Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Synthetic EEG Generation using Diffusion Models for Motor Imagery Tasks

Henrique de Lima Alexandre, Clodoaldo Aparecido de Moraes Lima

TL;DR

Brain-Computer Interfaces using EEG are hampered by data scarcity and variability, limiting robust classification. The authors introduce a diffusion probabilistic model (DDPM) to generate synthetic EEG channels for motor imagery tasks and compare it with a conditional Wasserstein GAN with Gradient Penalty (cWGAN-GP), using an end-to-end pipeline with $T=1000$ timesteps and a linear schedule from $\beta_{start}=0.0001$ to $\beta_{end}=0.02$. They demonstrate that DDPM-generated channels achieve lower mean squared error and higher Pearson correlation with real signals and preserve discriminative information for classifiers such as a CNN and U-Net, achieving accuracies above 95% on synthetic data. This work shows that synthetic EEG data can augment limited datasets for BCI applications, improving model generalization, and outlines future directions including multi-subject datasets, transformer-based temporal modeling, and real-time deployment.

Abstract

Electroencephalography (EEG) is a widely used, non-invasive method for capturing brain activity, and is particularly relevant for applications in Brain-Computer Interfaces (BCI). However, collecting high-quality EEG data remains a major challenge due to sensor costs, acquisition time, and inter-subject variability. To address these limitations, this study proposes a methodology for generating synthetic EEG signals associated with motor imagery brain tasks using Diffusion Probabilistic Models (DDPM). The approach involves preprocessing real EEG data, training a diffusion model to reconstruct EEG channels from noise, and evaluating the quality of the generated signals through both signal-level and task-level metrics. For validation, we employed classifiers such as K-Nearest Neighbors (KNN), Convolutional Neural Networks (CNN), and U-Net to compare the performance of synthetic data against real data in classification tasks. The generated data achieved classification accuracies above 95%, with low mean squared error and high correlation with real signals. Our results demonstrate that synthetic EEG signals produced by diffusion models can effectively complement datasets, improving classification performance in EEG-based BCIs and addressing data scarcity.

Synthetic EEG Generation using Diffusion Models for Motor Imagery Tasks

TL;DR

Brain-Computer Interfaces using EEG are hampered by data scarcity and variability, limiting robust classification. The authors introduce a diffusion probabilistic model (DDPM) to generate synthetic EEG channels for motor imagery tasks and compare it with a conditional Wasserstein GAN with Gradient Penalty (cWGAN-GP), using an end-to-end pipeline with timesteps and a linear schedule from to . They demonstrate that DDPM-generated channels achieve lower mean squared error and higher Pearson correlation with real signals and preserve discriminative information for classifiers such as a CNN and U-Net, achieving accuracies above 95% on synthetic data. This work shows that synthetic EEG data can augment limited datasets for BCI applications, improving model generalization, and outlines future directions including multi-subject datasets, transformer-based temporal modeling, and real-time deployment.

Abstract

Electroencephalography (EEG) is a widely used, non-invasive method for capturing brain activity, and is particularly relevant for applications in Brain-Computer Interfaces (BCI). However, collecting high-quality EEG data remains a major challenge due to sensor costs, acquisition time, and inter-subject variability. To address these limitations, this study proposes a methodology for generating synthetic EEG signals associated with motor imagery brain tasks using Diffusion Probabilistic Models (DDPM). The approach involves preprocessing real EEG data, training a diffusion model to reconstruct EEG channels from noise, and evaluating the quality of the generated signals through both signal-level and task-level metrics. For validation, we employed classifiers such as K-Nearest Neighbors (KNN), Convolutional Neural Networks (CNN), and U-Net to compare the performance of synthetic data against real data in classification tasks. The generated data achieved classification accuracies above 95%, with low mean squared error and high correlation with real signals. Our results demonstrate that synthetic EEG signals produced by diffusion models can effectively complement datasets, improving classification performance in EEG-based BCIs and addressing data scarcity.
Paper Structure (21 sections, 4 equations, 4 figures, 4 tables)

This paper contains 21 sections, 4 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Dataset
  5. Data Preprocessing
  6. Channel Selection for Synthetic Reconstruction
  7. Diffusion Model Architecture
  8. Baseline: cWGAN-GP for Synthetic EEG Generation
  9. Classifiers
  10. K-Nearest Neighbors (KNN)
  11. Logistic Regression (LR)
  12. Convolutional Neural Network (CNN)
  13. U-Net
  14. Evaluation Protocol
  15. Experimental Setup
  16. ...and 6 more sections

Figures (4)

  • Figure 1: EEG channels selected for synthetic reconstruction (highlighted in red). These channels (Fp1, Fp2, AF3, AF4, F7, F8, T7, and T8) were chosen due to their relatively lower relevance in motor imagery tasks, based on previous literature tang2022channelyu2020crossalotaiby2015review. The remaining channels represent areas typically considered highly relevant for capturing motor imagery-related neural activity.
  • Figure 2: Forward process (Diffusion) in a Simulated EEG signal
  • Figure 3: Reverse process (Denoising) in a Simulated EEG signal
  • Figure 4: Visual comparison of original and synthetic EEG signals for eight channels. The left column shows signals generated by the DDPM model, while the right column shows results from the cWGAN-GP model. Each plot presents a smoothed 6–12 second interval with real signals in solid lines and synthetic signals in dashed lines.