August 20, 2025Biomedical Signal Processing and Control4 citationsOpen Access

EGDN-KL: Dynamic graph–deviation network for EEG anomaly detection

INIlyes NaidjiUniversity of Biskra ATAhmed TibermacineUniversity of Biskra ITImad Eddine TibermacineSapienza University of Rome

Structured PICO

Population

Two public EEG datasets containing healthy cortical rhythms and anomalies (epileptiform spikes, abnormal rhythmic bursts, elevated beta power)

Intervention

EGDN-KL (structure-learning graph neural model coupling explicit Kullback-Leibler divergence term with attention-guided LSTM forecasting)

Comparator

Graph Deviation Networks (GDN), EGDN-Naïve, ResCNN, BiLSTM-Attention and GCN baselines

Outcome

Anomaly detection performance (Recall, Specificity, F1, Accuracy)

The EGDN-KL model delivers state-of-the-art EEG anomaly detection by capturing both temporal dynamics and inter-channel connectivity.

Abstract

Electroencephalography (EEG) anomalies are sustained or transient deviations – such as epileptiform spikes, abnormal rhythmic bursts, or pathologically elevated beta power – from the statistical regularities observed in healthy cortical rhythms. We introduce EGDN–KL, a structure-learning graph neural model that couples an explicit Kullback–Leibler divergence term with attention-guided LSTM forecasting to pinpoint such deviations. Evaluated on two public datasets, EGDN–KL achieves Recall 89.4%, Specificity 97.7%, F 1 88.3%, and Accuracy 87.3% using its optimal configuration (11 channels, Top- K = 10 ). These results surpass Graph Deviation Networks (GDN), EGDN-Naïve, ResCNN, BiLSTM-Attention and GCN baselines, whose best published accuracies range from 58.2% to 82.4%. By capturing both temporal dynamics and inter-channel connectivity, EGDN–KL delivers state-of-the-art anomaly detection while localizing the neural generators that underpin pathological activity.

Ask AI

Helpful

Bookmark

View Full Paper