Ocular Correction ICA
Contents
1. Downloads amp Support Ocular Correction ICA by Dr Markus Plank EEG recordings are often compromised by noise resulting from muscular activity environmental disturbances blinks saccades smooth pursuit and other eye movements Electro Ocular EOG artifacts are particularly severe since eye movements can hardly be suppressed over a sustained period of time In order to overcome the obvious problem of losing trials when rejecting data portions contaminated by EOG artifacts several methods have been proposed to attenuate EOG processes and consequently correct the contaminated EEG data one of which is Independent Component Analysis ICA Makeig Bell Jung amp Sejnowski 1996 ICA separates the EEG signal mixtures recorded at the scalp into temporally maximally independent component time courses allowing for the removal of artifactual processes While the selection of artifactual components is subjective and up to the user Jung and colleagues 1998 showed that ocular components representing eye movements and blinks have characteristic patterns in both their time courses as well as their topographies which can be used as selection Interestingly ICA based attenuation of ocular artifacts succeeds even in the absence of designated criteria Figure 1 eye channels This is a major advantage over regression based algorithms for example of Gratton Coles and Donchin 1983 A amplitude time ap O 5 ro2. G calculates the contribution of each component to the Global Field Power of all channels used for ICA during blink intervals Here both the activation profiles as well as the channel projections of the components are evaluated This option does not require the VEOG channel to be fed into ICA For all methods the components will be ranked according to their score with respect to the selected ocular measure Afterwards component scores are compared to the threshold specified in the field Total Value to Delete Depending on this threshold one or more components will be pre selected as ocular and subsequently marked for removal in red in the semi automatic interactive view of the transformation 4 Removal of ocular components Analyzer will then open the component activations from step 2 in an interactive view where ocular components selected for rejection are marked in red The interactive view allows for an evaluation and further fine tuning of the automatic component selection accomplished in step 3 After clicking Finish the marked components will be removed from the data You can utilize various visualization tools in the interactive view in order to make informed decisions In the following will explain a couple of these in more detail 1 2 The Interactive View Prior to examining the contents of the interactive view recommend checking the Operation Infos of the newly created Ocular Correction
3. ICA node in particular whether the algorithm has converged Only in this case the ICA decomposition and the displayed data are valid and can be interpreted As can be seen in Figure 3 the interactive view displays the component FOO 200 pV F01 200 pV Figure 3 A Component activations scaled by a La 200 pV F24 200 pV Pde TEESE I4 factor of 75 of components Foo Fo1 F11 and F24 as well as upper and lower VEOG channel activity Blinks are represented by interval markers highlighted in blue B Table containing the criterion score of all components for VEOG and HEOG activity as well as a color code on whether the component will be removed red or retained F00 green In the current example components have P been sorted based on their VEOG criterion score Veogu aaa C Topography of the component highlighted in 200 pV Sige made D the criterion table The map represents the iie F normed component projections towards the m channels unit less weights units can be neglected VeogL _ D Display settings In this part of the interactive 200 pV C ia view you can adjust the information displayed Blink S 21 Blink S 2 Blink www brainproducts com in the main window for example the component scaling with respect to the channel amplitudes page 2of4 activations labeled by F followed by a number in the main view In the current example inte
4. nchin E 1983 A new method for off line removal of ocular artifact Electroencephalogr Clin Neurophysiol 55 4 468 484 Hillyard S A amp Galambos R 1970 Eye movement artifact in the CNV Electroencephalogr Clin Neurophysiol 28 2 173 182 Jung T P Humphries C Lee T W Makeig S McKeown M J Iragui V amp Sejnowski T J 1998 Extended ICA removes artifacts from electroencephalographic recordings Advances in Neural Information Processing Systems 10 10 894 900 Makeig S Bell A J Jung T P amp Sejnowski T J 1996 Independent component analysis of electroencephalographic data Advances in Neural Information Processing Systems 8 145 151 Overton D A amp Shagass C 1969 Distribution of eye movement and eyeblink potentials over the scalp Electroencephalogr Clin Neurophysiol 27 5 546 Schl gl A Keinrath C Zimmermann D Scherer R Leeb R amp Pfurtscheller G 2007 A fully automated correction method of EOG artifacts in EEG recordings Clin Neurophysiol 118 1 98 104 page 4of 4
5. of the electric fields generated by the corneo retinal dipole Whereas the cornea is electrically positive the retina is negative As a result rotations of the corneo retinal dipole differentially interfere with cortically generated electric fields over the scalp Interestingly even sleep EEG recordings are contaminated with this type of horizontal EOG artifacts Schlogl et al 2007 A second source of EOG artifacts arises from contact of the eyelid with the negative cornea during blinks eliciting a burst of negativity Overton amp Shagass 1969 EOG amplitude has been found to be attenuated approximately with the square of the distance to the eyes with frontal channels being affected most severely Croft amp Barry 2000 Due to principles of volume conduction artifacts spread throughout all layers of cortex skull and tissue and are present at any scalp site where they interfere with the detection and analysis of cortical responses of interest Hillyard and Galambos 1970 were the first to report such modulations of the auditory CNV in all recorded scalp EEG channels whereas upward eye movements caused the CNV to be positive downward eye movements resulted in a negative CNV www brainproducts com Brain Products Press Release December 2013 Volume 49 References Croft R J amp Barry R J 2000 Removal of ocular artifact from the EEG a review Neurophysiol Clin 30 1 5 19 Gratton G Coles M G amp Do
6. on Artifact Rejection This step is suggested for marking non stereotypical artifacts such as punctual bursts EMG activity reactions rare non EEG glitches across all channels and other of electromyographic whole body startle artifacts ICA might separate each of these into separate components massively reducing the number of components left for capturing remaining processes Since data portions that have been marked as Bad Intervals are disregarded by Ocular Correction ICA please select rejection criteria such that blink periods are not marked am emphasizing this here since blinks and other eye movements are supposed to be used by the transformation Ocular Correction ICA 1 1 The Transformation Once your data fulfills these requirements you can access the transformation via Transformations gt Artifact Rejection Reduction gt Ocular Correction ICA The core steps of this transformation are visualized in Figure 2 you can find more details on these steps in the dedicated chapter of the Analyzer User Manual strongly recommend applying the transformation in semiautomatic mode for every history file since only then you will be able to visually inspect and adjust the automatically by Analyzer component selection accomplished 1 2 3 identify ocular Peg components VA remove ocular components place use D gt applyiCA Ba blink markers Figure 2 Steps of the transformation Ocular Correc
7. rval markers were selected for marking blink periods The table in the upper right corner lists the criterion score of all components for VEOG and HEOG activity The red and green cells in the first column of the table represent the components that will be removed or retained respectively components that are marked in red have been categorized as ocular components and will be removed components marked in green will be kept By double clicking the colored cell you can change its assignment You can sort the table based on the VEOG and HEOG criterion scores by clicking the respective field in the table header also see Figure 3 generally suggest dedicating some time to the careful examination of all components red and green one by one following the sorting order with respect to VEOG and HEOG criterion scores separately Below the table you can find a topographic mapping view of the component currently selected in the criterion table The the component projections unit less weights towards the channels map represents an interpolated representation of By using the options available below the mapping view you can scale the component activations with respect to the channel amplitudes ICA Scaling or use the drop down list to switch between the component activations ICA Components the cleaned corrected channel data after removing the components currently selected as ocular Correction or the simultaneous display of
8. t time Figure 1 Prototypical channel projections topographies left and activation profiles time courses right of Independent Components representing vertical eye movements and blinks A and horizontal eye movements B Please note that due to the nature of the ICA algorithm topographies and or time courses could also be inverted in polarity Now here s where the transformation Ocular Correction ICA as implemented in BrainVision Analyzer 2 comes into play It is a specialized version of the general ICA transformation which identifies and pre selects components with time courses that are potentially associated with eye movements and blinks Before delving into the intricacies of the transformation parameters and settings would like to raise your attention www brainproducts com Brain Products Press Release December 2013 Volume 49 BRAIN SION professional AN ALYZ E pe towards the crucial necessity of proper data pre processing for optimal reliable and valid ICA results In detail data should have undergone at least the following steps e Filtering High pass filtering low cutoff is recommended for removal of slow drifts resulting for example from slow body movements or sweating which ICA cannot handle well In case of higher sampling rates it might also make sense to apply a low pass filter high cutoff This prevents higher frequency content from using up additional components e Raw Data Inspecti
9. ta intervals or even the whole data provided that it can be afforded with respect to memory constraints Please keep in mind that Bad Intervals are neglected by ICA The ICA procedures available to you e g Infomax Fast ICA in a previous Support Tip http www brainproducts com productdetails php id 17 amp tab 3 The resulting ICA and Inverse ICA weight matrices can also be saved to text files have already been addressed 3 Criterion based identification of ocular components Once ICA has been accomplished the extracted components are evaluated in terms of their consistency and similarity to the EEG data For identification of components related to VEOG activity only the time intervals limited by blink markers are used By contrast for identification of components related to HEOG activity all data is used There are three methods available each using different measures a Sum of Squared Correlations with VEOG HEOG is a correlative score between the component activations and the activity of the VEOG HEOG channels selected in step 1 This option does not require the VEOG HEOG channels to be fed into ICA b Relative VEOG HEOG Variance calculates the share of each Brain Products Press Release December 2013 Volume 49 ICA component in the variance of the selected ocular channel activation This method requires the specified VEOG HEOG channel s to be also fed into ICA c Global Field Power only available for VEO
10. the projections of all components towards the channels Topographies If you select Correction from the drop down list you can display the corrected channel data after removing the ocular components Figure 4 When combining this selection with Brain Products Press Release December 2013 Volume 49 the option Overlay with Original Data you can compare the data before red graphs and after black graphs removing the currently selected ocular components in the main window This option can come in quite handy when clarifying whether specific components should be removed or retained since you can observe the immediate effects of your selection As you can see the key to properly attenuating ocular artifacts in your data in fact is to make careful conscious and informed decisions about the ocular nature of a component based on its criterion score time course and topography Ocular components ideally reflect processes purely based on eye movements while at the same time containing absolutely no brain based EEG signals As soon as you are Satisfied with your component selection you can click Finish in order for the changes to take effect ICA of BrainVision Analyzer 2 is a fantastic easy to use tool to Taken together the transformation Ocular Correction diminish the effects of eye movements in your EEG recordings When the recommended semiautomatic mode you have access to all based on data dri
11. tion ICA In detail the steps are as following 1 Blink marker placement At this stage a specified eye or scalp channel is scanned for blinks and blink markers are placed When not using existing markers two detection algorithms are available The Mean Slope algorithm of Gratton et al 1983 will detect any high amplitude activity in the scanned channels potentially including non blink artifacts By contrast the Value Trigger algorithm has been optimized for the detection of prototypical blink patterns Please note that horizontal eye movements will not be detected and marked page 10f4 2 ICA decomposition In this step you define the amount of data to be fed into ICA both in terms of the number of channels and time points Channels that are not selected for ICA will be displayed below the component activations which might be quite helpful for the comparison of component VEOG and HEOG channel activity Regarding the minimum number of data points samples activations and actual there is a simple rule of thumb You should use at least as many data points as the number of channels squared times 20 less than 64 channels or times 30 more than 64 channels Of course this is only a minimum recommendation which in practice often is not sufficient as the data should include the relevant information content to be decomposed for example enough eye movements and blinks Therefore suggest to use longer representative da
12. ven methods you go for the comfortable visualizations of the interactive view which of the components for render the fine tuning and manual adjustment automatically pre selected ocular each dataset transparent efficient and convenient hope that this brief introduction into Ocular Correction ICA made you curious about what else there is to know about the transformation and how it can be used for your particular research endeavors For any further questions regarding the transformation please contact us at Ssupport brainproducts com Ocular Correction ICA Interactive Mode g X ICA Components Blinks Fp1 100 pV Fpz 7 100 pV Fp2 100 pV F00 P U F3 l Figure 4 E Selecting Correction from the 100 pv F Show Normed Mappings E drop down list displays the cleaned corrected rs channel data after removing the four ocular components marked red in the criterion table F4 aa y oy in the main window Additionally activating the check box Overlay with Original Data allows 100 pV Finish Cancel l for a comparison of the data before red and after para ati ri S22 Blink black removing the ocular components F www brainproducts com page 30f4 Further reading Physiological basics of ocular artifacts Electro oculographic EOG artifacts result from two processes First rotations of the eyeballs alter the orientation
Download Pdf Manuals
Related Search