Sleep is a fundamental pillar for maintaining human physical and mental health, significantly influencing the proper functioning of various bodily systems. Sleep disorders are prevalent issues that, if left undiagnosed and untreated, can have severe consequences for individuals. One of the most common sleep-related breathing disorders is sleep apnea, which, if not identified, not only increases the risk of cardiovascular diseases and cognitive impairments but also imposes a substantial financial and economic burden on healthcare systems. Although clinical methods such as polysomnography (PSG) are considered the gold standard for diagnosing sleep apnea due to their accuracy, they are costly, time-consuming, and require attendance at specialized centers. Moreover, the potential for human error in interpreting results may affect their reliability.
Electrocardiography (ECG) signals, composed of components such as the P wave, QRS complex, and T wave, provide critical information regarding the heart’s electrical activity. During sleep apnea episodes, these signals exhibit variations in heart rate, leading to complex fluctuations in their temporal and frequency-domain characteristics. Compared to traditional machine learning-based approaches, deep neural networks (DNNs) offer a more effective means of analyzing these intricate patterns, significantly enhancing the accuracy of sleep apnea detection. This approach not only improves diagnostic precision but also substantially reduces the need for complex feature engineering processes.
In this study, a novel deep neural network-based method is developed for the automatic detection of sleep apnea using ECG signals. The dataset utilized is sourced from the well-established PhysioNet Apnea-ECG database. Following filtering and normalization, meaningful features are extracted from the signals. To enhance detection accuracy, a hidden Markov model (HMM) is employed in the post-processing stage. By leveraging temporal dependencies an