hemodialysis machine and biochemical indexes of a patient corresponding to the hemodialysis machine based on a pipeline and a filter; optimizing the long-short-term memory network based on a doliolaria group optimization algorithm to obtain a monitoring model; s2, preprocessing the operation data, the operation parameters and the biochemical indexes to obtain a training set, a verification set and a real-time data set; s3, training the monitoring model based on the training set and the verification set to obtain a real-time monitoring model; s4, inputting the real-time data set into a real-time monitoring model to obtain blood coagulation grades of the pipeline and the filter; s5, the corresponding in-vitro pipeline and filter blood coagulation monitoring system executes an early warning action based on the early warning rule and the blood coagulation level; the blood coagulation level of the pipeline and the filter is monitored through the interrelation of the operation data and the operation parameters of the hemodialysis machine and the biochemical indexes of the patient, and the recognition accuracy of the blood coagulation event is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent medical devices, and particularly to an in vitro pipeline and filter blood coagulation monitoring method and system. Background Art

[0002] With the development of artificial intelligence technology, more and more medical devices have been upgraded intelligently to achieve more convenient and accurate identification and monitoring, reducing the workload of staff; during continuous renal replacement therapy, blood clots often occur in the pipeline and filter corresponding to the hemodialysis machine, and in severe cases, it may even cause blockage, resulting in the inability to fully release the power of the hemodialysis machine; based on this, during continuous renal replacement therapy, it is necessary to monitor the pipeline and filter. Currently, most of them are manual monitoring, and the renal replacement therapy process usually needs to be carried out continuously for 24 hours, at least 4 to 5 hours. The time cycle is long, and it is difficult for staff to maintain high concentration of attention for a long time, making it difficult to detect and handle blood clotting events in the early stage and latent period in a timely manner.

[0003] Chinese Patent, Publication No.: CN117274708A, Publication Date: December 22, 2023, discloses a dialysis kettle blood clotting point identification and alarm method and related methods and systems, including: S1, obtaining a plurality of venous kettle images, performing image preprocessing on the venous kettle images to obtain a data set; S2, calculating the blood clotting area of the venous kettle images according to the blood clotting standard, and marking the actual labels on the venous kettle images according to the blood clotting area to obtain an updated data set; S3, constructing a convolutional neural network, setting the objective function for network training, and training the convolutional neural network with the data set to obtain a trained convolutional neural network model; finally, using the trained convolutional neural network model to identify the venous kettle images to be identified, obtaining the identification result, and calculating the ratio of the identification result to the venous kettle images to be identified to obtain the blood clotting grade; this invention reduces the workload of staff through the identification of image data, but the proportion of the latent period and early stage of the blood clotting event in the corresponding image is small and difficult to detect, especially in the latent period, there are even no graphics related to the blood clotting event in the image, resulting in low accuracy of blood clotting event identification. Summary of the Invention

[0004] The object of the present invention is to address the problem that the prior art is limited by images and results in low accuracy in identifying coagulation events. A method and system for monitoring coagulation of an extracorporeal pipeline and filter are proposed. Based on the operation data, operation parameters of the pipeline corresponding hemodialysis machine, and biochemical indexes of the patient corresponding to the hemodialysis machine, a training set, a validation set, and a real-time data set are established. The long short-term memory network is optimized based on the salp swarm optimization algorithm to obtain a monitoring model. The monitoring model is trained based on the training set and the validation set to obtain a real-time monitoring model. The real-time data set is input into the real-time monitoring model to obtain the coagulation level. The corresponding extracorporeal pipeline and filter coagulation monitoring system performs warning actions based on warning rules and the coagulation level. The present invention monitors the coagulation level of the pipeline and filter through the mutual connection of the operation data, operation parameters of the hemodialysis machine, and the biochemical indexes of the patient, and performs corresponding warning actions, significantly improving the accuracy of identifying coagulation events.

[0005] In a first aspect, a technical solution provided in an embodiment of the present invention is a method for monitoring coagulation of an extracorporeal pipeline and filter, including the following steps: S1. Obtain the operation data, operation parameters of the hemodialysis machine corresponding to the pipeline and filter, and the biochemical indexes of the patient corresponding to the hemodialysis machine; optimize the long short-term memory network based on the salp swarm optimization algorithm to obtain a monitoring model; S2. Preprocess the operation data, operation parameters, and biochemical indexes to obtain a training set, a validation set, and a real-time data set; S3. Train the monitoring model based on the training set and the validation set to obtain a real-time monitoring model; S4. Input the real-time data set into the real-time monitoring model to obtain the coagulation level of the pipeline and filter; S5. The corresponding extracorporeal pipeline and filter coagulation monitoring system performs warning actions based on warning rules and the coagulation level.

[0006] In this solution, the pipeline and filter are specifically the pipeline and filter of a hemodialysis machine. Therefore, the operation data, operation parameters of the hemodialysis machine, and the biochemical indexes of the patients corresponding to the hemodialysis machine are all related to the coagulation events of the pipeline and filter. The operation data at least includes transmembrane pressure, arterial pressure, venous pressure, filter pressure, and pre-filter pressure. The operation parameters at least include blood flow rate and ultrafiltration rate. The biochemical indexes at least include platelet count, prothrombin time, and activated partial thromboplastin time. However, there are many types and large volumes of operation data, operation parameters, and biochemical indexes. It is necessary to preprocess the operation data, operation parameters, and biochemical indexes, establish the corresponding relationships of various types of data based on the time scale, and integrate the various data into a whole to obtain a training set, a validation set, and a real-time data set. It should be noted that the training set and the validation set are obtained from historical data, and the real-time data set is obtained from real-time data. Secondly, since the operation data, operation parameters, and biochemical indexes are all continuous data, with continuous data characteristics hidden inside, the long short-term memory network has the ability to extract the long-term dependence relationships in such time series. In order to make the long short-term memory network fit the data in the training set, validation set, and real-time data set, and improve the performance of the long short-term memory network in processing the operation data, operation parameters, and biochemical indexes of the hemodialysis machine, the hyperparameters of the long short-term memory network are optimized based on the salp swarm optimization algorithm to obtain a monitoring model, and the monitoring model is trained based on the training set and the validation set to obtain a real-time monitoring model, effectively improving the ability of the real-time monitoring model to extract the continuous data characteristics of the operation data, operation parameters, and biochemical indexes. Furthermore, the real-time monitoring model predicts the coagulation grade of the corresponding pipeline and filter according to the continuous data characteristics of the data in the real-time data set. The coagulation grade represents the probability of a coagulation event occurring in the pipeline and filter in the future period of time, and potential coagulation events can be accurately detected in the early stage and latent period of the coagulation event. Accordingly, the extracorporeal pipeline and filter coagulation monitoring system performs warning actions based on the warning rules and the coagulation grade, and the corresponding staff can conduct pre-inspections and processing in advance according to the warning actions, effectively ensuring that the power of the hemodialysis machine can be fully released during the working process.

[0007] Preferably, in S1, the specific process of optimizing the long short-term memory network based on the salp swarm optimization algorithm to obtain a monitoring model is as follows: S11. Take the initial hyperparameters of the long short-term memory network as individuals to establish a population, and establish an objective function based on the mean square error of the operation data, the mean square error of the operation parameters, and the mean square error of the biochemical indexes. S12. Calculate the fitness of the individuals based on the objective function, and divide the population into leaders and followers based on the fitness. S13. Update the leaders based on the leader update formula of the salp swarm optimization algorithm, and update the followers based on the follower update formula of the salp swarm optimization algorithm. S14. Determine whether to end the optimization based on the number of updates of the leader and the iteration threshold. If the number of updates is less than the iteration threshold, continue the optimization and execute S12. If the number of updates is greater than or equal to the iteration threshold, end the optimization and execute S15; S15. Sort the fitness values of the individuals, and adjust the hyperparameters of the long short-term memory network based on the individual with the smallest fitness value to obtain the monitoring model.

[0008] Preferably, in S2, the operation data at least includes historical operation data and real-time operation data, the operation parameters at least include historical operation parameters and real-time operation parameters, and the biochemical indicators at least include historical biochemical indicators and real-time biochemical indicators; Among them, the historical operation parameters, historical operation data, and historical biochemical indicators are preprocessed to obtain a training set and a validation set; the real-time operation data, real-time operation parameters, and real-time biochemical indicators are preprocessed to obtain a real-time data set.

[0009] In this solution, since the training set and the validation set are used to train the monitoring model, the training set and the validation set are constructed based on historical operation data, historical operation parameters, and historical biochemical indicators. The training effect of the monitoring model is ensured by the characteristics of the historical data where the occurrence of blood coagulation events is known. At the same time, the data characteristics of historical data and real-time data are the same, ensuring the matching degree between the monitoring model and real-time data. The real-time data set is used to predict the probability of blood coagulation events occurring in the corresponding pipeline and filter, that is, the blood coagulation level. The real-time data set is constructed based on real-time operation data, real-time operation parameters, and real-time biochemical indicators, effectively improving the accuracy of the monitoring model in predicting the blood coagulation time.

[0010] Preferably, in S2, the specific process of preprocessing the operation data, operation parameters, and biochemical indicators to obtain a training set, a validation set, and a real-time data set is as follows: S21. Filter out the outliers in the historical operation data, historical operation parameters, and historical biochemical indicators based on a preset sliding window to obtain a historical outlier-removed data set; Filter out the outliers in the real-time operation data, real-time operation parameters, and real-time biochemical indicators based on a preset sliding window to obtain a real-time outlier-removed data set; S22. Fill in the missing values in the historical outlier-removed data set based on the monitoring model and the missing data filling criterion to obtain a historical complete data set; Fill in the missing values in the real-time outlier-removed data set based on the monitoring model and the missing data filling criterion to obtain a real-time complete data set; S23. Normalize the historical complete data set based on the normalization criterion to obtain a historical standard data set, and normalize the real-time complete data set based on the normalization criterion to obtain a real-time standard data set; Synchronously, based on the salp swarm optimization algorithm, extract the features of the data in the historical complete dataset to obtain historical data features, and based on the salp swarm optimization algorithm, extract the features of the data in the real-time complete dataset to obtain real-time data features; S24. Organize the historical data features and the historical standard dataset to obtain a training set and a validation set, and organize the real-time data features and the real-time standard dataset to obtain a real-time dataset.

[0011] In this solution, due to the large variety, large volume, and high complexity of the collected data, in order to process it efficiently and accurately, based on a sliding window, dynamically intercept part of the data in the operating data, operating parameters, and biochemical indicators, and then calculate the upper and lower limits of the part of the data based on the interquartile range method. Determine the normal data range from the upper and lower limits, mark the data outside the normal data range as outliers and delete them to obtain a dataset without outliers. At this time, there are a large number of missing values in the dataset without outliers. When the number of missing values occupies a certain proportion in the dataset without outliers, it will reduce the accuracy of the overall data and cannot comprehensively describe the data characteristics. Considering that the data is continuous and changes according to a certain law, and the monitoring model can capture the long-term dependencies in the data, then use the monitoring model to interpolate and fill the missing values in the dataset without outliers according to the missing data filling criterion to obtain a complete dataset. At this time, the data in the complete dataset has different units due to different types, that is, there is a dimension problem. In order to eliminate the influence of the dimension on subsequent data processing, use the standard score formula to normalize the data to obtain a standard dataset. It should be noted that the standard score formula may not necessarily match the data in the complete dataset, and it is necessary to use the salp swarm optimization algorithm to optimize the parameters in the standard score formula, that is, the mean and standard deviation. At the same time, in order to extract the potential features of the data in the standard dataset and facilitate the subsequent data processing process, based on the salp swarm optimization algorithm, extract the features of the data in the complete dataset to obtain data features. Finally, organize the data features and the standard dataset to obtain a training set, a validation set, and a real-time dataset; in addition, the data includes real-time data and historical data, and the training set and the validation set are constructed based on historical data, and the real-time dataset is constructed based on real-time data. Using the salp swarm optimization algorithm to extract data features can establish corresponding time tags based on the time series of the data, such as the start point of the training set, the end point of the training set, the start point of the validation set, the end point of the validation set, the start point of the real-time dataset, and the end point of the real-time dataset, which is convenient for the establishment of the training set, the validation set, and the real-time dataset.

[0012] Preferably, in S22, the specific process of filling the missing values in the historical dataset without outliers based on the monitoring model and the missing data filling criterion to obtain a historical complete dataset is as follows: Count the missing values in the historical dataset without outliers, and classify the missing values based on the time scale of the historical dataset without outliers to obtain a classified missing set; Determine whether to fill based on the number of missing values and the missing value threshold in the subset of the classification missing set. If the number of missing values is less than or equal to the missing value threshold, do not fill. If the number of missing values is greater than the missing value threshold, select data from the outlier-removed dataset based on the time scale of the corresponding subset as the input of the monitoring model to obtain the filled subset; Organize the filled subset and the outlier-removed dataset to obtain the historical complete dataset.

[0013] In this solution, each missing value corresponds to a time point, and whether it is operation data, operation parameters, or biochemical indicators, they all change within a certain range according to a fixed rule. Therefore, the occasional missing of a few data points does not affect the judgment of the overall data characteristics. Thus, the missing values with adjacent time points on the time scale are divided into a set to obtain the classification missing set, that is, the missing values in each set are continuous. When the number of missing values is greater than the missing value threshold, it proves that there is no data in the time period corresponding to the corresponding set. Select the adjacent set for the time period to form the subset to be filled. The monitoring model fills the missing values according to the data characteristics in the subset to be filled to obtain the filled subset. Finally, insert the filled subset into the outlier-removed dataset according to the time series to obtain the complete dataset. Secondly, the establishment process of the real-time complete dataset is the same as that of the historical complete dataset.

[0014] Preferably, in S23, the specific process of extracting the features of the data in the historical complete dataset based on the salp swarm optimization algorithm is as follows: Calculate the pressure fluctuation entropy, pressure range, change rate, extreme value frequency, and pressure correlation matrix based on the data in the historical complete dataset, and establish individuals based on the pressure fluctuation entropy, pressure range, change rate, extreme value frequency, and pressure correlation matrix to obtain a population; Set the maximization of information entropy gain as the objective function of the salp swarm optimization algorithm, and input the population into the salp swarm optimization algorithm to obtain the historical data features.

[0015] In this solution, the pressure fluctuation entropy is calculated using the sample entropy formula, the pressure range is calculated based on the maximum and minimum values, the change rate is calculated based on the change increment and the corresponding time, the extreme value frequency is statistically obtained, and the pressure correlation matrix is calculated using the correlation coefficient formula. Secondly, the extraction process of the real-time data features is the same as that of the historical data features.

[0016] Preferably, in S3, the specific process of training the monitoring model based on the training set and the validation set to obtain the real-time monitoring model is as follows: S31. Input the training set into the monitoring model to obtain the training results, and verify the accuracy of the training results based on the validation set. If the training results match the validation set, the accuracy verification is successful, and the corresponding monitoring model is marked as the real-time monitoring model. If the training results do not match the validation set, the accuracy verification fails, and S32 is executed; S32. Calculate the loss value of the training results based on a preset loss function, adjust the parameters of the monitoring model backward based on the loss value, and execute S31 based on the adjusted monitoring model.

[0017] In this solution, since the training set and the validation set are historical data, it is known whether a coagulation event has occurred in the corresponding pipeline and filter, and the validation set is used to verify the accuracy of the training results obtained by the monitoring model processing the training set. Based on this, the starting point of the time series of the validation set should be the ending point of the time series of the training set. Then, the data in the validation set will directly reflect the coagulation level of the data in the training set in the next period of time. Furthermore, based on the difference between the validation set and the training results, that is, the loss value calculated based on the loss function and the training results, the key nodes that cause the loss value in the monitoring model can be identified through backpropagation, such as batch size, gradient, etc. These key nodes can be adjusted for optimization, either manually or using the salp swarm optimization algorithm, and the adjustment process is continuously looped until the accuracy verification is successful, and the corresponding monitoring model is marked as the real-time monitoring model.

[0018] Preferably, in S4, the specific process of inputting the real-time data set into the real-time monitoring model to obtain the coagulation level of the pipeline and filter is as follows: Input the real-time data set into the real-time monitoring model to obtain the coagulation probability of the real-time data set; Based on the coagulation probability, the first coagulation threshold, the second coagulation threshold, and the third coagulation threshold, judge the coagulation level of the pipeline and filter. If the coagulation probability is less than the first coagulation threshold, it is determined that there is no coagulation crisis in the pipeline and filter. If the coagulation probability is greater than or equal to the first coagulation threshold and less than the second coagulation threshold, it is determined that the coagulation level of the pipeline and filter is mild coagulation. If the coagulation probability is greater than or equal to the second coagulation threshold and less than the third coagulation threshold, it is determined that the coagulation level of the pipeline and filter is moderate coagulation. If the coagulation probability is greater than or equal to the third coagulation threshold, it is determined that the coagulation level of the pipeline and filter is severe coagulation; Organize the situations of no coagulation crisis, mild coagulation, moderate coagulation, and severe coagulation of the pipeline and filter to obtain the coagulation level.

[0019] Preferably, in S5, the specific process of the corresponding extracorporeal pipeline and filter coagulation monitoring system performing warning actions based on the warning rules and the coagulation level is as follows: Select a warning action from the warning rules based on the coagulation level. When the coagulation level is that there is no coagulation crisis in the pipeline and filter, the extracorporeal pipeline and filter coagulation monitoring system continues to operate in the current state; When the coagulation level is mild coagulation, the extracorporeal circuit and filter coagulation monitoring system controls the corresponding warning light to start and edits a warning message; When the coagulation level is moderate coagulation, the extracorporeal circuit and filter coagulation monitoring system controls the corresponding warning light to start and edits a warning message; When the coagulation level is severe coagulation, the extracorporeal circuit and filter coagulation monitoring system controls the corresponding warning light and alarm to start and edits a warning message.

[0020] In this solution, for different coagulation levels, the lights of the warning lights are different. For example, when there is no coagulation crisis in the circuit and filter, there is no light, or green light is adopted. When the coagulation is mild, the light is yellow. When the coagulation is moderate and severe, the light is red. Specifically, it can be achieved by controlling different small light bulbs or the same small light bulb to be powered on. After editing the warning message, it can be notified to relevant staff in the form of text messages, emails, etc., and can also be displayed on the corresponding display screen, such as the screen of the hemodialysis machine. And the warning messages corresponding to different coagulation levels contain different suggestions. For example, when the coagulation is mild, it is recommended to adjust the anticoagulant dose. When the coagulation is moderate, it is recommended to increase the anticoagulant dose and take other intervention measures, such as adjusting the blood flow rate. When the coagulation is severe, it is recommended to replace the filter and adjust the treatment plan, etc.

[0021] On the other hand, another technical solution provided in the embodiments of the present invention is an extracorporeal circuit and filter coagulation monitoring system, including: an Internet of Things acquisition module, an edge computing module, a data analysis module, and a warning module; The Internet of Things acquisition module is used to acquire the operation data, operation parameters of the hemodialysis machine, and the biochemical indexes of the patients corresponding to the hemodialysis machine, and transmit the operation data, operation parameters, and biochemical indexes to the edge computing module; The edge computing module preprocesses the data transmitted by the Internet of Things acquisition module, constructs a monitoring model based on the salp swarm optimization algorithm and the long short-term memory network, and transmits the preprocessed data and the monitoring model to the data analysis module; The data analysis module determines the coagulation level of the pipeline and filter corresponding to the hemodialysis machine based on the data transmitted by the edge computing module and the monitoring module, and transmits the coagulation level to the warning module; The warning module selects and executes a warning action based on the warning rule and the coagulation level transmitted by the data analysis module.

[0022] The beneficial effects of the present invention: In this application, due to the large variety and volume of hemodialysis machine operation data, hemodialysis machine operation parameters, and biochemical indicators of the corresponding patients of the hemodialysis machine, preprocess the operation data, operation parameters, and biochemical indicators, establish the corresponding relationships of various types of data based on the time scale, and integrate the various data into a whole to obtain a training set, a validation set, and a real-time data set. It should be noted that the training set and the validation set are obtained from historical data, and the real-time data set is obtained from real-time data, and the end time of the training set is the start time of the validation set, which is convenient for verifying the performance of the monitoring model when training the monitoring model later for adjustment; Secondly, since the operation data, operation parameters, and biochemical indicators are all continuous data, with continuous data characteristics hidden inside, the long short-term memory network has the ability to extract long-term dependence relationships in this time series. In order to make the long short-term memory network fit the data in the training set, validation set, and real-time data set, and improve the performance of the long short-term memory network in processing hemodialysis machine operation data, operation parameters, and patient biochemical indicators, optimize the hyperparameters of the long short-term memory network based on the salp swarm optimization algorithm to obtain a monitoring model, and train the monitoring model based on the training set and the validation set to obtain a real-time monitoring model, effectively improving the ability of the real-time monitoring model to extract the continuous data characteristics of the operation data, operation parameters, and biochemical indicators. Furthermore, the real-time monitoring model predicts the coagulation grade of the corresponding pipeline and filter according to the continuous data characteristics of the data in the real-time data set. The coagulation grade represents the probability of a coagulation event occurring in the pipeline and filter in the future period of time, and potential coagulation events can be accurately detected in the early stage and latency period of the coagulation event. The corresponding in vitro pipeline and filter coagulation monitoring system performs warning actions based on the warning rules and the coagulation grade, significantly improving the recognition accuracy of coagulation events. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Other features, objects, and advantages of the present invention will become more apparent by reading the detailed description of the non-limiting embodiments with reference to the following drawings. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components.

[0024] Figure 1 It is a schematic flow chart of a method for monitoring coagulation of an in vitro pipeline and filter; Figure 2 It is a schematic structural diagram of a system for monitoring coagulation of an in vitro pipeline and filter. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] To make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only the best embodiments of the present invention, which are only used to explain the present invention and do not limit the protection scope of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present invention.

[0026] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of the operations (or steps) can be implemented in parallel, concurrently, or simultaneously. In addition, the order of the operations can be rearranged. The process can be terminated when its operations are completed, but it can also have additional steps not included in the drawings; the process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0027] Embodiment 1: As Figure 1 shown, this embodiment provides an in vitro pipeline and filter blood coagulation monitoring method, including the following steps: S1. Obtain the operation data, operation parameters of the corresponding hemodialysis machine, and biochemical indexes of the patient corresponding to the hemodialysis machine based on the pipeline and filter; the operation data at least includes transmembrane pressure, arterial pressure, venous pressure, filter pressure, and pre-filter pressure, the operation parameters at least include blood flow rate and ultrafiltration rate, and the biochemical indexes at least include platelet count, prothrombin time, and activated partial thromboplastin time; Optimize the long short-term memory network based on the salp swarm optimization algorithm to obtain a monitoring model; S11. Establish a population with the initial hyperparameters of the long short-term memory network as individuals, and establish an objective function based on the mean square error of the operation data, the mean square error of the operation parameters, and the mean square error of the biochemical indexes; S12. Calculate the fitness of the individuals based on the objective function, and divide the population into leaders and followers based on the fitness; S13. Update the leaders based on the leader update formula of the salp swarm optimization algorithm, and update the followers based on the follower update formula of the salp swarm optimization algorithm; S14. Determine whether to end the optimization based on the update times of the leaders and the iteration threshold. If the update times are less than the iteration threshold, continue the optimization and execute S12. If the update times are greater than or equal to the iteration threshold, end the optimization and execute S15; S15. Sort the fitness of the individuals in ascending order, and adjust the hyperparameters of the long short-term memory network based on the individual with the minimum fitness to obtain a monitoring model.

[0028] In this embodiment, the initial hyperparameters of the long short-term memory network at least include the number of hidden layer nodes, the learning rate, and the weight coefficient. Among them, the number of hidden nodes is usually set within 8 to 64, the learning rate is generally 0.001, and the weight coefficient is set according to experience. Multiple individuals can be obtained by randomly combining the hyperparameters, and then a population is formed. A target function is established based on the mean square error of the operation data, the mean square error of the operation parameters, and the mean square error of the biochemical indexes. It should be noted that the operation data, operation parameters, and biochemical indexes only serve as auxiliary data for the establishment of the monitoring model and can act alone in the establishment process of the monitoring data. However, in order to enable the monitoring model to adapt to the working environment of the corresponding hemodialysis machine, the operation data, operation parameters, and biochemical indexes act together in the establishment process of the monitoring model. The sum of the mean square error of the operation data, the mean square error of the operation parameters, and the mean square error of the biochemical indexes is used as the target function, and the weight coefficients of different mean square errors are set by using the expert scoring method. Then, the fitness of the individual is calculated based on the target function. The individual with the minimum fitness is marked as the leader, and the rest of the individuals are marked as followers. The leader is updated based on the leader update formula of the salp swarm optimization algorithm, and the followers are updated based on the follower update formula of the salp swarm optimization algorithm. During the update process, the boundary constraint conditions are used to limit the update of the leader and the followers to ensure that the corresponding parameters are within a reasonable range. The boundary constraint conditions are set based on the characteristics of the long short-term memory network hyperparameters and can be set as dynamic constraint boundaries here. That is, during the first optimization iteration, the boundary constraint conditions are [-1, 1], and during the last optimization iteration, the boundary constraint conditions are [-0.1, 0.1]. The change amount of the boundary constraint conditions at each iteration needs to be calculated based on the iteration threshold. If the iteration threshold is 400, the difference between the first upper boundary 1 and the last upper boundary 0.1 is divided by the iteration threshold 400 to obtain the change amount of the boundary constraint conditions at each iteration. The boundary constraint conditions gradually decrease by the change amount until the last iteration. Among them, the training set and the validation set are obtained from historical data, and the real-time data set is obtained from real-time data.

[0029] It should be noted that in order to improve the accuracy of the training set, validation set, and real-time data set, the training set, validation set, and real-time data set are established based on the time series. In S2, the operation data at least includes historical operation data and real-time operation data, the operation parameters at least include historical operation parameters and real-time operation parameters, and the biochemical indexes at least include historical biochemical indexes and real-time biochemical indexes.

[0030] S2. Preprocess the operation data, operation parameters, and biochemical indexes to obtain a training set, a validation set, and a real-time data set; S21. Filter out the outliers in the historical operation data, historical operation parameters, and historical biochemical indexes based on a preset sliding window to obtain a historical outlier-removed data set; Filter the outliers in the real-time operation data, real-time operation parameters, and real-time biochemical indicators based on a preset sliding window to obtain a real-time outlier-removed data set; S22. Fill in the missing values in the historical outlier-removed data set based on the monitoring model and the missing data filling criterion to obtain a historical complete data set; Fill in the missing values in the real-time outlier-removed data set based on the monitoring model and the missing data filling criterion to obtain a real-time complete data set; Specifically, count the missing values in the historical outlier-removed data set, and classify the missing values based on the time scale of the historical outlier-removed data set to obtain a classified missing set; Judge whether to fill in based on the number of missing values and the missing value threshold in the subset of the classified missing set. If the number of missing values is less than or equal to the missing value threshold, do not fill in. If the number of missing values is greater than the missing value threshold, select the data in the outlier-removed data set as the input of the monitoring model based on the time scale of the corresponding subset to obtain a filled subset; Organize the filled subset and the outlier-removed data set to obtain a historical complete data set; Synchronously, count the real-time missing values in the real-time outlier-removed data set, and classify the real-time missing values based on the time scale of the real-time outlier-removed data set to obtain a real-time classified missing set; Judge whether to fill in based on the number of real-time missing values and the missing value threshold in the real-time subset of the real-time classified missing set. If the number of real-time missing values is less than or equal to the missing value threshold, do not fill in. If the number of real-time missing values is greater than the missing value threshold, select the real-time data in the real-time outlier-removed data set as the input of the monitoring model based on the time scale of the corresponding real-time subset to obtain a real-time filled subset; Organize the real-time filled subset and the real-time outlier-removed data set to obtain a real-time complete data set; S23. Normalize the historical complete data set based on the normalization criterion to obtain a historical standard data set, and normalize the real-time complete data set based on the normalization criterion to obtain a real-time standard data set; Synchronously, extract the features of the data in the historical complete data set based on the salp swarm optimization algorithm to obtain historical data features, and extract the features of the data in the real-time complete data set based on the salp swarm optimization algorithm to obtain real-time data features; Specifically, calculate the pressure fluctuation entropy, pressure range, change rate, extreme value frequency, and pressure correlation matrix based on the data in the historical complete data set, and establish individuals based on the pressure fluctuation entropy, pressure range, change rate, extreme value frequency, and pressure correlation matrix to obtain a population; Set the maximum information entropy gain as the objective function of the salp swarm optimization algorithm, and input the population into the salp swarm optimization algorithm to obtain historical data features; Synchronously, calculate the real-time pressure fluctuation entropy, real-time pressure range, real-time change rate, real-time extreme value frequency, and real-time pressure correlation matrix based on the real-time data in the real-time complete dataset, and establish an individual based on the real-time pressure fluctuation entropy, real-time pressure range, real-time change rate, real-time extreme value frequency, and real-time pressure correlation matrix to obtain a real-time population; Set the maximization of information entropy gain as the objective function of the salp swarm optimization algorithm, and input the real-time population into the salp swarm optimization algorithm to obtain real-time data features; S24. Organize the historical data features and historical standard datasets to obtain a training set and a validation set, and organize the real-time data features and real-time standard datasets to obtain a real-time dataset.

[0031] In this embodiment, since the collected data is of various types, large volume, and high complexity, in order to process it efficiently and accurately, some data in the operation data, operation parameters, and biochemical indicators are dynamically intercepted based on a sliding window. Then, the upper and lower limits of the partial data are calculated based on the interquartile range method, and the normal data range is determined by the upper and lower limits. The data outside the normal data range is marked as outliers and deleted to obtain a dataset without outliers. At this time, there are a large number of missing values in the dataset without outliers. Each missing value corresponds to a time point, and whether it is operation data, operation parameters, or biochemical indicators, they all change according to a fixed pattern within a certain range. When the number of missing values occupies a certain proportion in the dataset without outliers, it will reduce the accuracy of the overall data and cannot comprehensively describe the data characteristics. Therefore, the missing values adjacent in the time scale are grouped into a set to obtain a classified missing set, that is, the missing values in each set are continuous. When the number of missing values is greater than the missing value threshold, it proves that there is no data in the time period corresponding to the set. Select the adjacent set for the time period to form a subset to be filled. The monitoring model fills the missing values according to the data characteristics in the subset to be filled, that is, the long-term dependence relationship in the data, to obtain a filled subset. Finally, the filled subset is inserted into the dataset without outliers according to the time series to obtain a complete dataset. At this time, the data in the complete dataset has different units due to different types, that is, there is a dimension problem. In order to eliminate the influence of the dimension on the subsequent data processing, the data is normalized using the standard score formula to obtain a standard dataset. It should be noted that the standard score formula may not necessarily match the data in the complete dataset, and the parameters in the standard score formula, that is, the mean and standard deviation, need to be optimized using the salp swarm optimization algorithm. The specific standard score formula is as follows: In the formula, z is the normalized data, x is the original data before normalization, μ is the mean, and σ is the standard deviation; At the same time, in order to extract the potential features of the data in the standard dataset for subsequent data processing, first calculate the pressure fluctuation entropy using the sample entropy formula. The sample entropy formula is as follows: In the formula, S is the pressure fluctuation entropy, A is the number of pairs of matching vectors with a length of m + 1, B is the number of pairs of matching vectors with a length of m, m is the vector length, r is the capacity set to 0.2. If the corresponding pressure data is [50, 53, 56], taking m as 2, the corresponding vectors are X1 = [50, 53] and X2 = [53, 56]. Calculate the distance d between the vectors, specifically: d[X1, X2] = max(|53 - 50|, |56 - 53|) = 3; If the distance d is greater than the capacity r, then the vector X1 and the vector X2 do not match. Count the number of matching times of each vector, and divide the number of matching times by the corresponding vector length m to obtain the number of pairs of matching vectors; Calculate the pressure range based on the maximum and minimum values of the pressure data. The specific formula for the pressure range is: P j = P max - P min ; In the formula, P j is the pressure range, P max is the maximum value of the corresponding pressure data, P min is the minimum value of the corresponding pressure data; Calculate the change rate based on the change increment and the corresponding time. The specific formula for the change rate is: In the formula, Δv is the change rate corresponding to the pressure data, and Δp is the change increment of the pressure data within the time Δt; The extreme value frequency can be directly counted. Calculate the pressure correlation matrix based on the correlation coefficient formula. The specific correlation coefficient formula is: In the formula, C i,j is the pressure correlation coefficient, cov(p i , p j ) is the covariance between the i-th pressure data p i and the j-th pressure data p j , σp i is the standard deviation of the i-th pressure data p i , σp j is the standard deviation of the j-th pressure data p j ; After obtaining the correlation coefficients of all the pressure data, form the pressure correlation matrix according to the corresponding order; Extract the features of the data in the complete dataset based on the salp swarm optimization algorithm to obtain data features. Finally, organize the data features and the standard dataset to obtain a training set, a validation set, and a real-time dataset. Additionally, the data includes real-time data and historical data, and the training set and the validation set are constructed based on historical data, while the real-time dataset is constructed based on real-time data. Using the salp swarm optimization algorithm to extract data features can establish corresponding time tags based on the time series of the data, such as the start point of the training set, the end point of the training set, the start point of the validation set, the end point of the validation set, the start point of the real-time dataset, and the end point of the real-time dataset, facilitating the establishment of the training set, the validation set, and the real-time dataset.

[0032] S3. Train the monitoring model based on the training set and the validation set to obtain a real-time monitoring model. S31. Input the training set into the monitoring model to obtain a training result, and verify the accuracy of the training result based on the validation set. If the training result is consistent with the validation set, the accuracy verification is successful, and the corresponding monitoring model is marked as the real-time monitoring model. If the training result is inconsistent with the validation set, the accuracy verification fails, and S32 is executed. S32. Calculate the loss value of the training result based on a preset loss function, adjust the parameters of the monitoring model backward based on the loss value, and execute S31 based on the adjusted monitoring model.

[0033] In this embodiment, since the training set and the validation set are historical data, it is known whether coagulation events have occurred in the corresponding pipelines and filters, and the validation set is used to verify the accuracy of the training result obtained by the monitoring model processing the training set. Based on this, the start point of the time series of the validation set should be the end point of the time series of the training set. Then, the data in the validation set will directly reflect the coagulation level of the data in the training set in the next time period. Furthermore, based on the difference between the validation set and the training result, that is, the loss value calculated based on the loss function and the training result, the key nodes that cause the loss value in the monitoring model can be identified by backpropagation, such as batch size, gradient, etc., and these key nodes can be adjusted for optimization, which can be adjusted manually or using the salp swarm optimization algorithm, and the adjustment process is continuously looped until the accuracy verification is successful, and the corresponding monitoring model is marked as the real-time monitoring model. The loss function uses a multi-objective optimization loss function, and the specific form of the loss function is: L total = λ1L MAE + λ2L Focal ; In the formula, L total is the loss value, λ1 is the weight of the mean absolute error loss, L MAE is the mean absolute error loss, λ2 is the weight of the focal loss, and L Focal is the focal loss.

[0034] S4. Input the real-time data set into the real-time monitoring model to obtain the coagulation grade of the pipeline and filter; Input the real-time data set into the real-time monitoring model to obtain the coagulation probability of the real-time data set; Based on the coagulation probability, the first coagulation threshold, the second coagulation threshold, and the third coagulation threshold, judge the coagulation grade of the pipeline and filter. If the coagulation probability is less than the first coagulation threshold, it is determined that there is no coagulation crisis in the pipeline and filter. If the coagulation probability is greater than or equal to the first coagulation threshold and less than the second coagulation threshold, it is determined that the coagulation grade of the pipeline and filter is mild coagulation. If the coagulation probability is greater than or equal to the second coagulation threshold and less than the third coagulation threshold, it is determined that the coagulation grade of the pipeline and filter is moderate coagulation. If the coagulation probability is greater than or equal to the third coagulation threshold, it is determined that the coagulation grade of the pipeline and filter is severe coagulation; Organize the pipeline and filter without coagulation crisis, mild coagulation, moderate coagulation, and severe coagulation to obtain the coagulation grade.

[0035] In this embodiment, the coagulation probability is actually a number between 0 and 1, which can be expressed as a corresponding percentage. The first coagulation threshold is 0.2, that is, 20%, the second coagulation threshold is 0.5, that is, 50%, and the third coagulation threshold is 0.8, that is, 80%.

[0036] S5. The corresponding extracorporeal pipeline and filter coagulation monitoring system performs warning actions based on the warning rules and the coagulation grade; Select warning actions in the warning rules based on the coagulation grade. When the coagulation grade is that there is no coagulation crisis in the pipeline and filter, the extracorporeal pipeline and filter coagulation monitoring system continues to run in the current state; When the coagulation grade is mild coagulation, the extracorporeal pipeline and filter coagulation monitoring system controls the corresponding warning light to start and edits warning information; When the coagulation grade is moderate coagulation, the extracorporeal pipeline and filter coagulation monitoring system controls the corresponding warning light to start and edits warning information; When the coagulation grade is severe coagulation, the extracorporeal pipeline and filter coagulation monitoring system controls the corresponding warning light and alarm to start and edits warning information.

[0037] In this embodiment, corresponding to different blood coagulation levels, the warning lights have different lights. When there is no blood coagulation crisis in the pipeline and filter, there is no light, or green light is adopted. When there is mild blood coagulation, the light is yellow. When there is moderate or severe blood coagulation, the light is red. Specifically, it can be achieved by controlling different small light bulbs or the same small light bulb to be powered on. After editing the warning information, it can be notified to relevant staff in the form of text messages, emails, etc., and can also be displayed on the corresponding display screen, such as the screen of a hemodialysis machine. And the warning information corresponding to different blood coagulation levels contains different suggestions. For example, when there is mild blood coagulation, it is recommended to adjust the anticoagulant dose. When there is moderate blood coagulation, it is recommended to increase the anticoagulant dose and take other intervention measures, such as adjusting the blood flow rate. When there is severe blood coagulation, it is recommended to replace the filter and adjust the treatment plan, etc.

[0038] On the other hand, as Figure 2 shown, another technical solution provided in the embodiment of the present invention is an in vitro pipeline and filter blood coagulation monitoring system, including: an Internet of Things acquisition module, an edge computing module, a data analysis module, and a warning module; the Internet of Things acquisition module is used to acquire the operation data, operation parameters of the hemodialysis machine, and the biochemical indexes of the patient corresponding to the hemodialysis machine, and transmit the operation data, operation parameters, and biochemical indexes to the edge computing module; The edge computing module preprocesses the data transmitted by the Internet of Things acquisition module, constructs a monitoring model based on the salp swarm optimization algorithm and the long short-term memory network, and transmits the preprocessed data and the monitoring model to the data analysis module; The data analysis module judges the blood coagulation level of the pipeline and filter corresponding to the hemodialysis machine based on the data transmitted by the edge computing module and the monitoring module, and transmits the blood coagulation level to the warning module; The warning module selects and executes a warning action based on the warning rule and the blood coagulation level transmitted by the data analysis module. In this embodiment, the Internet of Things acquisition device acquires data such as operation data, operation parameters, and biochemical indexes by connecting to devices such as pressure sensors and biochemical analyzers of the corresponding pipeline and filter through the Internet of Things.

[0039] This embodiment at least has the following substantial effects: In this embodiment, due to the large variety and volume of the operation data, operation parameters, and biochemical indicators of the hemodialysis machine for the corresponding patients, the operation data, operation parameters, and biochemical indicators are preprocessed, the corresponding relationships of various types of data are established based on the time scale, and the various data are integrated into a whole to obtain a training set, a validation set, and a real-time data set. It should be noted that the training set and the validation set are obtained from historical data, the real-time data set is obtained from real-time data, and the end time of the training set is the start time of the validation set, which is convenient for verifying the performance of the monitoring model when training the monitoring model later for adjustment; secondly, since the operation data, operation parameters, and biochemical indicators are all continuous data and there are continuous data characteristics hidden inside, the long short-term memory network has the ability to extract the long-term dependence relationship in this time series. In order to make the long short-term memory network fit the data in the training set, the validation set, and the real-time data set and improve the performance of the long short-term memory network in processing the operation data, operation parameters, and biochemical indicators of the hemodialysis machine, the hyperparameters of the long short-term memory network are optimized based on the salp swarm optimization algorithm to obtain a monitoring model, and the monitoring model is trained based on the training set and the validation set to obtain a real-time monitoring model, effectively improving the ability of the real-time monitoring model to extract the continuous data characteristics of the operation data, operation parameters, and biochemical indicators. Furthermore, the real-time monitoring model predicts the coagulation grade of the corresponding pipeline and filter according to the continuous data characteristics of the data in the real-time data set, and the coagulation grade represents the probability of a coagulation event occurring in the pipeline and filter in the next period of time, which can accurately detect potential coagulation events in the early stage and latency period of the coagulation event. Accordingly, the extracorporeal pipeline and filter coagulation monitoring system performs warning actions based on the warning rules and the coagulation grade, significantly improving the recognition accuracy of coagulation events.

[0040] The above specific implementation manners are the preferred implementation manners of the present invention, which do not limit the specific implementation scope of the present invention. The scope of the present invention includes but is not limited to this specific implementation manner. Any equivalent changes made according to the shape, structure, and method of the present invention are within the protection scope of the present invention.

Claims

1. An in vitro pipeline and filter blood coagulation monitoring method, characterized in that, It includes the following steps: S1. Obtain the operation data, operation parameters of the corresponding hemodialysis machine and the biochemical indexes of the patient corresponding to the hemodialysis machine based on the pipeline and filter; Optimize the long short-term memory network based on the salp swarm optimization algorithm to obtain a monitoring model; S2. Preprocess the operation data, operation parameters and biochemical indexes to obtain a training set, a validation set and a real-time data set; S3. Train the monitoring model based on the training set and the validation set to obtain a real-time monitoring model; S4. Input the real-time data set into the real-time monitoring model to obtain the coagulation grade of the pipeline and filter; S5. The corresponding extracorporeal pipeline and filter coagulation monitoring system performs warning actions based on the warning rules and the coagulation grade.

2. The in vitro pipeline and filter blood coagulation monitoring method according to claim 1, characterized in that In the above S1, the specific process of optimizing the long short-term memory network based on the salp swarm optimization algorithm to obtain a monitoring model is as follows: S11. Establish a population with the initial hyperparameters of the long short-term memory network as individuals, and establish an objective function based on the mean square error of the operation data, the mean square error of the operation parameters and the mean square error of the biochemical indexes; S12. Calculate the fitness of the individuals based on the objective function, and divide the population into leaders and followers based on the fitness; S13. Update the leaders based on the leader update formula of the salp swarm optimization algorithm, and update the followers based on the follower update formula of the salp swarm optimization algorithm; S14. Judge whether to end the optimization based on the update times of the leaders and the iteration threshold. If the update times are less than the iteration threshold, continue the optimization and execute S12. If the update times are greater than or equal to the iteration threshold, end the optimization and execute S15; S15. Sort the fitness of the individuals in descending order, and adjust the hyperparameters of the long short-term memory network based on the individual with the minimum fitness to obtain a monitoring model.

3. The extracorporeal pipeline and filter blood coagulation monitoring method according to claim 1, characterized in that, In the above S2, the operation data at least includes historical operation data and real-time operation data, the operation parameters at least include historical operation parameters and real-time operation parameters, and the biochemical indexes at least include historical biochemical indexes and real-time biochemical indexes; Among them, preprocess the historical operation parameters, historical operation data and historical biochemical indexes to obtain a training set and a validation set; Preprocess the real-time operation data, real-time operation parameters and real-time biochemical indexes to obtain a real-time data set.

4. The in vitro pipeline and filter blood coagulation monitoring method according to claim 3, characterized in that, In the above S2, the specific process of preprocessing the operation data, operation parameters and biochemical indexes to obtain a training set, a validation set and a real-time data set is as follows: S21. Filter out the outliers in the historical operation data, historical operation parameters and historical biochemical indexes based on a preset sliding window to obtain a historical outlier-free data set; Filter out the outliers in the real-time operation data, real-time operation parameters and real-time biochemical indexes based on a preset sliding window to obtain a real-time outlier-free data set; S22. Fill in the missing values in the historical outlier-free data set based on the monitoring model and the missing data filling criterion to obtain a historical complete data set; Fill in the missing values in the real-time outlier-free data set based on the monitoring model and the missing data filling criterion to obtain a real-time complete data set; S23. Normalize the historical complete data set based on the normalization criterion to obtain a historical standard data set, and normalize the real-time complete data set based on the normalization criterion to obtain a real-time standard data set; Synchronously, based on the salp swarm optimization algorithm, extract the features of the data in the historical complete dataset to obtain historical data features, and based on the salp swarm optimization algorithm, extract the features of the data in the real-time complete dataset to obtain real-time data features; S24. Organize the historical data features and the historical standard dataset to obtain a training set and a validation set, and organize the real-time data features and the real-time standard dataset to obtain a real-time dataset.

5. An in vitro pipeline and filter blood coagulation monitoring method according to claim 4, characterized in that, In S22, the specific process of filling the missing values in the historical outlier-removed dataset based on the monitoring model and the missing data filling criterion to obtain a historical complete dataset is as follows: Count the missing values in the historical outlier-removed dataset, and classify the missing values based on the time scale of the historical outlier-removed dataset to obtain a classified missing set; Judge whether to fill based on the number of missing values in the subset in the classified missing set and the missing value threshold. If the number of missing values is less than or equal to the missing value threshold, do not fill. If the number of missing values is greater than the missing value threshold, select the data in the outlier-removed dataset as the input of the monitoring model based on the time scale of the corresponding subset to obtain a filled subset; Organize the filled subset and the outlier-removed dataset to obtain a historical complete dataset.

6. The in vitro pipeline and filter blood coagulation monitoring method according to claim 4, characterized in that, In S23, the specific process of extracting the features of the data in the historical complete dataset based on the salp swarm optimization algorithm to obtain historical data features is as follows: Calculate the pressure fluctuation entropy, pressure range, change rate, extreme value frequency, and pressure correlation matrix based on the data in the historical complete dataset, and establish individuals based on the pressure fluctuation entropy, pressure range, change rate, extreme value frequency, and pressure correlation matrix to obtain a population; Set the maximization of information entropy gain as the objective function of the salp swarm optimization algorithm, and input the population into the salp swarm optimization algorithm to obtain historical data features.

7. The extracorporeal pipeline and filter blood coagulation monitoring method according to claim 1, characterized in that In S3, the specific process of training the monitoring model based on the training set and the validation set to obtain a real-time monitoring model is as follows: S31. Input the training set into the monitoring model to obtain a training result, and verify the accuracy of the training result based on the validation set. If the training result is consistent with the validation set, the accuracy verification is successful, and the corresponding monitoring model is marked as a real-time monitoring model. If the training result is inconsistent with the validation set, the accuracy verification fails, and execute S32; S32. Calculate the loss value of the training result based on a preset loss function, adjust the parameters of the monitoring model backward based on the loss value, and execute S31 based on the adjusted monitoring model.

8. The in vitro pipeline and filter blood coagulation monitoring method according to claim 1, wherein In S4, the specific process of inputting the real-time dataset into the real-time monitoring model to obtain the blood coagulation grade of the pipeline and the filter is as follows: Input the real-time dataset into the real-time monitoring model to obtain the blood coagulation probability of the real-time dataset; Judge the coagulation level of the pipeline and filter based on the coagulation probability, the first coagulation threshold, the second coagulation threshold, and the third coagulation threshold. If the coagulation probability is less than the first coagulation threshold, it is determined that there is no coagulation crisis in the pipeline and filter. If the coagulation probability is greater than or equal to the first coagulation threshold and less than the second coagulation threshold, it is determined that the coagulation level of the pipeline and filter is mild coagulation. If the coagulation probability is greater than or equal to the second coagulation threshold and less than the third coagulation threshold, it is determined that the coagulation level of the pipeline and filter is moderate coagulation. If the coagulation probability is greater than or equal to the third coagulation threshold, it is determined that the coagulation level of the pipeline and filter is severe coagulation; Organize the coagulation levels of no coagulation crisis, mild coagulation, moderate coagulation, and severe coagulation of the pipeline and filter to obtain the coagulation level.

9. The in vitro pipeline and filter blood coagulation monitoring method according to claim 1, characterized in that In the S5, the specific process of the corresponding extracorporeal pipeline and filter coagulation monitoring system performing warning actions based on the warning rules and coagulation levels is as follows: Select warning actions in the warning rules based on the coagulation level. When the coagulation level is no coagulation crisis in the pipeline and filter, the extracorporeal pipeline and filter coagulation monitoring system continues to operate in the current state; When the coagulation level is mild coagulation, the extracorporeal pipeline and filter coagulation monitoring system controls the corresponding warning light to start and edits warning information; When the coagulation level is moderate coagulation, the extracorporeal pipeline and filter coagulation monitoring system controls the corresponding warning light to start and edits warning information; When the coagulation level is severe coagulation, the extracorporeal pipeline and filter coagulation monitoring system controls the corresponding warning light and alarm to start and edits warning information.

10. An in vitro pipeline and filter blood coagulation monitoring system, applicable to an in vitro pipeline and filter blood coagulation monitoring method as described in any one of claims 1-9, characterized in that, It includes: an Internet of Things acquisition module, an edge computing module, a data analysis module, and a warning module; The Internet of Things acquisition module is used to acquire the operation data, operation parameters of the hemodialysis machine, and the biochemical indexes of the patients corresponding to the hemodialysis machine, and transmit the operation data, operation parameters, and biochemical indexes to the edge computing module; The edge computing module preprocesses the data transmitted by the Internet of Things acquisition module, constructs a monitoring model based on the salp swarm optimization algorithm and the long short-term memory network, and transmits the preprocessed data and the monitoring model to the data analysis module; The data analysis module judges the coagulation level of the pipeline and filter corresponding to the hemodialysis machine based on the data transmitted by the edge computing module and the monitoring module, and transmits the coagulation level to the warning module; The warning module selects and executes warning actions based on the warning rules and the coagulation level transmitted by the data analysis module.

Images