Subdural electroencephalograms (SEEGs) are of great value in localizing pri
mary epileptogenic regions in patients undergoing evaluation for focal rese
ctive epilepsy surgery. The data segments which contain a transition from i
nterictal to ictal activity carry the most critical diagnostic information.
Computer signal extraction within this transition period represents a chal
lenging signal processing problem. In this work a two-step method is presen
ted to extract early ictal activity. In the first step we employ a nonlinea
r signal decomposition technique in the wavelet domain to separate SEEG dat
a into ictal and background components. In the second step we use time-freq
uency analysis and a novel integration algorithm to extract the desired inf
ormation. Our experiments on clinically recorded data indicate that this me
thod is highly effective allowing us to reveal important hidden features in
the data which could not otherwise be observable. (C) 2001 Biomedical Engi
neering Society.