Statistical Analysis of Electrocardiogram Data

Abstract

In this thesis we focus on statistical analysis of electrocardiogram data. These data record the electrical activity of the heart muscle. The data used in this thesis were provided by Dr. Raimond Wong from Hamilton Regional Cancer Centre (HRCC). The number of independent cases is small (6 cases), but each electrocardiogram contains over 400000 plotting points. Three electrocardiograms came from cancer patients while the other 3 came from healthy volunteers. We conduct statistical analysis in two stages: extraction of feature vectors and clustering analysis of feature vectors. During the first stage, we define 7 statistics that capture important features of the electrocardiogram data. Then these 7 features are separately used in a univariate way to classify the electrocardiogram data into two groups as patients and volunteers. Results show that some of the features can separate the electrocardiogram data well, but others can not do the job well. During the stage of clustering analysis using the 7 features in a multivariate way, we employ three methods of clustering analysis: hierarchical clustering analysis, K-means clustering analysis, and Andrews plot clustering analysis. Results show that hierarchical clustering analysis and K-means clustering analysis misclassify one of the subjects. Andrews plot clustering analysis however successfully classify all the subjects. The first two methods are more objective while the latter requires more judgement. Note that the limited number of independent cases available does not support general conclusions, but our study suggest some potential for the methods discussed.