Artificial shrinking of channel location during segmentation to avoid missing channels

[mpompolas]

This is a Brainstorm visualization issue/feature.

Brainstorm's EEG topography for a single time point looks like this:

One thing to notice is the placement of some channels on the perimetry of the skull.

The graphics of the perimeter are not getting updated while we scroll though the recording in time; only inside of it. The perimeter is always a solid line.

When I create the datasets, the annotations also don't include the perimeter.
This is not problematic during the creation of the BIDS NIFTI datasets.

However, this is problematic during evaluation.
In order to get from ivadomed NIFTI segmentations to Brainstorm events, I monitor the values of the pixels of each channel, and then they are converted to events in time.
Since the perimeter pixels are never included in the annotations, no segmentation is returned on those positions, therefore the perimeter channels are never annotated.

To alleviate this problem, I "shrink" the coordinate system by a few pixels relative to the center of the topography disc.
This is shown on the following figure:

The new positions are saved on a new file, without changing the original EEG/MEG datasets. The new positions are used during conversion from Ivadomed model annotations to Brainstorm events.

Although this is a "hardcoded" approach, it shouldn't affect the end result too much since the topography is interpolated from the channel positions in the first place.

Alternatives

A different approach will be to change the way that Brainstorm is handling these visualizations.