Fusion of image and signal processing for the detection of obstructive sleep apnea

Patients with obstructive sleep apnea (OSA) syndrome experience repeated periods of apnea and arousal during sleep. A condition which in short term leads to excessive daytime sleepiness and in the long term may have clinical consequences such as stroke and cardiovascular abnormalities. Although complex equipment can be used to screen for sleep apnea, the screening tests are often expensive, inconvenient for the patient, and time-consuming to be manually analysed. This research investigates methods for automating sleep apnea screening using low-cost off-body cameras. Polysomnography video recordings of twenty one patients, 11 with OSA, and 10 `normals´ who were referred for suspected OSA, were analysed with the objective to differentiate the two groups. The proposed technique is based on motion estimation in videos using two successive video frames. The complexities of motion signals from the video data were analysed by calculating sample entropy over multiple time scales. The sample entropy values providing the best separation between the OSA and non-OSA groups were chosen using the Bhattacharyya distance and were then used as the input to a support vector machine classifier. The classification results both on the training and validation data indicate that patients with OSA can be differentiated from patients without OSA with 90% accuracy.