Selective Channel-Quality–Guided EEG Denoising for Brain Disorder Prediction

Electroencephalography (EEG) denoising is a critical preprocessing step for mitigating artifacts and improving signal quality. Although many algorithms achieve substantial noise reduction, their impact on downstream tasks such as classification remains underexplored. Furthermore, most existing datasets and benchmarks rely on single-channel simulations, which limits their applicability to real-world high-density EEG systems with dozens to hundreds of channels. Extending single-channel denoising methods to multi-channel recordings often incurs prohibitive computational costs. To address these challenges, we propose a selective channel-denoising framework that dynamically identifies high-noise channels and applies denoising only to those channels. Unlike indiscriminate full-channel denoising, our approach enhances computational efficiency while preserving task-relevant neural information. Experiments on multiple public EEG datasets demonstrate the proposed method achieves superior brain disease prediction accuracy to full-channel denoising, while substantially reducing runtime and memory consumption. These results highlight selective, channel-quality-guided denoising as a scalable and effective strategy for EEG preprocessing in clinical settings.