Real-time Ventricular Fibrillation Detection for Pervasive Health Monitorinng

Open Access
Kim, Jungyoon
Graduate Program:
Information Sciences and Technology
Doctor of Philosophy
Document Type:
Date of Defense:
December 17, 2013
Committee Members:
  • Chao Hsien Chu, Dissertation Advisor
  • Chao Hsien Chu, Committee Chair
  • Dongwon Lee, Committee Member
  • Sencun Zhu, Committee Member
  • Kyusun Choi, Committee Member
  • Energy Consumption
  • Ventricular Fibrillation
  • Real-time Detection
  • Wearable Health Monitoring System
  • Factor Analysis
Real-time ECG monitoring has been an attractive method for early detection of unexpected ventricular fibrillation (VF) disease in pervasive health monitoring. Numerous algorithms and wearable devices have been developed and evaluated to meet this emerging need. Most of them were conducted based-on the ideal environment for verifying their practical usage. However, in order to realize the actual application of VF detection in pervasive health monitoring, the VF detection system is required to not only improve the efficiency and effectiveness of the various factors, but also integrate the hardware and software and energy saving. In this dissertation, we explore a new approach to detect VF attacks using VF algorithms processed on an embedded microcontroller, and analyze various factors affecting the detection effectiveness and efficiency. We first examine research issues of VF detection in pervasive health monitoring. We propose a methodology for performing real-time VF detection using an embedded microcontroller to ensure timeliness and save battery power. We develop an integrated environment as a testbed for explorations of this study. As a benchmark, we adapt five light-weighted algorithms and a filtering process from open literature to test the VF detection capability of these algorithms. Evaluations of this research confirm that with some adaptation the conventional filtering process and detection algorithms can be efficiently deployed in a microcontroller with good performance results while saving battery power. Among the five algorithms we considered, the time delay (TD) method has the best overall performance, a relatively short calculation time and reasonably low power consumption. Early research has been predominately focused on developing effective algorithms for VF detection and most of the evaluations are conducted offline with pre-filtered data sets. The performance of these algorithms is highly dependent on several factors such as data filtering methods, data scaling, data extraction window size, detection algorithms and system parameters; therefore, understanding their impact on performance in real-time setting is important. We conduct experiments comparing the performance of five detection algorithms using the popular Creighton University (CU) database and measure their performance in terms of sensitivity, specificity, positive predictivity, accuracy and computational time. We have also examined the performance against an aggregated measure called receiver operating characteristic (ROC) curve. This study shows that (1) detection algorithm, data filtering and window size all have significant impact on VF detection, (2) it is important to select the proper threshold value that affects the trade-off of the performance metrics, and (3) among the five algorithms that we evaluated, Time Delay (TD) outperformed other algorithms, even with different window sizes and different filtering methods and its performance was not much impacted by these factors. Third, we develop an energy consumption model and verify its performance for the integrated VF detection of body area network (BAN). The conventional wireless sensing nodes of the BAN capture the raw electrocardiograph (ECG) signal and transmit the sampled data through wireless communication. This method not only increases wireless data traffic significantly, but also consumes a lot of energy. Since the wireless communication process consumes much more energy than processing in the embedded microcontroller, the processing-based method, which minimizes wireless communication and integrates the light-weight algorithm, is one of the best solutions for the remote VF detection in the BAN. We develop an integrated environment for evaluating the performance of VF detection algorithms and energy consumption simultaneously. As a benchmark, five light-weighted VF detection algorithms and the filtering process are programmed on an embedded microcontroller for the detection of VF abnormality. The total ratio of energy save is 93.68% per 508 seconds data between the real-time full data transmission and the proposed VF detection methodology using time delay (TD) algorithm. Fourth, we propose a new algorithm, called an extended real-time time-delay (ERTD), for real-time VF detection. We conduct experiments comparing the performance of TD and ERTD using the popular Creighton University (CU) database and measure their performance in terms of sensitivity, specificity, positive predictivity, accuracy, algorithm complexity and energy consumption. We have also examined the performance against an aggregated measure called receiver operating characteristic (ROC) curve and the area under curve (AUC). This study shows that (1) ERTD algorithm improves 2.18% of AUC comparing with TD, (2) The energy consumption of ERTD algorithm is improved up 10.56% than using TD in 508 seconds of energy measuring time, and (3) the algorithm complexity of ERTD is much reduced than TD.