Title: TGSD: Topology-Guided State-Space Diffusion for EEG Spatial Super-Resolution

Abstract:

While low-density Electroencephalography (EEG) is well-suited for wearable and Internet of Things (IoT) brain-sensing applications, the sparse sampling of electrodes often fails to provide adequate spatial data for characterizing neural activity across different brain regions. EEG spatial super-resolution seeks to reconstruct dense-channel EEG data from these sparse recordings; however, this task is fraught with difficulties. These challenges stem from the fact that missing channels usually occur at the whole-channel level, the full electrode layout’s spatiotemporal dependencies are frequently overlooked, and the transformation from sparse to dense signals is intrinsically ambiguous.

To overcome these obstacles, we introduce TGSD, a framework for EEG spatial super-resolution that utilizes topology-guided state-space diffusion. TGSD begins by using a Hierarchical Spatial Prior Encoder to acquire topology-aware priors across the entire electrode layout. This is achieved by merging local geometric relationships with contextual information at the region level. Leveraging these priors alongside sparse observations, a Conditional State-Space Diffusion Reconstructor iteratively synthesizes the missing-channel signals via reverse diffusion. This process alternates between temporal and channel-wise state-space modeling, thereby capturing long-range temporal dynamics and inter-channel dependencies within a single, unified architecture.

Evaluations conducted on the SEED and PhysioNet MM/I datasets indicate that TGSD consistently surpasses leading baseline methods. This superiority is evident in both reconstruction accuracy and downstream classification results across various super-resolution factors. These findings highlight the efficacy of integrating conditional diffusion with topology-aware spatial priors to improve low-density EEG sensing in wearable and IoT contexts. The official implementation code can be accessed at https://github.com/jtggz/TGSD.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC