Classification of Welch Periodograms Using Deep Learning Methods: A Case Study on the Prediction of Sleep Apnea

DergiPark İngilizce

This study proposes a novel approach for the diagnosis of sleep apnea using deep learning analysis of Welch Pediograms derived from electrocardiogram (ECG) signals. Despite affecting between 3% and 17% of the general adult population, depending on diagnostic criteria used, sleep apnea remains underdiagnosed due to limitations in traditional diagnostic methods, particularly polysomnography, which is costly, requires specialized facilities, and may disrupt natural sleep patterns. Our research addresses these challenges by developing an automated diagnostic system that analyzes frequency spectrum variations in RR intervals through Welch Pediograms. These visualizations, derived from Holter monitor recordings, effectively capture the characteristic frequency-domain changes in heart rate variability associated with sleep apnea episodes. Using Convolutional Neural Networks (CNNs), our methodology processes these spectral representations to identify patterns indicative of sleep apnea. This approach offers significant advantages including non-invasive monitoring, home-based testing capability, and improved patient comfort. Our findings contribute to both the methodological development in sleep medicine and the practical implementation of accessible diagnostic tools for sleep apnea, potentially improving early detection rates and patient outcomes.