Portal de Periódicos da CAPES

High-Powered Ocular Artifact Detection with C-LSTM-E

2022; Springer Science+Business Media; Linguagem: Inglês

10.1007/978-3-031-17618-0_34

1611-3349

Ian McDiarmid-Sterling, Luca Cerbin,

Tactile and Sensory Interactions

Electroencephalography (EEG) is a technique for examining brain waves through recording devices placed on the scalp. During EEG signal collection, unwanted ocular artifacts (OAs) are frequently introduced and must be removed before the EEG signal can be effectively used. Many deep learning approaches to identifying and correcting OAs attempt to balance prediction accuracy and power consumption, but we introduce a novel high-power ensemble of a Convolutional Neural Network (CNN) and a Long-Short Term Memory network (LSTM), C-LSTM-E. We compare the overall accuracy of C-LSTM-E to previously introduced methods for OA identification and correction, and discover that for certain prevalancies of OAs, C-LSTM-E outperforms previously introduced models. While C-LSTM-E is slightly less accurate than the state-of-the-art OA correction model, it does not require a channel selection algorithm and is robust to changes in OA prevalance. C-LSTM-E is the first CNN and LSTM ensemble method for OA identification.