摘要
随着医学成像技术的不断进步,有价值的数据越来越多,但是当前还无法充分利用这些数据,从而会造成资源浪费。基于此,笔者将SPECT诊断报告作为研究对象,将数据进行预处理后传入构建的BERT-BiLSTM-CRF框架,旨在实现诊断报告中病灶的自动提取。实验结果表明,该方法能够有效完成核医学诊断文本的疾病表征自动提取,获得的精确度、召回率和F1-Score分别为92.21、94.14和93.17。
With the continuous advancement of medical imaging technology,there are more and more valuable data,but these data cannot be fully utilized at present,which will cause a waste of resources.Based on this,the author takes the SPECT diagnosis report as the research object,and transfers the data into the constructed BERT-BiLSTM-CRF framework after preprocessing,aiming to realize the automatic extraction of the lesions in the diagnosis report.The experimental results show that the method can effectively complete the automatic extraction of disease characterizations of nuclear medicine diagnosis texts,and the obtained accuracy,recall rate and F1-Score are 92.21,94.14 and 93.17,respectively.
作者
张淋均
ZHANG Linjun(Key Laboratory of China's Ethnic Languages and Intelligent Processing of Gansu Province,Northwest Minzu University,Lanzhou Gansu 730000,China;Key Laboratory of Dynamic Streaming Data Computing and Applications,Northwest Minzu University,Lanzhou Gansu 730000,China)
出处
《信息与电脑》
2021年第5期87-89,共3页
Information & Computer
基金
西北民族大学中央高校基本科研业务费专项资金资助研究生项目(项目编号:Yxm2019117)。
