Feature extraction and classification of Electrooculography signals from Locked-In patients

Four ALS patients were recorded while controlling an interface for yes/no answering. This work proposes the use of different feature extraction methods for classification of EOG signal. The challenge stems from the need to extract patterns from patients with different clinical conditions. Time and frequency domain feature extraction methods are proposed by the use of Discrete Cosine Transform, Autoencoder and Complex Wavelet Transform, with accuracy up to 75 92 94 98 % in different patients.