Hi all, I'm currently trying to analyse resting state EEG data from a group of PDD patients. This was a fairly trivial task in the controls, but for some of the patients their low frequency patterns do not conform to what is expected. From their notes it would seem that this was resting tremor. I was wondering if anybody else had experienced this and knew of a good way to clean it. My main aims are to avoid rejecting too much data, and to ensure that the alpha and high theta bands are no longer contaminated.
Thanks
I do not have any medical knowledge but have some knowledge of signal processing. I am not suggesting a particular solution but would rather suggest a line of attack that can also help in the discussion if any that ensues on this thread.
From the information given I am treating this problem as a learning problem. There are many ways of going on with this problem. One of them is the adaptive filtering. This is a paradigm in which we require previous data to classify the new test cases, but this may not work if you are looking for a very general solution.
Other thing that can be done is to use data driven techniques things like EMD (Empirical Mode Decomposition), EWT (Empirical Wavelet Transform) so that you can decompose the data into frequency ranges. For this you will need the in depth physical understanding of the phenomena in question and then you may be able to define a particular range where you may find the data of interest (can either be the one you want to keep or discard).
Hope it is useful in some way
Have you tried an independent component analysis of a section of recording to identify the particular component related to the tremor, and then treat it as artefact removal (in the same way you would for eye blinks, ecg artifacts etc) and simply remove that component from the subsequent analysis. As a tremor it should have a reasonably defined frequency and should be not difficult to identify.
We're also analyzing some resting state EEG data from Parkinson's patients, and we have also been seeing higher levels of EMG noise in the PD patients when compared against the healthy controls. It would be interesting to hear more about your research, and the potential for developing algorithms to filter out tremor-based artifacts!
As you surely know, ICA is the gold standard for artefact removal and it works nicely with some of them, such as blinking artefacts. Conway and his colleagues also used Runica to remove walking artefacts (see liknk). So I think it's definitely a method worth trying.
However, tremor should be a regular oscillation, which would correspond to a relatively narrow artefact in the frequency domain. Based on the fact that empirical mode decomposition (EMD) (or ensemble empirical mode decomposition; EEMD) are designed to identify oscillatory components within the signals –with no a priori information on their characteristics–, either of them may work better than ICA. You can find EMD and EEMD toolboxes for download online (see the second link).
By the way do you think the tremor artefact you see on the signals are due to head tremor, or to the mechanical "transmission" of the hand oscillations that make the EEG electrodes "shake"?
Let us know what you find, I'm also interested in these type of questions.
Good luck!
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3424763/
http://rcada.ncu.edu.tw/research1.htm
Hi Nik. I've spent the last few years analysing EEG data from children with tics and ADHD so I appreciate how difficult it is to deal with movement artefact! I successfully removed motor artefact from my data with ICA in BVA and I would highly recommend this approach. It sounds like yours is a constant oscillation of some sort? Can it be removed with a notch filter at the frequency it oscillates?
Hi Lizzie. Thanks for your suggestion. So the need here I think actually goes slightly beyond just removing it., The more I look at the data the more I am uncertain of exactly what is causing it. My guess is that it is related to pathology, because I am not seeing it in controls. However, because it over laps with the classic range for resting tremor there is the possibility that this is a mix of two or more things (both cortical and not). Ideally I would like to be able to isolate this with ICA and to then analyse the data either with it in (knowing that it is not problematic), or with it to one side (knowing that it is not related to attention).
I think that adding a notch filter could be damaging, like I've said the expected pattern is for us to notice slowing of the alpha rhythm into the theta rhythm (ie bonanni). So if there does happen to be a cortical reason for this particular pattern then notching it out removes our ability to answer anything related to the fact that this is the dominant peak in the spectrum between eyes open and eyes closed.
There are also the occasional bursts of typical emg noise which can of course be removed using ica or just taken out. I'm currently struggling to fit some kind of classification to these patterns for future use though.
Hi Nik. This sounds like a really interesting (if difficult) problem! I agree with you re notch filters; fine for removing electrical line noise but a bit quick and dirty for more complex oscillatory patterns. I'd be interested to hear your solution to the tremor/neuronal activity issue. In terms of classifying the motor artefacts, I found that I had to do this on a participant-by-participant basis because the movements varied a lot between subjects. I don't know if that would be the same for your tremor patients.
- Badges
- Science topic
- Similar topics
- Biological Science
- Physiology
- Biosignals
- Biomedical Signal Processing