DGIST Develops AI Technology to Separate Brain Signals... Walking Prediction Performance Rises from 0.54-0.79 to 0.98
Detects Signal Changes in Response to Deep Brain Stimulation... Potential for Personalized Treatment use

Performance in predicting which foot a patient with Parkinson's disease will step forward with based solely on brain signals has improved significantly. When artificial intelligence (AI) was used, prediction performance rose to 0.98, up from 0.54-0.79 with existing analysis methods. This was achieved by removing background signals mixed in with brain waves and precisely analyzing only movement-related signals. The technology is also expected to be used in personalized Parkinson's disease treatments that adjust electrical stimulation according to a patient's condition.


On October 27, the Daegu Gyeongbuk Institute of Science and Technology (DGIST) announced that a research team led by Professor Choi Jiwoong of its Department of Electrical Engineering and Computer Science had developed NONOS, a deep-learning technology that separates background signals from rhythms in brain signals, in collaboration with Kumoh National Institute of Technology, Korea University, and the Korea Brain Research Institute.

A comparative overview of existing and proposed technologies for analyzing brain-signal rhythms. Existing methods have been limited by their inability to properly remove background signals from brain waves or capture momentary changes. NONOS directly extracts and removes background signals, enabling the oscillatory components of brain waves to be analyzed with high temporal precision. Image courtesy of the research team.

A comparative overview of existing and proposed technologies for analyzing brain-signal rhythms. Existing methods have been limited by their inability to properly remove background signals from brain waves or capture momentary changes. NONOS directly extracts and removes background signals, enabling the oscillatory components of brain waves to be analyzed with high temporal precision. Image courtesy of the research team.

View original image

Removing Background Signals from Brain Waves and Capturing Even Momentary Changes


Electrical signals generated in the brain contain both "rhythms (oscillatory components)" that repeat at regular intervals and underlying "background signals (non-oscillatory components)." It is similar to picking out one person's voice in a cafe where many people are talking.


"Beta waves," which oscillate dozens of times per second, are particularly important for people with Parkinson's disease. Excessive beta-wave activity is associated with symptoms such as slowed movement, making it a useful indicator for analyzing patients' motor condition and changes in response to treatment.


The problem is that existing analysis methods can mistakenly interpret changes in background signals as changes in beta waves. Other analysis methods that remove background signals also process signals by aggregating them over a period of time, making it easy to miss changes that occur briefly.


NONOS, developed by the research team, takes advantage of the fact that the background component of brain signals decreases in a consistent pattern as frequency increases. The AI identifies and removes background signals, leaving only the rhythmic components. This makes it possible to analyze even momentary changes in brain waves while preserving temporal resolution.


Improved Walking Prediction Performance and Potential for Personalized Electrical Stimulation Therapy


When the research team analyzed brain signals from people with Parkinson's disease, the system predicted which foot they would step forward with at a performance score of 0.98. This was higher than the 0.54-0.79 achieved by existing analysis methods.


The difference was also confirmed in experiments involving mice with a model of Parkinson's disease. When deep brain stimulation, which delivers electrical stimulation to the brain, was administered, beta-wave strength decreased sharply. NONOS sensitively detected this change, which had been difficult to identify with existing analysis methods.

The research team (from left): DGIST Professor Choi Jiung, Professor Kim Hanjun of Kumoh National Institute of Technology, and Kim Jinmo, a student in the integrated master's and doctoral program in the Department of Electrical Engineering and Computer Science at DGIST. Photo provided by DGIST.

The research team (from left): DGIST Professor Choi Jiung, Professor Kim Hanjun of Kumoh National Institute of Technology, and Kim Jinmo, a student in the integrated master's and doctoral program in the Department of Electrical Engineering and Computer Science at DGIST. Photo provided by DGIST.

View original image

Deep brain stimulation is a treatment that alleviates Parkinson's disease symptoms by delivering electrical stimulation through electrodes implanted in the brain. Accurately assessing a patient's brain state can help provide appropriate stimulation.


However, this study only evaluated the performance of a technology for analyzing brain signals. It has not yet been shown to improve treatment outcomes in actual patients, nor has it been developed to the point where it can operate in real time in implantable brain devices. The research team plans to reduce the computational load and develop the technology for real-time analysis.


"We have shown that rhythms and background signals mixed together in brain signals can be separated without losing temporal resolution," said Professor Choi Jiwoong of DGIST. "If the technology is further developed to operate in real time in implantable brain devices, it could help improve the accuracy of personalized deep brain stimulation therapy by adjusting electrical stimulation to a patient's condition."



The study also involved research teams led by Professor Kim Hanjun of Kumoh National Institute of Technology, Professor Jin Kyunghwan of Korea University, and Dr. Ra Jongcheol of the Korea Brain Research Institute. The findings were published in NeuroImage, an international journal in the field of neuroimaging.


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing