Data Augmentation for Seizure Prediction with Generative Diffusion Model

TL;DR

This paper tackles imbalanced EEG data in seizure prediction by introducing DiffEEG, a diffusion-model–based data augmentation method guided by STFT spectrograms to generate diverse, distribution-consistent preictal samples. By coupling a forward diffusion process with a learned reverse denoising process and conditioning on time-frequency features, DiffEEG expands the data manifold beyond simple transformations. Across five classifiers and two public datasets, DiffEEG consistently improves predictive metrics, with Multi-scale CNN achieving near state-of-the-art performance (CHB-MIT: Sens 95.4%, FPR 0.051/h, AUC 0.932; Kaggle: Sens 93.6%, FPR 0.121/h, AUC 0.822). The method demonstrates strong generality for EEG seizure prediction, suggesting substantial practical impact for reliable, private, and robust clinical predictions, while also highlighting areas for future improvement such as computational efficiency and cross-subject transferability.

Abstract

Data augmentation (DA) can significantly strengthen the electroencephalogram (EEG)-based seizure prediction methods. However, existing DA approaches are just the linear transformations of original data and cannot explore the feature space to increase diversity effectively. Therefore, we propose a novel diffusion-based DA method called DiffEEG. DiffEEG can fully explore data distribution and generate samples with high diversity, offering extra information to classifiers. It involves two processes: the diffusion process and the denoised process. In the diffusion process, the model incrementally adds noise with different scales to EEG input and converts it into random noise. In this way, the representation of data can be learned. In the denoised process, the model utilizes learned knowledge to sample synthetic data from random noise input by gradually removing noise. The randomness of input noise and the precise representation enable the synthetic samples to possess diversity while ensuring the consistency of feature space. We compared DiffEEG with original, down-sampling, sliding windows and recombination methods, and integrated them into five representative classifiers. The experiments demonstrate the effectiveness and generality of our method. With the contribution of DiffEEG, the Multi-scale CNN achieves state-of-the-art performance, with an average sensitivity, FPR, AUC of 95.4%, 0.051/h, 0.932 on the CHB-MIT database and 93.6%, 0.121/h, 0.822 on the Kaggle database.

Figures (9)

• Figure 1: The distribution illustration of down-sampling, sliding windows, recombination and DiffEEG. Each cluster represents the preictal data of a seizure. The clusters of different preictal data exhibit both similar and distinct distribution. (a) Down-sampling of interictal samples may cause the loss of useful information. (b) When using the sliding windows, the distribution of generated samples is between the front and rear samples. (c) The recombination is implemented within each seizure, because samples recombined by segments from different seizures have poor authenticity. The distribution of recombined samples is at the center of the three component samples and thus limited by them. (d) Instead, DiffEEG could fully explore the feature space and expand the distribution to outward area. The generated samples connect different clusters into a whole.
• Figure 2: The diffusion and reverse process in diffusion model.
• Figure 3: DiffEEG Architecture.
• Figure 4: The structure of the modified EEGNet.
• Figure 5: (a) The total preictal and interictal duration of the CHB-MIT database; (b) The total preictal and interictal duration of the Kaggle database.