Entropy-based EEG time interval selection for improving motor imagery classification

Abstract

Classification of different motor imagery tasks using electroencephalogram (EEG) signals is challenging, since EEG presents individualized temporal and spatial characteristics that are contaminated by noise, artifacts and irrelevant mental activities. In most applications, the EEG time interval on which feature extraction algorithms operate is fixed for all subjects, whereas the start time and the duration of motor imagery-based brain activities can vary from subject to subject. To improve the classification accuracy, this paper proposes a novel entropybased algorithm to accurately identify the time interval that motor imagery has been performed. The proposed algorithm searches through different time intervals across trials and finds the one with minimum irregularity. The hypothesis behind the proposed algorithm is that when motor imagery is performed, the activities of the neurons in the motor cortex tend to become more synchronized and less irregular. We evaluate our proposed algorithm using a publicly available motor imagery-based BCI dataset. The experimental results show that the proposed algorithm selects the EEG intervals leading to superior BCI performance compared to fixed EEG intervals that are commonly used for all subjects.

Regularized RKHS-Based Subspace Learning for Motor Imagery Classification

“…This concentration constraint is a very tiring mental task for the subjects [ 2 ]. The BCIs system uses a fixed time interval for all subjects, which is considered as one of the drawbacks of this model [ 11 ]. The MI experiment depends on the subject and there is no way to define exactly when the effect of motor imagery appears after the cue [ 12 ], i.e., it can appear immediately after the cue or after a period of time.…”