DETAILS OF PUBLIC EEG DATABASES
There are three well-known epilepsy datasets reported in literature and online available for research purposes. These datasets include the University of Bonn dataset, the CHB-MIT dataset, and the European Epilepsy dataset. The first two are freely available online and the last dataset needs to be paid for. The detail of these datasets is given below.
University of Bonn Dataset
This EEG database is publically available database provided by the University of Bonn as acquired by Andrzejak et al.35 It comprises five datasets denoted A, B, C, D, and E. Each dataset contains 100 single-channel EEG segments with duration of 23.6 seconds totaling 4097 samples per channel; see Table 4.1 for description. Each dataset is available as a zip file containing 100 TXT files (ASCII code). Datasets were recorded with
Table 4.1 Description of Bonn datasets Datasets File name Description
Datasets |
File name |
Description |
A |
Z001.txt to Z100.txt |
Sets A and B contain surface EEG recordings |
B |
O001.txt to O100.txt |
of five healthy subjects with eyes open (EO) and eyes closed, respectively. |
C |
N001.txt to N100.txt |
Datasets C, D, and E comprise EEG readings of five epileptic patients with sensors at various spatial locations. Dataset C contains EEG recordings from the hippocampal formation in the hemisphere opposite the epileptogenic zone. |
D |
F001.txt to F100.txt |
Dataset D contains EEG recordings of the epileptogenic zone. Both C and D readings were recorded during seizure-free periods. |
E |
S001.txt to S100.txt |
Set E is a collection of epileptic seizure activity recorded from the hippocampal focus. |
128-channel electrodes using an average reference. Pathological activity from datasets C, D, and E and eye movement artifacts had been previously removed. For data acquisition, 12-bit analog-to-digital converters were used with a sampling frequency of 173.61 Hz. Raw EEG data had been passed from a band-pass filter ranging from 0.53 to 40 Hz with a 12 dB/ oct filter roll-off.35 Hence, EEG data is comprised of 5 (classes) X 100 (observations/class) X 4097 (23.6 seconds/observation).