Generative AI Enables EEG Super-Resolution via Spatio-Temporal Adaptive Diffusion Learning

TL;DR

This work addresses the need for high spatial resolution EEG without the cost and discomfort of high-density setups by reconstructing HR EEG from LR data. It introduces STAD, a diffusion-based framework guided by a spatio-temporal conditioning module and a multi-scale Transformer denoising module, aided by a pre-trained EEG autoencoder to learn latent representations. STAD delivers superior SR EEG reconstruction over existing methods, and improves downstream tasks such as epilepsy-related classification and brain source localization, across multiple channel-scalings. The approach paves the way for cost-effective, high-resolution EEG in clinical and neurotechnological applications, with potential extensions to personalized, multimodal brain imaging and real-time BCI systems.

Abstract

Electroencephalogram (EEG) technology, particularly high-density EEG (HD EEG) devices, is widely used in fields such as neuroscience. HD EEG devices improve the spatial resolution of EEG by placing more electrodes on the scalp, which meet the requirements of clinical diagnostic applications such as epilepsy focus localization. However, this technique faces challenges, such as high acquisition costs and limited usage scenarios. In this paper, spatio-temporal adaptive diffusion models (STAD) are proposed to pioneer the use of diffusion models for achieving spatial SR reconstruction from low-resolution (LR, 64 channels or fewer) EEG to high-resolution (HR, 256 channels) EEG. Specifically, a spatio-temporal condition module is designed to extract the spatio-temporal features of LR EEG, which are then used as conditional inputs to direct the reverse denoising process. Additionally, a multi-scale Transformer denoising module is constructed to leverage multi-scale convolution blocks and cross-attention-based diffusion Transformer blocks for conditional guidance to generate subject-adaptive SR EEG. Experimental results demonstrate that the STAD significantly enhances the spatial resolution of LR EEG and quantitatively outperforms existing methods. Furthermore, STAD demonstrate their value by applying synthetic SR EEG to classification and source localization tasks, indicating their potential to substantially boost the spatial resolution of EEG.

Figures (10)

• Figure 1: Comparative illustration of low-density and high-density EEG Devices: advantages and disadvantages. The picture on the right shows the buddhist monk Barry Kerzin meditating with a high-density EEG device, picture from hanrath2019finite.
• Figure 2: The architecture of STAD aimed at generating SR EEG from LR EEG.
• Figure 3: Training Phase of STAD
• Figure 4: Quantitative comparison results of STAD and existing EEG SR methods using four metrics. (a): The PCC performance across various methods. (b): The MAE performance across various methods. (c): The NMSE performance across various methods. (d): The SNR performance across various methods. The synthetic results of STAD demonstrate the highest quality across these four quantitative evaluation metrics.
• Figure 5: Quantitative comparison between different scaling factors, including 2, 4, 8, 16. (a): The PCC performance across different scaling factors. (b): The MAE performance across different scaling factors. (c): The NMSE performance across different scaling factors. (d): The SNR performance across different scaling factors.