Improved time--frequency analysis of atrial fibrillation signals using spectral modeling

IEEE Trans Biomed Eng. 2008 Dec;55(12):2723-30. doi: 10.1109/TBME.2008.2002158.


In patients with atrial fibrillation (AF), the fibrillatory frequency trend and the time-dependent spectral characteristics can be investigated using a spectral profile technique. The spectral profile is updated by fitting each short-time spectrum. The aim of this study is to develop model-based means for stricter control on the update of the spectral profile. A spectral model defined by a superposition of Gaussian functions is suggested for describing the fundamental and harmonics of the atrial waves during AF, thereby accounting for basic characteristics of the typical AF spectrum. The model parameters are obtained from weighted least squares fitting of the model to the observed spectrum. The method was tested on simulated signals as well as on 48 ECG recordings from 15 patients with persistent AF. Using simulated signals, we assessed the accuracy in terms of magnitude and width of the spectral peaks. For SNR=0 dB, the maximum normalized error was less than 0.2 when estimating magnitude of both the fundamental and the harmonics, whereas it was less than 0.15 for the fundamental and 0.7 for the harmonics with respect to the estimation of the width. We observed a marked improvement while tracking the main fibrillatory frequency as the error was reduced by more than 50% in comparison with the original method. Analyzing ECGs, reliable spectral profiles were obtained in all recordings, even in those cases (5/48) that were not well characterized by the original method.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Atrial Fibrillation / diagnosis
  • Atrial Fibrillation / physiopathology*
  • Data Interpretation, Statistical
  • Electrocardiography / methods*
  • Fourier Analysis
  • Heart Atria / physiopathology
  • Heart Conduction System / physiopathology
  • Humans
  • Least-Squares Analysis
  • Normal Distribution
  • Pattern Recognition, Automated / methods
  • Signal Processing, Computer-Assisted*