Jaap van der Aar's mission over the past four years has been to combine unobtrusive wearables with specialized AI to improve how these conditions are monitored. One sleep disorder that Van der Aar has focused on is REM sleep behavior disorder (RBD), aiming to give clinicians better tools to recognize and treat this rare condition.
"Nightswimming deserves a quiet night" goes the opening line of the1993 song by American alternative rock band R.E.M. While night swimming may not be for everyone, a quiet night's sleep is one thing people value.
Sleep is not uniform though, and there are different sleep stages during a sleep cycle. One sleep stage of this cycle is REM sleep, short for Rapid Eye Movement. This is the sleep stage where vivid dreams take place.
REM sleep behavior disorder
An important aspect of REM is that one's muscles are temporarily paralyzed (known as muscle atonia), which prevents someone from acting out their dreams, such as running a marathon or swimming across an ocean. However, when people suffer from REM sleep behavior disorder (RBD) their muscles remain active.
"RBD is a sleep disorder where someone acts out their vivid dreams," says Jaap van der Aar. "This happens because their muscles aren't paralyzed, unlike in normal REM, and can lead to someone kicking, talking, and punching in their sleep."
The gold standard
Van der Aar just defended his PhD thesis, which he carried out at TU/e's Advanced Sleep Monitoring Group in the department of Electrical Engineering and in collaboration with Kempenhaeghe Center for Sleep Medicine, as part of the Eindhoven MedTech Innovation Center (e/MTIC) ecosystem.
His research focused on improving the accuracy of the models used to help monitoring sleep disorders, in an age where measuring sleep data is veering towards more comfortable sensors.
"The gold standard in hospitals is polysomnography (PSG), where sleep stages are evaluated every 30 seconds. The central issue is that patients need to wear a host of sensors and sleep in a hospital lab. Plus, it only captures a single night of data and does not give any long-term insight into their sleep. Nowadays, there is a shift towards using wearables such as headbands to measure brain activity or smartwatches to measure heart rate and movement."
Link to Parkinson's disease
For this PhD research, Van der Aar looked at ways to optimize the models in wearables for disorders like insomnia, apnea, and REM sleep behavior disorder (RBD).
"We now know that almost everyone who displays RBD eventually develops Parkinson's disease or a related neurodegenerative disease later in life," says Van der Aar. "It's the strongest so-called prodromal marker or indicator of such diseases. Accurate measurement of the sleep of those with RBD symptoms can facilitate earlier formal detection and treatment for a patient."
However, a critical issue with early detection and treatment is that only a small percentage of the general population is affected by the disorder.
"RBD affects less than 1% of the general population, mainly affecting males over 50 years old. But as current AI models are mainly trained on healthy young adults or on those with more prevalent sleep disorders, using these models to evaluate data from potential RBD patients can be insufficient. They often fail to classify the sleep stages in sleep disorders, as the sleep patterns look very different from those used to train the models."
Overcoming the small data problem
In general, Van der Aar would like to have hundreds - if not thousands - of sleep recordings to accurately train any AI models, but he started his work with a number far lower than this.
"For my study, we started a clinical study with 72 patients to evaluate wearable EEG headbands and wrist-worn photoplethysmography (PPG) smartwatches. In the beginning, the patients wore these in the comfort of their home for a week, and then afterwards, in the clinic, we measured their sleep stages using PSG," says Van der Aar.
Sleep terms of note
Polysomnography (PSG)
This is the gold standard sleep study. In PSG, a patient is required to wear a host of sensors that monitor brain activity (via EEG) and heart rhythm (ECG). The sleep stages of the patient in question are monitored every 30 seconds. A PSG sleep study is performed overnight in a clinical or hospital setting.
Electroencephalography (EEG)
The brain is a hive of activity, even when one is asleep. This activity is in part related to the creation of electrical signals by the neurons in the brain. EEG is an approach that users sensors to measures these electrical signals. Such signals can be measured using one or more electrodes placed on the skull of a patient.
Photoplethysmography (PPG)
Changed in blood volume in tissue can be measured by detecting how much light is absorbed by the tissue. This non-invasive optical-based technology is known as PPG. In PPG, an LED emits light that is partially absorbed by tissue. A detector beside the LED measures the amount of light that is reflected by the tissue. When blood moves through tissue it absorbs more light. This leads to a decrease in the light reaching the detector, and also produces a signal related to pulse rate. PPG measurements is integrated into many commercial smartwatches.
(Photo: iStockphoto)
As it turns out, 72 patients are not enough to train the AI models needed by Van der Aar and his colleagues. So, they turned to the SOMNIA dataset, a large dataset of sleep recordings from thousands of patients. Such a dataset can be used to help train a new AI model, such as that Van der Aar required in his study.
"These larger datasets have been used to train a general model, after which the AI models are adapted to the wearable sensors and rarer sleep disorders in the 72 patients in my study," notes Van der Aar.
As he wanted to measure sleep data for a long period of time, Van der Aar switched focus to the PPG smartwatches. "These are more suitable and easier to wear for the patient."
Early detection counts
In effect, Van der Aar examined the potential for improving sleep stage monitoring using smartwatches and did so in the patient's own home.
"Early detection for RBD can be greatly facilitated by the accurate monitoring of patients who may have RBD in the first place. This data can then help the clinician to make an accurate formal diagnosis of RBD. As a result, it may be possible to diagnose Parkinson's disease earlier."
"All of those in the study had been referred to Kempenhaeghe with an undiagnosed sleep disorder and wanted to pinpoint which sleeping disorder they had. We took sleep measurements to help train our AI model, but not to diagnose the patients - that's up to the clinicians."
Of the 72 patients, several were eventually diagnosed with insomnia, obstructive sleep apnea, night terrors, and other sleeping disorders. Just two of the cohort were diagnosed with RBD.
Hypnodensity help
As outlined in a previous article , the Advanced Sleep Monitoring Group at TU/e is embracing the use of hypnodensity graphs to analyze sleep data.
"Hypnodensity graphs don't label a 30-second window of sleep with a single stage. Instead, such graphs consider the probabilities of different sleep stages in these windows," says Van der Aar. "We found that using these graphs can capture new information about the sleep of RBD patients that cannot be identified using traditional sleep scoring."
Huge opportunities
According to Van der Aar, there is incredible potential for his research on wearables to monitor sleep in combination with advanced AI-models adapted to small data sets to help many people with common and rare sleep disorders.
"Thanks to these AI-models that can handle specific sensors, specific individuals, and specific disorders such as RBD, my dream is that sleep monitoring becomes easier for all to do and moves from the hospital to the home. There are huge opportunities with this research to make a difference."
Next up for Van der Aar is a combined postdoctoral position at the TU/e and Radboud UMC in Nijmegen. "For me, the most impact can be made when operating between the academic and clinical settings. That's big for me, and I'm excited for what's to come."
Yet, thanks to the work Van der Aar and his colleagues in the Advanced Sleep Monitoring Group at TU/e, those struggling with sleep disorders such as RBD will in the future receive timely detection and treatment so that they can have a quiet night's sleep. And like everyone, they deserve it.
