Mar 16, 2019, 2:29:51 PM3/16/19
We have been collecting and annotating data from Temple University
Hospital since 2012. The data goes back to 2002. We have collected over
30,000 sessions to date, and will probably hit 40,000 sessions by the
end of the current project. It is an impressive collection -
particularly given the amount of seizure data we have annotated.
It was always in our plan to acquire data from other hospitals,
particularly data with seizure events. Our experiments thus far have
shown a decent amount of compatibility between data collected at Temple
on Natus equipment, and data collected at Duke on Nihon Kohden
equipment. That is good news, but a more compelling argument would be to
include data from a variety of hospitals.
Over the past 5 years we have had numerous discussions with potential
collaborators on acquiring data from other hospitals. These discussions
always start off very enthusiastically, but, to be brutally honest, we
never close the deal and get the data. Institutional concerns always
seem to get in the way despite the authors' claims that the data will be
released into the public domain. The authors have, as yet, not delivered.
So, for better or for worse, TUH EEG remains the only unencumbered large
database of clinical EEG recordings. Many of you are doing research with
this data and it is having a significant impact. But we need a lot more
data if we are doing to push machine learning beyond its current levels
We are once again going to ask for your help in identifying new sources
of data. It is really important that we broaden our database beyond one
hospital, or the project is likely to come to an end in about 18 months.
So, if you have access to data you are willing to share, or know people
who are interested in providing such data, let's talk.