A Deep-Learning-Based Smart Healthcare System for Patient's Discomfort Detection at the Edge of Internet of Things
A Deep-Learning-Based Smart Healthcare System for Patient's Discomfort Detection at the Edge of Internet of Things
复制标题
DOI：
10.1109/jiot.2021.3052067
复制
发表时间：
2021-07-01
影响因子：
10.6
通讯作者：
Piccialli, Francesco
中科院分区：
文献类型：
--
作者：
Ahmed, Imran;Jeon, Gwanggil;Piccialli, Francesco
物联网（IoT）广泛支持智能医疗领域;结合计算机视觉、机器和深度学习技术;为自动化患者不适监测/检测系统提供快速准确的服务。传统的病人监护系统通常由可穿戴传感器和基于视觉的方法组成。在这篇文章中，提出并实现了一个基于物联网的无创自动化患者不适监测/检测系统，使用基于深度学习的算法。该系统基于IP摄像机设备;在不使用任何可穿戴设备的情况下检测患者身体的运动和姿势。Mask-RCNN方法用于提取患者身体上的不同关键点。这些检测到的关键点，然后转化为六个主要的身体器官，使用数据挖掘的关联规则。此外，为了分析患者的不适，测量检测到的关键点坐标信息。最后，距离和时间阈值被应用于将运动分类为与正常或不适条件相关联。这些关键点信息还用于确定躺在床上的患者的姿势。连续监测患者的身体位置和姿势，基于此区分舒适和不舒适水平。对于实验评估，记录覆盖两个患者的床的不同视频序列。实验结果表明，该系统的价值，实现了94%的真阳性率和7%的假阳性率。
The Internet of Things (IoT) widely supports the smart healthcare field; combined with computer vision, machine, and deep learning techniques; it provides fast and accurate services for automated patient discomfort monitoring/detection systems. Traditional patient monitoring systems are commonly composed of wearable sensors and vision-based methods. In this article, an IoT-based noninvasive automated patient's discomfort monitoring/detection system is presented and implemented, using a deep-learning-based algorithm. The system is based on an IP camera device; the patient's body's movement and posture are detected without using any wearable devices. The Mask-RCNN method is employed for the extraction of different key points on the patient body. These detected key points are then transformed into six major body organs using association rules of data mining. Furthermore, for analyzing the patient's discomfort, detected key point coordinates information is measured. Finally, the distance and the temporal threshold are applied to classify movements as either associated with normal or discomfort conditions. These key points information is also used to determine the postures of the patient lying on the bed. The patient's body position and posture are continuously monitored, based on which comfort and discomfort level are discriminated. For experimental evaluation, different video sequences are recorded covering two patient's beds. The experimental results show the proposed system's worth by achieving a true-positive rate of 94% and a false-positive rate of 7%.