Indexado en

Licencia y uso

Citaciones

Altmetrics

Grant support

This work has been partially funded by Spanish project PID2020-114867RB-I00, (MCIN/AEI and ERDF-"A way of making Europe"), Universidad Politecnica Salesiana 034-02-2022-03-31 and by Predoctoral Research Grants 2015-AR2Q9086 of the Government of Ecuador through SENESCYT.

Análisis de autorías institucional

Compartir

Enhancing P300 Detection in Brain-Computer Interfaces with Interpretable Post-processing of Recurrent Neural Networks

Publicado en:ADVANCES IN COMPUTATIONAL INTELLIGENCE, IWANN 2017, PT II. 14259 25-36 - 2023-01-01 14259(), DOI: 10.1007/978-3-031-44223-0_3

Autores: Oliva, Christian; Changoluisa, Vinicio; Rodriguez, Francisco B; Lago-Fernandez, Luis F

Afiliaciones

Resumen

Brain-computer interfaces (BCIs) are innovative systems that allow individuals to communicate with external devices without physical movements. These systems commonly use Event-Related Potentials (ERPs), particularly P300, as the signal control. However, despite their wide acceptance, there are still issues to be resolved, such as inter- and intra-subject variability. To address this challenge, we propose a novel approach based on post-processing the output of a Recurrent Neural Network using a Post-Recurrent Module (PRM). The PRM processes the temporal information extracted from the recurrent layer to make the final decision. This work shows that simple approaches, such as a reduce-max operation or a logistic regression layer, can improve the balanced accuracy by more than 9% compared to state-of-the-art results. Our findings also contribute to the interpretability of RNNs since we have deepened the internal mechanisms of the model through an extensive analysis of the PRM layer. Overall, this study enhances the performance of ERP-based BCIs.

Palabras clave