In order to develop techniques to predict cardiac arrest, long-term study of electrocardiogram (ECG) data needs to be done to detect changes in electrical activity of diseased hearts. In the past, limitations of computing power and storage space restricted the duration of long-term studies to several days. However, with today's technological advancement, data collection can be extended to months or years. The goal of this thesis research is to evaluate several alternatives for distributing the analysis of ECG data over multiple processors. Parallel algorithms utilizing Correlation Waveform Analysis (CWA) were implemented to compare individual heartbeats and form heartbeat templates. The purpose of the templates is to exhibit the different heartbeat morphologies encountered in the data. The processing is done on a Linux Beowulf Cluster using the standardized Message Passing Interface (MPI) libraries. In the thesis, the results of four different parallel approaches are compared, and their performance is evaluated.
展开▼
