Recognition And Analysis Of Medical Records Based On Deep Learning

In order to solve the problem of the balance of medical insurance,my country is promoting and implementing the diagnosis-related group(Diagnosis-Related Groups,DRG)medical insurance payment model.DRG is a case classification scheme,which mainly classifies cases by diagnosis based on the homepage of hospitalized medical records,and provides hospitals with medical insurance payment standards based on the classification results.At present,most hospitals only provide DRG with scanned images of the homepage of hospitalized medical records,which brings certain difficulties to DRG research.Therefore,researchers must first extract the key information from the homepage images of hospitalized medical record,and to extract key information,researchers first need to complete efficient and accurate text recognition.There are two specific problems in the text recognition of the homepage images of hospitalized medical record:first,because the homepage images of hospitalized medical record are large and the texts are dense,the general recognition model is difficult to accurately locate the text,which will cause errors in text recognition and losses of text areas;second,the homepage images of hospitalized medical record contains the table.The table area must be located and the key information extraction template of the table must be designed.Otherwise,the diagnosis information in the table and other recognition results will be mixed,which will affect the subsequent DRG grouping.This paper focuses on the above two issues to carry out related research.We improved the existing text detection model and text recognition model based on deep learning,and designed an extraction template for the homepage images of hospitalized medical record of a hospital.Finally,we developed a set of text recognition and analysis system for the homepage images of hospitalized medical records.The main work and results of the paper are as follows:1.By improving the DBNet text detection model based on deep learning,we have improved the accuracy of text detection on the homepage pictures of hospitalized medical records.Specifically,we assigned more weights to pixels in the text area,and adopted a more complex feature fusion mechanism to improve the feature extraction ability of the model for dense texts.Finally the model can accurately detect the position of the text in the homepage image of hospitalized medical record,compared with the DBNet model,the detection accuracy on the 200 test set of homepage images of hospitalized medical record increased by 2.1%to 97.2%.2.By improving the CRNN text recognition model based on deep learning,we have improved the accuracy of text recognition of the pictures on the homepage of hospitalized medical records.Specifically,we introduced a feature fusion mechanism for the CRNN model to improve the ability to extract detailed features of the text,so that the recognition accuracy of the 20,000 single-line text image test set of the homepage of hospitalized medical record increased by 1.5%to 95.8%.3.We specially designed a set of key information extraction templates for the homepage images of a hospital hospitalized medical record,and successfully extracted the table content from the text recognition results automatically and reproduced it in the form of an Excel table.Since the structure and information contained in the hospitalized medical record homepage of different hospitals are not very different,we can use only a slight modification to the template to extract the form information from the homepage images of hospitalized medical record of other hospitals.4.Finally,we designed and developed a set of text recognition and analysis system for the homepage images of hospitalized medical records,and completed the key function test.The test results show that the system can accurately extract the key text information from the homepage images of hospitalized medical record uploaded remotely,and automatically perform DRG grouping according to the extraction results,which provides a certain reference for the hospital to determine the medical insurance payment standard in the future.