Automatic Sleep Stage Determination by Conditional Probability: Optimized Expert Knowledge-based Multi-Valued Decision Making

Abstract

The aim of this study is to develop a knowledgebased automatic sleep stage determination system which can be optimized for different cases of sleep data at hospitals. The main methodology of multi-valued decision making includes two modules. One is a learning process of expert knowledge database construction. Visual inspection by a qualified clinician is utilized to obtain the probability density functions of parameters for sleep stages. Parameter selection is introduced to find out optimal parameters for variable sleep data. Another is automatic sleep stage determination process. The decision making of sleep stage is made based on conditional probability. The result showed close agreement comparing with the visual inspection. The developed system is flexible to learn from any clinician. It can meet the customized requirements in hospitals and institutions.