Method and system for maintaining vascular access intelligent management information system

Abstract

The invention relates to a vascular access intelligent management information system maintenance method and system, and belongs to the technical field of information system maintenance. The method comprises the following steps: verifying historical sample data, constructing an abnormal probability model and a state monitoring model, and obtaining an abnormal probability threshold value and a monitoring score threshold value; capturing patient information and operation data in real time to obtain the abnormal probability of real-time sample data, and judging whether the operation data reaches the standard or not; capturing the state data, obtaining the monitoring score judgment dichotomy label of the real-time sample data, and sending out a state judgment result. The abnormal probability and the monitoring score of the sample data are improved by constructing the abnormal probability model and the state monitoring model, and the abnormal probability threshold and the monitoring score threshold are obtained by adopting the box plot method, so that the real-time data stream is preliminarily judged, the dichotomous label is autonomously obtained through the state data, and the real-time data stream is subjected to classification. The timeliness of judging the health state of the vascular access by medical staff is improved, and the stability and the reliability of the intelligent management information system for the vascular access are maintained at the same time.

Description

Technical Field

[0001] The present invention belongs to the technical field of information system maintenance, and in particular relates to a maintenance method and system for a vascular access intelligent management information system. Background Art

[0002] With the rapid development of medical informatization, vascular access intelligent management information systems are playing an increasingly important role in clinical applications. Such systems significantly improve medical efficiency and safety monitoring by integrating patient data, monitoring vascular access status, and providing abnormal warnings.

[0003] However, with the continuous expansion of data volume, higher requirements are being placed on the maintenance of the intelligent vascular access management information system. The real-time and reliability of the data stream verification collected by the system are poor, and maintenance work cannot keep up with the changing needs of the system. Therefore, it is urgent to provide a maintenance method and system for the intelligent vascular access management information system. By verifying the data stream of the management information system in real time, the stability and reliability of the management information system can be improved. Summary of the Invention

[0004] The purpose of the present invention can be achieved through the following technical solutions: A first aspect of the present disclosure provides a method for maintaining a vascular access intelligent management information system, comprising the steps of: Process historical sample data and verify sample data, build an abnormal probability model and a condition monitoring model through the verified sample data to obtain abnormal probability thresholds and monitoring score thresholds; Capture patient information and operation data in real time and obtain the abnormal probability of real-time sample data through the abnormal probability model, determine whether the operation data meets the standards, and provide a judgment prompt; Capture status data and obtain monitoring scores of real-time sample data through the status monitoring model. Determine binary classification labels based on the monitoring scores that meet the standards, and issue status judgment results after vascular access is established. The method of obtaining an abnormal probability threshold and a monitoring score threshold comprises the following steps: An abnormality prediction model is constructed based on a deep reinforcement learning algorithm. Patient information, operation data, and binary classification labels are input into the abnormality prediction model for training to obtain the abnormal probability of vascular access establishment. A condition monitoring model is constructed through an improved spatiotemporal graph convolutional network. The condition data is input into the condition monitoring model for training to obtain the monitoring score of the condition data. The abnormal probability model and status monitoring model are used to obtain the abnormal probability and monitoring score of each historical sample data. The abnormal probability and monitoring score are analyzed respectively using the box plot method. The outlier boundaries are determined based on the calculated quartiles, and the respective outlier boundaries are set as the abnormal probability threshold and monitoring score threshold.

[0005] Furthermore, the processing of historical sample data includes the steps of: Data collection and preprocessing: Obtain historical sample data of vascular access establishment, including patient information, operation data, status data, and binary classification labels, and clean, normalize, and time-series align the historical sample data.

[0006] Furthermore, the patient information includes name, gender, age, hospitalization number and diagnosis information; the operation data includes time, operator information, location, number of times, disinfection and dressing materials used; the status data includes blood routine, infusion rate, infusion resistance, external catheter length and body temperature; the two-category label is a record of whether medical staff have any abnormalities in the establishment of vascular access.

[0007] Furthermore, obtaining the abnormal probability of vascular access establishment includes the steps of: Taking patient information, operation data, and binary classification labels as input, the patient information's identity identifiers are desensitized, and then the independent component analysis algorithm is used to reduce the dimensionality of the input data. By separating the independent non-Gaussian components in the data, the data with cumulative contribution greater than the contribution threshold is screened out as the execution prediction data; The execution prediction data is input into the deep reinforcement learning network in the abnormal probability model, and an adaptive genetic algorithm is used to initialize a policy space for the deep reinforcement learning network. The strategy space contains J individuals, each of which represents an operation strategy, and each individual is configured with an exploration probability and an update step size; Construct a reward function based on the mean squared error loss function of the anomaly probability model and the binary classification labels, set the reward threshold and maximum iteration rounds, and perform iterative optimization on the policy space; When the reward function value is higher than the reward threshold or reaches the maximum iteration round, the iteration is stopped and the operation strategy corresponding to the optimal individual is output; The acquired operation strategy is fed back to the deep reinforcement learning network to predict the abnormal probability of establishing vascular access for the sample data.

[0008] Furthermore, the calculation formula of the reward function is: ; in, represents the reward function value, represents the number of input samples, Indicates the The actual value of the samples, Indicates the The predicted value of the sample, Indicates the The mean square error term of samples, represents the trade-off coefficient, Indicates the probability of abnormal path establishment.

[0009] Furthermore, the monitoring score of obtaining status data includes the steps of: The improved spatiotemporal graph convolutional network includes a time slicing layer, a spatial convolution layer, an abnormal pattern clustering layer, an attention mechanism layer and an output layer; Extract time-frequency domain features from the state data, and perform multi-dimensional annotation on the extracted time-frequency domain features to generate annotated state data; The multi-dimensional annotation includes first-level annotation, second-level annotation and third-level annotation. The first-level annotation is the complication type annotation, the second-level annotation is the treatment measure annotation of the corresponding complication, and the third-level annotation is the abnormal treatment status annotation of the corresponding complication. The time-frequency domain features are used to capture the state change rules after the vascular access is established in the time dimension and frequency dimension, and to capture abnormal states; The labeled state data is input into the time slicing layer of the improved spatiotemporal graph convolutional network, and the labeled state data is segmented to generate a time series feature matrix; The time series feature matrix is input into the spatial convolution layer of the improved spatiotemporal graph convolutional network to perform preliminary spatial feature extraction and generate initial abnormal features of the state data; The spatial convolution layer converts the input data into discriminative initial abnormal features through the dependency relationship of the state data; The initial abnormal features are input into the abnormal pattern clustering layer of the improved spatiotemporal graph convolutional network, and the K-means algorithm is used to cluster the features belonging to the same abnormal pattern to form M abnormal pattern feature subspaces; The abnormal pattern feature subspace is input into the attention mechanism layer of the improved spatiotemporal graph convolutional network for feature weighted fusion to obtain the abnormal fusion features of the state data; The state data anomaly fusion features are input into the output layer of the improved spatiotemporal graph convolutional network, and the anomaly score of each state data is calculated by the Softmax function.

[0010] Furthermore, the calculation formula of the abnormality score is: ; in, For the The abnormality score of the state data, Indicates the The weight coefficient of each abnormal type, Represents input data In the The characteristic response value under each abnormal type, is the normalization factor, The number of exception types.

[0011] Furthermore, the determination of whether the operation data meets the standards includes: The abnormal probability model is used to obtain the abnormal probability of real-time sample data based on patient information and operation data. Based on the abnormal probability threshold, it is determined whether the patient information and operation data meet the conditions. When the conditions for establishing the pathway are not met, an early warning is issued to medical staff to adjust the operation data.

[0012] Furthermore, the issuing of the status determination result after the vascular access is established includes: The monitoring score of real-time sample data is obtained through the status data through the status monitoring model. The status data is judged to be up to standard based on the monitoring score threshold. When the monitoring score meets the standard, a binary classification label is generated based on the monitoring score.

[0013] A second aspect of the present disclosure provides a maintenance system for a vascular access intelligent management information system, which applies a maintenance method for a vascular access intelligent management information system as described above, including a data verification module, a model building module, and an early warning reminder module; The data verification module is used to preprocess the historical sample data of the system and verify the sample data through the abnormal probability threshold and monitoring score threshold of the sample data obtained by the abnormal probability model and the state monitoring model; The model building module is used to build an abnormality probability model and a condition monitoring model based on the verified historical sample data, using a machine learning algorithm or a statistical analysis model, combined with binary classification labels for supervised learning to complete the model building; The early warning reminder module is used to input real-time patient information and operation data into the abnormal probability model, calculate the abnormal probability of the current operation, and trigger an early warning if the abnormal probability exceeds the threshold; and after establishing a vascular access, input the real-time status data into the status monitoring model to generate a monitoring score, and compare it with the threshold to determine whether it meets the standard, and automatically generate a binary classification label for reminder.

[0014] The beneficial effects of the present invention are: The present invention first constructs an abnormal probability model and a status monitoring model by utilizing the historical sample data of the vascular access intelligent management information system, improves the abnormal probability and monitoring score of all sample data, then uses the box plot method to obtain the abnormal value boundary based on the abnormal probability and monitoring score to complete the verification of the historical sample data, and obtains the abnormal probability threshold and the monitoring score threshold. Then, the abnormal probability is obtained from the real-time data stream collected by the system for each vascular access establishment to make a preliminary judgment, and the operation data of the medical staff is adjusted and auxiliary reminders are provided to improve the efficiency of the medical staff and reduce the error rate of the operation. Finally, after the access is established, the status data is used to independently obtain a two-category label using the status monitoring model, thereby improving the timeliness of the medical staff's judgment on the health status of the vascular access, while maintaining the stability and reliability of the vascular access intelligent management information system. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] To facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0016] Figure 1 A schematic diagram of the steps of a maintenance method for a vascular access intelligent management information system provided by an embodiment of the present invention; Figure 2 A schematic structural diagram of a maintenance system of a vascular access intelligent management information system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0018] This embodiment provides a maintenance method for a vascular access intelligent management information system, such as Figure 1 As shown, the following steps are included: S1. Process historical sample data and verify the sample data. Use the verified sample data to build an abnormal probability model and a condition monitoring model to obtain an abnormal probability threshold and a monitoring score threshold, including the following steps: Data acquisition and preprocessing: Obtain historical sample data of vascular access establishment, which includes patient information, operation data, status data, and binary classification labels. Clean, normalize, and time-align the historical sample data to eliminate noise interference and unify resolution and timestamps.

[0019] It should be noted that the patient information includes patient identity and basic information such as name, gender, age, hospitalization number, and diagnosis information; the operation data includes process data of access establishment such as time, operator information, location, number of times, disinfection and dressing materials used; the status data includes blood routine, infusion rate, infusion resistance, external catheter length, and body temperature; the two-category label is a record of whether medical staff have any abnormalities in the establishment of vascular access, such as 0 for no occurrence and 1 for occurrence. Abnormalities include infection, thrombosis, catheter dysfunction, bleeding, etc.

[0020] Abnormality probability model construction: Based on the deep reinforcement learning algorithm, an abnormality prediction model is constructed. Patient information, operation data, and binary classification labels are input into the abnormality prediction model for training to obtain the abnormal probability of vascular access establishment. Specifically, it includes: Patient information, operation data, and binary classification labels are used as input. The patient information's identity identifier is desensitized, and the independent component analysis algorithm is used to reduce the dimensionality of the input data. By separating the independent non-Gaussian components in the data, the data with a cumulative contribution greater than the contribution threshold is selected as the execution prediction data. The execution prediction data is input into the deep reinforcement learning network in the abnormal probability model. At the same time, an adaptive genetic algorithm is used to initialize a policy space for the deep reinforcement learning network. The policy space contains J individuals, each of which represents a possible operation strategy (i.e., the operation action selected by the doctor under a specific state). The policy space is the set of candidate solutions searched by the algorithm. It continuously evolves through operations such as crossover and mutation of the genetic algorithm, combined with the trial-and-error learning of deep reinforcement learning, and ultimately screens out the optimal strategy that adapts to the environment (such as the abnormal probability prediction requirements).

[0021] Each individual is configured with an exploration probability and an update step size; the exploration probability is a hyperparameter unique to each individual, used to balance "using known strategies" and "exploring new strategies" to prevent premature convergence, and the update step size is used to dynamically adjust the parameter update amplitude.

[0022] The reward function is constructed based on the mean square error loss function of the abnormal probability model and the binary classification label. The calculation formula of the reward function is: ; in, Represents the reward function value, which is an indicator used to comprehensively evaluate the model's prediction performance and abnormal probability perception ability. The larger the reward function value, the better the model performance. represents the number of input samples, Indicates the The actual value of the samples, Indicates the The predicted value of the sample, Indicates the The mean square error term of samples is used to quantify the local prediction error, Represents the trade-off coefficient, which is used to adjust the weight ratio of the mean square error term and the abnormal probability term in the reward function. Indicates the probability of abnormal path establishment. close to 1 (high anomaly probability), Approaching 0, it has no significant effect on the reward. Close to 0 (low anomaly probability), A negative number reduces the reward function value, forcing the model to reasonably improve the abnormal probability prediction.

[0023] Set the reward threshold and maximum iteration rounds to perform iterative optimization on the policy space; When the reward function value is higher than the reward threshold or reaches the maximum iteration round, the iteration is stopped and the operation strategy corresponding to the optimal individual is output; The acquired operation strategy is fed back to the deep reinforcement learning network to predict the abnormal probability of establishing vascular access for the sample data.

[0024] Condition monitoring model construction: The condition monitoring model is constructed through an improved spatiotemporal graph convolutional network. The condition data is input into the condition monitoring model for training to obtain the monitoring score of the condition data. Specifically, the following are included: The improved spatiotemporal graph convolutional network includes a time slicing layer, a spatial convolution layer, an abnormal pattern clustering layer, an attention mechanism layer and an output layer; Time-frequency domain features are extracted from the status data, and the extracted time-frequency domain features are multi-dimensionally labeled to generate labeled status data; the multi-dimensional labeling includes first-level labeling, second-level labeling and third-level labeling, the first-level labeling is the complication type labeling, the second-level labeling is the corresponding complication treatment measure labeling, and the third-level labeling is the corresponding complication treatment status abnormality labeling.

[0025] The time-frequency domain features are used to capture the state change rules after the channel is established in the time dimension and frequency dimension, and can more sensitively capture signs of abnormal states.

[0026] The labeled state data is input into the time slicing layer of the improved spatiotemporal graph convolutional network, and the labeled state data is segmented to generate a time series feature matrix; The time series feature matrix is input into the spatial convolution layer of the improved spatiotemporal graph convolutional network to perform preliminary spatial feature extraction and generate initial abnormal features of the state data; the spatial convolution layer converts the input data into more discriminative initial abnormal features based on the dependency relationship of the state data; The initial abnormal features are input into the abnormal pattern clustering layer of the improved spatiotemporal graph convolutional network, and the features belonging to the same abnormal pattern are clustered using the K-means algorithm to form M abnormal pattern feature subspaces; each abnormal pattern feature subspace corresponds to a different abnormal type and can be subdivided by number; The abnormal pattern feature subspace is input into the attention mechanism layer of the improved spatiotemporal graph convolutional network for feature weighted fusion to obtain the abnormal fusion features of the state data; The abnormal fusion features of the state data are input into the output layer of the improved spatiotemporal graph convolutional network, and the abnormal score of each state data is calculated through the Softmax function; The calculation formula of the abnormality score is: ; in, For the The anomaly score of each state data is used to quantify the probability or confidence of the state data being abnormal. The higher the anomaly score, the greater the possibility that the state data is abnormal. Indicates the The weight coefficient of each abnormal type, Represents input data In the The characteristic response value under each abnormal type, is a normalization factor used to normalize the weighted summation result to a reasonable range. The number of exception types.

[0027] Threshold setting: The abnormal probability model and status monitoring model are used to obtain the abnormal probability and monitoring score of each historical sample data. The abnormal probability and monitoring score are analyzed separately using the box plot method. The outlier boundaries are determined based on the calculated quartiles, and the respective outlier boundaries are set as the abnormal probability threshold and monitoring score threshold.

[0028] It is understandable that the box plot is a data visualization tool based on quantiles, which can intuitively display the distribution range, median, quartiles and outliers of the data. In this embodiment, the quartiles are calculated and the outlier boundaries are determined for the abnormal probability and monitoring score respectively, and the abnormal sample data are eliminated based on the outlier boundaries to complete the sample data verification, and its outlier boundaries are set to the range values of reasonable sample data.

[0029] S2: Real-time capture of patient information and operation data to obtain abnormality probability, determine whether the operation data meets the standards, and provide judgment prompts, including: The abnormal probability model is used to obtain the abnormal probability of real-time sample data based on patient information and operation data. Based on the abnormal probability threshold, it is determined whether the patient information and operation data meet the conditions. When the conditions for establishing the pathway are not met, an early warning is issued to medical staff to adjust the operation data.

[0030] It should be noted that the maintenance method of this embodiment autonomously evaluates the conditions for establishing vascular access based on the real-time data circulating in the vascular access intelligent management information system according to the set judgment threshold based on the processing of historical sample data, thereby providing early warning assistance to medical staff, improving the processing efficiency of medical staff and the verification rate of system circulating data.

[0031] S3. Capture status data to obtain monitoring scores. Determine binary labels based on the monitoring scores that meet the standards. Issue the status judgment results after vascular access is established. Specifically, the following are included: The monitoring score of real-time sample data is obtained through the status data through the status monitoring model. The status data is judged to be up to standard based on the monitoring score threshold. When the monitoring score meets the standard, a binary classification label is generated based on the monitoring score.

[0032] It should be noted that the monitoring score is used to evaluate the credibility of the status data and to check whether there are any abnormalities in the collection of status data. Only when the evaluation is passed can the monitoring score be used to complete the classification according to the boundary value. The boundary value is selected through the relationship between the monitoring score and the binary classification label during the model training process.

[0033] This embodiment also provides a maintenance system for the vascular access intelligent management information system, such as Figure 2 As shown, it includes a data verification module, a model building module and an early warning module.

[0034] The data verification module is used to preprocess the historical sample data of the system and verify the sample data through the abnormal probability threshold and monitoring score threshold of the sample data obtained by the abnormal probability model and the state monitoring model.

[0035] Specifically, the abnormal probability threshold and the monitoring score threshold are set by the box plot method based on the abnormal probability and monitoring score of the sample data obtained by the abnormal probability model and the condition monitoring model.

[0036] The model building module is used to build an abnormality probability model (for assessing operational risks) and a status monitoring model (for assessing the health status of vascular access) based on verified historical sample data. The model building is completed by using a machine learning algorithm or a statistical analysis model combined with binary classification labels for supervised learning.

[0037] The early warning reminder module is used to input real-time patient information and operation data into the abnormal probability model, calculate the abnormal probability of the current operation, and trigger an early warning if the abnormal probability exceeds the threshold; and after establishing a vascular access, input the real-time status data into the status monitoring model to generate a monitoring score, and compare it with the threshold to determine whether it meets the standard, and automatically generate a binary classification label for reminder.

[0038] The present invention first constructs an abnormal probability model and a status monitoring model by utilizing the historical sample data of the vascular access intelligent management information system, improves the abnormal probability and monitoring score of all sample data, then uses the box plot method to obtain the abnormal value boundary based on the abnormal probability and monitoring score to complete the verification of the historical sample data, and obtains the abnormal probability threshold and the monitoring score threshold. Then, the abnormal probability is obtained from the real-time data stream collected by the system for each vascular access establishment to make a preliminary judgment, and the operation data of the medical staff is adjusted and auxiliary reminders are provided to improve the efficiency of the medical staff and reduce the error rate of the operation. Finally, after the access is established, the status data is used to independently obtain a two-category label using the status monitoring model, thereby improving the timeliness of the medical staff's judgment on the health status of the vascular access, while maintaining the stability and reliability of the vascular access intelligent management information system.

[0039] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for maintaining a vascular access intelligent management information system, characterized by: Including steps: Process historical sample data and verify sample data, build an abnormal probability model and a condition monitoring model through the verified sample data to obtain abnormal probability thresholds and monitoring score thresholds; Capture patient information and operation data in real time and obtain the abnormal probability of real-time sample data through the abnormal probability model, determine whether the operation data meets the standards, and provide a judgment prompt; Capture status data and obtain monitoring scores of real-time sample data through the status monitoring model. Determine binary classification labels based on the monitoring scores that meet the standards, and issue status judgment results after vascular access is established. The method of obtaining an abnormal probability threshold and a monitoring score threshold comprises the following steps: An abnormality prediction model is constructed based on a deep reinforcement learning algorithm. Patient information, operation data, and binary classification labels are input into the abnormality prediction model for training to obtain the abnormal probability of vascular access establishment. A condition monitoring model is constructed through an improved spatiotemporal graph convolutional network. The condition data is input into the condition monitoring model for training to obtain the monitoring score of the condition data. The abnormal probability model and status monitoring model are used to obtain the abnormal probability and monitoring score of each historical sample data. The abnormal probability and monitoring score are analyzed respectively using the box plot method. The outlier boundaries are determined based on the calculated quartiles, and the respective outlier boundaries are set as the abnormal probability threshold and monitoring score threshold.

2. The maintenance method of a vascular access intelligent management information system according to claim 1, characterized in that: The processing of historical sample data includes the following steps: Data collection and preprocessing: Obtain historical sample data of vascular access establishment, including patient information, operation data, status data, and binary classification labels, and clean, normalize, and time-series align the historical sample data.

3. The maintenance method of a vascular access intelligent management information system according to claim 2, characterized in that: The patient information includes name, gender, age, hospitalization number and diagnosis information; the operation data includes time, operator information, location, number of times, disinfection and dressing materials used; the status data includes blood routine, infusion rate, infusion resistance, external catheter length and body temperature; the two-category label is a record of whether medical staff have any abnormalities in the establishment of vascular access.

4. The maintenance method of a vascular access intelligent management information system according to claim 1, characterized in that: The method of obtaining the abnormal probability of establishing vascular access comprises the steps of: Taking patient information, operation data, and binary classification labels as input, the patient information's identity identifiers are desensitized, and then the independent component analysis algorithm is used to reduce the dimensionality of the input data. By separating the independent non-Gaussian components in the data, the data with cumulative contribution greater than the contribution threshold is screened out as the execution prediction data; The execution prediction data is input into the deep reinforcement learning network in the abnormal probability model, and an adaptive genetic algorithm is used to initialize a policy space for the deep reinforcement learning network. The strategy space contains J individuals, each of which represents an operation strategy, and each individual is configured with an exploration probability and an update step size; Construct a reward function based on the mean squared error loss function of the anomaly probability model and the binary classification labels, set the reward threshold and maximum iteration rounds, and perform iterative optimization on the policy space; When the reward function value is higher than the reward threshold or reaches the maximum iteration round, the iteration is stopped and the operation strategy corresponding to the optimal individual is output; The acquired operation strategy is fed back to the deep reinforcement learning network to predict the abnormal probability of establishing vascular access for the sample data.

5. The maintenance method of a vascular access intelligent management information system according to claim 4, characterized in that: The calculation formula of the reward function is: ； in, represents the reward function value, represents the number of input samples, Indicates the The actual value of the samples, Indicates the The predicted value of the sample, Indicates the The mean square error term of samples, represents the trade-off coefficient, Indicates the probability of abnormal path establishment.

6. The maintenance method of a vascular access intelligent management information system according to claim 1, characterized in that: The monitoring score of obtaining status data includes the following steps: The improved spatiotemporal graph convolutional network includes a time slicing layer, a spatial convolution layer, an abnormal pattern clustering layer, an attention mechanism layer and an output layer; Extract time-frequency domain features from the state data, and perform multi-dimensional annotation on the extracted time-frequency domain features to generate annotated state data; The multi-dimensional annotation includes first-level annotation, second-level annotation and third-level annotation. The first-level annotation is the complication type annotation, the second-level annotation is the treatment measure annotation of the corresponding complication, and the third-level annotation is the abnormal treatment status annotation of the corresponding complication. The time-frequency domain features are used to capture the state change rules after the vascular access is established in the time dimension and frequency dimension, and to capture abnormal states; The labeled state data is input into the time slicing layer of the improved spatiotemporal graph convolutional network, and the labeled state data is segmented to generate a time series feature matrix; The time series feature matrix is input into the spatial convolution layer of the improved spatiotemporal graph convolutional network to perform preliminary spatial feature extraction and generate initial abnormal features of the state data; The spatial convolution layer converts the input data into discriminative initial abnormal features through the dependency relationship of the state data; The initial abnormal features are input into the abnormal pattern clustering layer of the improved spatiotemporal graph convolutional network, and the K-means algorithm is used to cluster the features belonging to the same abnormal pattern to form M abnormal pattern feature subspaces; The abnormal pattern feature subspace is input into the attention mechanism layer of the improved spatiotemporal graph convolutional network for feature weighted fusion to obtain the abnormal fusion features of the state data; The state data anomaly fusion features are input into the output layer of the improved spatiotemporal graph convolutional network, and the anomaly score of each state data is calculated by the Softmax function.

7. The maintenance method of a vascular access intelligent management information system according to claim 6, characterized in that: The calculation formula of the abnormality score is: ； in, For the The abnormality score of the state data, Indicates the The weight coefficient of each abnormal type, Represents input data In the The characteristic response value under each abnormal type, is the normalization factor, The number of exception types.

8. The maintenance method of a vascular access intelligent management information system according to claim 1, characterized in that: The determination of whether the operation data meets the standards includes: The abnormal probability model is used to obtain the abnormal probability of real-time sample data based on patient information and operation data. Based on the abnormal probability threshold, it is determined whether the patient information and operation data meet the conditions. When the conditions for establishing the pathway are not met, an early warning is issued to medical staff to adjust the operation data.

9. The maintenance method of a vascular access intelligent management information system according to claim 1, characterized in that: The status determination result after the vascular access is established includes: The monitoring score of real-time sample data is obtained through the status data through the status monitoring model. The status data is judged to be up to standard based on the monitoring score threshold. When the monitoring score meets the standard, a binary classification label is generated based on the monitoring score.

10. A maintenance system for a vascular access intelligent management information system, using a maintenance method for a vascular access intelligent management information system according to any one of claims 1 to 9, characterized in that: Including data verification module, model building module and early warning module; The data verification module is used to preprocess the historical sample data of the system and verify the sample data through the abnormal probability threshold and monitoring score threshold of the sample data obtained by the abnormal probability model and the state monitoring model; The model building module is used to build an abnormality probability model and a condition monitoring model based on the verified historical sample data, using a machine learning algorithm or a statistical analysis model, combined with binary classification labels for supervised learning to complete the model building; The early warning module is used to input real-time patient information and operation data into the abnormal probability model, calculate the abnormal probability of the current operation, and trigger an early warning if the abnormal probability exceeds a threshold; After establishing the vascular access, the real-time status data is input into the status monitoring model to generate a monitoring score, which is then compared with the threshold to determine whether it meets the standard, and a binary classification label is automatically generated for reminder.

Images