A domestic research team presented a method that analyzes brainwave signals with artificial intelligence (AI) to distinguish the degree of pain.
Ahn Jin-woong, principal researcher at the DGIST Industrial AX Innovation Headquarters, said on the 26th that his team, together with Professor Jeon Seong-chan's team at GIST, developed AI technology that objectively classifies pain intensity using brainwaves. The findings were published in the May issue of the Institute of Electrical and Electronics Engineers (IEEE) Transactions on Neural Systems and Rehabilitation Engineering, an international journal in the field of rehabilitation engineering.
People feel pain differently even under the same stimulus. Hospitals mainly use the visual analog scale (VAS), in which patients directly indicate their pain intensity with numbers. For example, "no pain at all" is 0 and "the worst pain imaginable" is 10, and the patient chooses a number that reflects their condition.
This method is simple and widely used, but it has limits. The criteria for feeling pain vary by person, and even the same person may answer differently depending on the situation. In particular, objective evaluation is more difficult for patients with impaired consciousness, children, and older patients who find it hard to accurately express their pain verbally.
To mitigate these problems, the research team focused on brainwave signals. Brainwaves are signals that measure the brain's electrical activity at the scalp and change during processes of stimulation or sensory processing. The team developed a technology in which AI analyzes brainwave data generated under various thermal stimuli to classify pain intensity.
Conventional AI models often trained by using patients' self-reported pain scores directly as ground-truth data. However, self-reported pain scores can involve individual variability and subjective judgment. To reduce this issue, the team had two AI models compare each other's predictions. Only data deemed relatively reliable by both models were selected for training.
In validation using brainwave data from 41 people, the newly developed model outperformed existing models. It also maintained relatively stable prediction performance under new stimulus conditions not used in training.
Principal Researcher Ahn Jin-woong said, "This study addresses the bias of subjective self-report labels, a longstanding issue in brainwave-based pain analysis," adding, "We aim to integrate various biosignals and develop this into a pain AI platform that can be used in real clinical settings."
References
IEEE Transactions on Neural Systems and Rehabilitation Engineering (2026), DOI: https://doi.org/10.1109/TNSRE.2026.3692232