Synthesizing Epileptic Seizures: Gaussian Processes for EEG Generation

TL;DR

This work introduces GP-EEG, a principled, interpretable framework for generating long-horizon, multi-channel epileptic EEG by combining SVD-based dimensionality reduction, regime-aware Gaussian process modeling, and a domain-adaptive Conv-LSTM VAE. The pipeline explicitly handles non-stationarity via Changepoint-driven GP surrogates and preserves inter-channel structure through patient-specific loadings, augmented by a discrete kernel-state dynamics module and Poisson-based regime timing. Empirical results on CHB-MIT and Siena show GP-EEG produces realistic synthetic EEG and provides substantial benefits for seizure-detection training when used for data augmentation or in train-on-synthetic/test-on-real setups. The approach offers a scalable, interpretable alternative to end-to-end deep generators, with potential to improve benchmarking, robustness testing, and reproducibility in clinically sensitive EEG research.

Abstract

Reliable seizure detection from electroencephalography (EEG) time series is a high-priority clinical goal, yet the acquisition cost and scarcity of labeled EEG data limit the performance of machine learning methods. This challenge is exacerbated by the long-range, high-dimensional, and non-stationary nature of epileptic EEG recordings, which makes realistic data generation particularly difficult. In this work, we revisit Gaussian processes as a principled and interpretable foundation for modeling EEG dynamics, and propose a novel hierarchical framework, \textit{GP-EEG}, for generating synthetic epileptic EEG recordings. At its core, our approach decomposes EEG signals into temporal segments modeled via Gaussian process regression, and integrates a domain-adaptation variational autoencoder. We validate the proposed method on two real-world, open-source epileptic EEG datasets. The synthetic EEG recordings generated by our model match real-world epileptic EEG both quantitatively and qualitatively, and can be used to augment training sets.

Figures (16)

• Figure 1: Overview of GP-EEG, the proposed multi-stage framework for patient-specific synthetic EEG generation, described in Section \ref{['sec:full_pipeline']}. When time series are colored, dark blue corresponds to real seizures, dark orange to the output of Stage 2 and 6 prior to 'EEGification', and yellow to that of Stage 6 following 'EEGification'.
• Figure 2: Qualitative comparison of real and synthetic seizure data on the Siena (top) and CHB-MIT (bottom) datasets. For each dataset, the upper row shows t-SNE embeddings (perplexity=30) and the lower row shows KDE of amplitude distributions.
• Figure 3: Sample seizures on a 4 second interval. The first row is a real seizure segment from patient 23 of the CHB-MIT dataset, and rows 2, 3 and 4 are synthetic output produced by GP-EEG, COSCI-GAN, TimeVAE, and ImagenTime. Each column is a different realization.
• Figure 4: Sample seizures on a 4 second interval. The first row is a real seizure segment from patient 16 of the CHB-MIT dataset, and rows 2, 3 and 4 are synthetic output produced by GP-EEG, COSCI-GAN, TimeVAE, and ImagenTime. Each column is a different realization.
• Figure 5: Sample synthetic seizure segment, obtained using GP-EEG, on a 53 seconds interval.