Abstracts

RATIONALE: There is convincing evidence indicating that use of nonlinear dynamics in human intracranial EEG records yields a decrease of system complexity during seizures. Such a decrease is reminiscent of deterministic chaotic attractor behavior. Although descriptors derived from the correlation integral show ictal changes that parallel classical EEG interpretation they may also identify such changes in EEGs deemed interictal by human experts. Others have claimed identification of such changes occurring up to tens of minutes prior to electroclinical seizure onset which may be of predictive value. We investigated whether such changes are reliable. METHODS: We analyzed 10 intracranial EEG records obtained during preoperative evaluation in patients suffering from intractable epilepsy. We ran a 20-sec-long sliding window throughout each record starting at 1 hr prior to electroclinical seizure onset and terminating during the seizure. We calculated Dcg, the coarse-grained correlation dimension as earlier reported by us (Van der Heyden et al., Clin Neurophysiol 1999;110:1726-40) for 17 embedding dimensions, defined a saturability index (DcgS), and obtained a plateauing value at DcgS. We compared the values for Dcg, DcgS, and plateauing against the corresponding ones from phase-randomized surrogates of the same signals and the judgment of an expert EEGer. RESULTS: Consistent low DcgS values with plateauing were identified only during either electroclinical or electrographic seizures of several seconds' duration in all records. Dcg identified spurious low values caused by either high amplitude sharp transients or artifacts. DcgS and the expert EEGer rejected such false positives. CONCLUSIONS: DcgS increases the reliability of Dcg calculations by eliminating false positive detection of low-dimensional chaotic attractor-like behavior in the interictal EEG record without loss of detection power. DcgS parallels the human expert's interpretation of the transition to ictal EEG better than Dcg does. DcgS is the first of a series of measures we are developing to increase Dcg reliability in epilepsy.