연세대 윤경호 교수팀, MRI·뇌파 결합해 뇌 신경 활동 위치 추정하는 AI 개발

Professor <b>Yun Kyung-ho</b> from Ewha University and his research team have developed an artificial intelligence (AI) technology that combines MRI and electroencephalography (EEG) to estimate the location of brain neural activity in real-time. This innovative approach improves upon existing EEG-based brain activity localization techniques by incorporating each individual's distinct brain anatomy directly into the AI, thereby enhancing both the accuracy and generalization performance, especially for new patients.

Electroencephalography, commonly known as EEG, measures the brain's electrical activity using electrodes attached to the skull, providing measurements in milliseconds. It is widely utilized for diagnosing brain disorders and analyzing brain functions. However, traditional EEG methods struggle to accurately pinpoint the exact location of neural activity within the brain, particularly because people have varying brain sizes and shapes, making it difficult to apply models trained on one person's data to others.

To address this challenge, Professor Yun's team developed a multimodal deep learning model that combines individual MRI-derived brain structure information with EEG's temporal and frequency data. By simulating various brain neural activities and EEG signals based on individuals' brain structures, they trained the AI to estimate the location of neural activity in three dimensions without any additional training for new patients.

Experimental results showed that the proposed model achieved high accuracy in estimating the location of brain neural activity and maintained real-time processing capabilities. Furthermore, the model demonstrated stable application to MRI data from new individuals and outperformed EEG-based approaches in estimating brain neural activity locations when validated using actual electrical stimulation data.

This research highlights the significance of integrating both EEG signals and individual brain structures into AI learning, as well as generating a large-scale EEG dataset through simulations for AI training. The team also applied conformal prediction to provide not only the estimated location of neural activity but also an indication of the uncertainty of the prediction.

Professor Yun expressed his vision of evolving this research into an EEG foundation model that can learn diverse brain structures and neural activities across different individuals, ultimately extending to the development of a brain digital twin—a virtual representation of an individual's brain structure and function in a virtual space.

This technology would support the diagnosis and monitoring of neurological disorders like epilepsy, as well as personalized brain function analysis and precision medicine. The research was published in the international academic journal 'Computer Methods and Programs in Biomedicine.'