Automated Rapid Eye Movement Detector for Sleep Microstructure Classification: An Adaptive Quasi-Random Shot-Grouping Approach
Salut!
My name is Willie. I am a posdoctoral fellow at University of Montréal and the Centre for Advanced Research in Sleep Medicine, under the supervision of Dr. Julie Carrier and Dr. Jean-Marc Lina. My research centres on comprehending the pathophysiology and evolution of dream enactment behaviors, such as REM sleep behavior disorder. With limited tools available to facilitate the exploration of the associated pathophysiological mechanisms within REM sleep, we have dedicated our recent efforts into tool development. In this poster, we are introducing one of these tools: an adaptive quasi-random shot-grouping system, designed to detect rapid eye movements (REM) and classify REM sleep microstructures. Our goal is to apply these tools for future research into exploring potential biomarkers in REM sleep for patients with dream enactment behavior.
Hope to see you at my poster. :)
ps. Don't be shy to stop by even if it is just to say hi.
Feel free to contact me by email: chun.william.yao.cnmtl@ssss.gouv.qc.ca
Automated Rapid Eye Movement Detector for Sleep Microstructure Classification: An Adaptive Quasi-Random Shot-Grouping Approach
Rapid eye movement (REM) sleep, commonly treated as a uniform state, consists of two microstructures: phasic (with REM events) and tonic (without REM events). Due to the time-consuming nature of visual scoring without automated REM detection, there has been limited assessment of the distinct neurophysiological mechanisms underlying these microstructures in humans.
The study was conducted using five polysomnography recordings (with one designated for testing), randomly selected from the Montréal REM Sleep Behavior Disorder (RBD) cohort, capitalizing on the abundance of artifacts resulting from RBD. We developed an automated quasi-random shot-grouping system to detect REM events for microstructure classification. Mimicking the Human Genome Project, the algorithm emulates visual REM scoring, treating the baseline electrooculographic signal (EOG) value as a “DNA strand” with EOG deflections as “bases”. The extracted features are then utilized to differentiate REM events from artifacts, employing adapted thresholds that account for potential biases introduced during preprocessing. The iterative greedy algorithmic process adapts the information acquired from prior iterations and used it in the following iteration of signal preprocessing with the exhaustive inventory of REM events as output.
The REM microstructure classification system demonstrated high precision (91.1%) while also maintaining a negative predictive value of 88.9%, with less than 20 seconds required per recording on average. Beyond REM detection and microstructure classification, the system extends its utility to event-based REM window estimation without the conventional time constraint (e.g., 3-second epoch).
With its high efficiency, the proposed automated system merits consideration for facilitating future studies on human REM sleep microstructures.
(Interactive PDF during presentation)