ºÚÁϸ£ÀûÍø

Traditionally, chest-worn accelerometers have been limited to detecting posture—particularly whether a patient is sleeping on their back, a known risk factor for OSA. This research dramatically expands their utility.

Advanced signal processing techniques and neural networks allow the extraction of detailed physiological signals from subtle chest movements. These include respiratory effort and instantaneous heart rate—two core components of sleep monitoring. Importantly, these methods remain robust even in patients with sleep-disordered breathing, a population where signal noise and variability are typically high.

By transforming raw motion data into rich physiological information, the accelerometer evolves from a simple positional tool into a powerful diagnostic sensor.

Teaching machines to understand sleep

A key innovation in this work of lies in the application of machine learning. Neural networks, trained on large datasets and fine-tuned for accelerometry signals, enable accurate estimation of both sleep stages and the AHI.

This is particularly significant because sleep staging has traditionally required complex and intrusive setups like polysomnography (PSG). Here, sleep architecture derived from accelerometry shows substantial agreement with PSG results, suggesting that reliable sleep analysis can be achieved with far less burden on patients.

At the same time, the ability to estimate AHI without airflow measurements addresses a longstanding limitation in non-CPAP therapies. This means that treatments like positional therapy can finally be evaluated with the same level of rigor as CPAP.

A feedback loop for every therapy

One of the most impactful implications of this research is the introduction of a feedback loop for therapies that previously lacked one. Devices used in positional therapy can now be equipped to monitor residual OSA, providing both patients and clinicians with actionable insights.

This has two major benefits. First, ineffective treatments can be identified and adjusted—or replaced—much sooner. Second, successful treatments can be validated with greater confidence, improving adherence and long-term outcomes.

In essence, therapy becomes dynamic rather than static, guided by continuous physiological feedback instead of assumptions.

Expanding beyond sleep: integration and accessibility

Another strength of this technology is its adaptability. Because chest-worn accelerometers are already widely used in medical devices, integration is relatively straightforward. For example, extended Holter monitors—commonly used for cardiac monitoring—could be enhanced to include sleep apnea screening.

This convergence of technologies could significantly increase the detection of previously undiagnosed OSA, particularly in patients who would not otherwise undergo sleep studies. It also reduces the need for complex, expensive, and resource-intensive diagnostic procedures.

A step toward smarter, patient-centered care

Ultimately, this thesis represents a shift toward more accessible and patient-centered sleep medicine. By enabling continuous, unobtrusive monitoring of both sleep and breathing, it empowers earlier diagnosis, more precise treatment, and better long-term health outcomes.

Patients stand to benefit in multiple ways: undiagnosed OSA can be identified, ineffective therapies can be corrected, and overall sleep quality can be better understood and improved. What was once confined to specialized sleep labs may soon become part of everyday healthcare technology.

The message is simple but powerful: meaningful sleep insights no longer require complex equipment—just smarter use of the signals already available.

Read more about the research of Sebastiaan Overeem, Merel van Gilst and their team on new ways to monitor sleep disorders. Besides that there were the PhD defenses ‘Deep generative modelling in sleep diagnostics’ of Hans van der Gorp in January 2026 and ‘Optimizing Automated Sleep Staging’ of Jaap van der Aar in December 2025. 

Title of PhD thesis: . Supervisors: Prof. Sebastiaan Overeem (ºÚÁϸ£ÀûÍø), Dr. Ruud van Sloun (ºÚÁϸ£ÀûÍø), and Dr. Pedro Fonseca (Philips).