Electrocardiographic recordings of ventricular fibrillation (VF) appear cha
otic. Previous attempts to characterize the chaotic nature of VF have relie
d on peak-to-peak intervals [Witkowski et al., Phys. Rev. Lett. 1995;75(6):
1230-3; Garfinkel et al., J. Clin. Investig. 1997;99(2):305-314; Hastings e
t al., Proc. Natl. Acad. Sci. USA 1996,93:10495-9], the frequency spectrum
[Goldberger et al., 1986;19:282-289] or other derived measures [Kaplan and
Cohen, Circ. Res. 1990;67:886-92], with results that demonstrate some chara
cteristics of chaos, We have sought to determine whether VF is chaotic rath
er than random and whether the waveform can be described quantitatively usi
ng the tools of fractal geometry. We have constructed an attractor, measure
d the correlation dimensions, estimated the embedding dimension and measure
d Lyapunov exponents. When the digitized waveform is analyzed directly, VF
exhibits nonrandom, chaotic behavior over a decade of sampling frequency. W
ithin the scaling range we have estimated the Hurst exponent, and the self-
similarity dimension of the VF waveform, supporting the presence of chaotic
dynamics. Furthermore, these characteristics are measurable in a porcine m
odel of VF under different recording conditions, and in VF recordings taken
from human subjects immediately prior to defibrillation. Analyses of the H
urst exponents and self-similarity dimensions are correlated with the durat
ion of VF, which may have clinical applications. (C) 2000 Elsevier Science
Ireland Ltd. All rights reserved.