USC researchers build an AI ring that senses stress with light instead of electrodes
The research prototype outperformed an electrical sensing method in study tests. Its on-device AI result is distinct from the strongest classification scores.
Loading page…
The research prototype outperformed an electrical sensing method in study tests. Its on-device AI result is distinct from the strongest classification scores.
Listen to this story
USC’s prototype points to a different design tradeoff for wearable stress tracking: optical sensing avoids the electrode-and-gel interface while combining sweat, heart-rate, temperature and motion signals to distinguish psychological stress from heat or exercise. The study reports strong results on specific classification tasks, not one universal accuracy score; autonomous on-device inference reached 87.53%, while the longer-wear comparison used a gel-based EDA setup. A 40-person, yearlong office validation is planned. The ring remains research-stage, with no consumer launch date announced.
For stress-versus-relaxation classification, the optical ring scored 92.3% accuracy, compared with 77.95% for EDA.
In short-term tests across stress, relaxation, movement and heat-induced sweating, it reached 100% accuracy versus 74.58% for EDA.
After 12 hours, the ring retained about 80% accuracy versus 57% for the tested gel-based EDA setup; this is not a comparison with every electrical sensor.
A ring that continuously tracks stress must do more than notice sweating: it must separate psychological strain from heat and movement. USC researchers have demonstrated a light-based prototype that did that in study tests. But its strongest classification scores and its autonomous, on-device AI result are different measures—not one all-purpose accuracy rating.
The work appeared November 13 in Science Advances and was described by USC on November 18. Led by assistant professor Yasser Khan and doctoral student Md Farhad Hassan, the team built a research ring that measures sweat activity beneath the skin using near-infrared light, alongside other physiological signals.
Conventional electrodermal activity sensing, or EDA, measures changes in how readily skin conducts electricity as sweat glands activate. Its weakness is the skin-electrode connection. Movement can disturb readings, and conductive gel can dry during extended wear. The paper notes that dry electrodes and alternating-current systems can reduce some long-term interface problems.
The researchers call their alternative optodermal activity, or ODA. It uses two near-infrared LEDs: one at 960 nanometers, sensitive to water changes associated with sweat, and another at 1,250 nanometers as a reference. Comparing the reflected light lets the ring track subsurface sweat activity without electrodes or conductive gel.
The same optical system also records photoplethysmography, a light-based measurement of blood-flow changes that supplies heart-rate information. Temperature and motion sensors add context. Together, these signals help the system distinguish psychological stress from physical activity or heat-induced sweating, rather than treating every sweat response as the same event.
The paper reports 92.3% accuracy for distinguishing stress from relaxation, versus 77.95% for EDA. In short-term tests covering four conditions—stress, relaxation, physical movement and heat-induced sweating—the optical approach reached 100%, compared with 74.58% for EDA. Those figures describe specific classification tasks in the study.
Optical-ring binary classification accuracy reported in the paper.
Short-term classification across stress, relaxation, movement and heat-induced sweating.
The paper separately reports autonomous accuracy using the ring’s neural network.
For autonomous operation, the ring runs a neural network locally to classify its signals. The paper reports 87.53% accuracy for that on-device inference. USC says local processing avoids sending raw biometric data to an external server for classification, keeping that processing step on the wearable.
Extended wear produced another useful comparison. According to USC, after 12 hours the ring retained approximately 80% accuracy distinguishing stress from relaxation. The gel-based EDA comparison fell to 57% as the gel dried and readings became less reliable. That result concerns the tested gel-based setup, not every electrical stress sensor.
Light does not eliminate every sensing problem. The researchers are working to reduce interference from excessive sweat on the skin’s surface and to broaden testing across cognitive and psychosocial stressors. These are continuing development goals, rather than capabilities established by the headline classification scores.
Khan plans a yearlong validation study with 40 participants in an office environment, USC says. That planned study would extend the work beyond the reported tests into sustained workplace monitoring. The university describes continuous tracking as a way to identify daily stress patterns that occasional check-ins or self-reports might miss.
For now, the ring remains a research technology. A patent application is underway, and USC says it may become available for commercial licensing through its Stevens Center for Innovation—not that a consumer release has been scheduled.
Loading discussion...
Join the conversation
Explain whether continuous feedback would feel useful or become another worry.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.