Spike-Wave Index Prediction in Electroencephalography of Children with Self-limited Epilepsy with Centrotemporal Spikes by Using a Transformer with Spatial and Temporal Attention

<title>Abstract</title> <p> <bold>Objective:</bold> To address various issues associated with electroencephalography (EEG) interpretation for pediatric epilepsy diagnosis, this study proposes a deep learning model for automatically quantifying the spike-wave index (SWI), a critical marker representing the percentage of spike-and-slow-wave discharges during non-rapid eye movement (NREM) sleep and a key biomarker for detecting electrical status epilepticus during sleep (ESES) and atypical evolution in self-limited epilepsy with centrotemporal spikes (SeLECTS). <bold>Methods:</bold> An EEG database derived from 221 children diagnosed with SeLECTS was constructed. We propose a spatial-temporal attention transformer for evaluating the spike-wave index (STATFS) with deep learning methods to automatically classify EEG-based data and predict the SWI. The model demonstrates optimal performance by more effectively extracting features from the original signal, yielding better classification results in the low-dimensional space. <bold>Results:</bold> STATFS outperformed existing deep learning models in SWI prediction tasks, demonstrating enhanced ability to extract spatiotemporal features from raw EEG signals. Compared with traditional manual interpretation, STATFS enabled real-time SWI prediction within seconds. <bold>Conclusion:</bold> STATFS supports reliable clinical decision-making and offers valuable assistance for early intervention. STATFS addresses the low efficiency and subjectivity of EEG interpretations, thereby significantly conserving medical resources; providing convenience; and offering prospects for personalized treatments with substantial clinical. </p>