Large language models for automated sleep staging

Sleep holds a lot of information about the human body. Tracking your sleep can provide valuable insights into your sleep patterns, including duration, quality, and stages of sleep. This information can help one understand if they are getting enough restorative sleep or if there are disruptions that need addressing. Sleep is closely linked to overall health. By monitoring your sleep patterns, you can identify potential health issues such as sleep disorders (e.g., sleep apnea, insomnia) or underlying medical conditions that may affect your sleep quality. If these sleep disorders are detected and treated early, many major cardiovascular diseases, neurological diseases and mental health disorders can be prevented. We evaluated four LLMs—Llama8b-8192, Llama-70b, Mixtral, and Gemma2-9b-it—on sleep stage classification. Llama-8b-8192 achieved the highest accuracy at 44.4%, highlighting the limitations of general-purpose LLMs for medical time-series classification. We have tried to use its intelligence in identifying patterns between various PSG data signal channels and sleep stages.