Combination of AI components for biosignal processing application to sleep stage recognition

We present a novel approach to combining artificial intelligence components for biomedical signal processing. The modular algorithm mimics the step-by-step type procedure of a human expert and includes the two assessment steps most important for sleep stage scoring, pattern recognition in electrophysiological signal channels and rule evaluation for classifying the current sequence of patterns. The application of sleep stage scoring is a complex task in medical informatics. The AR-TISANA (artificial intelligence in sleep analysis) algorithm we have developed provides high rates of correspondence with the results produced by human experts. Additional features are the transparent decision-making process and information about the detailed structure of sleep. This has been achieved by utilizing neural networks for pattern recognition and neuro-fuzzy systems for rule evaluation. The AI components chosen to perform these two classification steps were particularly successful due to their individual strengths.