Abstracts

RATIONALE: A new method for accurate, automated time-frequency-energy (TFE) analysis of signals of arbitrary origin, known as the ITD, provides a means for filtering and decomposition of nonstationary or nonlinear signals that improves upon conventional (Fourier-based) filtering. The ITD decomposes an input signal into individual waves that can be separately analyzed, classified, and reassembled into component signals which have certain desired characteristics. The purpose of this preliminary study was to investigate the usefulness of the ITD method for the decomposition of EEG signals in applications such as seizure detection.
METHODS: Twenty seizures recorded intracranially from 10 subjects (2 per subject; contained within 10 min. recordings) were analyzed. Training segments consisting of the entire seizure (marked by expert visual scoring) and an interictal sample (4 min.) were extracted from the first recording for each subject. The second [dsquote]test[dsquote] segment from each subject was analyzed replacing the wavelet-based filtering step in the FHS algorithm (Osorio et. al., Epilepsia 1998; 39:615) with the nonlinear ITD-based filtering.
RESULTS: All 10 test seizures were detected with the new approach, and the ITD-based filter improved detection signal-to-noise ratio over the generic FHS algorithm. Detection times were comparable between the two methods. Expert visual review of the ITD-based filtering results determined that this new approach significantly improves upon conventional (linear) digital filtering in the decomposition of EEG into seizure and non-seizure components.
CONCLUSIONS: The ITD method for TFE analysis of nonstationary and nonlinear signals (such as the EEG) improves upon conventional filtering approaches for signal decomposition and analysis. In this study, we have demonstrated the utility of this method for seizure detection. Application of ITD to seizure prediction is being investigated and will be discussed.
Support: NIH SBIR [pound]1R43NS39240-01