□ A research team led by Professor Ji-Woong Choi of the Department of Electrical Engineering and Computer Science at DGIST (President Kunwoo Lee), in collaboration with research teams led by Professor Han-Joon Kim of Kumoh National Institute of Technology, Professor Kyong Hwan Jin of Korea University, and Dr. Jong-Cheol Rah of the Korea Brain Research Institute, has developed "NONOS," a deep-learning-based analysis technology that precisely separates "rhythms" from the underlying "background signals" in brain signals with high temporal resolution.
□ Electrical signals generated in the brain contain a mixture of "rhythms (oscillatory components)," which repeat at regular intervals, and underlying "background signals (non-oscillatory components)." Isolating and analyzing these rhythmic components from brain signals is similar to picking out one person's voice amid the humming noise of a crowded café. In particular, in the brains of patients with Parkinson's disease, the strength of a rhythm known as "beta waves," which oscillate dozens of times per second, becomes excessively high. This is associated with symptoms of slowed movement and is used as an indicator for assessing treatment effectiveness.
□ The problem is that conventional analysis methods cannot properly distinguish rapid changes in rhythms from those in background signals. The most widely used filtering techniques calculate the strength of beta waves without accounting for background signals, potentially mistaking changes in background signals for changes in rhythms. More recent methods that remove background signals analyze signals by grouping them into fixed time windows, but these methods can easily miss brief changes in rhythms, just as fast-moving objects appear blurred in photographs taken with long exposure times. Consequently, it has been difficult to achieve both "accurate separation" of rhythms and background signals and "detection of rapid changes" simultaneously.
□ The research team focused on the following characteristic of background signals: their power decreases exponentially with increasing frequency, following a specific functional form. NONOS is an AI model trained to take brain signals recorded over time as input and extract only the background components that follow this functional pattern. By subtracting the extracted background signals from the original signals, only the pure rhythmic components remain, making it possible to detect rapid changes in both components without losing temporal resolution.
□ In fact, when the research team analyzed the rhythmic components extracted by NONOS from the brain signals of patients with Parkinson's disease, the performance of predicting which foot a patient would step forward with (with values closer to 1 indicating greater accuracy) reached 0.98, a significant improvement over conventional methods (0.54–0.79). In experiments using mice modeling Parkinson's disease, NONOS was also the only method that sensitively detected the rapid decrease in beta-wave strength following deep brain stimulation (electrical stimulation therapy), which had been difficult to identify using conventional methods.
□ "The significance of this study lies in demonstrating that rhythms and background signals mixed together in brain signals can be separated without losing temporal resolution," said Professor Ji-Woong Choi of DGIST (Director of the Brain Engineering Convergence Research Center). "If we further develop NONOS to operate in real time on implantable brain devices by reducing its computational requirements, it will contribute to improving the accuracy of personalized deep brain stimulation therapy, which adjusts electrical stimulation according to each patient's condition."
□ "NONOS has the advantage of being able to separate the periodic and aperiodic components of various biological signals that change rapidly over time with high temporal resolution, and thus it will be useful for analyzing a wide range of time-varying signals beyond the examples presented in the paper," said Professor Han-Joon Kim of Kumoh National Institute of Technology.
□ Meanwhile, this research was supported by the National Research Foundation of Korea's STEAM Research Program (BRIDGE Convergence R&D) and Brain Science Leading Convergence Technology Development Program, as well as the Ministry of Education's Glocal Lab program. Professor Ji-Woong Choi of DGIST and Professor Han-Joon Kim of Kumoh National Institute of Technology served as co-corresponding authors, while Jinmo Kim, a student in the integrated master's–doctoral program in DGIST's Department of Electrical Engineering and Computer Science, participated as the first author. The findings were published in NeuroImage, a top-tier international journal in the field of neuroimaging.
