光学字符识别对卒中结构化随访表格识别效果的对比研究  

Comparative Study on Recognition Effect of Optical Character Recognition on Structured Follow-Up Form of Stroke

作　　者：王驰[1] 王伟[1] 邱玉发[1] 唐冬梅[1] 舒张 Wang Chi;Wang Wei;Qiu Yu-fa;Tang Dong-mei;Shu Zhang(Stroke Center,Taicang First People's Hospital,Taicang 215400,Jiangsu Province,China)

机构地区：[1]太仓市第一人民医院卒中中心,江苏太仓215400

出　　处：《中国社区医师》2022年第29期164-166,共3页Chinese Community Doctors

基　　金：苏州市青年科技项目(kjxw2018064)。

摘　　要：目的:研究不同光学字符识别(OCR)方案在真实环境中对纸质卒中随访表格的识别效果,探索OCR技术在结构化卒中随访表格电子化中的可行性。方法:收集太仓市第一人民医院2019-2020年社区、乡镇人群心脑血管病危险因素纸质随访表,根据图像采集质量分为正常、角度不佳、光线不佳。文字识别分别采用通用OCR和百度自定义模板文字识别(IOCR),输出文字经人工校对与原始数据对比,计算错误率。结果:通用OCR的正常、角度不佳、光线不佳识别率分别为99.3%、86.1%、97.1%,IOCR的正常、角度不佳、光线不佳识别率分别为99.0%、93.7%、98.1%。通用OCR与IOCR总体识别率比较,差异有统计学意义(P<0.05)。正常、角度不佳、光线不佳图片的通用OCR识别率比较,差异有统计学意义(P<0.05);正常、角度不佳、光线不佳图片的IOCR识别率比较,差异有统计学意义(P<0.05)。结论:结构化表格通过模板构建的IOCR识别方案可以提高识别率,在正常采集角度下能够基本满足临床需求,极大地提高医生的录入效率,形成电子档案,说明图像采集角度是影响识别率的重要因素,为数据的进一步利用提供基础。Objective:To explore the recognition effect of different optical character recognition(OCR)schemes on paper stroke follow-up forms,and to explore the feasibility of OCR technology in the electronic identification of structured stroke follow-up forms.Methods:The paper follow-up forms of community and township population with risk factors for cardiovascular and cerebrovascular diseases in the Taicang First People's Hospital from 2019 to 2020 were collected and divided into normal group,bad angle group and bad light group according to the quality of image acquisition.Text recognition adopted general OCR and Baidu OCR Text Recognition(IOCR).The output text is manually proofread and compared with the original data to calculate the error rate.Results:The recognition rates of normal,bad angle and bad light images were 99.3%,86.1%and 97.1%by the general OCR,and were 99.0%,93.7%and 98.1%by the IOCR,with statistical difference(P<0.05).There was a statistical difference in the overall recognition rate between the general OCR and the IOCR(P<0.05).There was a statistical difference in the recognition rate of normal,bad angle and bad light images by IOCR(P<0.05).Conclusion:Structured forms can improve the recognition rate through the template construction IOCR recognition scheme,basically meet the clinical needs under the normal acquisition angle,greatly improve the input efficiency of the physicians,and formulate electronic archives,suggesting that image acquisition angle is an important factor affecting the recognition rate,which provide the basis for further application of data.

关 键 词：光学字符识别 卒中 结构化表格 

分 类 号：R743.3[医药卫生—神经病学与精神病学] TP391.41[医药卫生—临床医学]