Detecting components of an ECG signal for sonification

In recent state-of-the-art electocardiogram (ECG) studies, many authors mention that they had to manually correct automatically detected peaks or exclude artifact-loaded segments from the automatically annotated data they were studying. Given the importance of accurate feature detection for signal analysis, this is clearly a limiting factor. Our investigation into the use of sonification for analysis of ECG data for medical and diagnostic purposes is also hampered by the lack of such a reliable ground truth. In order to be able to undertake a comparative analysis of sonification and numerical techniques, we are investigating ways to improve algorithmic feature detection, particularly more robust algorithms for the detection of important landmarks in the signal in the presence of noise, whilst accounting for the variability in the very nature of the signal. This paper is a work-in-progress report of our efforts to date.