Abstracts

INVESTIGATIONS INTO THE ORIGIN OF LONG RANGE TEMPORAL CORRELATIONS IN HUMAN INTRACRANIAL EEG

Abstract number : 1.071
Submission category :
Year : 2004
Submission ID : 4172
Source : www.aesnet.org
Presentation date : 12/2/2004 12:00:00 AM
Published date : Dec 1, 2004, 06:00 AM

Authors :
1Matt Stead, 1Gregory A. Worrell, and 2Brian Litt

We and others have shown that long range temporal correlations (LRTCs) exist in the energy time series of human intracranial EEG. The energy time series can be generated in multiple ways such as squaring, absolute value, or an envelope function with only minor differences in outcome. However on 20 minute segments of EEG the raw signal does not show this characteristic persistence. As LRTCs can arise in multiple ways in a time series, we pursued a series of investigations to reveal the nature of the LRTC in human EEG. Intracranial EEG data were collected from a series of patients at our centers who were implanted with chronic depth electrodes as part of a presurgical evaluation of their seizure disorders. LRTCs were calculated with detrended fluctuation analysis (DFA). Analyses of the energy time series included comparisons of high and low frequency bands; between seizure-onset, and contralateral seizure-remote electrodes; and between sleep, wake, and awake preseizure states. Analyses of raw EEG included 8 to 24 hour segments. Simulation data was generated and analyzed by superimposing regular sinusoidal signals on a background of simulated data with LRTC of known scaling constant. The raw EEG showed a plateau, absence of scaling behavior, beginning at time windows of about one second, but restored scaling at windows greater than one hour. The asymptopic scaling seen in the EEG voltage time series can be recovered from short time windows ([sim]15 minutes) using the energy time series. The simulated data provided a model of the raw EEG with replication of the plateau and restored scaling in the energy time series. We consistently found greater scaling constants in the low frequency bands compared to the high frequency bands, but no significant differences between behavioral states or brain regions studied. Hippocampal EEG recordings demonstrate robust LRTC that extend over hours of recording time. We show that to accurately recover the scaling behavior prolonged continuous raw EEG recordings (hours in duration) are required, and previous studies reporting scaling constants from short time series must be interpeted with caution. However, investigation of the energy time series appears to extract the scaling behavior from more limited time series ([sim] 10 minutes). The pronounced LRTCs in the energy transform of human EEG are present in the raw signal, but are obliterated in windows less than about one hour, possibly due to the regular oscillations generally considered to be the relevant signal. The origin and biophysical role of LRTC in hippocampal neuronal dynamics remains unclear.