Attention Aggregation U-Net++ (Aaunet++) and Diffusion Based Variational Autoencoder (Dvae) for Epileptic Seizure Detection

This is the authors' abstract. We don't add key points for this paper.

To monitor brain electrical activity and identify unusual spikes or asymmetries that can point to a seizure, electroencephalograms (EEG), are mostly utilised in epileptic seizure detection.However, it takes a lot of time for experts to visually examine an EEG, and their diagnoses could even conflict.For EEG diagnoses, an automated computer-aided diagnosis is therefore required.The Multivariate Variational Mode Decomposition (MVMD) algorithm for EEG signal denoising is suggested in this study.Construction of multivariate oscillations is based on the assumption that the same frequency element must be present in all channels of the input EEG data.Recovery of signals that have inherent multivariate modulated oscillations from multivariate input EEG data, a variational optimisation problem is developed for signal denoising.Attention Aggregation U-Net++ (AAUNET++) model has been developed for feature extraction.The original multi-level skip connections of U-net++ are replaced in the feature extraction process by attention aggregation, which is introduced to capture global contextual information.Diffusion Based Variational Autoencoder (DVAE) is introduced for seizure detection in EEG signals.DVAE integrates the probabilistic latent representation capability of a Variational Autoencoder (VAE) with the progressive noise modeling of a Denoising Diffusion Probabilistic Model (DDPM) to achieve robust feature learning and enhanced noise suppression in EEG.Using the Children's Hospital Boston-Massachusetts Institute of Technology (CHB-MIT) scalp EEG database, the recommended method attains superior efficiency by precision (PRE), recall/sensitivity, F-measure (F-score), and accuracy (ACC).