A Novel Study on Classifying Smokers in Apnea Patients Using Physiological Signals Processing

Millions of people worldwide are afflicted with a common sleep disease called apnea. Smoking affects the severity and course of therapy of apnea and is a risk factor for it. As a result, precise classification of smokers from non-smokers is necessary for efficient diagnosis and personalized care. This innovative study offers a thorough method for classifying smokers from non-smokers based on physiological indicators obtained from polysomnographic recordings of apnea patients, including nasal pressure, thorax effort, and chin electromyogram (Chin-EMG). We employed various machine learning classifiers, including XGBoost, logistic regression, k nearest neighbors, random forest, and support vector machines, on features extracted from polysomnographic recordings of apnea patients (equal number of smokers and non-smokers) randomly selected from the APPLES dataset. We evaluated the classifiers across different stages of sleep ($\mathbf{W}, \mathbf{N} 1, \mathbf{N} 2, \mathbf{N} 3, ~ R E M$) and apnea subtypes (OSA, CSA, mixed), and hypopnea events. Results demonstrate exceptional performance, with accuracy, and f1-score ranging from $\mathbf{7 5 \%}$ to $\mathbf{9 5 \%}$ for the EMG, thorax effort, and nasal pressure signals across all sleep stages and apnea subtypes. Notably, classification performance was highest during deep sleep (N3), suggesting that patients are more vulnerable during this stage of sleep and emphasizing the need for improved treatment. Our results show even stronger classification metrics to discriminate smokers from non-smokers compared to previous studies. Our findings will aid in the creation of personalized treatment plans and automated tools for diagnosing sleep apnea. Improved patient outcomes are made possible by the newfound insights, which enable early detection and focused therapies.