SleepVLM: Explainable and Rule-Grounded Sleep Staging via a Vision-Language Model
Researchers have developed SleepVLM, a novel vision-language model designed for automated sleep staging. This model not only achieves state-of-the-art accuracy in classifying sleep stages from polysomnography images but also provides explainable, clinician-readable rationales based on established medical criteria. Independent evaluations have confirmed the model's reasoning quality, suggesting it could enhance the trustworthiness and auditability of automated sleep staging in clinical settings. The team has also released a new dataset, MASS-EX, to support further research in interpretable sleep medicine. AI
IMPACT Introduces explainable AI to sleep staging, potentially increasing clinical trust and adoption of automated systems.