I'm still working on improving source visualization during a seizure in real-time.
The issue is as follows:
- Loreta with the current EEG data appears stroboscopic and difficult to interpret (strobo).
- In an attempt to enhance this, and somehow following the steps outlined in the SEEG tutorial, I downsampled the original signal to 2Hz and then interpolated the time (downsampled). While it looks better, it's not entirely accurate.
What I would like to achieve is:
- Implement a 1-second sliding average that works as a temporal smoother.
- Display time-frequency activity dynamically.
- Any other of your suggestions, which would be greatly appreciated.
I am not sure about the nature of the issue you are raising: I encourage you to specify better the properties of the phenomenon of interest you are trying to detect/visualize. If you downsample your data at 2Hz, you won't be able to detect faster fluctuations, which might be constituents of the seizure onset mechanism -- again, it depends on the science question you are asking the data. For instance, you may want to derive the envelope of the faster activity that may be building up at the onset of the seizure. To that end, you could extract the Hilbert transformation of the source time series in the frequency bands of interest and then visualize such envelopes on the cortex. Note that the epileptogenicity tutorial details an approach in that vein.
But again, define your objectives as specifically as possible before you head in the analyses.
Dear @Sylvain ,
Thanks for your prompt reply.
I'm attempting to replicate the epileptogenicity analyses (just like in the tutorial) but using scalp EEG instead of intracranial electrodes, in a patient with a focal, secondarily generalized seizure (see fig below)
Based on scalp projections, your suggestion about the Hilbert transform would come out really handy, but for some reason, I cannot display time-freq activity in the cortex (screenshot)
I would really appreciate it f you could help me work this out.
You need to run the Hilbert transform of the cortex time series, not that of the EEG sensors.
Many thanks, @Sylvain.
I can't believe it was actually that simple.
It looks amazing now.