You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@uima.apache.org by Jukka Zitting <ju...@gmail.com> on 2007/06/01 13:52:11 UTC

Using UIMA for EEG analysis?

Hi,

I'm trying to develop a system for automatically detecting various
types of brain activity based on raw EEG data. I have gigabytes of raw
data that I want to analyze, and I'm wondering if I could use the UIMA
framework in this task.

The high level requirements is that given the raw EEG data the
analysis system should produce a set of annotations that indicate
which parts of the EEG data indicate certain kinds of brain activity
like wake/sleep, REM/non-REM, etc. The typical approach is to use
relative strengths of selected frequency bands for the classification,
but I'm also experimented with self-organizing maps and other
auto-adapting mechanisms in an attempt to increase the accuracy of the
annotations.

So far I've used custom code (both standalone applications and Matlab
plugins) to manage things, but it seems like UIMA would be a nice
framework for handling such operations. I guess I could implement both
the frequency band and more advanced analyzers as UIMA analysis
engines.

Do you think UIMA would be a good match for my needs? Are there any
(public) examples of doing something similar? Good pointers on where I
should start?

BR,

Jukka Zitting

Re: Using UIMA for EEG analysis?

Posted by Eddie Epstein <ea...@gmail.com>.
Hi Jukka,

UIMA has specifically been designed to support analysis of arbitrary
data types. In addition to a flexible type system for creating an
appropriate analysis "language" for analytics to communicate with each
other, UIMA has other features that should be applicable to your
domain. In particular, support for analyzing large artifacts,
including access to remote data and re-segmentation of artifact(s)
into new artifacts that can be passed along to specified analytics,

Other advantages of using UIMA would be to reuse existing UIMA
components for analyzing free text associated with an EEG image, and
for components that make it easy to create search indexes to find
content matches based on the analysis.

The best place to start is to develop a vision for the overall
application you would like to have. For a new analysis domain such as
this, there are not many type system definitions available to reuse,
but designing the data model is generally the first detailed UIMA
design step, even before deciding the modularity of analytics.

Eddie Epstein