Dear Jan,
Markus Bauer (CCed) should be able to give you more specific advice on saccades as he spent a lot of time removing them also with the SPM tool and his code for detection of saccades is included in SPM12 artefact tool.
But to answer your questions:
1) Indeed polarity is not well defined for SVD components as each component has two corresponding vectors and if their polarities are flipped together there will be no effect on the outcome. This should not matter for the purposes of artefact removal as what matters is just the spatial topography and not the polarity.
2) I think your sequential procedure is OK, although it might be possible perhaps to manage with less topographies by computing them from blinks and saccades combined. In BESA from which this approach originates this kind of sequential procedure is routinely done.
Best,
Vladimir
> On 10 Feb 2015, at 17:33, Jan <[log in to unmask]> wrote:
>
> I am extracting spatial components (SVD via spm_eeg_spatial_confounds) for right saccades and left saccades separately from averaged data epoched around the onset of the saccades. I know from the literature how the topographies of the most prominent artefacts should look like. Most importantly, the topography of right saccade artefacts should be the inverse of left saccade artefacts. However, a good deal of my subjects do not show this pattern in the topographies of the extracted components, i.e. for most of my subjects the components for right and left saccades look a like. They look reasonable, but the polarity is wrong.
>
> Now I am wondering if the topographies of the extracted components can actually be interpreted in the same way as I would interpret topographies of ERP data? Basically, the topographies of extracted components are just spatial filters, right?! I guess my problem boils down to the question whether the polarity of the topographies that are displayed at the end of the call to spm_eeg_spatial_confounds is meaningful?
> When I reconstruct the data using only the first component, for instance, the waveforms on the scalp show the appropriate polarity ( after the call to spm_svd: U(:,1)*L(1,1)*V(:,1)' ). When I reconstruct the 'raw' wave form of the component ( L(1,1)*V(:,1)' ), the polarity is sometimes inverted as compared to the ERP on the scalp. Am I doing something wrong? Or is everything okay, and SVD simply does not care about polarity of the components, but the weighting by the 'spatial confound' vector takes care of appropriate polarity?
>
> In the end, when I use the extracted spatial confound topographies to correct my data, the saccade artefacts seem to be removed. However, I am not sure if I am doing something wrong along the way, and remove not just artefacts but distort my actual EEG data.
>
> Another quick question that is related: Since I want to remove blink artefacts from my data as well, I apply a sequential procedure to clean my data. I remove blink artefacts first, and then use the blink-artefact-free data to extract the components for left saccades remove them, and then take care of right saccades. I assumed that this procedure would be more appropriate than estimating all three types of artefact components on the raw data, given that I am using SVD to estimate them. Is this sequential procedure correct, or rather not necessary and an estimation of all artefact components on the raw data would be fine?
>
> I would really appreciate if you could give me some advice on these matters, or if you could share some of your experience with saccade artefact removal. Thanks a lot in advance.
>
> cheers,
> Jan
>
> --
> Jan Herding, M.Sc.
> Freie Universität Berlin
> Neurocomputation and Neuroimaging Unit
> Habelschwerdter Allee 45
> 14195 Berlin
> Room JK 25/212
> Tel: +49-(0)30-838-56693
> [log in to unmask]
>
> Graduate Student at:
> GRK 1589/1: "Sensory Computation in Neural Systems"
> Bernstein Center for Computational Neuroscience Berlin
> Philippstr. 13, Haus 6
> 10115 Berlin
>
|