Non-Parametric Spectral Analysis of Slow Wave Activity in Resting-State EEG as a Biomarker of Dementia Risk

Waldemar Bauer, Mariusz Pelc, Katarzyna Bialas, Tomasz Kajdanowicz, Aleksandra Kawala-Sterniuk

Abstract: Slow wave activity in electroencephalography (EEG) is emerging as a promising early biomarker for neurodegenerative disorders, including Alzheimer's Disease (AD) and dementia. In this study, we applied non-parametric spectral analysis methods to automate the detection of slow wave activity and stratify dementia risk using resting-state EEG data from 79 middle-aged participants in the PEARL-Neuro Database. Power spectral density (PSD) was estimated using Welch's method, employing Hann-windowed 4 second epochs with 50 % overlap to ensure robust handling of the non-stationary and noisy nature of EEG signals. Relative power was calculated for canonical frequency bands: delta (0.5 – 4 Hz), theta (4 – 8 Hz), alpha (8 – 13 Hz), and beta (13 – 30 Hz). Additionally, delta/beta and theta/beta power ratios were derived as potential biomarkers of cortical dysregulation. To account for non-normality in the data, we used non-parametric statistical methods, including Spearman's rank correlation, to assess relationships between absolute and relative EEG power across all recording sites and clinical scores related to attention deficit and hyperactivity. Our results revealed that increased delta and theta power, along with elevated delta/beta and theta/beta ratios—particularly in frontal and parietal regions—were significantly associated with higher dementia risk profiles. These findings demonstrate that non-parametric spectral analysis of resting-state EEG provides a reliable and scalable approach for identifying early neurophysiological markers of dementia risk, supporting its potential utility in both clinical and research settings.

External IDs:dblp:conf/icdm/BauerPBKK25