自然語言處理之緊急救護派遣對話分析：以重大創傷分類為例

Natural Language Processing based Analysis of the Emergency Medical Service Dispatch Communication: A Case Study for Major Trauma

歐芷妍(Chih-Yen Ou)

指導教授 ： 陳柏華

國立臺灣大學/工學院/土木工程學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201904445

摘要

緊急救護派遣為一重要之系統，需在短時間內提供處於生命威脅之病患醫療上之協助。在派遣過程中，派遣員依賴報案人之陳述作救護車派遣決策，因此雙方的溝通是否有效，將影響緊急救護之效能。然而，目前在創傷案件的報案中，派遣員不易從報案人之用詞，分辨患者之創傷嚴重程度。因此本研究提出一自動化方法，辨識報案電話中所述之患者是否為重大創傷。此方法主要由兩部分組成，首先用Deep Speech 2國語語音辨識器，將報案電話內容轉譯成文字，再透過支持向量機（SVM）分類案件。研究結果發現，在車禍案件中，若報案人提及“呼吸”及“快點”，則該案件是重大創傷案件之機會較高。在分類器的表現上，敏感度（sensitivity）以及特異度（specificity）都高於八成，因此透過本研究轉移學習（transfer learning）所訓練出的語音辨識器，以及利用派遣對話內容之文字所訓練出之創傷程度分類器，組成了一有效之自動化嚴重程度辨識系統，具協助實務派遣流程之可能性。

關鍵字

緊急救護服務 ； 緊急救護報案電話 ； 重大創傷 ； 國語語音辨識系統 ； 文字分類器 ； 自動化嚴重程度辨識系統

並列摘要

The Emergency Medical Service (EMS) Dispatch is a critical system which assists patients with life threatening conditions. In emergency calls, the communication between the dispatcher and the caller plays a vital role dispatch decision-making. The dispatcher depends on the caller’s description of the patient’s conditions. However, there is no specific phrases in the current practice that can recognize the severity of trauma cases. This study proposed a methodology for automatic severity recognition to identify major trauma cases through emergency calls. The approach is consist of two processes: a Mandarin speech recognizer, Deep Speech 2, for audio to text transcription; and the Support Vector Machine (SVM) for text classification. This study discovered that if the caller said “呼吸” (breath) or “快點” (hurry), the case may tend to be a major trauma. The sensitivity and specificity of the classifier were both above 80%. Through deep learning for the transcription and the SVM for classification, this study has implemented an effective severity recognition system. The system has the potential to assist the dispatch system in the real world.

並列關鍵字

Emergency Medical Service ； Emergency Call ； Major Trauma ； Mandarin Speech Recognition System ； Text Classification ； Automatic Severity Recognition System

參考文獻

[1] C. D. Deakin, S. England, D. Diffey, and I. Maconochie, "Can ambulance telephone triage using NHS Pathways accurately identify paediatric cardiac arrest?," Resuscitation, vol. 116, pp. 109-112, 2017.

[2] S. Viereck et al., "Recognising out-of-hospital cardiac arrest during emergency calls increases bystander cardiopulmonary resuscitation and survival," Resuscitation, vol. 115, pp. 141-147, 2017.

[3] S. N. Blomberg et al., "Machine learning as a supportive tool to recognize cardiac arrest in emergency calls," Resuscitation, 2019.

[4] S. P. Jones et al., "The identification of acute stroke: an analysis of emergency calls," International Journal of stroke, vol. 8, no. 6, pp. 408-412, 2013.

[5] M. Riou et al., "‘She’s sort of breathing’: What linguistic factors determine call-taker recognition of agonal breathing in emergency calls for cardiac arrest?," Resuscitation, vol. 122, pp. 92-98, 2018.