Brain Sci. Issa MF 1 2 , Juhasz Z 1. Blinks and eye movements generate large amplitude peaks that corrupt EEG measurements. Independent component analysis ICA has been used extensively in manual and automatic methods to remove artifacts. By decomposing the signals into neural and artifactual components and artifact components can be eliminated before signal reconstruction. Unfortunately, removing entire components may result in losing important neural information present in the component and eventually may distort the spectral characteristics of the reconstructed signals.
|Published (Last):||8 November 2007|
|PDF File Size:||4.42 Mb|
|ePub File Size:||9.43 Mb|
|Price:||Free* [*Free Regsitration Required]|
That is, instead of a collection of simultaneously recorded single-channel data records, the data are transformed to a collection of simultaneously recorded outputs of spatial filters applied to the whole multi-channel data.
These spatial filters may be designed in many ways for many purposes. In the original scalp channel data, each row of the data recording matrix represents the time course of summed in voltage differences between source projections to one data channel and one or more reference channels thus itself constituting a linear spatial filter. After ICA decomposition, each row of the data activation matrix gives the time course of the activity of one component process spatially filtered from the channel data.
In the case of ICA decomposition, the independent component filters are chosen to produce the maximally temporally independent signals available in the channel data. These are, in effect, information sources in the data whose mixtures, via volume conduction, have been recorded at the scalp channels.
The mixing process for EEG, by volume conduction is passive, linear, and adds no information to the data. On the contrary, it mixes and obscures the functionally distinct and independent source contributions.
These information sources may represent synchronous or partialy synchronous activity within one or possibly more cortical patch es , else activity from non-cortical sources e.
The following example, from Onton and Makeig , shows the diversity of source information typically contained in EEG data, and the striking ability of ICA to separate out these activities from the recorded channel mixtures.
View full-size version of this data image. Fifteen seconds of EEG data at 9 of scalp channels top panel with activities of 9 of independent components ICs, bottom panel. While nearby electrodes upper panel record highly similar mixtures of brain and non-brain activities, ICA component activities lower panel are temporally distinct i. Compare, for example, IC1 and IC3, accounting for different phases of eye blink artifacts produced by this subject after each visual letter presentation grey background and ensuing auditory performance feedback signal colored lines.
Compare, also, IC4 and IC7, which account for overlapping frontal Hz theta band activities appearing during a stretch of correct performance seconds 7 through Typical ECG and EMG artifact ICs are also shown, as well as overlapping posterior Hz alpha band bursts that appear when the subject waits for the next letter presentation white background For comparison, the repeated average visual evoked response of a bilateral occipital IC process IC5 is shown in red on the same relative scale.
Clearly the unaveraged activity dynamics of this IC process are not well summarized by its averaged response, a dramatic illustration of the independence of phase-locked and phase-incoherent activity.
To test this function, simply press OK. We detail each entry of this GUI in detail below. Only runica, which calls runica. To use the fastica algorithm Hyvarinen et al.
Details of how these ICA algorithms work can be found in the scientific papers of the teams that developed them. In general, the physiological significance of any differences in the results or different algorithms or of different parameter choices in the various algorithms have not been tested -- neither by us nor, as far as we know, by anyone else.
Applied to simulated, relatively low dimensional data sets for which all the assumptions of ICA are exactly fulfilled, all three algorithms return near-equivalent components. Note about fastica: Using default parameters, this algorithm quickly computes individual components one by one.
However, the order of the components it finds cannot be known in advance, and performing a complete decomposition is not necessarily faster than Infomax. Thus for practical purposes its name for it should not be taken literally. Also, in our experience it may be less stable than Infomax for high-dimensional data sets.
Very important note: We usually run ICA using many more trials that the sample decomposition presented here. In our experience, the value of k increases as the number of channels increases. However, to find components, it appears that even 30 points per weight is not enough data.
In general, it is important to give ICA as much data as possible for successful training. Can you use too much data? This would only occur when data from radically different EEG states, from different electrode placements, or containing non-stereotypic noise were concatenated, increasing the number of scalp maps associated with independent time courses and forcing ICA to mixture together dissimilar activations into the N output components.
The bottom line is: ICA works best when given a large amount of basically similar and mostly clean data. Supported Systems for binica: To use the optional and much faster binica, which calls binica.
The executable file must also be accessible through the Unix user path variable otherwise binica. Please contact us to obtain the latest source code to compile it on your own system. Channel types: It is possible to select specific channel types or even a list of channel numbers to use for ICA decomposition.
Use the channel editor to define channel types. Initial learning rate will be 0. Learning rate will be multiplied by 0. Online bias adjustment will be used. Removing mean of each channel Final training data range: Starting weights are the identity matrix Sphering the data Beginning ICA training We also recommend the use of collections of short epochs that have been carefully pruned of noisy epochs see Rejecting artifacts with EEGLAB. In the commandline printout, the angledelta is the angle between the direction of the vector in weight space describing the current learning step and the direction describing the previous step.
If, on the other hand, the learning rate were too low, the angle would be near 0 degrees, learning would proceed in small steps in the same direction, and learning would be slow. The default annealing threshold of 60 degrees was arrived at heuristically, and might not be optimum.
Note: the runica Infomax function returns two matrices, a data sphering matrix which is used as a linear preprocessing to ICA and the ICA weight matrix.
For more information, refer to ICA help pages i. If you wish, the resulting decomposition i. This is because PCA specifically makes each successive component account for as much as possible of the remaining activity not accounted for by previously determined components -- while ICA seeks maximally independent sources of activity.
This is useful since, apart from ideally radially oriented dipoles on the cortical surface i. That is, the ordering, scalp topography and activity time courses of best-matching components may appear slightly different. This is because ICA decomposition starts with a random weight matrix and randomly shuffles the data order in each training step , so the convergence is slightly different every time.
Is this a problem? At the least, features of the decomposition that do not remain stable across decompositions of the same data should not be interpreted except as irresolvable ICA uncertainty. Differences between decompositions trained on somewhat different data subsets may have several causes.
We have not yet performed such repeated decompositions and assessed their common features - though this would seem a sound approach. Instead, in our recent work we have looked for commonalities between components resulting from decompositions from different subjects.
It is similar to the window we used for plotting ERP scalp maps. Simply press OK to plot all components. Note: This may take several figures, depending on number of channels and the Plot geometry field parameter. An alternative is to call this functions several times for smaller groups of channels e. The following topoplot. Note that the scale in the following plot uses arbitrary units.
The included function viewprops makes a display like that below, but with component type labels -- and clicking on a component will pop up a window with an expanded set of component property measures, as well as the estimated probabilities of each component being of each type.
The main criteria to determine if a component is 1 cognitively related 2 a muscle artifact or 3 some other type of artifact are, first, the scalp map as shown above , next the component time course, next the component activity power spectrum and, finally given a dataset of event-related data epochs , the erpimage.
In the window above, click on scalp map number 3 to pop up a window showing it alone as mentioned earlier, your decomposition and component ordering might be slightly different.
Note: Usually, the best-fitting sphere is a cm or more above the plane of the nasion and ear canals. From the commandline, topoplot. The headrad value should normally be kept at its physiologically correct value 0. The distance of the electrode positions from the vertex, however, is proportional to their great circle distance on the scalp to the vertex. This keeps the electrodes on the sides of the head from being bunched together as they would be in a top-down view of their positions.
This great-circle projection spreads out the positions of the lower electrodes. In the plot, they appear spread out, whereas in reality they are bunched on the relatively narrow neck surface. The combinations of top-down and great-circle projections allows the full component projection or raw data scalp map to be seen clearly, while allowing the viewer to estimate the actual 3-D locations of plot features. This binica. The function should automatically use the spline file you have generated when plotting ERP 3-D scalp maps.
Select one ore more components below and press OK. You may use the Matlab rotate 3-D option to rotate these headplots with the mouse.
Else, enter a different view angle in the window above. Studying and removing ICA components To study component properties and label components for rejection i. The difference between the resulting figure s and the previous 2-D scalp map plots is that one can here plot the properties of each component by clicking on the rectangular button above each component scalp map.
For example, click on the button labeled 3. This component can be identified as an eye artifact for three reasons: The smoothly decreasing EEG spectrum bottom panel is typical of an eye artifact; The scalp map shows a strong far-frontal projection typical of eye artifacts; And, It is possible to see individual eye movements in the component erpimage. Eye artifacts are nearly always present in EEG datasets.
They are usually in leading positions in the component array because they tend to be big and their scalp topographies if accounting for lateral eye movements look like component 3 or perhaps if accounting for eye blinks like that of component 10 above.
Since this component accounts for eye activity, we may wish to subtract it from the data before further analysis and plotting. Now press OK to go back to the main component property window. Another artifact example in our decomposition is component 32, which appears to be typical muscle artifact component.
This components is spatially localized and show high power at high frequencies Hz and above as shown below. Many other components appear to be brain-related Note: Our sample decomposition used in this tutorial is based on clean EEG data, and may have fewer artifactual components than decompositions of some other datasets. The main criteria for recognizing brain-related components are that they have: Dipole-like scalp maps, Spectral peaks at typical EEG frequence is i.
The component below has a strong alpha band peak near 10 Hz and a scalp map distribution compatible with a left occipital cortex brain source.
Improved EOG Artifact Removal Using Wavelet Enhanced Independent Component Analysis.
Current computers are fast enough to allow easy confirmation and adjustment of suggested rejections by visual inspection, which our eegplot. We therefore favor semi-automated rejection coupled with visual inspection, as detailed below. In practice, we favor applying EEGLAB rejection methods to independent components of the data, using a seven-stage method outlined below. These data are not perfectly suited for illustrating artifact rejection since they have relatively few artifacts! Nevertheless, by setting low thresholds for artifact detection it is possible to find and mark outlier trials. Though these may not necessarily be artifact related, we use them here to illustrate how the EEGLAB data rejection functions work. We encourage users to make their own determinations as to which data to reject and to analyze using the set of EEGLAB rejection tools.
ECG Artifact Removal from Surface EMG Signal Using an Automated Method Based on Wavelet-ICA.
Vogor Next, the components that are determined to be related to the ocular artifacts are projected back to be subtracted from EEG signals, producing the clean EEG data eventually. The results show that the proposed model is effective in removing OAs and meets the requirements rejecton portable systems used for patient monitoring as typified by the OPTIMI project. In the data presented here, the algorithm performed very similar to human experts when those were given both, the topographies of the ICs and their respective activations in a large amount of trials. EEG artifacts reduction by multivariate empirical mode decomposition and multiscale entropy for monitoring depth of anaesthesia during surgery. Finally, we propose an algorithm, which uses eye tracker information to objectively identify eye- artifact related ICA-components ICs in an automated manner.