|
Neuro Traces
(www.neurotraces.com)
|
||
|
|
The whole is more
than
the sum of its parts.
(Aristotle)
|
|
|
A neurophysiological tour of Internet sites
The seed. A very good article
The February issue of the EEG Journal of 2001 includes an article titled
"Epileptic seizures are characterized by signal complexity"
by Bergey and Franaszczuk. The authors try to characterize the complexity of the EEG in epileptic seizures by comparing the amount of "atoms" necessary to generate the EEG signal during different moments of a seizure. The article contains amazing representations of mathematical systems and epileptic seizures. In the figures, the "energy" is distributed among different frequencies located in time by plotting "boxes" in a bidimensional graph. The result is quite fine and, probably, you are interested in knowing more about this procedure. This page describes the process of using Internet to know more about this algorithm. In their article, Bergey and Franaszczuk describe the procedure as following: "In 1993, Mallat and Zhang developed a relatively fast algorithm (matching pursuit) to compute such decompositions". It is a good start point to know more about this method. When we click `related articles' in Pubmed we find another paper of the same group, on which "matching pursuit algorithm" is applied to seizures originating from the mesial temporal lobe (the whole text of both articles and several other very interesting papers can be downloaded from
Epilepsy Research Laboratory
). The reference cited ("Matching pursuit with time-frequency dictionaries") is appropriate to begin our tour.
First of all, we would like to know more about Mallat. So, we try Google and find out that he is one of the key persons in the field of signal processing using wavelets. He wrote a book ("A wavelet tour in Signal Processing") and some contents of his book are accessible in two packages of software: "Wavelab" (a toolbox of Matlab) and Lastwave (a set of routines written in C). The article contains a very technical description that we do not fully understand. It includes some figures of "time frequency energy distributions" and in the abstract we can read: "we introduce an algorithm, called matching pursuit that decomposes any signal into a linear expansion of waveforms chosen in order to best match the signal structures". Perhaps we could try to use the software previously found. In our workplace nobody has access to Matlab but perhaps we could find some demo to understand better the content of the article. The power of GNU.When we search in Google the words "matching pursuit gnu" something interesting appears. One page calls our attention. It includes terms such as "home page", "gnu/linux" and "gtk" . We soon learn that Guimauve is a "graphical user interface to compute matching pursuit decomposition on a signal" based on LastWave (do you remember that Mallat included this software in his page?). Since Lastwave is GPL licensed, it can be adapted to fill different needs. We have a very powerful tool that marks the difference: a distribution of Linux that includes all that we need to download and use Guimauve. We opt for the download of the package in RPM format (so we will not need to compile it) and a few minutes later we have a new package installed and running in our computer. To enter the data, we only need a file in ASCII but we are at home and we would like to experiment with real data. Perhaps some of the results of Bergey and Franaszczuk could be reproduced on a superficial EEG or on a sleep recording. The abundance of PhysionetNow, we need physiological data and there is a site that contains a lot of them: Physionet . When we access to Physiobank we can opt for different formulas to get data. We can download the files (and convert them to ASCII with rdsamp), we can receive data by e-mail, or we can see the data in our browser and save the content as a text file. We get data from "slp01a", a file containing a polysomnographic recording, and are able to check Guimauve. To our surprise, the program indicates which column we want to analyze, so our file in ASCII can be directly processed (once eliminated the first two lines with an editor) and we are able to experiment something with this new algorithm. The result can be found here. In summaryThis long history or any other similar would have been more improbable (even impossible) five years ago. The access to documents indexed by their content, the access to bibliography in a field at which we are not directly involved or the access to raw physiological data would have been much more difficult. Even, concepts about free flow of ideas and methods were less developed. Nowadays, an evening at home affords us to understand more deeply some of the work of the John Hopkins unit of epilepsy .
Jesus Olivan MD
We are very interested in knowing your comments, opinions or experiences about this topic (English, Spanish and French languages can be understood). We will try to answer any question.
|
||
|
|
||