文献类型：会议论文

英文题名：Automated Clinical Summary Generation via Integrating Structured and Unstructured Data

作者：Fu, Jiaojiao[1];Yang, Bowen[2,3];Guo, Yi[1];Zhou, Yangfan[2,3];Wang, Xin[2,3]

机构：[1]East China Univ Sci & Technol, Sch Informat Sci & Engn, Shanghai, Peoples R China;[2]Fudan Univ, Shanghai Key Lab Intelligent Informat Proc, Shanghai, Peoples R China;[3]Fudan Univ, Sch Comp Sci, Shanghai, Peoples R China

会议论文集：12th CCF Big Data Conference

会议日期：AUG 09-11, 2024

会议地点：Shandong University, Qingdao, PEOPLES R CHINA

主办单位：Shandong University

语种：英文

外文关键词：Automatic text summarization; Clinical data; Data-to-text; Discharge summary

摘要：Automatically generating clinical texts can significantly reduce the time physicians spend on clinical data recording, which is particularly important for developing countries where physicians are extremely busy due to a severe shortage. This work automatically generates discharge summaries as a case to explore the methods and feasibilities of automatic clinical text summarization. Existing work typically uses either structured or unstructured data alone to generate discharge summaries. However, the content generated often has issues such as being overly verbose, lacking focus, or omitting significant information, especially key indicators and medications. This work innovatively proposes a data integration-based clinical text generation approach, using content generated from unstructured clinical data as the basis and supplementing it with text generated from structured clinical data. This study utilizes advanced natural language processing algorithms and models to create clinical texts. It addresses the challenges of lacking datasets suitable for fine-tuning pre-trained models and combining the advantages rather than the disadvantages of both types to produce discharge summaries. Experimental results show that the structured supplementation approach can effectively improve the generation of clinical texts. This work demonstrates that clinical texts generated using existing natural language processing technologies still do not meet the demands of medical practice, pointing out the need to develop further text generation technologies tailored to the characteristics of clinical data.