Simulating Brain Signals: Creating Synthetic EEG Data via Neural-Based Generative Models for Improved SSVEP Classification

Abstract

Despite significant recent progress in the area of Brain-Computer Interface (BCI), there are numerous shortcomings associated with collecting Electroencephalography (EEG) signals in real-world environments. These include, but are not limited to, subject and session data variance, long and arduous calibration processes and predictive generalisation issues across different subjects or sessions. This implies that many downstream applications, including Steady State Visual Evoked Potential (SSVEP) based classification systems, can suffer from a shortage of reliable data. Generating meaningful and realistic synthetic data can therefore be of significant value in circumventing this problem. We explore the use of modern neural-based generative models trained on a limited quantity of EEG data collected from different subjects to generate supplementary synthetic EEG signal vectors, subsequently utilised to train an SSVEP classifier. Extensive experimental analysis demonstrates the efficacy of our generated data, leading to improvements across a variety of evaluations, with the crucial task of cross-subject generalisation improving by over 35% with the use of such synthetic data.

Critical and Ictal Phases in Simulated EEG Signals on a Small-World Network

“…By generating this synthetic data, we aim to train machine learning (ML) algorithms for predicting seizures in patients with epilepsy. Due to the difficulty in acquiring high-quality patient data, and the relative rarity of seizures events compared with interictal intervals, the ability to generate synthetic data may alleviate an important bottleneck in seizure prediction methods (Aznan et al, 2019 ). Rapid advances in computational power permit more realistic full HH simulations that were considered not practical only a few years ago.…”

EpilepsyGAN: Synthetic Epileptic Brain Activities With Privacy Preservation

“…In the specific case of brain signals, the application of GANs to the generation of realistic synthetic signals has obtained very limited success so far: [29] generated EEG-like signals, without demonstrating the quality of the synthetic data in any specific task or pathology detection. [30] generated synthetic EEG data to augment existing real training sets for Brain-Computer Interfaces, but they did not evaluate on fully synthetic training sets. [31] used a GAN to upsample the spatial resolution of EEG signals and, despite the improvement in visual quality, the resulting training set produced a degradation of 4-9% of accuracy in a mental imagery classification task in comparison to the original training set.…”

“…The nonlinear features extracted are: sixth and seventh level sample entropy [54] for k = 0.2 and k = 0.35; third, fourth, fifth, sixth and seventh level permutation entropy [55] for n = 3, n = 5 and n = 7; third, fourth, fifth, sixth and seventh level, as well as raw signal, Shannon, Renyi and Tsallis entropies. The power features are: total power, total and relative band power in the bands delta [0.5,4] Hz, theta [4,8] Hz, alpha [8,12] Hz, beta [13,30] Hz, gamma [30,45] Hz as well as in the bands [0,0.1] Hz, [0.1,0.5] Hz, [12,13] Hz. After the features are extracted, the target training set is used to train the random forest classifier and the resulting classifier is evaluated against the test set.…”

PlethAugment: GAN-Based PPG Augmentation for Medical Diagnosis in Low-Resource Settings

“…In the medical domain, the work in [21] uses a 1D convolution-based GAN to generate electroencephalogram (EEG) brain signals. Inspired by this work, others generate synthetic epileptic brain activity signals [22] and EEG signals [23] specifically to improve classification models. Others [24] use a GAN to generate an open-source a privacy-protected vital sign dataset.…”