Abstracts

Dynamical entrainment of short-time Lyapunov exponents (STLmax) has been reported to predict seizures (see, e.g., [1]). However, STLmax and other related methods of estimating Lyapunov exponents are notoriously sensitive to choices of parameters used in their computation, computationally expensive, and difficult to interpret when applied to complex data such as EEG/ECoG. We investigate the relationship between STLmax and power spectral density (PSD) changes in ECoG, since the latter is a well understood measure for quantifying signal characteristics., Data previously analyzed with STLmax in [1] were used in this study and STLmax computation parameters were set to match those used in [1]. We compared STLmax values computed using interictal, ictal, and postictal epochs of ECoG against phase randomized surrogate data. These surrogate signals were constructed to have the same PSD as the ECoG data, while removing all underlying nonlinear dynamical structure., The STLmax values computed from ECoG were indistinguishable from those obtained from surrogate data. Plotting STLmax of ECoG epochs vs. 20 surrogate values resulted in a linear correlation (corr. coeff. = .916; Fig. 1)., STLmax changes, when applied to ECoG, appear to be primarily due to PSD changes in the signal and not to underlying dynamics. This implies that STLmax may be replaced with other measures that are much easier to compute and interpret, such as those based on PSD. Moreover, our findings imply that claims of seizure prediction based upon detecting changes in nonlinear system dynamics (as measured with Lyapunov exponents) should test the ability of the approach in comparison to surrogate data.
[1] Iasemidis et al., Clin. Neurophys. 116 (2005) 532-544.[figure1], (Supported by NIH/NINDS grant #5R01NS046602-02.)