An Automatic Sleep Scoring System for Accurate Estimation of Various Sleep Measures

Abstract

Human beings spend approximately one-third of their lives sleeping Sleep not only plays a vital role in our health but also affect our memory attention and metabolic function However having a good sleep quality is not easy for the sleep disorders especially the insomnia patients The clinical use Polysomnography (PSG) to record sleep physiological signals all-night from these patients The recording will be manual sleep scoring by the expert for diagnosis Except for the sleep stage the expert also references the objective sleep measurements to assess the sleep quality and symptom Since manual scoring is a very subjective and time-consuming work there are many automatic sleep scoring methods have been proposed Although these methods have a good performance in the sleep scoring the accuracy of sleep measurements is rarely a concern in their system Moreover the most system only fit on the subjects with good sleep efficiency the bad sleep efficiency is rare In this study we propose a system with accurate estimation of sleep measurements and fit on the healthy subjects and the insomnia subjects The main classification method of the system is a rule-based decision tree which combines the expert knowledge Generally the automatic sleep scoring method may not consider the differences between groups Our study merges the expert’s observation of the healthy and the insomnia subjects and proposes two suitable models respectively Moreover an automatic selecting method that could choose the proper model is also proposed The agreement between the expert and the system is 85 38% and the mean bias of the sleep measurements including sleep efficiency sleep onset time wake after sleep onset and total sleep time are only 1 32 min -1 34 min -4 06 min and 5 4 min respectively In the future the elderly can be integrated into the system to improve the system By doing the needs assessment with experts the system can be modified into the one that can be used clinically

Date of Award	2017 Aug 24
Original language	English
Supervisor	Sheng-Fu Liang (Supervisor)