Disease-specific intrinsic frequency alterations in electroencephalographic signals of Alzheimer’s disease, frontotemporal dementia, and Parkinson’s disease

Resting-state electroencephalography (EEG) is widely used for clinically assessing neurodegenerative disorders; however, conventional spectral analyses such as discrete Fourier transform (DFT)-based amplitude spectra may not fully capture disease-specific alterations in intrinsic oscillatory dynamics, particularly under nonstationary conditions. We aimed to evaluate whether dynamic mode decomposition (DMD), a data-driven method capable of extracting intrinsic frequencies from nonstationary, multichannel EEG signals, can sensitively identify disease-related neural dynamics in Alzheimer’s disease (AD), frontotemporal dementia (FTD), and Parkinson’s disease (PD). Resting-state EEG datasets were re-analysed: a dementia dataset (AD, n = 36; FTD, n = 23; healthy participants, n = 29), a PD dataset (PD, n = 27; healthy participants, n = 27), and a cross-dataset validation cohort (AD, n = 35; healthy participants, n = 32).

DMD yielded dynamic modes (DMs), whose intrinsic frequencies formed temporal frequency DM (tfDM) features for classification with a linear support vector machine and nested cross-validation. Here we show that the DM frequency distributions differ from the amplitude spectra of the same EEG signals. Classification using tfDM features significantly outperforms that using amplitude features in dementia (62.16% vs 57.34%) and PD (71.11% vs 61.85%; both p