Informed Bootstrap Augmentation Improves EEG Decoding

TL;DR

EEG decoding is hindered by noise and trial-to-trial variability, especially in data-limited or complex paradigms. The authors propose a reliability-based weighted bootstrapping augmentation that samples and averages trials according to ERP-derived informativeness, improving the quality of augmented samples for within-subject decoding. Across a Sentence Evaluation paradigm, weighted augmentation yields higher decoding accuracy (up to 71.25%) than unweighted approaches, demonstrating more robust and discriminative EEG representations. This approach offers a simple, data-efficient path to extend EEG decoding performance in challenging experimental settings and can be extended to subject-level analyses and adaptive weighting in future work.

Abstract

Electroencephalography (EEG) offers detailed access to neural dynamics but remains constrained by noise and trial-by-trial variability, limiting decoding performance in data-restricted or complex paradigms. Data augmentation is often employed to enhance feature representations, yet conventional uniform averaging overlooks differences in trial informativeness and can degrade representational quality. We introduce a weighted bootstrapping approach that prioritizes more reliable trials to generate higher-quality augmented samples. In a Sentence Evaluation paradigm, weights were computed from relative ERP differences and applied during probabilistic sampling and averaging. Across conditions, weighted bootstrapping improved decoding accuracy relative to unweighted (from 68.35% to 71.25% at best), demonstrating that emphasizing reliable trials strengthens representational quality. The results demonstrate that reliability-based augmentation yields more robust and discriminative EEG representations. The code is publicly available at https://github.com/lyricists/NeuroBootstrap.

Figures (3)

• Figure 1: EEG within-subject sentence-type decoding pipeline. The first three PCs were extracted as spatial features ($\bm{\hat{\bm{E}}}$) through PCA across channels from the preprocessed EEG ($E$). Augmented trials ($\bm{\hat{\bm{E}}_{aug}}$) were then generated using the bootstrapping method.
• Figure 2: Feature quality evaluation. A. Mean ERP difference (Z-Score) across subjects for each condition. B. Mean SNR (dB) across subjects for each TOI. Stars indicate significant differences between conditions (paired t-test, $N=137$, *** $p<0.001$, FDR-corrected).
• Figure 3: SVM based time-resolved within-subject EEG sentence-type decoding performance across conditions. Decoding accuracy is shown for $Bio$, $Int$, $BI_u$ with uniform bootstrapping, and for $BI_w$ with weighted bootstrapping. Shaded areas denote the standard error of the mean across subjects. Horizontal colored lines indicate time intervals with significant differences between paired conditions (using cluster-based permutation test, $N=137$, $p<0.05$).