Independent component analysis (ICA) has been shown to be useful when applied to electroencephalographic (EEG) data. By unmixing the channel recordings into statistically independent component processes, the components can capture anatomically and functionally distinct brain source processes and can also separate out non-brain artifacts in the data. However, using ICA for EEG analysis can seem quite difficult at first. Its effective use requires overcoming two main problems: first, cleaning the data sufficiently, and then correctly interpreting the results.
Here we intend address the second problem by creating an automated independent component classifier. This is not a new effort; a few other toolboxes already attempt this. In this project we are attempting to make two important improvements. The first is to make the labels maximally informative and reliable. The Swartz Center for Computational Neuroscience has accumulated a large amount of EEG data. We are attempting to make use of this data to create a classifier that recognizes several types of components and is reliable enough that its results may be used for automated data processing. We propose to do this by gathering and making use of a large number crowd-sourced component labels and building a maximally reliable model using machine learning methods. We also intend to make an online version for use in the Real-time EEG Source-mapping Toolbox (REST) to support Online Recursive ICA (ORICA).
So how do we need your help? We have more than half a million components from 8,000+ individual recordings. We are labeling a small portion of them, but most of the components will remain unlabeled. Therefore we are trying a crowd-sourced approach ‒ by asking you and the other members of the eeglablist to label as many components as you have time to label. Our machine learning approach will use all the labeled data as well as the still unlabeled data to optimize the classifier. We will release the final product as an EEGLAB plug-in available though the EEGLAB extension manager.