VAEEG: Variational auto-encoder for extracting EEG representation

• DOI: 10.1016/j.neuroimage.2024.120946
• Publication Date: 2024-11-19T08:33:41Z
• Authors: Tong Zhao; Yi Cui; Taoyun Ji; Jiejian Luo; Wenling Li; Jun Jiang; Zaifen Gao; Wenguang Hu; Yuxiang Yan; Yuwu Jiang; Bo Hong
• Keywords: Representation; Autoencoder

The electroencephalogram (EEG) exhibits characteristics of complexity and strong randomness. Existing deep learning models for EEG typically target specific objectives datasets, with their scalability constrained by the size dataset, resulting in limited perceptual generalization abilities. In order to obtain more intuitive, concise, useful representations brain activity, we constructed a reconstruction-based self-supervised model based on Variational Autoencoder (VAE) separate frequency bands, termed variational auto-encoder (VAEEG). VAEEG achieved outstanding reconstruction performance. Furthermore, validated efficacy latent three clinical tasks concerning pediatric development, epileptic seizure, sleep stage classification. We discovered that certain features: 1) correlate adolescent developmental changes; 2) exhibit significant distinctions distribution between seizures background activity; 3) show variations across different cycles. corresponding downstream fitting or classification tasks, extracted demonstrated superior Our can extract effective features from complex signals, serving as an early feature extractor tasks. This reduces amount data required simplifies models, streamlines training process.