Transformer-Based Sleep Stage Classification Enhanced by Clinical Information

The recent study on Transformer-based sleep stage classification marks a significant advancement in automated sleep analysis. Traditional manual sleep staging is labor-intensive and often inconsistent, but this new model leverages clinical metadata such as age, sex, and BMI, alongside expert annotations of sleep events, to enhance accuracy. Analyzing data from 8,357 participants in the Sleep Heart Health Study, the model demonstrated a notable improvement in performance, achieving macro-F1 and micro-F1 scores of 0.8031 and 0.9051, respectively. This contextual fusion approach not only surpasses the baseline performance of 0.7745 and 0.8774 but also outperforms multi-task alternatives, indicating a promising direction for developing context-aware sleep staging systems. The findings underscore the importance of integrating clinically relevant features into machine learning models, paving the way for more accurate and reliable sleep health assessments.