The Annual meeting of the American Epilepsy Society began in Seattle on Dec 5th and continues through the 9th.
One of the most exciting developments is a novel project designed to improve the ability to detect and predict seizures using EEG. Working with very large datasets of continuous EEG recorded over 18 months in naturally epileptic dogs (over 900 seizures) and 2 human subjects undergoing pre-surgical evaluation, stored in the cloud, an army of over 200,000 “computer nerds” in over 500 teams from all over the world have entered a competition through Kaggle.com to develop machine learning algorithms to tackle the hard problems of detecting and predicting seizures. This has broad applicability to developing novel treatment approaches to treatment of human epilepsy including responsive neurostimulation, optogenetics and just-in-time drug delivery.
As a result, in a 3 month period, competitors improved accuracy to up to 97.5% for seizure recognition ( 1st prize Michael Hills Australia).
The seizure prediction competition ran from August to November 2014. The top performers were a team from Australia and California who achieved 84% accuracy in predicting onset.
This is really exciting for both the results of the competition but also because of the novel way that open source collaborations have resulted in a rapid advance in this area of research and will serve as a model for interdisciplinary collaboration. The EEG datasets will be available to researchers on iEEG.org for future studies.