Date of Award
Thesis - Restricted
Master of Science (MS)
Electrical and Computer Engineering
Automated ECG interpretation has benefited patient monitoring by increasing medical vigilance, which allows clinicians the freedom to care for multiple patients while being acutely aware of the health status for individual patients. However, as healthcare demands increase and technology evolves, clinicians are presented with the challenge of caring for more patients while integrating more data. It is widely recognized that the technology originally designed to help synthesize data, such as ECG interpretation, has not kept pace with these clinical needs. Specifically, false ECG monitoring alarms that were once tolerable are now a burden and threat to patient safety. To counter these consequences, much research has been done to improve monitoring alarms. The work in this thesis focuses on improving ECG monitoring alarms by reducing the number of false positive critical arrhythmia alarms. Critical arrhythmia alarms include abnormal cardiac rhythms that require immediate clinical attention, such as ventricular tachycardia. It has been established that many false ECG alarms are due to signal artifact. The results of this work show that automatic detection of ECG artifact can be achieved in a manner that minimally impacts the detection of critical arrhythmias. In fact one statistical classifier designed in this work provides 100% sensitivity for detection of critical arrhythmias and 46% specificity for ECG artifact on a dataset of 255 ECG samples. Of the 100 classification features evaluated as part of this work, those extracted from a reconstructed signal using wavelet detail coefficients provided the best class separation. Features representing the signal's percent power in frequencies below 1.67 Hz and the correlation across coefficients from the continuous wavelet transform also allowed for class separation. The eclectic nature of these features suggests that information originating from a single domain is inadequate for distinguishing arrhythmic patterns from artifacts. Furthermore, the specificity demonstrates that the classifier may be used to suppress almost half of the false critical arrhythmia alarms. Alternatively, the classifier may be used to communicate signal quality issues to clinicians or rank ECG leads for improved machine-interpretation. Together, these applications work towards reducing false alarms while upholding the sensitivity of current monitoring systems to true arrhythmic events.
Messerges, Joanne L., "Identifying Electrocardiogram Artifact for Improved Clinical Alarms" (2008). Master's Theses (1922-2009) Access restricted to Marquette Campus. 4722.