Chapter 12: Multiple Datasets
| (MT) Chapter 11: Timefrequency decomposition
|| Tutorial Outline
||II.Multiple subject processing tutorial|
Processing Multiple Datasets
Processing multiple datasets sequentially and automatically is important for analysing data. While early versions of EEGLAB exclusively relied on command line scripting for processing multiple datasets, some automated processing is now available directly from the EEGLAB graphic user interface (gui). To explore this capability, you must first select several datasets. For this section, we will use data from the STUDY tutorial.
To work on multiple datasets at once, it is first necessary to select them. To do so, use menu item Dataset > Select multiple datasets
As in the screen view above, select a few datasets. Note that the behavior of EEGLAB will differ depending on your optional settings under File > Save memory. If you allow only one dataset to be present in memory at a time (see Memory options for more details), existing datasets will be automatically overwritten on disk. However, if you allow all datasets to be present in memory simultaneously, only the datasets in memory will be overwritten and their copies in disk files will not be affected (you may then select menu item File > Save current dataset(s) to save all the currently selected datasets).
EEGLAB functions available through the EEGLAB menu that can process multiple datasets can be seen in the Tools menu. When there are multiple current datasets, menu items unable to process multiple datasets are disabled. Currently (v6.x-), functions that can process multiple datasets include functions that resample the data, filter the data, re-reference the data, remove channel baselines, and run ICA. If all the datasets have the same channel locations, you can also locate equivalent dipoles for independent components of multiple datasets.
All available tools process data in a similar way. Upon menu selection, a menu window pops up (identical to the single dataset window) in which you may select processing parameters that are then applied to all the datasets. If the dataset copies on disk are overwritten (under the only-one-set-in-memory option), then a warning window will appear. For example, selecting the Tools > Change sampling rate menu item pops up the following interface.
To resample all datasets at 125 Hz, enter 125 then OK. If the current datasets have to be resaved to disk (because at most one dataset can be present in memory), the following warning window appears:
The graphic interface for running ICA is a bit more elaborate. Select menu item Tools > Run ICA. The following window will appear.
The statistical tools in EEGLAB for evaluating STUDY condition, session, and group measure differences assume that datasets for different conditions recorded in the same session share the same ICA decomposition, i.e. the same independent component maps. By default, pop_headplot.m will concatenate STUDY datasets from the same subject and session. For example, you may have several datasets time locked to different classes of events, constituting several experimental conditions per subject, all collected in the same session with the same electrode montage. By default (leaving the lowest check-box checked), pop_headplot.m will perform ICA decomposition on the concatenated data trials from these datasets, and will then attach the same ICA unmixing weights and sphere matrices to each dataset. Information about the datasets selected for concatenation will be provided on the Matlab command line before beginning the decomposition. </font>
In some cases, concatenating epoched datasets representing multiple conditions collected in the same session may involve replicating some portions of the data. For example, the pre-stimulus baseline portion of an epoch time locked to a target stimulus may contain some portion of an epoch time locked to a preceding nontarget stimulus event. Infomax ICA performed by pop_headplot.m and pop_headplot.m does not consider the time order of the data points, but selects time points in random order during training. Thus, replicated data points in concatenated datasets will only tend to be used more often during training. However, this may not bias the forms of the resulting components in other than unusual circumstances.
Some other blind source decomposition algorithms such as pop_headplot.m do consider the time order of points in brief data windows. The version of the pop_headplot.m function used in EEGLAB has been customized to avoid selecting data time windows that straddle an epoch boundary. To apply pop_headplot.m to concatenated datasets, however, the epoch lengths of the datasets are assumed to be equal.
If you wish (and have enough computer RAM), you may also ask pop_headplot.m to load and concatenate all datasets before running ICA. This will concatenate all the datasets in you computer main memory, so you actually need to have enough memory to contain all selected datasets. We do not recommend this approach, since it tacitly (and unreasonably) assumes that the very same set of brain and non-brain source locations and, moreover, orientations, plus the very same electrode montage exist in each session and/or subject.
After ICA decomposition of all selected datasets, you may use menu item File > Create Study > Using all loaded datasets to create a study using all loaded datasets (if you only want to use the dataset you selected, you will have to remove the other datasets from the list of datasets to include in the STUDY). Using a STUDY, you may cluster ICA components across subjects. See the multiple subjects processing tutorial for more details.
Concluding remark on data tutorial
This tutorial only gives a rough idea of the utility of EEGLAB for processing single-trial and averaged EEG or other electrophysiological data, the analyses of the sample dataset(s) presented here are by no way definitive. One should examine the activity and scalp map of each independent component carefully, noting its dynamics and its relationship to behavior, and to other component activities, to begin to understand its role and possible biological function.
Further questions may be more important: How are the activations of pairs of maximally independent components inter-related? How do their relationships evolve across time or change across conditions? Are independent components of different subjects related, and if so, how? EEGLAB is a suitable environment for exploring these and other questions about brain dynamics.
In the next tutorial, we show more about how to import data and events into EEGLAB datasets.