Efficient Deep Learning Model for Classification of Patients with Sleep Disorders from Healthy Individuals

We propose novel deep learning architectures, which are small and efficient in distinguishing individual sleep disorders from healthy people. We perform this classification in subject-independent mode, which shows its reliability in real-life scenarios. We propose models that utilize electromyogram, electrooculogram, and electrocardiogram and do not require electroencephalogram at all. This work uses a publicly available dataset called CAP Sleep database, which has overnight polysomnogram recordings from healthy subjects and patients with sleep disorders such as insomnia, narcolepsy, bruxism, nocturnal frontal lobe epilepsy, periodic leg movement syndrome, and sleep-disordered breathing. We have achieved F1 scores of 0.78 to 0.86 for the detection of different sleep disorders.