□ Daegu Gyeongbuk Institute of Science and Technology (DGIST, President Lee Kunwoo) announced that a research team led by Principal Researcher An Jinung at the DGIST Industrial AX Innovation Institute (he also serves as an adjunct professor in the Interdisciplinary Engineering), in collaboration with Professor Jeon Seong-chan’s team at Gwangju Institute of Science and Technology (GIST), developed technology that uses artificial intelligence (AI) to analyze electroencephalogram (EEG) signals triggered by thermal stimuli and objectively classify pain intensity.
□ As pain perception varies from person to person, previous methods relied heavily on the Visual Analogue Scale (VAS), a subjective scale expressed by patients. This resulted in inconsistent evaluations, even for the same stimulus, and posed significant limitations in accurately assessing pain for patients who have difficulty communicating, such as those with impaired consciousness, children or the elderly.
□ Principal Researcher An’s team developed a technology that uses AI to analyze EEGs generated during various thermal stimuli to classify pain intensity. Notably, shifting from conventional methods that simply learned from patients’ subjective pain scores, the team implemented an innovative algorithm in which two AI models compare their prediction results and selectively learn only from highly reliable data. Using this, the bias in pain expression, which varies from person to person, was effectively reduced.
□ Consequent to testing with EEG data from 41 participants, the model they developed demonstrated significant improvements in performance compared with conventional models and maintained stable predictions in new stimulus environments, in which the model had not been trained yet. Furthermore, they revealed that delta wave activity in the left and right anterior temporal lobes (F7 and F8) is closely associated with pain intensity, thereby establishing a neurophysiological basis for developing brain-based digital biomarkers.
□ “This study directly addresses the bias in subjective self-reported labels, which was the chronic limitation of EEG-based pain analysis,” stated An. “We intend to develop this into a universal pain AI platform that can be utilized in actual clinical settings by integrating various bio-signals.”
□ First author Jeong Ui-jin, a postdoctoral researcher, stated, “We hope this technology will be widely used for pain monitoring before and after surgeries, chronic pain tracking, and objective pain assessment in intensive care units,” adding, “Moving forward, we will devote ourselves to research so that it can be expanded into a brain-computer interface (BCI)-based real-time monitoring system.”
□ This study was conducted with support from the National Research Foundation of Korea’s Mid-Career Researcher Support Program and the Future Promising Convergence Technology Pioneer (Challenge Type) Program. The findings were published in the May issue of IEEE Transactions on Neural Systems and Rehabilitation Engineering, a prestigious international journal in rehabilitation engineering.
IEEE Transactions on Neural Systems and Rehabilitation Engineering
EEG-based Pain Classification via Sample Selection to Mitigate Subjective Label Bias
12-May-2026