Technical Perspective: Making Sleep Tracking More User Friendly

An exciting area of research in mobile and ubiquitous computing is the recent development of novel sensing systems capable of continuously tracking behavioral and physiological signals from individuals in their natural environment. Often referred to as digital biomarkers, these signals capture people's everyday routines, actions, and physiological changes that can explain outcomes related to health, cognitive abilities, and more.

A key behavioral biomarker is sleep, which is essential for human health, learning, cognitive abilities, and brain development. About one-third of adults suffer from some form of sleep disorder. However, current sleep-tracking options are mostly restricted to special sleep clinics or hospitals, where individuals are removed from their natural sleep environment and undergo polysomnography (PSG) that monitors brain activity via electroencephalography (EEG), eye movement via electro-oculography (EOG), and muscle activity using electromyography (EMG).

These bio-signals are combined to infer sleep duration, sleep quality, and stages of sleep (light, deep, and REM sleep). Beyond sleep, the ability to track these signals can lead to new types of brain-computer interfaces and the detection of alertness and interests, eating moments, autism onset, and more.

In the recent years, researchers in ubiquitous and mobile computing have pushed the boundaries of sensing digital biomarkers, and have created novel systems that can track important markers in the real world. Crucially, these systems are unobtrusive, meaning they do not require removing individuals from their natural environment or to be burdened with a bulky sensing setup in order to get reliable measurements. One of the main challenges in this domain is to balance the fidelity and accuracy of signals in the presence of natural usage variations with the burden that is placed on the users. Existing sleep and bio-signal sensing solutions that measure EEG, EOG, or EMG include wearable headbands, eyemasks, smart shirts, and wristbands that are not as cumbersome as PSG and are much less expensive, but still are not comfortable enough for individuals to use continuously over extended periods of time.

There has also been much work related to developing sleep trackers that leverage sensor signals and usage patterns from smartphones so that they require minimal effort from users. Such systems have been used to estimate sleep duration, interruptions, and even chronotype. More recently, contactless sensing approaches that use WiFi or Doppler Radar measures have been used to detect specific sleeping problems, such as sleep apnea or more general sleep quality, and sleep stages based on changes in breathing rate, heart rate, and movement. These contactless and smartphone-based methods place very low or no burden on the user. Moreover, they are generally reliable for coarser features of sleep such as duration and interruptions. However, they are less reliable for getting finer-grained information about sleep stages.

The Lightweight In-ear BioSensing (LIBS) system work detailed in the following paper provides a nice balance in terms of minimizing the burden on users and the granularity at which we can automatically track various measures of sleep. The custom flexible electrodes wrapped around off-the-shelf foam earplugs are able to pick up multiple signals of interest (EEG, EOG, EMG) from the ear canal, and are designed for user comfort, allowing individuals to sleep more naturally. More importantly, they can be used at home in a user's normal bedroom setting.

The LIBS approach is a great example of how to trade off signal quality and user burden.

Using signal separation and classification algorithms, the LIBS system can pull out the different bio-signals from the single mixed in-ear channel, derive relevant spectral and temporal features from these signals, and classify stages of sleep. The LIBS approach is a great example of how to trade off signal quality and user burden. Systems that focus on designing low-burden sensing and that push the boundary in terms of sensing granularity and resolution can significantly increase the adoption of mobile and ubiquitous solutions in the real world, especially in the realm of healthcare where fidelity and accuracy of the measures are important. This is even more important if diagnosis and treatment decisions are to be made based on the measurements.

LIBS is an elegant engineering solution to overcome real-world usability barriers, it is accurate, and it provides a reliable alternative to highly intrusive PSG-based measures. Of course, in order to assess how broadly applicable the system is will require more longitudinal testing across variable sleep environments and across people with different sleeping habits and in diverse age groups. Nonetheless, LIBS takes a significant step in the right direction, and is sure to inspire more solutions that can effectively balance accuracy and usability.