Comparing subset selection methods in multi-lead electrocardiogram data

Figures

• 

  Fig. 1. Visual of heartbeats from record I13 of the INCART Database (Tihonenko et al. 2008; Goldberger et al. 2000)

• 

  Fig. 2. Plots of individual ECG beats. a shows a synthetic beat with individual ECG segments labeled in P-QRS-T order. b shows an RR-interval from record I13 of the St. Petersburg INCART 12-Lead Arrhythmia Database (Tihonenko et al. 2008; Goldberger et al. 2000)

• 

  Fig. 3. 12-lead data matrix formation using data from record I07 of the INCART data set (Tihonenko et al. 2008; Goldberger et al. 2000). Each column corresponds to a sample in time from one of 12 simultaneously recorded ECG beats concatenated such that each row contains data from all 12 leads from a single RR-interval

• 

  Fig. 4. VCG representation example from record I07 of the INCART database (Tihonenko et al. 2008; Goldberger et al. 2000)

• 

  Fig. 5. Oversampling results on the train set for ranks \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k = \{1,2,...,20\}$$\end{document} and for each ECG data representation. Subfigures a–d show the results for single-lead, 12-lead, VCG, and VCG magnitude representations, respectively

• 

  Fig. 6. Normalization detection for non-oversampling algorithms on DS1 data representations. Subfigures a–d show the results for single-lead, 12-lead, VCG, and VCG magnitude representations, respectively