Can a Sensor Ring Read Skin Gases after a Meal?

Could differences in what we eat also appear in the trace compounds emitted by our skin? In 2026, Wenhao Ye and colleagues integrated tiny gas sensors into a ring and examined signal changes associated with food intake and exercise in six healthy people.

The study aimed to continuously read volatile organic compounds emitted by the skin using a wearable device. Volatile organic compounds are organic substances that readily enter the air. This was not a study of body-odor intensity perceived by the human nose or of fragrance preferences.

From large analytical instruments to wearable sensors

Detailed analysis of substances emitted by the skin uses techniques such as gas chromatography–mass spectrometry (GC–MS), which separates components and examines their molecular characteristics. Large instruments, however, cannot be carried around in everyday life.

The researchers therefore assembled semiconductors whose electrical resistance changes on exposure to compounds on a small chip. Rather than each sensor selecting only one specific molecule, the system combines responses from multiple sensors to infer which condition a signal resembles.

The chip uses structures containing arrays of tiny holes and different metal oxides. By combining upper and lower materials, the researchers created nine sensing elements with different responses. Packing this response diversity into a small area is a key technical feature.

The active device region occupies 0.0081 square millimeters, but the chip including its support is 3 mm square. It was integrated into a ring-shaped device together with flexible circuitry for wireless communication and signal processing, a display, and other components.

Starting with gases of known identity and concentration

Before testing people, the researchers examined responses to seven gases, including acetone, ethanol, and ammonia. They varied concentrations stepwise from very low levels and repeated measurements in dry air and at different humidities.

The result was a different combination of responses across the nine sensors for each gas. Machine learning classified these differences, while changes in the responses over time were used to estimate concentration. Identifying gas type and estimating quantity were evaluated as separate tasks.

They also tested classification and component-specific concentration estimation in binary mixtures drawn from five gases. This addressed conditions in which humidity and mixing complicate signals, but does not mean that every mixture around the skin could be identified.

Six participants consumed specified foods

The human experiment involved three healthy men and three healthy women recruited at the university. There were six dietary conditions: an alcoholic beverage, fruit, meat, carbohydrates, nuts, and vegetables. Examples included orange juice for fruit, chicken breast for meat, rice for carbohydrates, and lettuce for vegetables, with amounts specified in advance.

Participants wore the ring on one hand while signals were recorded wirelessly. In some measurements, volatile compounds emitted by the other hand were also collected for subsequent GC–MS analysis. This was a more restricted experiment than everyday eating, when people combine many different foods.

After eating, differences appeared in sensor-response magnitude, the time taken for responses to change, and the shapes of the curves. The researchers used these response patterns as clues for classifying the dietary condition under which they had been obtained.

What does 98.20% accuracy describe?

The dietary-classification accuracy of 98.20% reported in the paper came from a comparison of four machine-learning methods using participant 01’s data as an example. The best-performing method classified observations by reference to nearby data with similar responses.

This figure does not mean that the ring can identify what any new wearer has eaten with 98.20% accuracy. The study also showed that signals for dietary categories could be separated in an analysis pooling all six participants, but that is a different result from the accuracy obtained in one participant’s example.

Moreover, the main paper and supplementary material do not adequately describe how training and evaluation data were separated for dietary classification or whether performance was tested in previously unseen participants. Independent validation is needed to judge performance for people using the device for the first time in everyday life.

Skin compounds changed, but their origins were not established

GC–MS comparisons were conducted on skin volatiles collected from three participants. They confirmed differences in component signals between baseline samples collected before food intake and post-meal samples. This supports the possibility that eating changes the mixture of compounds emitted by the skin.

However, analyzing compounds with another instrument is not the same as identifying every individual molecule using the ring alone. The ring’s classifications use patterns of responses across multiple sensors.

The authors discuss how processes such as the liver’s handling of fructose or staged protein digestion in the stomach and intestine might explain differences in post-meal curves. These explanations are interpretations based on earlier studies. The ring measurements did not establish that a particular peak originated in a specific organ.

Exercise and alcohol-intake estimation were tested separately

For exercise, the researchers compared three states: rest, running, and subsequent recovery. Participants remained quiet for ten minutes, ran on a treadmill for 30 minutes, and then rested for another 30 minutes. Sensor signals changed over this sequence, and the three states could be classified within the study’s data.

Intake estimation was examined in a limited experiment varying the amount of alcohol consumed. Using a different AI model that handles time-dependent changes, agreement between actual intake and estimated values reached 0.991 on a measure called the coefficient of determination.

The coefficient of determination describes how well estimates fit variation in numerical values. It does not mean that “99.1% of meals were correctly identified,” nor that the quantity or calories of every food were measured with the same accuracy.

Toward objective clues for tracking diet

The authors emphasize the possibility of continuously investigating actual food intake and adherence to dietary plans using bodily signals rather than relying solely on self-reports. Miniaturized sensors could provide a new measurement tool for tracking day-to-day changes in nutrition research.

The study also prototyped a system that sends classification results to an app, where generative AI provides health-related guidance. However, it did not evaluate the medical accuracy of that guidance or whether it improves diet or health.

Only six people participated, and the foods and measurement conditions were limited. Further validation is needed to determine whether the system can handle individual physiological differences, mixed meals, ambient gases, and sensor changes during long-term use. It cannot yet be described as a tool for diagnosing disease or generally measuring an individual’s food intake.

The research received support from public research funding in China and Hong Kong, university funds, foundations, and other sources. The authors declared no other competing interests. The study was conducted with participant consent and university research-ethics approval.

Summary

In experiments involving six healthy people, a ring incorporating miniature sensors detected differences in skin-volatile signals associated with eating and exercise. Packing differently responsive sensing elements onto a small chip and analyzing their combined responses and changes over time made wearable measurement possible.

The study’s contribution is to demonstrate a route toward continuously examining trace compounds emitted by the body in settings closer to everyday life. If future work validates performance in new people and with complex diets, chemical signals from the skin could become an objective source of information for nutrition research and assessment of actual food intake.

References

Ye, W., Ding, R., Wang, C., et al.(2026). Miniaturized olfactory sensor chip-based AI-wearable biometric ring for human body metabolic odor analysis. Nature Communications, 17, 4541. DOI: 10.1038/s41467-026-70746-z. Full original article. © The Author(s) 2026. The original article is published under CC BY-NC-ND 4.0.