Neurofeedback Therapy Meets Transformers: Rewiring Sleep Disorders Through AI-Driven EEG Modulation

Abstract

Sleep disorders such as insomnia, sleep Apnea, and hypersomnia significantly impair neurophysiological functioning, yet conventional treatments like Cognitive Behavioral Therapy for Insomnia (CBT-I) remain resource-intensive and difficult to personalize. This study introduces a novel AI-powered neurofeedback simulation framework designed to detect dysregulated EEG frequency band activity across sleep stages and simulate targeted interventions. A Transformer-based model serves as the core component, offering a unique capability to model cross-epoch temporal dynamics and frequency-specific spectral patterns. Unlike traditional architectures that treat EEG epochs in isolation, the Transformer captures how EEG band activity evolves across the night, critical for identifying persistent dysregulation patterns and planning stage-specific interventions. Through its multi-head attention mechanism, the model can simultaneously monitor delta, theta, alpha, beta, and gamma fluctuations while preserving sleep architecture transitions using positional encoding. Dysregulated epochs are classified with 92% accuracy, and intervention simulations-such as beta suppression in N2 or delta enhancement in REM-led to measurable improvements: average WASO decreased by 23%, and Sleep Efficiency improved by 13%. This framework not only demonstrates the efficacy of Transformer-based temporal-spectral modelling in EEG but also lays the foundation for closed-loop, wearable-compatible, personalized neurofeedback systems for remote sleep therapy.