Skip to content

DM-RE2I: A framework based on diffusion model for the reconstruction from EEG to image

🔥INFO

Blog: 2025/07/19 by IgniSavium

  • Title: DM-RE2I: A framework based on diffusion model for the reconstruction from EEG to image
  • Authors: Hong Zeng, Wanzeng Kong et.al (Hangzhou Dianzi University, University of Yamanashi)
  • Published: October 2023
  • Comment: Biomedical Signal Processing and Control
  • URL: https://www.sciencedirect.com/science/article/abs/pii/S174680942300558X

🥜TLDR: TSConv + guided DDPM

Motivation

The paper aims to address the challenge of reconstructing high-quality, semantically accurate images from EEG signals—which suffer from low signal-to-noise ratio and individual variability—by proposing a novel diffusion-model-based framework (DM-RE21) that combines a robust EEG semantic feature extractor (EVRNet - one residual net) and a denoising diffusion module (EG-DDPM), overcoming the limitations of previous LSTM, CNN, VAE, and GAN-based approaches in terms of semantic fidelity, resolution, and generalizability.

Model

image-20250720161441608

EEG Encoder

image-20250720161844812

image-20250816230923101

Image Decoder

image-20250816230800336

image-20250816230858375