Emotions are hard to read from outside cues alone, and depression is even harder. We built an EEG-based brain-computer interface that pairs graph neural networks — which model the brain as a network of interacting electrodes — with transformer attention, which captures how those signals evolve over time. On benchmark EEG datasets the hybrid model performs competitively on both emotion recognition and depression assessment, suggesting a path toward affect-aware and mental-health-oriented BCI applications.
Emotion is a psychophysiological response to internal or external stimuli that strongly influences human perception, behaviour, and daily activities. Non-invasive technologies such as brain-computer interfaces (BCIs) have made emotion recognition more objective by measuring brain activity directly instead of relying only on observable cues.
Among available modalities, electroencephalography (EEG) provides high temporal resolution and is sensitive to subtle affective changes, which makes it suitable for monitoring both emotional responses and mental-health conditions. Our work proposes an EEG-based BCI system that applies deep-learning methods to recognize emotions and assess depression levels from recorded brain activity.
Experiments on benchmark EEG datasets demonstrate that the proposed approach achieves competitive performance across both tasks, indicating its potential for future affect-aware and mental-health-oriented applications.
A collaborative effort across the team — combining domain knowledge in BCI, signal processing, and deep learning.
BibTeX entry — copy directly into your reference manager.
@inproceedings{rajeswari2025eeg,
title = {EEG-Based Emotion and Depression Recognition Using a Hybrid GNN--Transformer Architecture},
author = {M, Rajeswari and Srinivasaiah, Santosh and Babu, Bhumika P and Geetha, S and Raj, Sanjitha Manya and J, Likhitha},
booktitle = {IEEE},
year = {2025},
publisher = {IEEE},
url = {https://ieeexplore.ieee.org/document/11481919}
}