东北大学学报:自然科学版 ›› 2020, Vol. 41 ›› Issue (7): 938-943.DOI: 10.12068/j.issn.1005-3026.2020.07.005

• 信息与控制 • 上一篇    下一篇

一种改进的医疗文本分类模型:LS-GRU

李强1, 李瑶坤2, 夏书月3, 康雁1,4   

  1. 1.东北大学 医学与生物信息工程学院, 辽宁 沈阳110169; 2.中国石油天然气管道工程有限公司, 河北 廊坊065000;3.沈阳医学院附属中心医院, 辽宁 沈阳110024; 4.深圳技术大学 健康与环境工程学院, 广东 深圳518118)
  • 收稿日期:2019-08-10 修回日期:2019-08-10 出版日期:2020-07-15 发布日期:2020-07-15
  • 通讯作者: 李强
  • 作者简介:李强(1989-),男,山东曲阜人,东北大学博士研究生; 康雁(1964-),男,辽宁沈阳人,东北大学教授,博士生导师.
  • 基金资助:
    国家科技部重点技术研发项目(2017YFC0114200); 国家重点研发计划项目(2018YFC1311900).

An Improved Medical Text Classification Model: LS-GRU

LI Qiang1, LI Yao-kun2, XIA Shu-yue3, KANG Yan1,4   

  1. 1. College of Medicine and Biological Information Engineering, Northeastern University, Shenyang 110169, China; 2. China Petroleum Pipeline Corporation, Langfang 065000, China; 3. The Central Hospital Affiliated to Shenyang Medical, Shenyang 110024, China; 4. College of Health Science and Environmental Engineering, Shenzhen Technology University, Shenzhen 518118, China.
  • Received:2019-08-10 Revised:2019-08-10 Online:2020-07-15 Published:2020-07-15
  • Contact: KANG Yan
  • About author:-
  • Supported by:
    -

摘要: 为了帮助低年资医生阅读胸部CT影像,并更加精确高效地为临床医生反馈影像报告结果,提出一种改进GRU深度学习框架LS-GRU,用来解决影像报告文本分类问题,即可以根据影像科医生描述,自动反馈给临床医生诊断建议.数据来源于呼吸科影像报告1168例,选择了两种描述相近的疾病(肺气肿和肺炎)进行分类,其中肺气肿患者报告大约652例,肺炎约516例.分别验证GRU、BiGRU及LSTM等模型,实验结果表明,LS-GRU模型分类更精确,且具有较高的鲁棒性.

关键词: 深度学习, 医疗文本分类, GRU, 慢阻肺, LSTM

Abstract: In order to help radiologists report the CT image results more accurately and effectively to the clinicians, an improved GRU deep learning framework LS-GRU was proposed to solve the classification of image report text, which can be automatically fed back to clinicians according to radiologists’ descriptions. The data was collected from more than 1168 cases of respiratory imaging reports. Two diseases(emphysema and pneumonia) with similar descriptions of radiologists were classified. About 652 cases of emphysema and 516 cases of pneumonia were reported. The GRU, BiGRU and LSTM models were validated, respectively. The results show that the LS-GRU model is more accurate and robust.

Key words: deep learning, medical text classification, GRU(gate recurrent unit), emphysema, LSTM(long-short term memory)

中图分类号: