Abstract
The electroencephalograph (EEG) source imaging (ESI) method is a non-invasive technique providing high temporal resolution imaging of brain electrical activity. However, its accuracy is often compromised by noise and source-irrelevant signals. This study introduces WPESI, a novel ESI method based on Wavelet Packet Transform (WPT) and subspace component selection. By decomposing EEG signals into subspaces and retaining only those highly correlated with the original signal, the method reconstructs a cleaner input for inverse calculations. In computer simulations, visual evoked potential (VEP) experiments, and clinical epilepsy cases, WPESI demonstrated superior localization accuracy compared to the standard sLORETA approach (OESI). The results highlight WPESI's potential for clinical applications, such as localizing epileptogenic foci from scalp EEG.
Introduction
EEG Source Imaging (ESI) is a vital tool for estimating the internal origins of electrical activity recorded at the scalp. It is used extensively in neuroscience and clinical medicine to map cognitive functions and identify the sources of epileptic seizures. Despite its utility, ESI faces a significant challenge: the "noise" problem. Scalp EEG signals are frequently contaminated by muscle artifacts, environmental interference, and redundant frequency components that obscure the underlying neural activity.
Standard inverse algorithms, such as sLORETA (referred to here as OESI), often struggle to distinguish meaningful signals from this noise, resulting in blurred or inaccurate cortical maps. This inaccuracy is exacerbated when using low-density electrode montages or in environments with a poor signal-to-noise ratio (SNR). This paper proposes a solution through a refined preprocessing pipeline that surgically removes irrelevant data before the imaging process begins.
The WPESI Methodology
The proposed WPESI (Wavelet Packet EEG Source Imaging) pipeline enhances accuracy through a four-step process designed to isolate relevant neural components:
- Wavelet Packet Transform (WPT): Unlike standard transforms that focus on low-frequency components, WPT decomposes both low and high frequencies into a comprehensive tree-like structure of "subspaces."
- Subspace Component Selection: The algorithm calculates the Pearson correlation coefficient ($r$) between each subspace and the original signal. Subspaces with a correlation below a threshold (set at 0.5) are discarded as noise.
- Signal Reconstruction: A "cleaned" EEG signal is reconstructed using only the retained, high-correlation subspaces.
- Inverse Solution: The reconstructed signal is applied to a realistic 3-layer Boundary Element Model (BEM) created from MRI data. The sLORETA algorithm then calculates the final current density distribution on the cortex.
Experimental Validation
The researchers validated WPESI across three distinct scenarios:
- Computer Simulations: 50 simulated dipole sources at varying depths were tested against Gaussian and pink noise at different SNRs (5 to 20 dB) and electrode densities (19 to 128 channels).
- Visual Evoked Potential (VEP) Data: Healthy subjects were exposed to a checkerboard stimulus to provide a "ground truth" for activity in the occipital cortex.
- Clinical Epilepsy Data: Data from three patients with refractory epilepsy were analyzed, with the seizure onset zone (SOZ) identified by clinical experts serving as the reference.
Key Results and Performance
WPESI consistently outperformed the standard OESI method across all tests. In simulations, WPESI reduced localization error by approximately 4 mm compared to OESI and by 24 mm compared to L1 norm methods.
In VEP experiments, WPESI produced more concentrated and focused activation clusters in the occipital lobe. Notably, while OESI's accuracy degraded as the number of averaged EEG trials decreased (lowering the SNR), WPESI remained stable and precise. Furthermore, WPESI showed remarkable robustness with only 19 electrodes, suggesting it can extract high-quality data from standard clinical setups.
In clinical applications, the activity sources estimated by WPESI aligned closely with surgically confirmed epileptogenic foci in all three patients, demonstrating its practical utility as a pre-surgical planning tool.
Conclusion
WPESI offers a significant advancement in EEG source imaging by effectively filtering out irrelevant signal components through WPT-based subspace selection. While future work may explore more complex multi-source scenarios and adaptive thresholding, the current study proves that WPESI is an easy-to-implement and highly accurate algorithm for non-invasive brain mapping and clinical diagnostics.
Summary updated