Enhancing Single Event-Related Potentials Through Preprocessing and Denoising

Extracting evoked potentials (EPs) from single trials in electroencephalography (EEG) remains a major challenge due to a characteristically low signal-to-noise ratio (SNR). This paper presents an enhanced denoising framework that combines multiresolution wavelet transform (MWT) with a statistical resampling technique. A key contribution is the introduction of an SNR-based preprocessing step that assesses individual trials and discards those with an SNR below 0 dB to prevent heavily corrupted data from degrading the analysis. Unlike traditional methods that rely on Gaussian noise assumptions, our approach utilizes empirical resampling to estimate optimal wavelet coefficient thresholds in a fully data-driven manner. Hard thresholding is subsequently applied to isolate transient neural events from background fluctuations. The method was validated using synthetic signals and real EEG recordings from ten subjects (aged 20–31 years) performing an Eriksen flanker task. Results from simulations demonstrated a significant mean SNR improvement of 13 dB. In real data applications, the error-monitoring components (Ne and Pe) were clearly identified at the single-trial level, with peak latencies observed at approximately 180 ms and 220 ms, respectively. This approach enables reliable single-trial EP analysis without the need for templates or multichannel recordings, offering a robust tool for brain–computer interfaces and clinical diagnostics.