Graph Attention Networks for Detecting Epilepsy from EEG Signals Using Accessible Hardware in Low-Resource Settings

TL;DR

The paper tackles under-diagnosed epilepsy in low-resource settings by developing a graph-based framework that processes resting-state EEG from affordable hardware. It models EEG as spatio-temporal graphs and employs GATv2 with edge-focused attention to detect epileptic patterns while enabling explainability through attention scores and Grad-CAM. The approach achieves superior AUROC performance over baselines and reveals consistent fronto-temporal connectivity biomarkers, demonstrating feasibility on Colab-trained models and Raspberry Pi deployment. This work paves the way for scalable, interpretable neurodiagnostic tools in LMICs using low-cost EEG devices. It also highlights the value of edge-aware graph learning for capturing heterogeneous brain connectivity in epilepsy.

Abstract

Goal: Epilepsy remains under-diagnosed in low-income countries due to scarce neurologists and costly diagnostic tools. We propose a graph-based deep learning framework to detect epilepsy from low-cost Electroencephalography (EEG) hardware, tested on recordings from Nigeria and Guinea-Bissau. Our focus is on fair, accessible automatic assessment and explainability to shed light on epilepsy biomarkers. Methods: We model EEG signals as spatio-temporal graphs, classify them, and identify interchannel relationships and temporal dynamics using graph attention networks (GAT). To emphasize connectivity biomarkers, we adapt the inherently node-focused GAT to analyze edges. We also designed signal preprocessing for low-fidelity recordings and a lightweight GAT architecture trained on Google Colab and deployed on RaspberryPi devices. Results: The approach achieves promising classification performance, outperforming a standard classifier based on random forest and graph convolutional networks in terms of accuracy and robustness over multiple sessions, but also highlighting specific connections in the fronto-temporal region. Conclusions: The results highlight the potential of GATs to provide insightful and scalable diagnostic support for epilepsy in underserved regions, paving the way for affordable and accessible neurodiagnostic tools.

Figures (5)

• Figure 1: Overall Pipeline: Data are collected with an affordable portable EEG in low-income countries (left), they are preprocessed and used to create brain connectivity matrices (center). Then, classified using graph attention networks, and the used weights are investigated to shed some lights about epilepsy biomarkers (right). In this figure the electrodes labels for the Epoc EEG device and the used GAT architecture are also shown.
• Figure 2: Attention weights related to the brain connectivity for the Nigeria dataset: (a) sagittal, (b) axial, and (c) coronal view, showing stronger weights in the fronto-temporal area. For the labels of the nodes, please refer to Figure 1. The color code marks the dark red connections as most relevant as those among AF3, AF4, F3, F4, FC5 and FC6.
• Figure 3: Attention weights related to the brain connectivity for the Guinea-Bissau dataset: (a) sagittal, (b) axial, and (c) coronal view, showing stronger weights in the fronto-temporal area. For the labels of the nodes, please refer to Figure 1. The color code marks the dark red connections as most relevant as those among AF3, AF4, F3, F4, FC5 and FC6.
• Figure 4: Attention weights at node level for the Nigerian dataset computed by Grad-CAM method. Here the nodes with highest weights are the last listed, which are for example AF3, AF4, F3, F4, F7, F8, FC5 and FC6.
• Figure 5: Attention weights at node level for the Guinea-Bissau dataset computed by Grad-CAM method. Here the nodes with highest weights are the last listed, which are for example AF3, AF4, F3, F4, F7, F8, FC5 and FC6.