DMD on EEG reveals brain disorder patterns with 70% consistency

Researchers Jacob Kang and Jong-Hyeon Seo have applied Dynamic Mode Decomposition (DMD) to EEG signals in a new arXiv paper (2608.02804), targeting high-frequency neural dynamics that are often obscured by noise and low-frequency activity. DMD, a data-driven method from fluid dynamics, extracts coherent spatiotemporal modes from time-series data. The team used high-frequency DMD modes from neurologically relevant EEG channels as features, building a feature table for each signal. After post-processing, a random distribution test revealed that approximately 70% of samples exhibited consistent high-frequency dynamics within a specific channel, indicating that these signals are not random noise but carry structured, reproducible neural activity.

In classification experiments, the researchers applied principal component analysis (PCA) to the feature table and found that the resulting components formed a consistent pattern that reliably separated the alcohol-dependent group from the control group. This demonstrates that DMD-derived high-frequency EEG features are not only stable but also diagnostically discriminative. The study's approach could be extended to other brain disorders such as epileptic seizures, where high-frequency oscillations are known markers. By providing a robust method to detect and characterize these signals, the work opens a path toward more sensitive EEG-based diagnostic tools. Potential real-world applications include non-invasive screening for addiction disorders, real-time seizure monitoring, and brain-computer interfaces that rely on distinct neural dynamics. The paper, with 6 figures and 3 tables, is available for replication and further research in quantitative biology and machine learning.