Characterization of entropy measures against data loss: application to EEG records

Annu Int Conf IEEE Eng Med Biol Soc. 2011:2011:6110-3. doi: 10.1109/IEMBS.2011.6091509.

Abstract

This study is aimed at characterizing three signal entropy measures, Approximate Entropy (ApEn), Sample Entropy (SampEn) and Multiscale Entropy (MSE) over real EEG signals when a number of samples are randomly lost due to, for example, wireless data transmission. The experimental EEG database comprises two main signal groups: control EEGs and epileptic EEGs. Results show that both SampEn and ApEn enable a clear distinction between control and epileptic signals, but SampEn shows a more robust performance over a wide range of sample loss ratios. MSE exhibits a poor behavior for ratios over a 40% of sample loss. The EEG non-stationary and random trends are kept even when a great number of samples are discarded. This behavior is similar for all the records within the same group.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Algorithms*
  • Artifacts*
  • Electroencephalography / methods*
  • Entropy
  • Humans
  • Information Storage and Retrieval / methods*
  • Reproducibility of Results
  • Seizures / diagnosis*
  • Sensitivity and Specificity