IESS-FusionNet: Physiologically Inspired EEG-EMG Fusion with Linear Recurrent Attention for Infantile Epileptic...

Area of Science:

• Medical technology
• Artificial intelligence
• Neuroscience

Background:

• Infantile Epileptic Spasms Syndrome (IESS) is a severe infant epilepsy linked to neurodevelopmental issues.
• Early diagnosis and treatment are vital to improve cognitive outcomes.
• Current automated detection methods struggle with complex EEG and EMG signals.

Purpose of the Study:

• To develop an accurate, real-time automated detection system for IESS using simultaneous EEG and EMG data.
• To introduce a novel deep learning framework, IESS-FusionNet, with physiologically inspired cross-modal fusion.

Main Methods:

• Developed IESS-FusionNet, an end-to-end dual-stream framework integrating Continuous Wavelet Transform, Spatio-Temporal Convolution, and Bidirectional Mamba.
• Employed a Cross Time-Mixing module based on RWKV architecture for efficient, time-decaying, bidirectional cross-modal integration.
• Validated the model on a clinical dataset of synchronized EEG-EMG recordings from infants with confirmed IESS.

Main Results:

• IESS-FusionNet achieved 89.5% accuracy, 90.7% specificity, and 88.3% sensitivity.
• The model significantly outperformed existing unimodal and multimodal detection baselines.
• The proposed cross-modal fusion mechanism used 60% fewer parameters than quadratic cross-attention, enabling real-time deployment.

Conclusions:

• IESS-FusionNet provides an accurate and computationally efficient solution for automated IESS detection.
• The physiologically inspired cross-modal fusion enhances performance and suitability for clinical applications.
• This technology holds promise for reducing diagnostic delays in IESS, potentially improving infant neurodevelopmental outcomes.