For at least 30 years, QRS detectors are published and compared. Yet the reliable detection of the QRS complex does not seem to be solved any time soon. Although many algorithms get published and compared, no clear best algorithm could be determined yet. As previous research showed, the performance often depends on the data set chosen for evaluation, the used quality metrics, and noise characteristics applied to the data. Besides all that, comparing a newly developed algorithm to existing ones needs a lot of time and effort. This hinders the development process of new QRS detectors.
A standardized and easy to use platform is needed to compare different QRS detectors. This platform needs to assess the detector’s performance depending on ECG signal characteristics such as noise or certain heartbeat types and combinations of them. It also has to evaluate the algorithms based on metrics that depict the algorithm’s performance. These metrics need to be displayed such that medical professionals can understand the strengths and weaknesses of QRS detectors to enable the choice of the most fitting QRS detector.
This thesis presents the parts to implement such a standard. The first contribution are metric definitions for both classification and regression metrics that do not have an implicit bias. The second part of the thesis deals with the selection of a data foundation that includes all the different data variants that can occur in practice. Finally, the developed standard was implemented in a prototypical application.