Detection of Event Related Patterns using Hilbert Transform in Brain Computer Interface

The feature extraction and classification of motor imagery (MI) tasks comprehend incentive issues in the scope of electroencephalogram (EEG) based brain computer interface (BCI). In this study, proposed method i.e. Hilbert transform (HT) is used for the detection of event-related patterns (ERPs) and the machine learning classifiers were implemented for the classification of both left and right hand MI tasks. The proposed method followed two steps: First, sensorimotor frequency band (8–30 Hz) related to MI activities were extracted from the signals. The important features i.e. band power (BP) and ERPs were extracted from the HT. The significant extracted features were fed into the different machine learning classifiers such as Naive Bayes (NB), linear discriminant analysis (LDA) and support vector machine (SVM). The classification accuracy (% CA), Cohen’s kappa coefficient (K) and area under the receiver operating characteristic (Auc) for all classifiers were evaluated. The proposed method was tested on publicly available BCI-competition 2008 Graz data set (II-b). Results show that the features extracted from the HT meets higher performance (% CA=82.22%, K=0.63 and Auc =0.81) in comparison with the conventional methods. Which is demonstrates that the proposed approach has capability to enhance BCI for real-time applications.