A CNN based graphical user interface controlled by imagined movements

An electroencephalogram (EEG) based brain-computer interface (BCI) enables the control of some external activity directly from the brain, without any physical movement/overt action. The external activity can be the cursor control of a computer or it can provide commands to the devices to perform certain functions. This work proposes a movement imagery (MI) based graphical user interface (GUI) for typing 26 English alphabets and tasks like food, water, medicine along with cancel and confirm commands. Convolutional Neural Network (CNN) is used to extract the spatial features from the recorded EEG signals. These features are fed to an ensemble-based extreme gradient (XG) boost classifier in a five-classification framework. By varying the hyper-parameters of the classification model, the highest accuracy of 84.7% for CNN and 92.87% for the cascaded structure of CNN and the XG boost classifier is achieved. The minimum execution time taken is 1.18 s for CNN and 3.24 s using both CNN and XG boost classifier. The work shows that it is possible to classify the information embedded in MI signals and can serve as a basis for an alternate communication channel to patients in advanced stages of Amyotrophic lateral sclerosis.