Step-Aware Residual-Guided Diffusion for EEG Spatial Super-Resolution

TL;DR

SRGDiff tackles the challenge of achieving high-fidelity EEG spatial super-resolution with lightweight low-density systems by reframing SR as dynamic conditional diffusion guided by low-density inputs. It introduces a step-aware Residual Direction Module and a Step-Aware Modulation Module that inject per-step, LD-derived cues into the reverse diffusion process, ensuring consistency with the LD observation while attaining high-density detail and preserving topology. Trained in two stages on high-density data and evaluated across SEED, SEED-IV, and Localize-MI with signal-, feature-, and downstream-level metrics, SRGDiff shows consistent improvements over strong baselines, including substantial NMSE reductions, PCC/SNR gains, and the lowest EEG-FID scores. The method delivers practical benefits for real-time EEG analysis and visualization, improving downstream emotion recognition and epileptic detection without requiring full-channel recordings.

Abstract

For real-world BCI applications, lightweight Electroencephalography (EEG) systems offer the best cost-deployment balance. However, such spatial sparsity of EEG limits spatial fidelity, hurting learning and introducing bias. EEG spatial super-resolution methods aim to recover high-density EEG signals from sparse measurements, yet is often hindered by distribution shift and signal distortion and thus reducing fidelity and usability for EEG analysis and visualization. To overcome these challenges, we introduce SRGDiff, a step-aware residual-guided diffusion model that formulates EEG spatial super-resolution as dynamic conditional generation. Our key idea is to learn a dynamic residual condition from the low-density input that predicts the step-wise temporal and spatial details to add and uses the evolving cue to steer the denoising process toward high-density reconstructions. At each denoising step, the proposed residual condition is additively fused with the previous denoiser feature maps, then a step-dependent affine modulation scales and shifts the activation to produce the current features. This iterative procedure dynamically extracts step-wise temporal rhythms and spatial-topographic cues to steer high-density recovery and maintain a fidelity-consistency balance. We adopt a comprehensive evaluation protocol spanning signal-, feature-, and downstream-level metrics across SEED, SEED-IV, and Localize-MI and multiple upsampling scales. SRGDiff achieves consistent gains of up to 40% over strong baselines, proving its superiority in the task of EEG spatial super-resolution. Moreover, topographic visualizations comparison and substantial EEG-FID gains jointly indicate that our SR EEG mitigates the spatial-spectral shift between low- and high-density recordings.

Figures (13)

• Figure 1: (a) Existing static guidance strategy vs. (b) our residual guidance strategy for EEG super-resolution, and (c) corresponding topographical maps of LD input, ESTformer output, GT HD EEG, and SRGDiff reconstruction.
• Figure 2: SRGDiff overview. (a) Overall architecture: Low-density EEG $X^{L}$ conditions the latent reverse process. RDM predicts a residual direction from $X^{L}$ and current decoder features. SMM provides step-aware affine parameters to fuse the residual and modulate activations. (b) Residual learning: At each step, the predicted residual guides denoising, and the residual derived from the forward noising process provides supervision via a residual loss.
• Figure 3: EEG-FID evaluation across three datasets compared with ESTformer and STAD.
• Figure 4: Visualization of EEG topographic maps between ground-truth and reconstructed EEG signals by ESTformer, STAD and SRGDiff.
• Figure 5: Ablation study performance comparison between SRGDiff and three variant models on the SEED dataset.