Dados do Trabalho

Título

The development of an algorithm for detection of sleep using signals of movement and snoring using the Biologix system

Introdução

Obstructive sleep apnea (OSA) is extremely common and largely underdiagnosed. Home sleep test that does not use EEG are alternative to traditional polysomnography (PSG). However, in contrast to PSG, the results are expressed in relation to total recording time and not corrected for sleep efficiency. Biologix system consists of a high resolution oximetry with a built in accelerometer that also collects snoring from the mobile phone and uses cloud computing. Biologix has been recently validated against polysomnography for OSA diagnosis.

Objetivo

To develop an algorithm based on Biologix signals of movement and snoring to discriminate sleep from awake time.

Métodos

We studied 268 patients with suspected OSA underwent standard laboratory PSG and Overnight Digital Monitoring (Biologix Sistemas Ltd., Brazil). Actigraphy and snoring signals served as input to the neural network that used PSG data as the gold standard. Sleep detection was based on a neural network that used the Tensorflow library. The population was divided in training (70%) and testing (30%).

Resultados

Sleep classification had 82.7% accuracy, 91.3% sensitivity and 58.0% specificity. The total sleep time (TST) obtained by the neural network correlated with PSG-TST (r=0.77, p < 0.001). The comparison between PSG-apnea hypopnea index and Biologix-Oxygen Desaturation index (ODI) using total recording time improved when using sleep time estimated from the neural network with a mean error of 5.4 vs 6.4 (p < 0.05) and a limit of agreement of 16.2 vs 17.3, respectively. Biologix-ODI using the estimated sleep time demonstrated good performance for moderate-to-severe OSA diagnosis, with 94.2% accuracy, 91.4% sensitivity and 100.0% specificity.

Conclusões

The developed algorithm presented good performance for detecting sleep. The algorighm improves the accuracy of Biologix system as compared with PSG for determination of OSA severity.

Palavras-chave

obstructive sleep apnea, automatic sleep classification, home sleep test