An EEG Signals-based Deep Spatio-Temporal Feature Extraction with Dual Attention for Detecting Parkinson’s Disease

Research overview

The text below is the source article’s original abstract under a verified reuse license. cc by

Abstract Parkinson's disease (PD) is a neurological disorder, causing irregular brain oscillations that can vary rapidly, impairing motor and cognitive abilities. It is also very common, progressive, and complex. High-temporal-resolution electrophysiological monitoring, such as EEG, can provide valuable insights into brain changes associated with PD. Early and effective detection is crucial for adequate management and treatment planning. This paper proposes a Deep Spatio-Temporal Feature Extraction-based Dual-Attention EEGNet (CSFE-DAE) framework for PD detection using EEG signals. Initially, Common Spatial Pattern (CSP) is employed to extract discriminative spatial features by maximizing the variance differences between PD and healthy control (HC) signals. These features are then fed into a Dual-Attention (DA) EEGNet, which integrates both temporal attention to emphasize informative time segments and spatial attention to focus on the most relevant EEG channels. Data augmentation techniques, such as the addition of Gaussian noise, were integrated to enhance robustness against sensor variability and improve generalization. The empirical analysis demonstrates that the proposed CSFE-DAE achieves a mean accuracy of 97.59% (± 0.47) using 5-fold cross-validation on the publicly available UC San Diego resting-state EEG dataset. Additionally, it is implemented on the independent OpenNeuro dataset, showing improvement in external validation and generalization. These results show that CSFE-DAE is a competent and lightweight diagnostic tool that outperforms state-of-the-art EEG-based PD diagnostic techniques.

Related topics