Abstract
INTRODUCTION: The integration of artificial intelligence into the diagnosis and management of sleep-disordered breathing presents a transformative opportunity to enhance clinical outcomes, particularly through novel methods like EEG data analysis. Leveraging advancements in auditory-linguistic modeling, this study aligns with the growing interest in innovative diagnostic technologies for sleep-related conditions as highlighted in the "Novel Technologies in the Diagnosis and Management of Sleep-Disordered Breathing" research topic. Traditional approaches in OSA screening often rely on polysomnography, which, despite its high accuracy, suffers from limited accessibility, cost, and patient comfort issues. Furthermore, these methods rarely incorporate insights from cognitive and auditory processing frameworks that could deepen diagnostic precision. METHODS: To address these gaps, we propose an AI-enabled screening methodology that utilizes EEG signals in conjunction with insights from English listening comprehension models. Our Auditory-Linguistic Hierarchical Transformer (ALHT) and the Context-Adaptive Dual Attention Mechanism (CADA) are applied to EEG feature extraction, offering a robust framework for analyzing sleep patterns while adapting to patient-specific and contextual variations. RESULTS: Experimental results demonstrate superior classification accuracy and adaptability in noisy environments. DISCUSSION: These outcomes showcase the model's ultimate potential in enhancing both accessibility and reliability in OSA diagnostics.