Learning from Heterogeneous EEG Signals with Differentiable Channel Reordering
Aaqib Saeed
David Grangier
Olivier Pietquin
Neil Zeghidour
[Paper]
[Poster]

Illustration of various forms of inconsistencies that can arise in EEG recordings and overview of the CHARM framework with a 1D convolutional classifier for EEG classification tasks.

Key contributions

  • We introduce a novel framework CHAnnel Reordering Module (CHARM) for training a single model across varying EEG collections that can differ both in number and location of electrodes.
  • Our differentiable module uses an attention mechanism on multichannel EEG signals, identifies the location of each channel from their content, and remaps them to a canonical order.
  • We perform experiments on four EEG classification datasets for the tasks of seizure, artifact detection, and abnormal EEG recordings.
  • We demonstrate the efficacy of via simulated shuffling and masking of input channels.
  • We also propose a channel masking and shuffling augmentation strategy for multi-channel input to improve robustness of standard models.

Abstract

We propose CHARM, a method for training a single neural network across inconsistent input channels. Our work is motivated by Electroencephalography (EEG), where data collection protocols fromdifferent headsets result in varying channel ordering and number, which limits the feasibility of transferring trained systems across datasets. Our approach builds upon attention mechanisms to estimate a latent reordering matrix from each input signal and map input channels to a canonical order. CHARM is differentiable and can be composed further with architectures expecting a consistent channel ordering to build end-to-end trainable classifiers. We perform experiments on four EEG classification datasets and demonstrate the efficacy of CHARM via simulated shuffling and masking of input channels. Moreover, our method improves the transfer of pre-trained representations between datasets collected with different protocols.


Main Results




Paper and Poster

A. Saeed, D. Grangier, and O. Pietquin, and N. Zeghidour
Learning from Heterogeneous EEG Signals with Differentiable Channel Reordering
(arXiv) | (Official Link)


[Bibtex]
[Poster]