Research On Intelligent Detection Methods For Arrhythmia In Dynamic Electrocardiogram And High-performance Implementation

As one of major diseases which seriously threaten the life and health of human beings,cardiovascular disease(CVD)has a high morbidity and mortality which grows with years.Over 80% of CVD cases are accompanied by cardiac arrhythmias,whose occurrence is considered as a high-risk factor inducing heart diseases and sudden cardiac death.Therefore,research on timely and accurate detection of arrhythmias is of vital importance when early prevention and medical interference for heart diseases and sudden cardiac death is concerned.Electrocardiogram(ECG)is an important tool for diagnosing arrhythmias.Most of present automatic detection methods for arrhythmia were based on ECG characteristic points detection,followed by feature extraction.A major drawback of these methods lies in that their accuracy greatly suffers from ECG signals with comparatively lower signal quality.Meanwhile,defect in these methods' efficiency is also exposed when a long-term ECG signal is analyzed,making them incompetent for supporting demands of mobile health services.Aiming to solve problems mentioned above,in this paper research on automatic detection for arrhythmia and its high-performance implementation is presented.Major contributions include:(1)An automatic detection method for arrhythmia based on artifact cancellation.Having analyzed common arrhythmia detection methods' deficiency in processing noisy ECG data collected by Holter monitoring,this paper introduces an artifact identification process,and proposes an automatic detection method for arrhythmia based on artifact cancellation,which could greatly reduce the risk of incorrect judgement for arrhythmia detection.The experimental results showed that the proposed method could effectively identify premature atrial beats and premature ventricular beats.Especially,the proposed method is superior to other competitive methods in detection of atrial premature beats.The classification performance of the proposed method is not lower than other methods on identification of premature ventricular beats.Furthermore,the arrhythmia classification performance of the proposed method could be more competitive due to its artifact cancellation mechanism.(2)An automatic atrial fibrillation detection method based on deep convolutional neural networks.As a short ECG signal failed to provide abundant rhythm information,most of present ECG automatic analysis methods show poor classification performance in detecting arrhythmias from short-time ECG signals collected by wearable devices.In this paper,a multi-scaled fusion deep convolutional neural networks(MS-CNN)was proposed to provide a solution to atrial fibrillation identification from single-lead short signals.By designing a two-way deep convolution network topology with different scales,the MS-CNN could capture the different characteristics of ECG data,and greatly improve the accuracy of the detection of atrial fibrillation.(3)Parallel automatic arrhythmia detection methods based on graphic processing unit(GPU).Focusing on the excessive computation time cost of present arrhythmia detection methods in the face of long-term ECG data from Holter,this paper proposes parallel automatic arrhythmia detection methods for both GPUs on cloud servers and mobile devices.The more detailed descriptions are following: firstly,time cost analysis of the sequential algorithm is concerned,this paper proposes a parallel automatic arrhythmia detection algorithm running on cloud servers which reorganises the pipeline and parallelizes the time-consuming parts of the sequential algorithm.Meanwhile,based on the concurrent programming paradigm of multiple threads and multiple streams on CPU/GPU,a concurrent parallel automatic detection algorithm for arrhythmia is implemented by introducing a message queue assignment mechanism;secondly,the architecture of the automatic detection algorithm for arrhythmia running on mobile devices is studied.By adjusting workgroup size,data vectorization,zero memory copy and other optimization technologies for mobile parallel computing,a parallel automatic detection algorithm for arrhythmia based on OpenCL framework is proposed,which makes the computing efficiency greatly improved and the energy consumption is greatly reduced.