By Gracié Roríguez

Sleep science has become crucial as researchers find strong links between sleep disruption and many health issues. As scientists learn more about the importance of sleep, the technology we use to measure and understand it has lagged behind. In order to better understand and diagnose sleep problems, researchers need to gather reliable real-world, long-term data to understand the cause and effect relationships of sleep on our health. 

Today’s status quo in sleep science: Polysomnography

A good night's sleep involves four main stages: awake, light sleep, deep sleep, and REM sleep. Our body cycles through these stages with different brain wave patterns, heart rate, temperature, and other signals. These signals can be used to identify the duration of each sleep stage and construct a map that reveals information about a person’s health.

Today, polysomnography (PSG) is the most common test used to monitor sleep architecture, using a set of sensors that pick up body signals while a person sleeps. One of these signals is electricity from the brain, captured by a technology called electroencephalography (EEG). Currently, PSG is used as the gold standard thanks to its precision and the wide variety of signals it can capture.

‍

The limitations of PSG

However, PSG is highly inconvenient and expensive to perform. People can only take this test in hospitals and sleep centers, where they have to spend the night in an unfamiliar bed with multiple uncomfortable sensors connected to their body. Unsurprisingly, the data collected from in-hospital PSG tests does not reflect a person’s usual sleep in the peace and quiet of home. Plus, people can only be kept in these facilities a few nights at most, limiting the amount of data that can be collected. 

Limited information leads to imprecise diagnosis and treatment. The shortcomings of PSG leave researchers, doctors and patients heavily underserved. Doctors and researchers have a pressing need to capture accurate data in real-world settings for long periods of time, using devices and methods that can be used by any adult or child sustainably without disrupting the person’s comfort. New technology that can maintain the accuracy and reliability of PSG, while overcoming its limitations, would be significantly beneficial for the medical community and patients alike.

The dawn of a new era in sleep science

At Neureka, we’re on a mission to bridge the divide, which is why we created the Neureka Sleep platform. Neureka Sleep harnesses the potential of noninvasive wearables to record sleep signals from the comfort of home. We have built a state-of-the-art algorithm that, once deployed, can process signals in real time and identify each sleep stage with top accuracy. 

Amazingly, our algorithm measures a person’s sleep architecture 83% as accurately as PSG, while collecting fewer signals from the body in a more comfortable way. Having sleep architecture recorded and analyzed every single night represents an unprecedented feat in the world of sleep science, which is bound to revolutionize how we understand sleep and diagnose health conditions sooner.

‍

‍

How our groundbreaking algorithm was born

The algorithm powering our technology was trained using 1,946 publicly available sleep recordings from the MESA study (Dean, Dennis et.al), the largest collection of data used to build sleep algorithms. From the 11 available signals, we gathered heart rate, motion, eye movement and blood oxygenation values that were already cleanly labeled with PSG-validated sleep stages from the MESA study, and trained our model to measure sleep stages just based on these 4 biometrics, thereby creating a leaner method for identification.

In order to do this with the greatest accuracy, we needed to break up this task in two steps. In the first step we challenge a neural network, which is a type of AI, to label the sleep sessions with 3 stages: awake and light sleep together as one stage; deep sleep as another stage; and REM sleep as the last stage. In the second step, another program distinguished between awake and light sleep, resulting in our four-stage classification. 

As next steps, to achieve leaner signal collection in the future, we set out to investigate the degree to which each signal contributes to the model. The most important factor to determine a sleep stage is heart rate variability, with its influencing factor being 4.47 times higher than any other signal collected. This result is consistent with other studies in this discipline (Korkalainen H et al, Yi R et al, Widasari ER et al).

‍

‍

Unprecedented accuracy compared to previous research

Our groundbreaking algorithm can take these four signals — heart rate, motion, eye movement and blood oxygenation— from the body and generate reliable, reproducible sleep staging with an accuracy rate of 83% when compared to PSG. Our results surpass all previous research that used the MESA database and other sleep stage classifications, putting our work as the most accurate that any algorithm using this database has gotten to replicating PSG results with such lean signals (Imtiaz). 

We reviewed previous research in this field, identifying 90 different machine learning techniques for sleep staging. All of the studies using these techniques got limited results: their average accuracy rate was 79%, with only 3 studies reaching above 80% (Imtiaz). Some of their limitations include only identifying 2 stages (asleep and awake) or considering far fewer sleep sessions due to inefficient processing. The studies with accuracy greater than ours use EEG, forcing them to be performed in hospitals or sleep centers.

‍

‍

Step into the future of sleep science with Neureka

Now we know a path forward exists with technology that addresses current unmet needs, a solution that is better and more convenient than existing alternatives. But how do we hope to bring this technological feat to real-world settings to benefit the lives of families with underserved needs related to sleep disruption?

One of the most underserved populations who would benefit from this technology are people with epilepsy, Alzheimer's, Parkinson’s and other neurological conditions. The quality of life of these populations could be significantly improved by providing them, their families and their doctors with real-world long-term information about their sleep architecture (American Neurological Association). Our flagship creation, Neureka Sleep, is already monitoring the sleep of families with epilepsy worldwide. Once deployed, our algorithm will generate unprecedented data to better understand people’s condition better than any current wearable platform in the field.

With this and other use cases, Neureka Sleep is heralding a new era in sleep science. As we step into the future of this field, the potential for real-world impact becomes palpable. Our platform is set to elevate families’ quality of life by providing invaluable, real-world, long-term insights into the sleep architecture of underserved populations: families facing sleep disruption due to neurological conditions. We’re giving life to our mission of revolutionizing sleep science from the comfort of home with the creation of this technology, standing as a testament to innovation, and a brighter, more informed tomorrow. 

How might the widespread adoption of advanced sleep monitoring technology reshape our understanding of sleep-related health issues on a global scale? We invite you to follow our journey as we crack the code on sleep and health disorders. 

To learn more about our algorithm, read our publication here.

If you’d like to know more about Neureka and follow our journey to change the future, you can join our social media handle @NeurekaAI.