Abstracts

Rationale: The ability to develop epileptic seizure detection and prediction algorithms, test them, and properly compare results to those attained by other researchers and other algorithms requires the availability of a common database. There is very limited availability of such databases and they are expensive to generate. We have developed and published a fully anonymized database of 'full-stay' ECoG recordings from subjects undergoing inpatient intracranial monitoring for epilepsy surgery evaluation at the University of Kansas' Comprehensive Epilepsy Center. Methods: A total of 1437 hr of multi-channel data (154 GB) from ten subjects (per subject mean 143.7 hr, std 53.3) was collected, with Human Subjects Committee approval, using conventional monitoring equipment (BMSI; Los Gatos, CA). The intracranial recordings were sampled at 240 Hz with 10 bits of precision (0.6 μV/bit). The database contains a total of 60 clinical seizures (per subject mean 6.0, std 2.5) and 1174 subclinical seizures (per subject mean 117.4, std 260.8). Electrode locations and event onsets and ends obtained via expert visual review are included. Results: The binary database, along with MATLAB® software to read, display, extract subsets, or convert to other formats are available on a portable USB2 HD to "check out and copy" for simply the cost of round-trip insured postage. See http://www.fhs.lawrence.ks.us/PublicECoG for details. Conclusions: High quality publicly available data are of significant value in the development of methods and devices to better understand, diagnose and treat intractable epilepsy. Development and publication of this database was supported by NIH/NINDS grant no. 3R01NS046602-03S1, and by Flint Hills Scientific, L.L.C. and Medtronic, Inc.