基於自然語言處理與類神經網路之智慧急診檢傷分級系統

Artificial Intelligence Triage System of Emergency Department Based on Natural Language Processing and Neural Network

蔣昱(Chiang, Yu)

指導教授 ： 黃紹華

國立交通大學/電機學院/電信工程系所/碩士(2020年)

摘要

為了解決急診室壅塞及人力分配不均之問題，本論文研發適合急診室使用之智慧檢傷分級系統。此系統藉由輸入急診病人之非結構化資料（主訴）及結構化資料（醫療、生理資訊）判斷病人應被判定之檢傷等級。本研究使用之資料為敏盛醫院急診病歷資料，我們將非結構化資料經過自然語言處理，與結構化資料同時送入神經網路進行判斷。本論文實驗結果顯示，交通大學語音實驗室之繁體中文剖析器與Word2vec詞嵌入在急診主訴文句上有最佳的效果，再經過BiLSTM神經網路能達到0.68的分類準確率。我們也使用同樣的模型架構預測病患經醫師會診後指派的檢驗項目，該模型的F1-score達到0.60。透過以上兩個類神經網路模型，急診室能有效減少護理師人力，並減少醫師尋找檢驗項目及開檢驗單的時間，加速急診的流程。

關鍵字

自然語言處理 ； 遞迴神經網路 ； 人工智慧 ； 檢傷分類 ； 醫療資訊

並列摘要

In order to solve the problems of congestion and uneven distribution of human resource in the emergency department, we propose the smart triage system suitable for the emergency department. This system uses the unstructured data (chief complaint) and structured data (medical and physiological information) of the emergency patient to determine the level of injury that the patient should be judged. The data used in this research is the emergency medical records of Min-Sheng General Hospital. Unstructured data is processed through natural language processing, and merged with structured data, and finally sent to the neural network for judgment. The experimental results of this paper show that NCTU Traditional Chinese Parser and Word2vec word embedding have the best results, the BiLSTM neural network can achieve a classification accuracy of 0.68. Furthermore, we use the same model architecture to predict patient inspections assigned by doctor after consultation. The F1-score of this model reaches 0.60. It can significantly reduce the human resource of nurses, and reduce the time for doctors to find medical order code and assign inspections. All in all, the proposed model speeds up the emergency department efficiency.

並列關鍵字

NLP ； RNN ； AI ； Triage ； Medical Information

參考文獻

[1] 天主教輔仁大學附設醫院， https://www.hospital.fju.edu.tw/Guide?FuncID=ERGPROC

[2] 林增記，2010，May. 12，急診掛號不再是先到先看--五級檢傷分類判別先後順序，Available: http://www.kmuh.org.tw/www/kmcj/data/9912/5.htm

[3] 張靜慧，2010，Oct. 01，康健雜誌，143期，擠到爆！誰來搶救急診室？，Available: https://www.commonhealth.com.tw/article/article.action?nid=62016

[4] 台灣急診檢傷與急迫度分級量表，Available: https://www.taccn.org.tw/upload/site_content_article/74/TTAS98121801.pdf

[5] X. Zhang, J. Kim, R. E. Patzer, S. R. Pitts, A. Patzer, and J. D. Schrager, “Prediction of Emergency Department Hospital Admission Based on Natural Language Processing and Neural Networks,” Methods of Information in Medicine, vol. 56, no. 05, pp. 377–389, 2017.