Search

CN-122000021-A - Catheter early warning system based on nursing system linkage response

CN122000021ACN 122000021 ACN122000021 ACN 122000021ACN-122000021-A

Abstract

The invention relates to the technical field of medical informatization and intelligent early warning, and discloses a catheter early warning system based on linkage response of a nursing system. The system comprises a core and associated elements for extracting catheter risk records through a data analysis module. The map construction and completion module maps the elements into a catheter care knowledge map, and generates map structure data after completion. The state evolution and calculation module performs depth traversal and coding on the graph data, and outputs a final state vector containing dynamic association information through a state evolution network. The high-association record screening module calculates the matching degree according to the matching degree, and screens out the high-association history record. The early warning decision generation module fuses the historical scheme and the current state vector, and generates a structured early warning containing a specific risk identifier and a linkage operation instruction through the early warning decision network. The system realizes the depth knowledge reasoning and dynamic evolution prediction of catheter risks, and improves the accuracy of early warning and the effectiveness of clinical decision support.

Inventors

  • SU QIANQIAN

Assignees

  • 宁波市第二医院

Dates

Publication Date
20260508
Application Date
20251226

Claims (10)

  1. 1. A catheter alert system based on a linked response of a care system, the system comprising: The data analysis module is used for acquiring and analyzing an original catheter risk record set from the nursing database and extracting core risk elements and associated environment elements; the map construction and complementation module is used for mapping the core risk elements and the associated environment elements into a pre-constructed catheter care knowledge map to generate corresponding initial risk map segments, carrying out knowledge complementation operation on the initial risk map segments to form a complete standard risk map, and further generating a map structure data representation reflecting topological connection relation among the risk elements; The state evolution and calculation module is used for performing depth traversal on the graph structure data representation, capturing multi-hop associated information and encoding the multi-hop associated information into an initial state vector with uniform dimension, inputting the initial state vector into a serialized state evolution network for processing, and outputting a final state vector after evolution; The high-association record screening module calculates the matching degree between different catheter risk records by using the final state vector after evolution, and screens a risk record set with high association degree according to the matching degree; The early warning decision generation module extracts a historical processing scheme text from the risk record set with high association degree, splices the historical processing scheme text with a final state vector of the current catheter risk record, and inputs spliced information to an early warning decision network, and the early warning decision network outputs a structured early warning instruction containing a specific risk identifier and a linkage operation command.
  2. 2. The catheter alert system based on a coordinated response of a care system of claim 1, wherein said parsing the raw catheter risk record set from the care database comprises: Identifying a text description field and a time series data field for each record in the original catheter risk record set; executing natural language segmentation processing on the text description field to obtain a plurality of description fragments; Identifying a named entity for each description fragment, and marking terms belonging to medical events, physical parameters, medical appliances and nursing operations, wherein the terms form the core risk factors; window segmentation is carried out on the time sequence data fields to form a plurality of time window data segments with equal length or unequal length; and extracting statistical features and trend features from each time window data segment, wherein the statistical features and the trend features form the associated environment elements.
  3. 3. The catheter alert system based on a coordinated response of a care system according to claim 1, wherein the performing a knowledge-completion operation on the initial risk profile segment comprises: Inputting the initial risk map segments into a pre-trained knowledge map embedding model to obtain vectorized embedding of each entity and relation in the initial risk map segments; Calculating semantic distances between the vectorized embedding of the initial risk map segments and the vectorized embedding of all sub-graph structures in the catheter care knowledge map; Selecting a target sub-graph structure with the minimum semantic distance from the catheter care knowledge graph; Comparing the initial risk map segment with the target sub-graph structure, and finding out a solid node set and a relation edge set which exist in the target sub-graph structure and are missing in the initial risk map segment; And merging the missing entity node set and the relation edge set into the initial risk map fragment to generate the complete standard risk map.
  4. 4. A catheter alert system based on a coordinated response of a care system according to claim 1 and wherein said depth traversing of said graph structure data representation comprises: Starting from a root node represented by the graph structure data, accessing all directly adjacent nodes by adopting a breadth-first traversal strategy, and recording a first traversal path; for each node accessed in the traversal process, acquiring all attribute information of the node in the catheter care knowledge graph, and converting the attribute information into attribute feature vectors; continuously accessing indirect adjacent nodes in the two-hop and three-hop ranges of each node along the first traversal path, and expanding the association path; Organizing all accessed nodes, attribute feature vectors thereof and relationship types represented by edges connecting the nodes into a node sequence and a relationship sequence with sequences according to the access sequence; and carrying out joint coding on the node sequence and the relation sequence through a sequence encoder to generate the initial state vector with unified dimension.
  5. 5. The system of claim 1, wherein the inputting the initial state vector into the serialized state evolution network for processing comprises: the processing process needs to go through the stages of state initialization, state enhancement, state fusion and state solidification in sequence, In a state initialization stage, the initial state vector is processed by a linear transformation layer and a nonlinear activation function to obtain a basic state vector; In the state enhancement stage, the basic state vector is simultaneously sent into a multi-head attention computing unit and a feedforward neural network, wherein the multi-head attention computing unit captures the dependency relationship between different parts in the basic state vector and outputs an attention enhancement vector, and the feedforward neural network carries out nonlinear transformation on the basic state vector and outputs a transformation enhancement vector; In the state fusion stage, adding the attention enhancement vector and the transformation enhancement vector element by element, and carrying out layer normalization processing on the addition result to obtain a fused state vector; In the state solidification stage, the fused state vector is processed through a gating circulation mechanism, information in the vector is selectively updated and reserved, and the final state vector after evolution is output.
  6. 6. The catheter alert system based on a coordinated response of a care system of claim 5, wherein the calculating a degree of matching between different catheter risk records using the evolving final state vector comprises: From the evolved final state vectors corresponding to all catheter risk records, optionally one is used as a query state vector; Calculating cosine similarity scores between the query state vector and each of the rest of the evolved final state vectors to obtain a series of similarity scores; and inputting all similarity scores into a soft maximum function for normalization processing, and taking the normalized result value as the matching degree.
  7. 7. The catheter alert system based on a coordinated response of a care system according to claim 6, wherein the screening the risk record set with high association according to the matching degree comprises: Presetting a matching degree threshold; Marking all catheter risk records with the matching degree larger than the matching degree threshold value as candidate association records; Sorting the candidate association records according to the matching degree from high to low; And selecting a preset number of candidate associated records ranked at the forefront to form the risk record set with high association degree.
  8. 8. The catheter alert system based on a coordinated response of a care system according to claim 1, wherein the extracting a history handling scenario text from the set of highly correlated risk records comprises: Reading the complete data storage entry of each record in the risk record set with high association degree; locating text fields for recording care interventions and treatment results from the complete data storage entries; Extracting original text content in the text field; And cleaning the original text content, and removing nonsensical symbols, stop words and repeated expressions in the original text content to form the standardized historical processing scheme text.
  9. 9. The catheter alert system based on a coordinated response of a care system according to claim 1, wherein the extracting the historical processing scheme text and stitching it with the final state vector of the current catheter risk record comprises: inputting the history processing scheme text into a text encoder to obtain a text feature vector with a fixed dimension; connecting the text feature vector with the final state vector after evolution of the current catheter risk record in a feature dimension to form an lengthened composite feature vector; And mapping the lengthened composite feature vector to the spliced information matched with the input dimension of the early warning decision network through a dimension reduction projection layer.
  10. 10. The catheter alert system based on a coordinated response of a care system according to claim 1, wherein the alert decision network outputting a structured alert instruction comprising a specific risk identification and a coordinated operation command comprises: the early warning decision network comprises a plurality of full-connection layers and an output layer; The spliced information sequentially passes through the plurality of full-connection layers, each layer is connected with a nonlinear activation function, and the output layer receives the output of the last full-connection layer and generates a multidimensional output vector through linear transformation; and interpreting the multidimensional output vector, mapping the first few dimensions of the vector into specific risk type codes, and mapping the last few dimensions of the vector into specific operation action codes, wherein the risk type codes and the operation action codes jointly form the structured early warning instruction.

Description

Catheter early warning system based on nursing system linkage response Technical Field The invention relates to the technical field of medical informatization and intelligent early warning, in particular to a catheter early warning system based on linkage response of a nursing system. Background In clinical care, effective early warning of catheter related risks is a key link for ensuring patient safety. Existing catheter pre-warning techniques rely mostly on rule matching or simple statistical analysis of structured fields in the care database. Such methods treat each risk record as an isolated event, with processing logic built on top of explicit, direct data correlation, lacking the integrated utilization of complex medical knowledge context behind the risk and indirect correlation factors. The prior art solutions have drawbacks. The model based on rules or shallow statistics cannot be fused with a professional catheter care knowledge system, so that granularity of risk identification is coarse, interpretation is poor, potential or complex risk modes are difficult to find, and the existing method generally ignores the networking topological relation existing between risk elements and the dynamic evolution characteristic of the networking topological relation along with time. The risk is not a static isolated indicator and may conduct and evolve in networks of multiple factors such as patient signs, care procedures, drug interventions, and the like. The modeling capability of the conventional technology on the graph structure association and the time sequence evolution rule is insufficient, so that early warning is delayed or misaligned, and prospective intervention decision cannot be supported. How to deeply structure and dynamically fuse the domain knowledge with real-time data and how to accurately model complex association and state evolution process among risk factors becomes a core problem to be solved for improving the intelligent level of a catheter early warning system. Disclosure of Invention The invention aims to provide a catheter early warning system based on linkage response of a nursing system, so as to solve the problems in the background technology. In order to achieve the above object, the present invention provides a catheter early warning system based on a linkage response of a nursing system, the system comprising: The data analysis module is used for acquiring and analyzing an original catheter risk record set from the nursing database and extracting core risk elements and associated environment elements; the map construction and complementation module is used for mapping the core risk elements and the associated environment elements into a pre-constructed catheter care knowledge map to generate corresponding initial risk map segments, carrying out knowledge complementation operation on the initial risk map segments to form a complete standard risk map, and further generating a map structure data representation reflecting topological connection relation among the risk elements; The state evolution and calculation module is used for performing depth traversal on the graph structure data representation, capturing multi-hop associated information and encoding the multi-hop associated information into an initial state vector with uniform dimension, inputting the initial state vector into a serialized state evolution network for processing, and outputting a final state vector after evolution; The high-association record screening module calculates the matching degree between different catheter risk records by using the final state vector after evolution, and screens a risk record set with high association degree according to the matching degree; The early warning decision generation module extracts a historical processing scheme text from the risk record set with high association degree, splices the historical processing scheme text with a final state vector of the current catheter risk record, and inputs spliced information to an early warning decision network, and the early warning decision network outputs a structured early warning instruction containing a specific risk identifier and a linkage operation command. Preferably, said parsing the original catheter risk record set from the care database comprises: Identifying a text description field and a time series data field for each record in the original catheter risk record set; executing natural language segmentation processing on the text description field to obtain a plurality of description fragments; Identifying a named entity for each description fragment, and marking terms belonging to medical events, physical parameters, medical appliances and nursing operations, wherein the terms form the core risk factors; window segmentation is carried out on the time sequence data fields to form a plurality of time window data segments with equal length or unequal length; and extracting statistical features and trend features from each tim