search

ivadomed / MEEG-Brainstorm

Repository for training MEEG datasets with the ivadomed framework in Brainstorm
1 stars 3 forks source link

Data augmentation to account for missing channels #2

Closed mpompolas closed 2 years ago

mpompolas commented 2 years ago

Each slice in time for the MEEG NIFTI files is created by an interpolation of the signals on the MEEG topography. However, many times some channels need to be rejected because they are very noisy, and the interpolation from the neighbouring channels is filling in for them.

In order to account for this common practice and generalize training better, some channels can be randomly rejected. Since this is done on the channel level (not the NIFTI level), this needs to be enabled on the Brainstorm side right before converting MEEG signals to NIFTIs.

TODO: Add a checkbox button for users to select if they want to enable this data augmentation.

THINGS TO CONSIDER: should it be only a single channels to be dropped out, or multiple?

mpompolas commented 2 years ago

Instead of a checkbox, I added a value entry. If an integer N is inserted by the user, each trial will have UP TO N channels dropped out.

This is done on the signals level, not on the image level. The interpolation from the neighbouring channels will cover their absence.

image