Energy-Efficient EEG-Based Autism Spectrum Disorder Detection Using a Hyperbolic Attention Neural Network.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Artificial Intelligence

Background:

• Wearable wireless systems are transforming e-health, with electroencephalography (EEG) crucial for early autism spectrum disorder (ASD) diagnosis.
• Current methods for EEG analysis in wearable devices are energy-intensive, limiting operational lifespan.
• There is a need for efficient, on-node processing for real-time ASD detection.

Purpose of the Study:

• To develop an energy-aware, sensor-based scheme for early childhood ASD detection using EEG signals.
• To reduce the energy consumption of wearable EEG monitoring systems.
• To enhance the accuracy and feasibility of real-time ASD detection.

Main Methods:

• On-node signal denoising using chaotic signal models.
• Feature extraction via dual tree discrete wavelet transform (DT-DWT) and lightweight feature selection using parrot optimization (PO).
• A novel Hyperbolic Cross-Head Attention-Based Neural Network (HyperCrossNet) with deep reversible learning and attention mechanisms, optimized by the Pied Kingfisher Optimization Algorithm (PKO).

Main Results:

• Achieved 99.92% classification accuracy, 99.91% recall, and 99.90% F1-score.
• Significantly reduced energy consumption for data transmission compared to conventional methods.
• Demonstrated the system's effectiveness for real-time wearable detection.

Conclusions:

• The proposed energy-aware system enables accurate and efficient on-node ASD detection from EEG signals.
• This approach supports long-term physiological monitoring with wearable wireless systems.
• The HyperCrossNet model offers a promising solution for early ASD diagnosis in e-health applications.