Intranet Feed RSS

Andrzejak RG, Schindler K, Rummel C (2012). Nonrandomness, nonlinear dependence, and nonstationarity of electroencephalographic recordings from epilepsy patients. Phys. Rev. E, 86, 046206

The Bern-Barcelona EEG database

Ralph G. Andrzejak, Kaspar Schindler, Christian Rummel
published 16. October 2012, page revised 11. June 2013
 
Overview: This page provides information about the source code, data, and results provided along with the manuscript [1]. If you use any of these resources, please make sure that you cite reference [1]. This will allow other researchers to locate the resources and the corresponding information. Links to the source code, data, and results can be found at the bottom of this page. Please read this page before using any of the resources. We suggest to refer to the resources as Bern-Barcelona EEG database
 
  
Source code: All Matlab source codes are included in the file ASR_Sources_2013_06_11.zip. (link at the bottom of this page) Before you start using the code you should read reference [1] and references therein to understand what the code is doing. To get started with the code you should open ASR_Main.m and read the comments in this code. Afterwards have a look at the files called from ASR_Main.m. Key references are given at the beginning of each source code. These coincide with those already included in reference [1]. The date in the name of the source code file (2013_06_11) indicates when we published this version. We have extracted this code from the one we used to carry out our study. We have tested the code, and to the best of our knowledge it has no bugs. Accordingly, we do not expect that we have to make any corrections to the source code. Nonetheless, we recommend that you check back to this page for possible revisions. In response to questions, we might add further comments to the source code. If you want to be informed about possible future releases, send an email to Ralph Andrzejak. You will then be included in a mailing list.
 
Previous versions of the source code
In the version ASR_Sources_2012_10_16.zip two source codes were missing (ASR_C.m and ASR_AF.m). These were now included in the version ASR_Sources_2013_06_11.zip
 
 
Notation: In the following we use the letters ‘F’ and ‘N’ to refer to focal and non-focal signals, respectively. These letters appear in the filenames and some headers of text files as specified below.
 
Data: Please refer to reference [1] for a detailed description of the acquisition, pre-processing and selection of the data. The entire data set is provided in several compressed zip-files (links at the bottom of this page): 
Data_F_Ind_1_750.zip
Data_F_Ind_751_1500.zip
Data_F_Ind_1501_2250.zip
Data_F_Ind_2251_3000.zip
Data_F_Ind_3001_3750.zip
Data_N_Ind_1_750.zip
Data_N_Ind_751_1500.zip
Data_N_Ind_1501_2250.zip
Data_N_Ind_2251_3000.zip
Data_N_Ind_3001_3750.zip
 
Files with F and N contain focal and non-focal signal pairs, respectively. Each Zip-file contains 750 individual text files. The number in the file name corresponds to the index of the signal pair contained in this file. Each text file contains one individual signal pair. The x-signal is contained in the first column, the y-signal is contained in the second column. The two columns are separated by commas. All files have 10240 rows. Subsequent rows correspond to subsequent samples. The files contain no headers.
 
We also provide a small subset of the recordings containing the first 50 signals only (Data_F_50.zip and Data_N_50.zip). This will allow you to have a quick look at the data before deciding whether you want to download the entire dataset.
 
ResultsResults are given in comma-separated text format. The files Results_F_All.txt and Results_N_All.txt contain the results for all focal and non-focal signal pairs, respectively. The structure of both files is identical. The first row of the files contain the header:
Results_F_All.txt has the header: ‘Index, SF, UxF, UyF, BF’.
Results_N_All.txt has the header: ‘Index, SN, UxN, UyN, BN’.
All subsequent 3750 contain the index of the signal pair and the results of the four hypotheses tests, separated by commas. The results of the hypotheses tests are 0 (test accepted) or 1 (test rejected). The index used in the result files corresponds to the one used in the name of the data files.
 
The first column (header ‘index’) contains the index of the signal pair, running from 1 to 3750. The second column (header ‘SF’ or ‘SN’ ) contains the results of the stationarity test for the pair of signals. The third column (header ‘UxF’ or ‘UxN’) contains the results of the randomness test for the signal x. The fourth column (header ‘UyF’ or ‘UyN’) contains the results of the randomness test for the signal y. The fifth column (header ‘BF’ or ‘BN’) contains the results of the nonlinear-independence test for the pair of signals x and y.
 
Important remark: If you rerun the analysis using our algorithms and our data, you will not get the identical results as provided here. The reason is an inherent stochastic component in these results: All test are based on surrogates. Surrogates are random signals. Each time you generate a surrogate, you get another realization of this random signal. Therefore, when you calculate e.g. the nonlinear prediction error for a set of surrogates, you obtain an independent identically distributed random sample (i.i.d.) from the surrogates’ distribution. If you run the test twice, you will get two independent random samples. Now suppose that the nonlinear prediction error for the original signal is close to the distribution of the surrogates’ nonlinear prediction errors. In that case, the original result might be just inside the surrogates’ results for one set of surrogates, and just outside the surrogates’ distribution for another set of surrogates. Therefore, in some cases you might get a rejection of a test which is listed as accepted in our results. Likewise, in other cases you might get an acceptance of a test which is listed as rejected in our results. Importantly, statistically you will get the same results. All rejection probabilities and conditioned rejection probabilities will be close to the ones provided here. The remaining differences are due to the fluctuations caused by the random nature of the surrogates described in this paragraph.
 
Key wordsWe add the following key words related to reference [1] to help people to find this page. EEG download page, electroencephalogram, epilepsy, intracranial EEG recordings, nonlinear signal analysis, nonlinear time series analysis, free EEG database, nonlinear prediction error source code, surrogate signals, surrogate source code, EEG download page Bonn, electroencephalographic recordings, open Matlab source codes