Hi François,
I have some questions about results I obtained when applying Compute SSP:Heartbeats.
First, after I finish the Detect Heartbeats process, even if I only type cardiac in the event name window, I do not only get “cardiac” events for my subjects, I also get some more (ex. “cardiac2”, “cardiac3”, etc.).
When I run Compute SSP:Heartbeats, I only enter cardiac like in the tutorial, and I wonder if all the cardiac events will be corrected or if I should write the other cardiac events like listed above.
Finally, I obtain corrected files that I think are very strange and I wonder if I applied the correction correctly. I attached some pictures concerning that as an example. I used existing SSP projectors (is that the problem?).
Thanks!
Maxime
Hi Maxime,
The event detection also classifies the events by shape, as described in the SSP tutorial:
http://neuroimage.usc.edu/brainstorm/Tutorials/TutRawSsp#Automatic_detection:_EOG
This can be due to a bad ECG signal quality. You can either group all those categories together before you calculate the SSP or use the “SSP: Generic” process and disable the classification.
The correction you get seems to indicate that there is something completely wrong in the way you calculated your projectors. Either you used the wrong parameters or the wrong process, or there are some previous projectors that make it capture something completely different.
What is relevant to evaluate the quality of the cleaning the topography of the SSP components and the distribution of the singular values (% values).
I would recommend you delete all your projectors, and start over from the beginning, as shown in the introduction SSP tutorial (cited before).
You can also read all the other tutorials that talk about SSP cleaning:
http://neuroimage.usc.edu/brainstorm/Tutorials/TutMindNeuromag#Heartbeats_and_eye_blinks
http://neuroimage.usc.edu/brainstorm/Tutorials/Resting#Pre-processing
http://neuroimage.usc.edu/brainstorm/Tutorials/Yokogawa#Artifacts:_Heartbeats_and_eye_blinks
http://neuroimage.usc.edu/brainstorm/Tutorials/Auditory#Heartbeats_and_eye_blinks
Try with different options: check or not the “Use existing SSP”, different window sizes, different frequency bands (you may have to exclude the alpha band from the eye blink SSP calculation if you have a lot of alpha and you see some pariental or occipatal sensors in your blink components).
What will give you good results if your cardiac markers are always exactly at the peak of the ECG is the option “Average” instead of “PCA” (using the SSP: Generic).
You have to play around with all those options for a while to get the sense of what they are doing and what is the most adapted to your recordings.
Every dataset has to be cleaned in a slightly different way, there is not a single procedure that will clean all the MEG recordings. The distribution of the noise generated by the eyes and the heart highly depends on the MEG system, the subject, the sampling frequency and the task.
Cheers,
Francois
Hi again François,
Ok, I tried some stuff based on the info you sent me. I merged all the cardiac events together. Also, some EEG channels’ type were marked as MISC, so I changed them to EEG. Finally, I tried some different SSP components until I got what I think is acceptable. I have attached some images for 3 subjects for before and after the correction. I decided not to correct for the eye blinks since my procedure consists of eye-closed wakefulness EEG. My objective is to calculate the PLV of each group of subjects by frequency bands and regions.
I would appreciate your advice to have an idea if the correction can now be considered valid. I mostly am not sure about subject C (I really can’t seem to get it better and I wonder if I should exclude such subjects). I would also like to know to what point noise affects these computations; does it change anything if I compute the SSP after having marked some channels as bad? With subject C, are the abnormalities mainly due to an amplification of the noise that was present before the computation?
Thank you!
Maxime
Hi Maxime,
It looks like your SSPs are doing a lot more than removing the heartbeats.
Please post screen captures the topographies of the components you use (button “display topography” in the the window “select active components”).
This is the best way to get a sense of what is going to be removed.
Please also post screen captures of this “select active components”, it is very informative to see the list of computed components.
Have you tried with the “average” option? This is probably the safest way to get good cardiac SSPs, if your events are all exactly at the peak of the ECG.
Regarding the screen captures you posted: you should look at the traces before/after with the same amplitude scale, if not it’s very difficult to estimate what happens.
For that, just turn off the button [AS] in the time series figure (= “Auto-scale”).
You can mark bad channels before or after the SSP calculation, it doesn’t matter.
You should definitely fix your data for eye movements. When your eyes are closed, they move a lot, then you record only the ocular muscles with half of your electrodes.
SSP and high-pass filters are two valid options.
Francois
Hi,
I have joined examples of the topographies, and of the active components.
I have tried the “average” option. It does not seem to change anything from the standard compute SSP: Heartbeats. I also sent a screenshot about that.
Finally, I added an image of the same subject with both cardiac and blinks corrections applied.
The other cardiac ssp component in active components list is the averaged one.
Thanks!
Maxime
Hi Maxime,
Have you checked if the heart was even generating any artifact?
In EEG, the cardiac artifact can be very low and may not require any specific cleaning…
I see a tag “NREM” on top of your recordings. Is you subject sleeping?
If this is the case, what do you mean by “blink”?
Sleep recordings require a different type of procedure (or any time the eyes are closed) because the eyes move in a completely different way.
I see you are in Montreal, you are welcome to come visit with your data on a laptop, we might be able to help you.
If you want to, send us an email (click on my name on this forum).
Cheers,
Francois