A multi-feature classification approach to detect sleep apnea in an ultrasonic upper airway occlusion detector system

Obstructive Sleep Apnea (OSA) is the most common form of sleep disorder breathing. It is estimated that this insidious disease affects 15% of the US adult population. Current procedure of diagnosing OSA requires polysomnography (NPSG) conducted in accredited sleep laboratories and the data getting scored by certified sleep technicians, a costly process that is not readily available in all areas. Ultrasonic techniques are increasingly used in the area of medical diagnosis and treatments due to their safety and economic costs. This paper investigates a feasibility study of a multi-channel ultrasonic OSA detection system. The approach utilizes wavelet-based as well as temporal and spectral features extracted from multiple ultrasound waves transmitted through patient's neck during sleep. Using NPSG data as gold standard, the proposed classifier makes a preliminary decision on the data sequence by labeling epochs as normal or apneic. A Finite State Machine (FSM) is employed to update the classified labels for a more robust detection. Experimental results on three sleep disordered patients suggest that it may be feasible to consider the proposed approach for an ultrasound based detection system.