基于ERNIE 2.0模型的肺结节病历结构化方法  

STRUCTURED METHOD OF PULMONARY SARCOID MEDICAL RECORDS BASED ON ERNIE 2.0 MODEL

作　　者：毛俊华[1] 钟臻[1] 陶书衡 徐佳鼎 马驰野 Mao Junhua;Zhong Zhen;Tao Shuheng;Xu Jiading;Ma Chiye(Shanghai Hospital of Traditional Chinese Medicine,Shanghai 200071,China;Shanghai Institute of Computing Technology Co.,Ltd.,Shanghai 200040,China)

机构地区：[1]上海市中医医院,上海200071 [2]上海市计算技术研究所有限公司,上海200040

出　　处：《计算机应用与软件》2025年第2期102-110,共9页Computer Applications and Software

摘　　要：采用手工阅读分析肺结节病历的方式容易产生实体遗漏和提取特征效率低下问题。为方便医生做肺结节病历的研究,基于ERNIE 2.0模型,对肺结节病历中有医学研究价值的疾病、异常检测结果、直径等实体进行抽取,处理成结构化文本,便于医生进行相关检索、统计与研究。实验结果表明,该模型具有深度剖析知识增强语义能力,具有更丰富的语料库,相较于固定规则,可以理解相对复杂的语义,有一定的泛化性,效果提升显著,F1值可达94%,优于BiLSTM(Bidirectional Long-Short Term Memory)和BERT模型的结果。The study of pulmonary nodule pathology using manual reading and analysis of pulmonary nodule medical records is prone to case entity omission and inefficient feature extraction.Based on the ERNIE 2.0 model,medical research valuable entities such as diseases,abnormal detection results,and diameters in pulmonary nodule medical records were extracted and processed into structured text for doctors to conduct relevant searches,statistics,and research.The experiments show that the proposed model has the advantages of deep dissection of knowledge to enhance semantic capability and a richer corpus,and can understand relatively complex semantics with certain generalization.With F1 value up to 94%,it has significant improvement in effectiveness than BiLSTM and BERT models.

关 键 词：病历结构化 命名实体识别 ERNIE 深度神经网络 语义理解 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]