Deep Learning and Machine Learning Decoders of EEG during Hand Movements
Aristotle University of Thessaloniki · 01.01.2024
DOI: 10.26262/heal.auth.ir.360509 ↗Kurz gefasst
Fachartikel in der Zeitschrift Aristotle University of Thessaloniki, erschienen am 01.01.2024. Verfasst von Despoina Markoglou.
Abstract (Originalsprache)
This study investigates advanced machine learning and deep learning methods for analyzing electroencephalographic (EEG) signals to improve the performance and reliability of Brain-Computer Interfaces (BCIs). Using the "Reach and Grasp movement dataset" from the "BNCI Horizon 2020" database, which includes EEG data from gel, wet, and dry recording systems, we explore the classification of two hand movements—palmar grasp and lateral grasp—along with the resting state. Six models were tested: three machine learning models (Bagged Tree, Gaussian SVM, Cosine KNN) and three deep learning models (Conformer, Robust CNN, Robust Conformer), with the Robust Conformer developed in this study to integrate features of CNN and transformer networks. CNNs in these models capture local data features, while transformers enhance the ability to learn long-term dependencies, contributing to the efficacy of EEG signal decoding. To evaluate model performance and generalizability, the Leave-One-Subject-Out (LOSO) method was used, training on data from all but one participant and testing on the excluded participant, iteratively repeated for each participant. Results showed that deep learning models outperformed machine learning models, with the Robust Conformer achieving the highest classification accuracy across all recording systems. The inclusion of transformers in the Robust Conformer was pivotal for enhancing classification accuracy, underscoring their value in EEG signal processing. This research advances BCI technology by developing a new deep learning model for effective EEG analysis, demonstrating the LOSO method’s effectiveness for generalization, and validating transformer networks as beneficial for EEG classification accuracy. The findings support the potential of these advancements to enable more sophisticated and reliable BCI systems for applications ranging from neurorehabilitation to assistive technologies.
Schlagwörter
Zitieren
APA
Markoglou, D. (2024). Deep Learning and Machine Learning Decoders of EEG during Hand Movements. Aristotle University of Thessaloniki. https://doi.org/10.26262/heal.auth.ir.360509
Vancouver
Markoglou D. Deep Learning and Machine Learning Decoders of EEG during Hand Movements. Aristotle University of Thessaloniki. 2024. doi:10.26262/heal.auth.ir.360509
BibTeX
@article{Markoglou2024,
author = {Markoglou, Despoina},
title = {Deep Learning and Machine Learning Decoders of EEG during Hand Movements},
journal = {Aristotle University of Thessaloniki},
year = {2024},
doi = {10.26262/heal.auth.ir.360509}
}
EUVIMED