SinTransNet: an EEG-based deep learning framework for infantile epileptic spasms syndrome detection.

Area of Science:

• Neurology
• Artificial Intelligence
• Signal Processing

Background:

• Infantile Epileptic Spasms Syndrome (IESS) is a severe infant epilepsy causing neurodevelopmental issues.
• EEG interpretation for IESS is complex, time-consuming, and prone to errors, delaying treatment.
• Current diagnostic methods struggle with the non-stationarity and complexity of EEG signals.

Purpose of the Study:

• To develop an automated deep learning framework, SinTransNet, for accurate and efficient detection of IESS from EEG signals.
• To address the limitations of manual EEG interpretation in diagnosing IESS.
• To improve timely therapeutic interventions for infants with IESS.

Main Methods:

• Proposed SinTransNet, a deep learning framework utilizing multi-band EEG decomposition, sinusoidal convolutions, and Transformer attention.
• Decomposed EEG signals into five frequency bands (δ, θ, α, β, γ) for feature extraction.
• Employed Transformer attention to capture inter-band correlations and long-range dependencies.

Main Results:

• SinTransNet achieved high performance on a proprietary dataset of 129 EEG recordings.
• Demonstrated an average accuracy of 85.69%, sensitivity of 80.55%, and specificity of 90.76% in detecting epileptic spasms.
• The framework effectively identified key oscillatory features and complex EEG patterns.

Conclusions:

• SinTransNet offers an automated and efficient solution for IESS identification from EEG.
• The proposed deep learning approach shows significant potential to enhance clinical workflows in pediatric neurology.
• Early and accurate diagnosis through SinTransNet can facilitate timely interventions and improve neurodevelopmental outcomes for infants.