add the possibility to compute the MEG (forward FEM) with a block of sensors

[tmedani]

Hello all,
@ftadel, to complete the previous PR, this new PR adds the option to compute the MEG FEM forward with a group of sensors, this will reduce the size of the transfer matrix, therefore the memory required for the computation, and avoid the computation crashes.
I have fixed the max number of sensors to 250 for each iteration.

The new duneuro panel includes a check box for this option

I have also validated the output results with this option.

@ftadel I have attached some questions to the scripts, let me know if there are any changes that I can do.

Thanks :),
Takfarinas

[ftadel]

250?
For a 256-electrode EEG channel, it means that it would run once with 250 and once with 6 sensors?
For an Elekta MEG: one execution with 250 and one with 56?
This sounds like a weird logic.

Have you tried to quantify the increase in computation time vs. the decrease in memory usage?
For example: when you divide the number of sensors by 2, by 4, by 8.
If you need to split this list, I guess the most optimal way to balance the computation time and the memory usage would be to split in equal blocks, isn't it?

I am not in favor of adding useless parameters that no one will ever modify. Maybe one check box that allows to divide in 2 or 4 could be useful for re-running when it crashes because of memory errors?

[tmedani]

250?
For a 256-electrode EEG channel, it means that it would run once with 250 and once with 6 sensors?
For an Elekta MEG: one execution with 250 and one with 56?
This sounds like a weird logic.

as discussed, this is not for EEG but for MEG sensors where four integrations points are used for magnetometers and 8 for gradiometers, so the total number of points for computation is around 8x250 ~= 2000

Have you tried to quantify the increase in computation time vs. the decrease in memory usage?
For example: when you divide the number of sensors by 2, by 4, by 8.

Yes, in some basic case, less block faster is the solution, with more block the process of reading/writing is multiplied.

Maybe one check box that allows to divide in 2 or 4 could be useful for re-running when it crashes because of memory errors?

ok, this is another useful option, then I can modify the code to this option

[ftadel]

In panel_duneuro/UpdatePanel, add something like:

if ~isempty(jCheckMegPerBlockOfSensor)
    jCheckMegPerBlockOfSensor.setEnabled(~jCheckSaveTransfer.isSelected());
    if jCheckSaveTransfer.isSelected()
        jCheckMegPerBlockOfSensor.setSelected(0);
    end
end

You could also rename jCheckMegPerBlockOfSensor into jCheckMegBlock.
These super long names are painful to read.

And remove this comment :)

[ftadel]

Delete this section, move it to UpdatePanel()

[ftadel]

Can you please fix the conflicts?

[ftadel]

Hmmm... I thought you had just posted something in the PR, Github sent me a notification...
But looking at it now, it looks like an old branch. Not sure what happened, sorry.

If this is outdated, please close the PR - thanks

[tmedani]

I have a newest version that I will clean and push asap