Sweaty hands, racing heart, rapid breathing: we most often experience these stress symptoms in social situations. This includes the critical interview as well as the bumpy blind date and the presentation that we have to give in front of the entire class. It is completely normal that we experience such psychosocial stress from time to time, but if it occurs too often it can become chronic. The result: health problems such as depression or cardiovascular diseases.

In order to protect people from chronic stress, we would need not only prevention but also reliable methods that can be used to measure a person’s stress level. So far, however, this has only been possible with blood and saliva samples and is therefore not very practical in everyday life. Researchers led by Robert Richer from the University of Erlangen-Nuremberg have now developed a method for the first time that can record social stress non-invasively and in real time.

The idea: When we feel stressed, our body posture and dynamics may also change – perhaps we move more than usual, less or differently. If we knew exactly how stress affects posture and trained artificial intelligence accordingly, these AI systems could detect in real time those people who are currently under stress.

In order to track down the typical human posture under stress, Richer and his colleagues first equipped several test subjects with small acceleration sensors on their heads, shoulders, chest, arms and legs, as well as hands and feet. The test subjects now had to complete two simulated job interviews and mental arithmetic tasks. In one run the examiners were friendly and supportive; in the other they did not react at all to the examinee and thus put him under pressure.

By evaluating the movement data from the two runs and comparing them with each other, the team was able to identify typical stress-related patterns in the subjects’ posture. In a next step, Richer and his colleagues trained an adaptive algorithm with this information and had it pick out in real time those test subjects who were typically stressed.

The result: In around 75 percent of cases, the AI ​​systems were able to correctly classify a person as stressed or relaxed, as the team reports. Stress was evident in all test subjects when they moved less overall or even completely froze for longer periods of time. For example, they crossed their hands in front of their body or stood straight as a rod instead of sometimes shifting their weight.

However, the extent to which someone freezes under stress is very individual, explains Richer: “A person who is generally more active will move more calmly in such a case, but may still have greater movement dynamics in comparison than a person who is generally less active emotional. But we were able to identify some form of movement reduction in all of them.”

In the future, the new method could help to examine stress without contact, to understand it better and thus to prevent it better. And that is urgently needed because: “We see that the burden of stress is increasing throughout society: This is reflected, for example, in the costs of health care and in the increasing number of absences from work,” explains senior author Nicolas Rohleder, also from the university Erlangen-Nuremberg. (Scientific Reports, 2024; doi: 10.1038/s41598-024-59043-1)

Source: Friedrich Alexander University Erlangen-Nuremberg

By Anna Manz

The original for this article “Are you stressed? Your posture reveals it” comes from scinexx.