Wearable sensor with AI detects fatigue and stress

Researchers at the National University of Singapore have developed a wearable sensor that enables continuous monitoring of fatigue and stress, even during movement. The system combines a novel hydrogel material with AI-based signal processing.

Fatigue and burnout pose significant risks, particularly in professions requiring sustained alertness. However, current diagnostic approaches rely largely on subjective questionnaires that are not suitable for continuous or real-time assessment.

A research team led by the National University of Singapore (NUS) has now introduced a wearable biosensor platform that enables objective, continuous monitoring of fatigue using cardiovascular signals. The findings were published in Nature Sensors.

Wearable devices typically struggle to deliver reliable data during daily activities. Motion artefacts caused by body movement, muscle activity and physiological interference often distort weak signals such as electrocardiograms (ECG) or blood pressure measurements.

Existing solutions usually address only limited noise sources. The NUS team developed a system that suppresses multiple types of interference simultaneously, improving signal quality under real-world conditions.

At the core of the system is a metahydrogel artefact-mitigating platform (MAP), which combines advanced materials with AI-driven signal processing. The hydrogel integrates two filtering mechanisms:

• Nanoparticle structures that absorb and scatter mechanical vibrations
• A glycerol-water electrolyte that filters electrical noise while preserving low-frequency heart signals

This approach improves ECG signal-to-noise ratio from 5.19 dB to 37.36 dB and increases peak detection accuracy from 52 percent to 93 percent. Blood pressure measurements achieved deviations as low as 3 mmHg, meeting ISO clinical standards.

"Compared with current commercial devices, our metahydrogel platform demonstrates superior performance, particularly under motion conditions where artefact suppression is critical. Current smartwatches typically achieve ECG signal-to-noise ratios of 10–20 dB, which can decrease by approximately 40 percent under motion due to artefacts and unstable contact. Our system achieves around 37 dB during daily activities," said Dr Tian Guo, first author of the study.

Using high-quality cardiovascular data collected by the wearable sensor, a deep-learning system was able to classify fatigue levels with 92 percent accuracy. In comparison, models trained on conventional data without the new platform achieved only 64 percent.

The system captures physiological changes linked to the autonomic nervous system, including heart rate variability and ECG waveform patterns, which are associated with fatigue and stress.

Beyond fatigue detection, the platform demonstrated applicability across multiple biosignals, including heart sounds, respiratory signals, brain activity, eye movement and voice recordings. This highlights its potential for broader use in neurophysiological monitoring and digital mental health applications.

The researchers aim to further develop the system in collaboration with clinicians and industry partners. "We hope to work closely with mental-health physicians to better understand what types of physiological data are most relevant in real-world settings, as well as the level of accuracy required to meet clinical needs. Clinicians can provide valuable insights to help us establish meaningful links between the data and pathological conditions," said Professor Ho Ghim Wei.

In parallel, efforts are underway to improve manufacturing processes and scalability to support future commercialization of the wearable sensor platform.

MEDICA-tradefair.com; Source: National University of Singapore