In this thesis we characterized morphology and diagnostic electrocardiogram (ECG) ST-segment feature-vector time series of international reference database LTST DB of 24-hour ambulatory records. To estimate the power of representation of transient ST-segment morphology changes using feature-vector time series of the Karhunen-Loève Transformation (KLT) and of the Transformation based on the Legendre Polynomials (LPT) in comparison to traditional representation of transient ST-segment diagnostic changes, which is based on measuring of the ST-segment level and slope in time domain, we used several metrics which include the residual error during reconstruction of the ST-segment using the first few coefficients of the KLT and LPT transformation, the Pearson and Spearman correlation coefficients between individual coefficients of both transformations and the ST-segment level and slope, and cross-correlation between individual KLT and LPT coefficients. On the basis of the results it is evident that the representation of transient ST-segment morphology changes with feature-vector time series in the LPT transformation space exhibits the highest representational power during intervals of transient ischemic ST-segment episodes in terms of the lowest mean residual errors, highest affinity to traditional representation of transient diagnostic ST-segment changes, and possibility of direct expert insight into clinically relevant categories of transient ST-segment morphology and diagnostic changes. In the scope of our work we also developed a graphic user interface WinECG for visualization and examination of raw ECG signals, average heart beats, feature-vector time series, numerical values of feature vectors, and expert annotations.
|
