Fuzzy Logic-Enhanced Deep Neural Networks for Sleep Apnea Severity Classification Through Snoring Detection


This research investigates the underestimated health implications of obstructive sleep apnea (OSA), a common condition affecting nearly 30 million individuals in the United States leading to significant health complications. The primary focus is on evaluating the severity of key symptoms through the development of a convolutional neural network (CNN). A CNN is trained to distinguish between snoring and non-snoring episodes providing a basis for calculating the snoring index, which quantifies the number of consecutive snoring episodes within an hour. Research has shown that both frequency and intensity of snoring can be correlated to the severity of sleep apnea.Complementing this analysis an oxygen saturation device connected to a raspberry pi is incorporated to measure blood oxygen levels (Spo2) and additional metric for assessing apneas and hypopneas. The combination of these three variables, snoring index, intensity and Spo2 form a basis for an approximate evaluation of the severity of sleep apnea symptoms. To achieve this, a fuzzy inference system is employed to categorize the severity of the symptoms into distinct levels: normal, mild, moderate, severe, and critical.



Audio classification, Convolutional Neural Networks, Fuzzy Logic, Sleep Apnea



Electrical and Computer Engineering