Neuronal avalanches for EEG-based motor imagery BCI: robustness of classification performance & validity of feature selection - Institut de Neurosciences des Systèmes Accéder directement au contenu
Poster De Conférence Année : 2024

Neuronal avalanches for EEG-based motor imagery BCI: robustness of classification performance & validity of feature selection

Résumé

Brain-Computer Interface (BCI) research is a growing field not yet fully explored. The current features that are used in this context rely on local measurements disregarding the propagation nature of brain functioning. To overcome these limitations, we proposed a powerful and original marker that captures the dynamic nature of the brain functioning: the neuronal avalanches, propagating cascades of bursts of activity among multiple brain regions. To assess their potential as BCI candidates, we tested neuronal avalanches using electroencephalography signals during resting state and a motor imagery task within a BCI protocol, in both source and sensor-space, relevant for online studies in a clinical setting. Neuronal avalanche analysis entails identifying significant signal excursions beyond a threshold and clustering them based on temporal proximity, providing insights into rich functional connectivity dynamics. To track the probability that an avalanche would spread across any two channels/region we built an avalanche transition matrix (ATM). We compared classification performance resulting from the use of ATM to the backmarker, Common Spatial Patterns. In both sensor and source-space our pipeline yielded significantly higher classification accuracy in most of subjects and a smaller inter-subject variability. We investigated the interpretability of these findings by comparing the selected features and the set of edges where a significant condition effect was observed in most of the subjects, a large correspondence between them conducted to an improved classification performance. Our results suggest that working in the sensor space provide enough spatial resolution to perform the classification with a simpler framework than in the source space. Furthermore, based on these observations, we propose integrating periodic and aperiodic features as a possible way to enhance task classification and neuronal avalanches as potential alternative features in BCI design.
Fichier principal
Vignette du fichier
Poster_Camilla_Mannino_CORTICO24.pdf (31.11 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04621936 , version 1 (24-06-2024)

Identifiants

  • HAL Id : hal-04621936 , version 1

Citer

Camilla Mannino, Mario Chavez, Pierpaolo Sorrentino, Marie-Constance Corsi. Neuronal avalanches for EEG-based motor imagery BCI: robustness of classification performance & validity of feature selection. Journées CORTICO 2024, May 2024, Nancy, France. ⟨hal-04621936⟩
0 Consultations
0 Téléchargements

Partager

Gmail Mastodon Facebook X LinkedIn More