Enhancing Photoplethysmography-Based Sleep Staging Models Through Temporal Context Optimization

Accurate sleep stage classification is essential for diagnosing sleep disorders and assessing sleep quality. While polysomnography (PSG) remains the gold standard, photoplethysmography (PPG) is a more practical alternative due to its affordability and widespread use in wearable devices. However, state-of-the-art sleep staging methods often require prolonged continuous signal acquisition, which is impractical for wearable devices due to high energy consumption. Shorter signal acquisitions are more energy-efficient but typically compromise accuracy. This study proposes an adapted sleep staging model based on top-performing state-of-the-art methods, optimized for use with shorter PPG segment sizes. We concatenate 30-second PPG segments over 15-minute intervals to incorporate extended contextual information, balancing feasibility and accuracy. Our approach achieved an accuracy of 0.75, a Cohen's Kappa of 0.60, and F1-Weighted score of 0.74. The proposed strategy consistently outperformed methods using short PPG segment sizes. These findings highlight the potential of context-aware approaches to improve sleep staging accuracy in energy-constrained wearable applications.