Single EOG channel performs well in distinguishing sleep from wake state for both healthy individuals and patients.

Using a single EOG channel, sleep-wake states of patients with different sleep disorders are accurately classified. We used polysomnography data of 27 patients (mixed apnea, periodic limb movement syndrome, sleep apnea-hypopnea syndrome, and dyssomnia) from DRMS-PAT and 20 healthy subjects from DRMS-SUB databases. We extracted a 67-dimensional feature vector, involving statistical features derived from ensemble empirical mode decomposition, approximate entropy, and relative powers in different frequency bands. Of these, the most relevant features are selected by exploiting mutual information between the features and corresponding labels. RUSBoost classifier is deployed to take care of the unbalanced data distribution. We achieved a high sensitivity of 97.5% and 95.3% as well as high specificity of 96.4% and 93.3% for sleep state in healthy and patients' groups, respectively. Ten-fold crossvalidation accuracies of 91.6% and 95% are achieved for patients and healthy individuals, respectively, using a single EOG channel. Clinical relevance-Accurate detection of sleep-wake states is crucial for the diagnosis of various sleep disorders including apnea-hypopnea syndrome and insomnia. Automated sleep-wake classification using EOG facilitates easy and convenient long-term sleep monitoring of patients without disturbing their sleep, thereby assisting the clinicians to analyze their sleeping patterns.