Temporal convolutional networks and transformers for classifying the sleep stage in awake or asleep using pulse oximetry signals

CASAL, RAMIRO; DI PERSIA, LEANDRO E.; SCHLOTTHAUER, GASTÓN

Journal of Computational Science

Elsevier

Lugar: Amsterdam; Año: 2022 vol. 59

1877-7503

Sleep disorders are very widespread in the world population andsuffer from a generalized underdiagnosis, given the complexity oftheir diagnostic methods. Therefore, there is an increasing interestin developing simpler screening methods. Pulse oximeter is an idealdevice for sleep disorder screenings, since it is a portable,low-cost and accessible technology. This device can provide anestimation of the heart rate (HR), which can be useful to obtaininformation regarding the sleep stage. In this work, we developed anetwork architecture in order to classify the sleep stage in awake orasleep using only HR signals from a pulse oximeter. The proposedarchitecture has two fundamental parts. The first part has the aim ofobtaining a representation of the by using temporal convolutionalnetworks. Then, the obtained representation is used to feed thesecond part, which is based on transformers, a model built solelywith attention mechanisms. Transformers are able to model thesequence, learning the transition rules between sleep stages. Theperformance of the proposed method was evaluated on the Sleep HeartHealth Study dataset, composed of 5000 healthy and pathologicalsubjects. The dataset was split into three subsets: 2500 fortraining, 1250 for validating and 1250 for testing. The overallaccuracy, specificity, sensitivity and Cohen?s Kappa coefficientwere 90.0%, 94.9%, 78.1%, and 0.73.