非接觸生命徵象監測的院前急救 輔助系統

Abstract

中文摘要 在緊急事故發生時，報案者常因情緒緊張或缺乏醫療知識而無法完整描述傷患狀況，導致院前急救決策延遲。本研究旨在開發一套智慧化緊急通報與救護輔助系統，以提升院前急救資訊傳遞效率與傷患評估之準確度。系統採用Web App 架構，使民眾可透過行動裝置即時上傳影像、語音及生理訊號等多種資料。影像部分利用YOLOv8-seg模型進行傷口與衣物破損區域辨識；語音資料透過語音轉文字技術並結合大型語言模型分析語音清晰度與患者主訴內容；生理訊號則透過 60 GHz毫米波雷達感測器MR60BHA2進行非接觸式量測，以取得患者之心率與呼吸速率等生命徵象。系統進一步透過Mamdani型模糊推論機制整合影像、語音與生命徵象等多模態資訊，依據傷口嚴重度、語音清晰度、主訴風險以及生命徵象狀態進行傷患評估，並將患者狀態分級為立即復甦、危急、緊急、次緊急與非緊急等五種等級。當系統判定為高危險等級時，將即時通知附近具急救資格之人員並同步將相關資訊傳送至救護車端，以協助院前急救人員提早掌握患者狀況並進行醫療資源調度。研究結果顯示，本系統能於事故現場快速取得結構化傷患資訊，並透過模糊推論進行即時傷患評估，具有提升院前急救效率與降低救援延誤風險之潛在應用價值。 Abstract In emergency accidents, callers often cannot clearly describe the condition of injured patients due to panic or lack of medical knowledge. This situation may delay decision-making in prehospital emergency care. Therefore, this study aims to develop an intelligent emergency reporting and rescue assistance system to improve the efficiency of information transmission and the accuracy of injury assessment in prehospital emergency situations. The proposed system is implemented as a Web App architecture, allowing users to upload images, voice recordings, and physiological data through mobile devices in real time. For image analysis, the YOLOv8-seg model is used to detect wounds and damaged clothing regions. Voice data are processed using speech-to-text technology combined with a large language model to analyze speech clarity and patient complaints. Physiological data are obtained through a 60 GHz millimeter-wave radar sensor (MR60BHA2), which performs non-contact measurement to capture vital signs, including heart rate and respiratory rate. The system further integrates multimodal information such as images, speech, and vital signs using a Mamdani-type fuzzy inference mechanism. Based on wound severity, speech clarity, complaint risk level, and vital sign conditions, the system performs injury assessment and classifies patient conditions into five levels: resuscitation, critical, urgent, less urgent, and non-urgent. When a high-risk condition is detected, the system immediately notifies nearby trained responders and simultaneously sends relevant information to ambulance personnel. This enables emergency responders to understand the patient’s condition earlier and perform better medical resource allocation. The results indicate that the proposed system can rapidly collect structured patient information at accident scenes and support real-time injury assessment through fuzzy inference. The system has potential to improve the efficiency of prehospital emergency care and reduce delays in emergency response.