Subject Selection Framework to Improve Personalised Models for Motor-Imagery BCIs via Wavelets and Graph Diffusion
Published in ICLR 2024 - Workshop on Learning from Time Series For Health (TS4H), 2024
Authors: Konstantinos Barmpas, Yannis Panagakis, Dimitrios Adamos, Nikolaos Laskaris and Stefanos Zafeiriou
[ICLR TS4H Page] - [Workshop Page] - [Repo]
Personalized electroencephalogram (EEG) decoders hold a distinct preference in healthcare applications, especially in the context of Motor-Imagery (MI) Brain-Computer Interfaces (BCIs), owing to their inherent capability to effectively tackle inter-subject variability. This study introduces a novel subject selection framework that blends ideas from discriminative learning (based on continuous wavelet transform) and graph-signal processing (over the sensor array). Through experimentation with a publicly available MI dataset, we showcase enhanced personalized performance for MI-BCIs. Notably, it proves particularly advantageous for subjects who initially demonstrated suboptimal personalized performance.
ICLR24 TS4H Poster
