Nox Research, the Analysis, Data, Automation, and Research team at Nox Medical, regularly presents its work at international conferences in the field of sleep medicine, AI, and data science. At this year’s SLEEP 2022 Meeting, Nox Research will present a new research poster. Titled, A novel deep learning AI method for estimating wake, NREM, and REM sleep from breathing and activity, the poster introduces a novel method for estimating sleep time by analyzing breathing parameters.
There has been a trend toward applying artificial intelligence (AI) to extract more information from sleep recordings. Here we present an AI model that estimates the states of wake, NREM, and REM from respiratory inductance plethysmography (RIP) and activity signals. There are known physiological changes in breathing between the three sleep states. The breathing-based model may be less susceptible to errors due to arrhythmias and heart rhythm medications than models relying on cardiac-derived signals. The model uses a special deep learning architecture which allows the model to learn important signal characteristics and preserves contextual temporal information that facilitates the sleep state prediction. The model was trained using manually scored sleep recordings from sleep centers in 5 countries. The final validation was carried out on a previously unseen external dataset from a separate sleep clinic to evaluate how the model would perform in the real world when it encounters new data from a new sleep clinic. The model shows high agreement on all sleep states in a cohort representative of the patient population visiting a sleep clinic in the United States. The model uses signals routinely collected as part of a type 3 home sleep apnea test (HSAT) and has the potential to enhance the value of such sleep tests.
Please join the Research Poster Presentation:
Tuesday, June 7
From 6:15 – 7:15 PM
Hall C2, Board #330
*Nox BodySleep is not available in the United States.
Photo: Nox Research presents a research poster at Vancouver’s 2019 World Sleep Congress.
Topic: Research & Publications