電子病歷在當今醫療系統中已被廣泛使用，成為紀錄病患訊息的主要媒介，電子病歷帶來諸多好處，例如：加速資訊傳播、減少實體儲存所需空間或是加快撰寫速度，但也因為其特性，而造成許多問題，例如：在醫療資訊系統內，複製貼上功能雖然可以加快醫護人員撰寫速度，但長期下來卻會產生冗餘資訊，造成醫護人員閱讀時的阻礙，導致醫療照護品質下降。 使用文本摘要方法，雖然可以有效精簡病歷篇幅，但長期累積下來，摘要後的病歷依然存在冗餘資訊，若能先識別病歷中的新資訊再進行摘要，將可降低冗餘資訊比例。過去研究對於新資訊識別方法多為針對單詞、字串樣式進行處理的文字層級(Word-level)，若能考慮語義層級(Sematic-level)，將能有更好的表現。 本研究使用語義等級(Semantic-level)方法，對文本進行處理，以文本中存在的概念利用二分法及相似度分數法作為判斷標準，進行新資訊識別，再與醫師標註之Gold Standard進行比較，衡量標註效果，最後將新資訊標註結果呈現於醫療決策支援系統中，使醫護人員能快速參考病歷並做出決策。;Electronic health records have been widely used in nowadays’ healthcare system and have become the essential intermedium for keeping the patients’ health records. It has so many advantages, such as accelerating the transmission of data, reducing physical space for storing notes, and bringing efficiency for the healthcare professionals to writing notes. On the other hand, the healthcare professionals can use copy and paste while conducting clinical notes, and it’ll create information redundancy. In the long run, that would be a huge obstacle for healthcare professional to read them and decline the quality of healthcare. Through the ordinary method of text summarization, the length of the medical records can be shortened, but the redundancy still remains. If the new information in each note could be identified preliminary, it’ll help to lower the portion of redundancy. Previous studies of new information identification mainly focus on word-level, if the semantic-level can also be considered, it may yield a better result. The purpose of this paper is using text-mining techniques to identify the relevant new information at semantic-level. We proposed two methods: 1) Concept occurrence and 2) Concept similarity score to annotate new information and then evaluating the performance with gold standards. Finally, visualize the results, make healthcare professionals reading clinical records more efficient, and achieving better decisions.
