Multi-Channel Vision Transformer for Epileptic Seizure Prediction

Abstract

Epilepsy is a neurological disorder that causes recurrent seizures and sometimes loss of awareness. Around 30% of epileptic patients continue to have seizures despite taking anti-seizure medication. The ability to predict the future occurrence of seizures would enable the patients to take precautions against probable injuries and administer timely treatment to abort or control impending seizures. In this study, we introduce a Transformer-based approach called Multi-channel Vision Transformer (MViT) for automated and simultaneous learning of the spatio-temporal-spectral features in multi-channel EEG data. Continuous wavelet transform, a simple yet efficient pre-processing approach, is first used for turning the time-series EEG signals into image-like time-frequency representations named Scalograms. Each scalogram is split into a sequence of fixed-size non-overlapping patches, which are then fed as inputs to the MViT for EEG classification. Extensive experiments on three benchmark EEG datasets demonstrate the superiority of the proposed MViT algorithm over the state-of-the-art seizure prediction methods, achieving an average prediction sensitivity of 99.80% for surface EEG and 90.28–91.15% for invasive EEG data.

Bi-Branch Vision Transformer Network for EEG Emotion Recognition

“…Transformer was first proposed in natural language processing [61], and since then it has been applied in other domains [62]. Vision Transformer (ViT) based on multi-head self-attention to patches of images has achieved outstanding results in the field of computer vision.…”

“…Transformer is a novel neural network architecture that was primarily created for natural language processing applications, in which multi-layer perceptron layers are utilized on top of multi-head attention mechanisms to capture the longrange dependencies in sequential input. Vision Transformer has recently demonstrated considerable promise in a variety of computer vision applications, such as picture classification and segmentation [62]. In perspective of this, we propose a new kind of ViT, the Two-branch Vision Transformer Network, which uses a different type of EEG feature representation.…”

Deep Optimized Electrodes and Frequency Bands in the Phase Space for Identification of Seizures

“…Seizure detection using EEG signals has witnessed remarkable advancements with the application of deep learning techniques [8,9,10,11]. In recent years, numerous studies have explored the utilization of sophisticated neural network architectures such as LSTM networks [12,13,6] networks and Transformers [5,14] for enhancing the accuracy and reliability of seizure detection and prediction. This section presents an overview of recent publications that exemplify the contributions made in this direction.…”

“…Recently, a lightweight Transformer architecture has been proposed to enhance accuracy in detecting seizure patterns by capturing local and global dependencies within EEG signals [5]. Similarly, [14] introduced a Multichannel Vision Transformer (MViT) for automated spatiotemporal spectral features learning in multi-channel EEG data to achieve state-of-the-art performance in seizure prediction. Researchers have also explored hybrid architectures that combine the strengths of LSTM and self-attention for classification and prediction from time-series data [13].…”

Efficient graph convolutional networks for seizure prediction using scalp EEG

“…In terms of algorithms, deep learning has become the most popular research method for seizure prediction. However, traditional deep learning models are often used, such as Convolutional Neural Networks (CNN) (Rosas-Romero et al, 2019 ; Sharan and Berkovsky, 2020 ; Wang et al, 2020 ; Li et al, 2021d ; Ozdemir et al, 2021 ; Usman et al, 2021 ), Recurrent Neural Networks (RNN) (Tsiouris et al, 2018 ; Li et al, 2021d ; Usman et al, 2021 ), and new deep learning models, such as multi-view CNN (Liu et al, 2019 ), multi-time scale CNN (Qi et al, 2021 ), semi-expanded CNN (Hussein et al, 2021 ), and Transformer (Hussein et al, 2022 ). Can only process Euclidean grid data, often EEG data or the feature is represented as a real number matrix.…”