Pressure Sensitive Mat: An Alternative Sensor to Detect Sleep-Related Breathing Disorders

Sleep Apnea (SA) is a common disorder that affects approximately 2% of middle-aged women and 4% of middle-aged men. It is characterized by repetitive cessation of breathing during sleep. SA has significant health and social consequences such as daytime sleepiness, impaired quality of life, and in the worst case, myocardial infarction and sudden cardiac death. It has been estimated that approximately 80% of individuals with moderate to severe SA syndrome have not been diagnosed. The lack of patient sleep histories has caused low identification of SA and referral rates, especially in primary care facilities. Moreover, due to the inadequate prevalence of overnight polysomnography (PSG) as a standard clinical test of SA, patients suspected of having this sleep disorder have to wait several months for diagnosis and treatment. The costly and time-consuming nature of PSG and the lack of sleep clinics have created a demand for suitable home-based health monitoring devices. Over the years, several devices have been developed to monitor sleep unobtrusively, while an individual is lying in bed. However, most of these devices would either disrupt the sleep of the patient or be disrupted by the patient during routine bed sheet changes. Pressure measurement using a Pressure Sensitive Mat (PSM) enables a non-contact approach for monitoring patient vital signs such as respiration rate. The PSM has the potential to replace obtrusive breathing sensors in the sleep lab and to be used as a pre-screening tool for patients suspected of having sleep apnea. This thesis proposes multiple algorithms applicable to PSM in order to assess sleep quality. First, fusion techniques are proposed to extract a breathing signal from PSM. Second, a wide range of machine learning approaches including a simple threshold-based algorithm, a linear support vector machine (SVM) and two deep learning methods (i.e., a temporal convolutional network (TCN) and a bidirectional long short-term memory (BiLSTM) network) are compared to find a good- iii performing method for automatically detecting central sleep apnea (CSA) events from PSM signals. The results show that the accuracy of the model with the best performance is 95.1% and it is achieved by the BiLSTM network. Finally, by applying SVM, personalized systems are optimized to investigate long-term sleep pattern changes such as central apnea index (CAI), bed occupancy (BO), day-clock, and night-clock from previously recorded data.