Sleep, Sleep Disorder
Conditions
Keywords
Sleep, Polysomnography, EEG, PPG, Wearable Device
Brief summary
In this study, a two-part recursive convolutional neural networks model was developed, extracting features for each epoch window independently from before and after sleep onset (epoch encoder), and then trained in the context of long-term relationships in the sleep process (sequence encoder), using an approach similar to human expert classification based on information from single-channel forehead EEG and PPG (IR, Green, Red). The classification is based on guidelines from the American Academy of Sleep Medicine and calculated six parameters: total sleep duration (TST), wake (W), N1, N2, N3, and REM. The validation study of the developed model and the device was conducted at the Sleep Disorders Centre of the Istanbul Medical Faculty using concurrent polysomnographic data from 305 male and female patients aged 18 to 65 years.
Interventions
DEVICESleep tracking device
In addition to polysomnography, a device containing EEG+PPG sensors for sleep classification was placed on the forehead, and another device containing PPG and accelerometer sensors was placed on the wrist. The wrist to which the device is attached is randomly assigned.
Sponsors
Analog Devices
PNAPS Health Informatics and Space Technologies Inc.
Study design
Observational model
OTHER
Time perspective
CROSS_SECTIONAL
Eligibility
Sex/Gender
ALL
Age
18 Years to 65 Years
Healthy volunteers
Yes
Inclusion criteria
* Participants suffering from sleep disorders * Participants sent to PSG test by neurologists, pulmonologists, psychiatrists, and otolaryngologists
Exclusion criteria
* Anyone who has been diagnosed as having a contagious skin disease * Participants who do not have consent to have an additional device in their forehead area * Incomplete of sleep measurement
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Sleep Stages Classification Accuracy | 4-5 months | The collected EEG data were classified according to Cohen's kappa (\>85), which is considered successful in the literature. Initially the open source codes YASA, tinysleepnet and attentionsleep have been implemented. These codes yielded kappa 0.64, accuracy 0.80, kappa 0.69, accuracy 0.79 and kappa 0.65, accuracy 0.78 respectively. The values obtained do not correspond to those reported in the classification articles. Subsequently, 29 participants from our own dataset were tested in these classifications as a preliminary test, with poor results. On an individual basis, the highest cappa score was 0.51. Development of our own classification system is in progress. |
| Interoception analysis from PPG data collected from facial skin | 4-5 months | According to our preliminary analyses, we found that the intermediary rhythm (0.12-0.18 Hz) associated with interoception is also present in sleep patients. In one participant, for example, a value of 0.19 was obtained as a ratio of total sleep time. In addition, an intermediary rhythm is observed in all stages of sleep, including wakefulness, light sleep, deep sleep and REM. |
Countries
Turkey (Türkiye)
Outcome results
None listed