QRS feature discrimination capability: quantitative and qualitative analysis

This paper presents the main results obtained from the analysis of features extracted from QRS complexes through the application of a simple methodology developed to quantitatively and qualitatively evaluate such features. A third party tool named tooldiag was used to analyze features extracted from a compact ECG arrhythmia database. Three feature extraction methods were evaluated time domain features extracted directly from QRS samples, QRS decomposition in a basis generated by Principal Components Analysis (PCA) and QRS decomposition in a simplified basis. Classification error estimation has shown features extracted by decomposition of QRS in the PCA generated basis to have the best discrimination capability: their classification error rate was 7% lower than that of features extracted by decomposition in the simplified basis and 33% lower than that of time domain features