Method and device for improving accuracy of OCR recognition of physical examination report

Abstract

The invention discloses a method and device for improving the accuracy of OCR recognition of a physical examination report, and the method comprises the following steps: report uploading, template recognition, OCR region recognition, semantic analysis recognition, verification and standardization, result output, and outputting the final result of the physical examination report after verification and standardization. The device comprises a report uploading module, a template recognition module, an OCR region recognition module, a semantic analysis recognition module, a verification and standardization module and a result output module. According to the method, the accuracy of physical examination report and invoice identification is improved based on the credibility evaluation function of the surrounding features.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence recognition, and in particular relates to a method and a device for improving the accuracy of OCR recognition of medical examination reports. Background technique [0002] At present, medical examination report identification and invoice identification have become a very important technology and means. This technology can greatly improve the efficiency and automation of claim verification and insurance, and medical examination report information analysis. The difference, all lead to a very low accuracy rate of the final recognition. [0003] Existing OCR extraction technology solutions, such as figure 1 As shown, it is mainly to obtain the characteristics and areas of the physical examination report, obtain the medical examination report template after comparing the source of the format, analyze and locate the extracted text, identify it, verify and standardize the target medical ...

AI Technical Summary

Problems solved by technology

There are the following shortcomings. At present, there are many kinds of medical examination reports, and simple models and algorithms cannot achieve a good accuracy rate. The current general method generally achieves an accuracy rate of about 50%. At present, the recognition of medical examination reports generally does not go through deep learning neural networks. Template discrimination and classification are generally obtained by comparing the categories of templates through simple similarity; the current algorithm cannot solve the problem of matching misplacement of detailed items in the medical examination report, for example, the white blood cell count of blood routine is recognized as urine Routine white blood cell counts, and such errors exist in a large number of physical examination reports; the current algorithm cannot solve the problem of wrong correspondence between items and values ​​in the medical examination report, for example, the white blood cell count is 100, but the actual item and value match wrong

Images