Conditional Diffusion-Based EEG Channel Reconstruction for Motor Imagery Decoding

This paper proposes a diffusion-based framework using Denoising Diffusion Probabilistic Models (DDPM) for EEG signal reconstruction in Motor Imagery (MI)-based Brain–Computer Interface (BCI) applications. The framework is integrated with CSP+LDA, FBCSP+LDA, and EEGNet decoding baselines to improve robustness while preserving decoding performance. The proposed method reconstructs predefined EEG channels from spatially adjacent signals to maintain inter-channel dependencies, and was evaluated using training sessions and an independent held-out test session. The work highlights areas for further improvement in EEG-based Motor Imagery decoding models in Latin America.