電子化醫療紀錄之信息萃取方法、應用及效能驗證

Abstract

醫療需求已隨著世界人口的上升而全球性的增加，加重了現有的醫療資源與醫療人力的負擔。隨著人類文明的發展，各種發明逐步的降低人類的勞力負擔並增加人類的生活品質，科技的進步已讓我們過著眾多科技圍繞與輔助的生活，然而醫療照護目前仍屬於高勞力密集的服務，在醫療需求增加的情況下，現有的醫療人力正處於不敷所需的邊界，若我們能如同工業革命般利用機器來輔助人類的醫療服務，或許能舒緩甚至解決醫療系統如此緊繃的情況。本研究專注於使用醫療資料庫內保藏的大量醫療經驗來輔助臨床醫師並期望能增加臨床人員的工作效率，透過建立文字摘要系統加速臨床人員掌握醫療紀錄中的關鍵資訊，並發展出一套模式將病人資訊轉變為電腦可分析的數據向量，其可用於醫療的事件預測、疾病推估、相似病情模式的檢索。隨著深度學習與醫療資訊科技的發展，醫療服務系統必將有嶄新的一面。

Nowadays, technology is almost everywhere and dramatically helps us; however, this penetration is still limited somewhere. Healthcare service is the one, and it may be the time to change. The increasing demand for healthcare is a worldwide issue, which might cause emergency department crowding. Labor-intensive healthcare services have faced high-intensity work pressure, and the issue of occupational burnout has become more prominent. We could take the concept of the Industrial Revolution to use machines to enhance physicians’ ability by leveraging medical experience in electronic health records. This study implemented the medical text summarization method to help people notice the critical words in lengthy records and established the medical concept embedding method to adapt the downstream event prediction and concept retrievals. Our proposed methods achieve the highest performance and have practical clinical applications. Deep learning methods will improve the clinical efficiency of doctors and promote healthcare services to turn a new page. (Portions of this dissertation have been published in two journals, and I have obtained permission from JMIR for the copyright of my article in the journals.)