Search

CN-115910359-B - Medical path accurate classification model generation method and system based on case feature labels

CN115910359BCN 115910359 BCN115910359 BCN 115910359BCN-115910359-B

Abstract

The invention discloses a medical path accurate classification model generation method and system based on case feature labels, wherein medical logs are firstly extracted to be structured logs, treatment event sequences of all cases are collected, then cases with similar medical behaviors are aggregated through track clustering, process mining is carried out on all clustering clusters, a process tree model is generated, circulation nodes are optimized, medical paths are extracted from the process tree model, then case feature label judgment is carried out on the extracted medical paths by adopting a trained neural network model cluster, finally, the case feature label judgment and the corresponding medical paths are integrated, and an available medical path accurate classification model is formed. The method combines the technologies of track clustering, process discovery, neural network and the like, fully plays the advantages of each technology, and can achieve the aim of accurately recommending medical paths according to the characteristics of cases, wherein the performance of the obtained model is far higher than that of a model generated by a single technology.

Inventors

  • CAO YONGZHONG
  • XUE JIE
  • LIU ZHIPENG

Assignees

  • 扬州大学

Dates

Publication Date
20260508
Application Date
20221014

Claims (9)

  1. 1. The medical path accurate classification model generation method based on the case feature label is characterized by comprising the following steps of: step 1, extracting a medical log into a structured log, wherein the structured log summarizes all treatment events which are collected and sequenced according to time; step 2, dividing a treatment path set in the medical log into a plurality of subsets according to the similarity of treatment event sequences through track clustering, wherein each subset corresponds to one cluster; step3, each cluster is excavated through a process discovery technology to obtain a plurality of process tree models; step 4, medical path extraction is carried out from each process tree model, and at least one medical path is extracted from each process tree model; step 5, inputting the medical path extracted in the step 4 into a trained neural network model cluster to judge case feature labels, wherein the case feature labels comprise case attributes associated with complications, each neural network model in the neural network model cluster is trained and obtained based on a training data set formed by the treatment event sequence obtained by processing in the step 1 and the selected case feature labels, the neural network model adopts a GRU-Attention model, the model adopts GRU as a neural network unit, and the Attention mechanism is used for further extracting treatment process features, and the extracted medical path is input into the trained GRU-Attention model to judge labels to be expressed as: ; Wherein the method comprises the steps of Is the number of extracted medical paths, m is the number of extracted medical paths, Is the result of the corresponding case feature label, An attention value representing the jth treatment event of the ith medical path, n being the maximum number of treatment events in the medical path; And 6, integrating the case feature label judgment and the corresponding medical path to generate a medical path accurate classification model based on the case feature label.
  2. 2. The medical path accurate classification model generation method based on case feature labels according to claim 1, wherein the specific process of extracting the medical log in step 1 comprises the following steps: step 1-1, counting by taking cases as units, namely summarizing the complete medical process of one medical treatment of a certain patient in a medical log; step 1-2, sorting by taking time as a unit, namely sorting all treatment methods of one case by time; And step 1-3, summarizing according to an XES format, wherein the XES file has a label hierarchical relationship of Log-Trace-Event, all medical logs correspond to Log labels, a certain case corresponds to Trace labels, and the Event corresponds to each treatment Event to obtain a structured Log.
  3. 3. The medical path accurate classification model generation method based on case feature labels according to claim 1, wherein the track clustering specific process in step 2 comprises: Step 2-1, calculating the Levenshtein distance between every two case treatment paths by adopting the Levenshtein distance as an index for measuring the similarity between the case treatment paths; And 2-2, clustering the case treatment paths according to the process similarity by using a Markov clustering algorithm.
  4. 4. The medical path accurate classification model generation method based on case feature labels according to claim 1, wherein the process discovery specific process in step 3 comprises: step 3-1, performing process discovery on each cluster in the step 2; Step 3-2, dividing the direct following graph by using a generalized mining algorithm in four dividing modes of exclusive, sequential, parallel and circulation until the direct following graph cannot be continuously decomposed; And 3-3, representing the decomposed process in a tree form, and generating a process tree model.
  5. 5. The medical path accurate classification model generation method based on case feature labels according to claim 1, wherein the extracting the medical path specific process in step4 comprises: step 4-1, replacing the circulating nodes in each process tree model obtained in the step 3 with parallel nodes; and 4-2, obtaining medical paths with set quantity or set proportion by random walk under the condition of conforming to the node characteristics of the process tree, such as exclusive, sequential and parallel, by utilizing the characteristics of the tree, and filtering the paths obtained by random walk by the intervals of the number of medical events in the set paths to obtain the medical paths meeting the conditions.
  6. 6. The medical path accurate classification model generation method based on case feature labels according to claim 1, wherein the specific process of generating the medical path model in step 6 comprises: Step 6-1, classifying the extracted medical path through GRU-Attention models corresponding to the case feature labels to obtain a plurality of case feature labels and corresponding medical event Attention tables, and combining and splicing to form a triplet Where ct is the case signature label set, p is the medical path, The attention table of the events in the medical path to the labels is that m is the number of the case feature labels in ct, and k is the number of the medical events in p; and 6-2, putting the triplets obtained by modeling all the extracted medical paths into a set to form a medical path accurate classification model based on case feature labels.
  7. 7. The medical path accurate classification model generation method based on case feature labels according to claim 6, wherein step 6 further comprises inputting one or more case feature labels to the medical path accurate classification model, the model outputting the corresponding one or more recommended medical paths and the attention value of each treatment event in the medical path to each label.
  8. 8. Medical path accurate classification model generation system based on case feature label, characterized by comprising: The medical treatment log collecting module is used for collecting medical treatment events according to the time sequence of each collected case; The clustering module is used for dividing the treatment path set in the medical log into a plurality of subsets according to the similarity of the treatment event sequences through track clustering, and each subset corresponds to one cluster; the process discovery module is used for excavating each cluster through a process discovery technology to obtain a plurality of process tree models; The path extraction module is used for extracting medical paths from the process tree models respectively, and each process tree model extracts at least one medical path; The system comprises a path extraction module, a label judgment module, a treatment process feature analysis module and a treatment process feature analysis module, wherein the path extraction module is used for extracting medical paths from the medical path extraction module, the medical path extraction module is used for inputting the medical paths extracted by the path extraction module into a trained neural network model cluster to carry out case feature label judgment, the case feature label comprises case attributes associated with complications, each neural network model in the neural network model cluster is obtained by training a training data set formed by a treatment event sequence obtained by a preprocessing module and a selected case feature label, the neural network model adopts a GRU-Attention model, the GRU is used as a neural network unit by the model, the Attention mechanism is used for extracting treatment process features, and the extracted medical paths are input into the trained GRU-Attention model to carry out label judgment and are expressed as follows: ; Wherein the method comprises the steps of Is the number of extracted medical paths, m is the number of extracted medical paths, Is the result of the corresponding case feature label, An attention value representing the jth treatment event of the ith medical path, n being the maximum number of treatment events in the medical path; and the model generation module is used for integrating the case feature label judgment and the corresponding medical path and generating a medical path accurate classification model based on the case feature label.
  9. 9. The case feature label-based medical path accurate classification model generation system of claim 8, further comprising a medical path recommendation module for inputting one or more case feature labels to the medical path accurate classification model, the model outputting a corresponding one or more recommended medical paths and an attention value for each label for each treatment event in the medical path; the process for generating accurate classification model of medical path comprises classifying the extracted medical path by GRU-Attention model corresponding to multiple case feature labels to obtain multiple case feature labels and corresponding medical event Attention list, and combining and splicing to form triplet Where ct is the case signature label set, p is the medical path, The attention table of the events in the medical path to the labels is that m is the number of the case feature labels in ct, and k is the number of the medical events in p; and putting the triplets obtained by modeling all the extracted medical paths into a set to form a medical path accurate classification model based on the case feature labels.

Description

Medical path accurate classification model generation method and system based on case feature labels Technical Field The invention belongs to the field of process mining and deep learning fusion of medical application, and particularly relates to medical path model construction of complex cases and complications. Background The clinical path is to establish a set of standardized treatment mode and treatment program for a certain disease, is a comprehensive mode related to clinical treatment, and takes evidence-based medical evidence and guidelines as guidance to improve the methods for treating tissues and disease management, and finally has the effects of standardizing medical behaviors, reducing variation, reducing cost and improving quality. It is an important way for medical institutions to provide 'proper, efficient, low-consumption, high-quality' medical services under the condition of limited sanitary resources. However, the clinical route is mainly used for diagnosing single diseases, and when complex cases are encountered, such as other diseases in the body of a patient, complications and complications may occur, the blind treatment scheme adopting the clinical route may have adverse effects, and the treating physician needs to adjust the treatment scheme empirically. The existing massive medical logs are utilized to generate reliable medical paths for complex cases, and the reliable medical paths become current research hotspots. The process mining technology which has been developed in recent years plays a remarkable role in the field of medical paths, and can discover the characteristics of the treatment process from a huge medical log, so that a medical process model is mined, and verification and guidance are performed for determining the treatment scheme for doctors. However, when dealing with medical logs with complex cases, conventional process discovery techniques do not achieve good results. The reason is that even for the same kind of diseases, the treatment paths have larger differences due to different case characteristics, but the traditional process discovery technology does not distinguish the diseases from the diseases, and the diseases are integrated into a process model through unified mining, so that the model is complicated and excessively generalized. In addition, the medical path extracted from the procedure model lacks means of classification and cannot be matched with the corresponding complex case. Disclosure of Invention Aiming at the problems of the existing methods, the invention aims to provide a medical path accurate classification model generation method and system based on case feature labels, which can effectively utilize the existing medical logs to provide corresponding treatment scheme methods for cases with different features and provide a reliable way for the application of complex cases. The medical path accurate classification model generation method based on the case feature labels comprises the following steps: step 1, extracting a medical log into a structured log, wherein the structured log summarizes all treatment events which are collected and sequenced according to time; step 2, dividing a treatment path set in the medical log into a plurality of subsets according to the similarity of treatment event sequences through track clustering, wherein each subset corresponds to one cluster; step3, each cluster is excavated through a process discovery technology to obtain a plurality of process tree models; step 4, medical path extraction is carried out from each process tree model, and at least one medical path is extracted from each process tree model; Step 5, inputting the medical path extracted in the step 4 into a trained neural network model cluster to judge case feature labels, wherein the case feature labels comprise case attributes associated with complications, and each neural network model in the neural network model cluster is obtained by training based on a training data set formed by the treatment event sequence obtained by processing in the step 1 and the selected case feature labels; And 6, integrating the case feature label judgment and the corresponding medical path to generate a medical path accurate classification model based on the case feature label. Further, the specific process of extracting the medical log in the step 1 includes: step 1-1, counting by taking cases as units, namely summarizing the complete medical process of one medical treatment of a certain patient in a medical log; step 1-2, sorting by taking time as a unit, namely sorting all treatment methods of one case by time; And step 1-3, summarizing according to an XES format, wherein the XES file has a label hierarchical relationship of Log-Trace-Event, all medical logs correspond to Log labels, a certain case corresponds to Trace labels, and the Event corresponds to each treatment Event to obtain a structured Log. Further, the specific track clustering process in the step 2