Explainable Computer-Aided Detection of Obstructive Sleep Apnea and Depression

Abstract

Obstructive Sleep Apnea Syndrome (OSAS) and Major Depressive Disorder (MDD) are common conditions associated with poor quality of life. In this work, we aim to classify OSAS and depression in patients with OSAS using machine learning techniques. We have extracted features from electrocardiograms (ECG), electroencephalograms (EEG), and breathing signals from polysomnography (PSG) at specific 5-minute intervals, where the participants' statuses are known, meaning we do not need breathing signals. These statuses include sleep stage, whether or not they have depression, or an apneic event has occurred. The PSGs were recorded from a total of 118 subjects with a 75/25 split for training and testing and the resultant features were used in sleep staging and classifying OSAS and depression in OSAS patients. Sleep staging was best done with random forest without feature selection, yielding an accuracy of 70.52 % and F1-Score of 69.99 %. The best classification performance of OSAS happened during deep sleep without feature selection and SVM, which yielded an accuracy of 98.36 % and F1-Score of 98.82 %. All sleep stages with Chi 2 ANN yielded an accuracy of 72.95 % and F1-Score of 73.43 % for classification of depression in OSAS patients. Results show promise in detecting OSAS and depression in OSAS patients, and the Bland-Altman plot shows that posterior probability provides comparable means of detecting OSAS to the apnea-hypopnea index (AHI). Besides detection of OSAS in depressed patients, this work serves to classify depression and give insights into relevant sleep stages to both of those conditions, allowing better planning for polysomnography.INDEX TERMS Electroencephalography (EEG), electrocardiography (ECG), obstructive sleep apnea syndrome (OSAS), depression, sleep staging, machine learning.The associate editor coordinating the review of this manuscript and approving it for publication was Humaira Nisar .with metabolic diseases, mood disorders, reduced cognitive performance, increased risk of accidents, depression, memory loss, cardiovascular complications, such as arrhythmias, coronary heart disease, heart failure, and strokes among other effects [1], [2], [3]. OSAS is a relatively common condition of sleep-disordered breathing, affecting 3-7 % of men and 2-5% of women in the general population [4], increasing to about 24 % and 9 % respectively when polysomnographic criteria are solely considered [5], [6].Depression or Major Depressive Disorder (MDD) is a common mental disorder characterized by reduced production of 110916

Comparing ANOVA and PowerShap Feature Selection Methods via Shapley Additive Explanations of Models of Mental Workload Built with the Theta and Alpha EEG Band Ratios

“…It is important to note that both the ANOVA F-score and Shapley-based feature selection methods have been utilized to analyze EEG data. These selection methods have been applied and compared in various situations, such as the diagnosis of Parkinson's disease [46], recognizing emotions [47], detecting sleep apnea and depression [48,49], and diagnosing schizophrenia [50]. Although there is a considerable amount of research comparing the ANOVA F-score and Shapley-based feature selection methods in different problem scenarios, there is limited research on comparing these feature selection methods for measuring the mental workload physiologically using EEG band ratios.…”

Effect of depression on phase coherence between respiratory sinus arrhythmia and respiration during sleep in patients with obstructive sleep apnea

“…Additionally, 32 controls without OSA or major depressive disorder are included from the Stanford Technology Analytics and Genomics in Sleep (STAGES) dataset (Zhang et al, 2018). Same data were analyzed and used in the following research (Moussa et al, 2022).…”