REM sleep behavior disorder, or RBD, is a condition that causes abnormal movements or brief repeated twitching during sleep, and occasional episodes of dream enactment.
THE CHALLENGE
RBD affects more than one million Americans and is almost always an early sign of Parkinson’s or dementia, often preceding other symptoms by 10-15 years. That means it presents an unprecedented opportunity for developing therapies against Parkinson’s disease or dementia and ultimately identifying those who would benefit from early prevention therapy.
However, RBD has been very difficult to diagnose.
“A simple screening question on whether or not people act out their dreams is poorly sensitive because many people with RBD rarely have full-blown dream enactment episodes but only have small twitches that they or their partners are not aware of,” explained Mount Sinai clinician Dr. Emmanuel During.
During is associate professor of neurology at the Icahn School of Medicine, where he sees patients at The Mount Sinai Integrative Sleep Center and specializes in movement and sleep disorders, as well as pulmonary medicine and critical care.
Mount Sinai Health System is one of the largest academic medical systems in the New York metro area, with 48,000 employees working across eight hospitals, more than 400 outpatient practices, more than 600 research and clinical labs, a school of nursing, and a leading school of medicine and graduate education.
“A simple screening question also is poorly specific because more common conditions – like sleep apnea or a form of restless legs movements during sleep – can cause symptoms of dream-enactment mimicking RBD,” During continued. “RBD questionnaires lack accuracy.”
The gold standard test for diagnosing RBD is an in-lab sleep test in a sleep center, also known as a polysomnogram, which measures muscle activity (increased in RBD) during REM sleep using muscle sensors, or electromyography.Â
But RBD has been very difficult to diagnose using this in-lab sleep test because it’s very difficult to interpret and subject to artifacts, to the point that even sleep experts can disagree on the diagnosis.
“Notably, although a video camera records any movements during sleep, the current standard for interpreting the test does not take any of that into account, and no automated tool currently exists for interpreting video data,” During noted.Â
“In fact, in most sleep centers, the video data is discarded after the sleep test and only the rest of the data collected – EEG, breathing signals, EKG, etc. – will be stored.”
Furthermore, he said, unless a person undergoing a sleep test is specifically suspected to have RBD by their sleep specialist – based on a history of dream enactment – the diagnosis is easily missed because in 99% of cases, the test is conducted to evaluate for sleep apnea, which does not require assessing muscle activity,” he added.
This leads to missing “incidental cases” of RBD, which occur in at least 1% of the adult population, according to studies. To address this need, the Mount Sinai research team developed a method for automating the diagnosis of RBD by analyzing video recordings of sleep during in-lab sleep tests.
PROPOSAL
The team developed an algorithm to automatically interpret the frequency and pattern of body movements detected during REM (rapid eye movement) sleep and determine whether, based on these movements, a person has RBD or not.
This had been done only by one group in Austria, but their study used a particular research-grade 3-D camera, which required being added to the standard sleep test hardware.
“No study before ours had tested video data routinely collected with the 2-D-infrared camera used in all clinical sleep labs around the world,” During said.
MEETING THE CHALLENGE
“We assembled a large dataset – larger than done in the prior study – comprising 81 sleep study recordings of patients with RBD (“cases”) and 91 without RBD (“controls”), including 63 with a range of other sleep disorders and 28 healthy sleepers,” During explained.
An optical flow computer vision algorithm automatically detected movements during REM sleep, from which features of rate (frequency), ratio (proportion of time in REM sleep showing movements), magnitude and velocity of movements, and ratio of immobility (a measure of the distribution pattern of movements in REM sleep) were extracted, he continued. From those five features, a machine-learning classifier was trained to differentiate RBD from other sleep conditions and normal sleep.
“We also were interested in testing the accuracy of the resulting classifier to detect RBD in patients who actually have RBD, but who, on the review of their sleep test, were not reported to move, based on manual review of video footages by a sleep expert,” During recalled. “A total of 11 such patients with RBD but no visible-with-the-human-eye movements were identified, and 71 patients with RBD with visible movements.”
RESULTS
The Mount Sinai team found, as they expected, that people with RBD exhibited an increased number of movements in REM sleep, particularly brief movements shorter than two seconds, including jerks or twitching known as myoclonus. Accuracies for detecting RBD ranged from 84.9% (with only two features) to 87.2% (with five features).
Combining all five features but only analyzing short (less than a 2-second duration) movements achieved the highest accuracy at 91.9%.
Of the 11 patients with RBD without noticeable movements during the sleep test, seven were correctly identified, or detected as having RBD, based on the Mount Sinai algorithm.
ADVICE FOR OTHERS
“This is the first study showing that a simple algorithm analyzing video recordings acquired during sleep tests, conducted under routine clinical care, can diagnose RBD and with a very high accuracy of 91.9%,” During noted. “This work improves prior frameworks by using a 2-D camera that is routinely used in sleep laboratories, and improving performance by adding only three features.
“This approach could be implemented in clinical sleep laboratories to facilitate and improve the diagnosis of RBD,” he concluded. “Coupled with automated detection of REM sleep, it should also be tested in the home environment, using conventional infrared cameras to detect and monitor RBD.”
Follow Bill’s HIT coverage on LinkedIn: Bill Siwicki
Email him:Â bsiwicki@himss.org
Healthcare IT News is a HIMSS Media publication
WATCH NOW: Seattle Children’s Chief AI Officer talks better outcomes through the technology
