Research On Home Telemonitoring System For The Elderly Based On Internet Of Things And Cloud Platform

In China,aging population continues to intensify,and empty nesters and the elderly living alone are widely concerned by the society.The research and development of home health telemonitoring technologies for the elderly is expected to alleviate the social problem.The development of Internet of Things（Io T）and cloud computing technologies has provided new opportunities for home health telemonitoring of the elderly.However,convenience,comfort,long-term effectiveness,and intelligence are still unmet technical requirements for home health telemonitoring of the elderly.The cutting-edge technologies,such as wearable devices,machine learning and artificial intelligence,can provide potential solutions.Therefore,in this dissertation,the home telemonitoring system for the elderly and the related key technologies based on Io T and cloud platform were studied;the Io T technologies including wearable electrocardiogram（ECG）devices and home gateways,were innovatively integrated with cloud computing technologies;by using machine learning and deep learning technologies,the intelligent diagnosis and classification algorithms for wearable ECG signals were designed;finally,a home telemonitoring system for the elderly was developed.The successful deployment of the designed system is expected to assist doctors and community caregivers in providing convenient,comfortable,long-term and accurate telemonitoring services for the elderly at home,as well as reduce the risk factors of heart disease,and improve home safety for the aged.The main contents and innovations are as follows.1)Research on intelligent method of arrhythmia classification based on machine learning and wearable single-lead ECG signals.A new arrhythmia classification method based on decision tree ensemble was proposed,which can classify single-lead ECG recordings into four classes:normal sinus rhythm,atrial fibrillation,other arrhythmias and noise.The proposed method included three steps:preprocessing,feature extraction and classification.The preprocessing included ECG signal filtering and R-wave position detection.In feature extraction,four groups of 31features were designed for arrhythmia classification,including the feature groups of atrial fibrillation,morphology,RR interval and noise.In classification,100-fold cross validation was used to train the decision tree ensemble classifier based on Ada Boost,and the feature importance and effectiveness were analyzed.Furthermore,by using gradient boosting method,17 key features were selected and the performance of decision tree ensemble classifier was improved.The proposed method achieved an F₁score of 0.82 in the Physio Net/Computing in Cardiology challenge 2017,ranking second with eight other algorithms.The proposed method effectively solved the problem of the performance degradation of the traditional methods on wearable ECG data,and realized the intelligent classification of wearable ECG signals.2)Research on intelligent method for detecting premature beats using deep learning with wearable single-lead ECG signals.A new method of premature beat detection based on wavelet synchrosqueezed transform（WSST）and deep neural network（DNN）was proposed,which can locate positions of premature ventricular beat（PVC）and supraventricular premature beat（SPB）in single-lead ECG signals.Firstly,the raw ECG data were processed using WSST,yielding 178 WSST coefficients per sample.Secondly,the coefficients were used to train the DNN model,which inputed time-series data of different length,and outputed the prediction probability of PVC and SPB for each sampling point.At last,the premature beat positions were calculated based on the predicted probability of the sampling points.In the China Physiological Signals Challenge 2020,the proposed method achieved the scores of PVC_err=-97913 and SPB_err=-95348 on the misdetection evaluation indexes of PVC and SPB,respectively,ranking fourth and second,respectively.The proposed algorithm can be used as a new method for detecting PVC and SPB from long-term wearable single-lead ECG data with high noise level.3)Research on intelligent method for arrhythmia detection from multi-lead ECG signals based on deep learning.A PVC detection DNN model and a AF detection DNN model using three standard limb leads ECG data were proposed.The PVC detection method effectively combined hand-crafted features and deep neural network;hand-crafted features included beat waveform and normalized RR interval;the DNN model of PVC detection was composed of one-dimensional convolution neural networks（1D CNNs）and gated recurrent unit（GRU）networks,which inputed variable-length time-series data of hand-crafted features,and outputed the PVC prediction probability for each beat.The DNN model of AF detection was composed of 1D CNNs,attention network and GRU.The model inputed any length of raw ECG signals and outputed the prediction probability of AF.The ensemble model of AF detection included single-lead models trained with ECG data of standard lead I,II and III,respectively,and three-lead models trained with ECG data of all standard leads.The results showed that the F₁ score of the PVC detection model and the AF detection ensemble model on comprehensive ECG databases were 0.984 and 0.942,respectively.The proposed method only used three standard leads of ECG data and achieved excellent performance on multiple-source databases,which effectively suppressed the problem of performance degradation of the existing methods on cross databases.4)Design of home telemonitoring system for the elderly based on Io T and cloud platform.The system architecture consisted of three parts:the wearable ECG acquisition device,home monitoring terminal and cloud server softwares.The wearable ECG devices included patch-type ECG collectors and wearable smart clothing,which can collect human ECG signals and acceleration signals in real time for a long time.A home gateway hardware and an Android App software were developed.The home gateway and Android phone were used as home monitoring terminals to provide user interface for the elderly and to communicate with cloud platform.Cloud server softwares included four levels:data transmission,data storage,data analysis and data application.Data transmission included web application programming interface,short message service,notification push and message broker.Data storage service included relational database and object storage,and was used to store personal information and ECG data.Data analysis was the deep learning-based arrhythmia detection algorithms deployed in the cloud platform,which can assist doctors in diagnosing heart diseases for the elderly quickly and accurately.Data application included an online doctor diagnosis software and a community real-time monitoring software,which can assist doctors and caregivers in the diagnosis of arrhythmia and real-time ECG and safety event monitoring for the elderly.The experimental results verified the reliability of data acquisition and remote transmission,and the effectiveness of system functions.The research of this dissertation on the home telemonitoring system for the elderly based on Io T and cloud platform,and its key technologies,has important academic significance and practical value for promoting the widespread application of wearable devices,Io T,cloud computing and artificial intelligence technologies in home care for the elderly.