CN-122025185-A - DRG grouping method based on artificial intelligence

CN122025185ACN 122025185 ACN122025185 ACN 122025185ACN-122025185-A

Abstract

The invention discloses a DRG grouping method based on artificial intelligence, which relates to the technical field of medical informatics, and the invention enables text characterization to be more robust to missing key information through the contrast learning training of mask medical records and complete medical records, and semantic features which can be used for grouping can be formed when the text is abbreviated or absent, thereby improving the depicting capability of clinical situations of complex cases; in addition, confidence score is introduced to carry out gating and weighted fusion on the complement features, so that the complement information participates in decision making when being trusted, automatic weight reduction or degeneration is carried out when not being trusted, only the initial features are used, disturbance of noise text on grouping results is reduced, output stability is kept, the mutual exclusion situation of diagnosis and operation combination and obvious contradiction situation of structured fields and text entities are marked and rejection or weight reduction is carried out through a coding consistency and conflict sample processing mechanism, training data distribution is more consistent, and the pulling of abnormal modes on model boundaries is reduced.

Inventors

Gong Yunhan
YU HONGJI
GUO XINGHUA

Assignees

人云控股有限公司

Dates

Publication Date: 20260512
Application Date: 20260224

Claims (10)

1. A DRG grouping method based on artificial intelligence, comprising: Step S1, obtaining structural data and medical record text of a target case; s2, carrying out standardized processing and coding embedding on the structured data, segmenting the medical record text, obtaining text feature vectors through a text encoder, and obtaining initial feature vectors of target cases; s3, inputting the text feature vector into a pre-trained feature completion model, and outputting a completion feature vector and a confidence score, wherein the confidence score is a numerical value representing the confidence degree of the completion feature vector on the semantic information of the current target case; s4, fusing the complement feature vector and the initial feature vector according to the confidence score to obtain a comprehensive feature vector; And S5, inputting the comprehensive feature vector into a pre-trained DRG grouping prediction model, and outputting a DRG grouping result of the target case.
2. The DRG grouping method based on artificial intelligence according to claim 1, wherein the structured data at least comprises a diagnosis code, a surgery or operation code and patient basic information, and the standardized processing comprises converting the diagnosis code and the surgery or operation code into standard codes under a preset coding system, and filling or marking missing values with preset rules.
3. The DRG grouping method based on artificial intelligence according to claim 1, wherein the training data of the feature completion model comprises complete medical record texts of training cases, and wherein corresponding mask medical record texts are generated for each complete medical record text according to a preset mask proportion, and the mask medical record texts are obtained by masking clinical key information in the complete medical record texts.
4. The DRG grouping method based on artificial intelligence according to claim 3, wherein training of the feature completion model comprises the steps of inputting the complete medical record text of the same case and the mask medical record text thereof into a text encoder of a shared parameter respectively to obtain a pair of representation vectors, introducing the representation vectors of different cases as negative samples, and enabling the pair of representation vectors of the same case to be adjacent in a feature space and far away from the negative sample representation vectors by optimizing a contrast loss function.
5. The method of claim 3, wherein the clinical key information comprises at least one of a text segment of symptoms, signs, test indicators and their values, visual conclusions, complications or complications, treatments or treatments, medication information, and a severity of illness description.
6. The DRG grouping method based on artificial intelligence according to claim 1, wherein the fusing comprises calculating feature residuals between the complement feature vectors and the initial feature vectors, wherein the feature residuals are difference vectors between the complement feature vectors and the initial feature vectors after alignment mapping, weighting the feature residuals with the confidence scores as weights, and adding the feature residuals with the initial feature vectors to obtain the comprehensive feature vectors.
7. The DRG grouping method based on artificial intelligence of claim 1, wherein the DRG grouping prediction model is a neural network model, and wherein a cross attention layer is set when processing the integrated feature vector, such that the supplemental feature vector is used to query the initial feature vector and generate an attention weighted representation to output the DRG grouping result.
8. The DRG grouping method based on artificial intelligence according to claim 1, further comprising generating explanatory information after outputting the DRG grouping result, wherein the generating explanatory information comprises determining a plurality of structured fields or text fragments with highest grouping decision contribution degree based on attention weight or gradient contribution degree of the DRG grouping prediction model, and outputting corresponding explanation texts, and the gradient contribution degree is a sensitivity representation of each feature dimension in the integrated feature vector output by the DRG grouping prediction model.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the steps of an artificial intelligence based DRG grouping method as claimed in any one of claims 1 to 8 when the processor executes the computer program.
10. A computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor performs the steps of an artificial intelligence based DRG grouping method as claimed in any one of claims 1to 8.

Description

DRG grouping method based on artificial intelligence Technical Field The invention relates to the technical field of medical informatics, in particular to a DRG grouping method based on artificial intelligence. Background In the current DRG payment practice, the case grouping is usually based on the structural elements in the medical records front page/settlement list, and comprises information of main diagnosis, main operation or operation, age, complications and the like, thus forming a disease group with similar resource consumption, the reliability of the grouping and payment results is obviously influenced by the data integrity, rationality and standardization of reporting codes, expense details and the like, and the related flow generally needs to audit and correct the conditions of core field deletion, inconsistent diagnosis and encoding and the like. Meanwhile, unstructured clinical texts (such as disease course records, discharge nodules and the like) in the electronic medical records contain information with finer granularity on disease evolution, treatment reactions, complication clues and the like, can be used for assisting diagnosis/operation coding, severity identification and resource consumption prediction, so that supplementary evidence is provided for intelligent grouping or grouping interpretation, and part of schemes in the prior art utilize natural language processing and a pre-training language model to learn semantic characterization from clinical texts, so that tasks such as DRG prediction or early prediction are realized. However, in a real clinical scene, the problems of short record, missing key elements or fuzzy expression and the like possibly exist in part of medical record texts due to the influence of factors such as inconsistent workload, recording habit, discharge settlement and medical record filing time points, so that the available text features for modeling are insufficient, and at the moment, the availability of the model for text information is reduced, and prediction is easier to rely on limited structured codes or a small number of fields, so that deviation is generated in the aspects of complication/complication severity identification, resource consumption level depiction and the like, and the difficulty of grouping result interpretation and auditing is increased. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. The invention provides a DRG grouping method based on artificial intelligence, which solves the problems that medical record texts are always short, AI grouping degenerates and depends on a structure code, and real resource consumption is difficult to reflect. In order to solve the technical problems, the invention provides the following technical scheme: In a first aspect, an embodiment of the present invention provides an artificial intelligence based DRG grouping method, which includes: Step S1, obtaining structural data and medical record text of a target case; s2, carrying out standardized processing and coding embedding on the structured data, segmenting the medical record text, obtaining text feature vectors through a text encoder, and obtaining initial feature vectors of target cases; s3, inputting the text feature vector into a pre-trained feature completion model, and outputting a completion feature vector and a confidence score, wherein the confidence score is a numerical value representing the confidence degree of the completion feature vector on the semantic information of the current target case; s4, fusing the complement feature vector and the initial feature vector according to the confidence score to obtain a comprehensive feature vector; And S5, inputting the comprehensive feature vector into a pre-trained DRG grouping prediction model, and outputting a DRG grouping result of the target case. The method for grouping DRG based on artificial intelligence is a preferable scheme, wherein the structured data at least comprises diagnosis codes, operation or operation codes and patient basic information, and the standardized processing comprises the steps of converting the diagnosis codes and the operation or operation codes into standard codes under a preset coding system and filling or marking missing values with preset rules. The DRG grouping method based on the artificial intelligence is characterized in that training data of the feature completion model comprises complete medical record texts of training cases, and corresponding mask medical record texts are generated for each complete medical record text according to a preset mask proportion, wherein the mask medical record texts are obtained by masking clinical key information in the complete medical record texts. The training of the characteristic completion model comprises the steps of respectively inputting the complete medical record text of the same case and the mask medical record text thereof into a text