Classification of multi-task ECoG signals using fisher's linear discriminant analysis

Objective. The electrocorticogram (ECoG) signals with rich motion information have been applied to brain-computer interface (BCI) study for many years. However, there are big challenges for BCIs to extract discriminative features and classify different motions from the brain signals. The classification performance of a BCI depends on the methods of features extraction and classifier. This study aims to obtain a good classification result of multi-task ECoG signals. Approach. In this paper, we extracted features of broadband gamma power using band pass filter and the Hilbert transform. Then we classified the multi-task ECoG signals (i.e. rest and movement, hand and facial motion) with two layer classifier using Fisher Linear Discriminant Analysis (FLDA) after the optimal channel subsets of task-related cortical locations selected from multi-channel ECoG signals. Results. Our results demonstrated the high classification performance of the methods with the max classification of 97% and the average accuracy rate of 85%. Significance. In summary, our results revealed that the considerably potential method of multi-task ECoG classification was applied to improve the performance of BCIs.