Redefining obstructive sleep apnea diagnosis: An attention augmented CNN-BiLSTM hybrid alternative to traditional PSG testing

Obstructive sleep apnea (OSA) is a common sleep disorder characterized by repeated episodes of upper airway obstruction during sleep. Despite its health risks, OSA often goes undiagnosed, as many individuals are unaware of the condition or misattribute its symptoms such as daytime fatigue or restlessness to poor sleep or lifestyle factors. Traditional diagnostic methods like polysomnography (PSG) being accurate are expensive, require multiple overnight hospital stays, and often cause patient discomfort, making early and widespread detection challenging. This study addresses these challenges and aims to develop an alternative approach for traditional methods by implementing different deep learning models for automated OSA detection. The proposed alternative approach integrates:•Proposing convolution neural network - bidirectional long short-term memory network (CNN-BiLSTM) hybrid architecture with necessary subset of physiological signals for accurate detection of OSA while minimizing computational overhead.•Integrating attention mechanisms to improve the model's capacity for relevant feature representation and classification performance.•Insights obtained from proposed hybrid architecture with analysis of variance (ANOVA) selected signals better for OSA detection.The model achieves high accuracy of 91.43 % and 95.36 % with F1-score of 0.86 and 0.75 respectively on two publicly available databases. The study emphasizes the need for alternative approaches for OSA and home-based detection methods. Given the global limited awareness of OSA, this study introduces an advanced, accessible, accurate diagnostic solution and outperforms traditional models.