 |
 |
Dear List, I come from a different world (biomechanics) to ask for your wisdom. I have high density electromyography signals (HDEMG) from the gastrocnemius muscle. These were recorded with a grid of 64 electrodes (8x8) (10mm inter-distance). My goal is to compare the spatial distribution (change) of EMG activity before and after some bouts of exercise using SPM. I did some basic pre-processing to the HDEMG signals according to what we usually do in biomechanics (band pass filters, RMS calculations). Then, I interpolated the signals in 2D to create a matrix of 64x64. With this data, I created .nii files (see attached) that can be opened (see attached), viewed and analysed in SPM12 (I have already done some 2nd level analyses). At this moment, I have some doubts about my next steps. I think that maybe 64x64 it’s too much interpolation, however, there is a good reference that recommend this for EEG (https://www.hindawi.com/journals/cin/2011/852961/). What do you think? I have been playing with different resolutions in my images. For instance, I have an image of 8x8. For some reason, when I display this in SPM12, the image appears “interpolated” but the actual information of the file confirms that it’s a matrix of 8x8 (image 1 attached). The same happens with images of 32x32. If I open this file in Matlab or in other .nii viewer, the image looks like how I expect. There is something that I am doing wrong in SPM? Understanding that smoothing is important to accommodate variability over subjects, should I do smoothing to my images anyway, considering that I have done some pre-processing to the signals? Anyway, I have been playing with the smoothing factors, and according to the matched filter theorem, I should use a quite large kernel (e.g. 20 – 30 mm) to match the size of my muscle´s activation (that covers a big area of the recording grid). Using this ends up with my images having a huge boundary effect (lower activity band around the edge of the image) (image 2 attached). Since I have lots of activation in one of the sides of my image, this might be a problem. There is any way to avoid this? Should I use a smaller kernel? Thanks for the advice Patricio