Using a Portable Device for Online Single-Trial MRCP Detection and Classification

Abstract

In the past decade, the use of movement-related cortical potentials (MRCPs) for brain computer interface-based rehabilitation protocols has increased manifolds. Such systems suffer severely from high frequency colored noise making it extremely difficult to recognize these signals with high accuracy on a single-trial basis. All previous work in this domain has mainly focused on offline systems using computing power of lab computers in which the detection of the MRCPs is done independent to the classification of the type of movement. The main focus of this work is to test the detection of the presence of the MRCP signal as well as its classification into different types of movements in a single online system (portable Raspberry Pi II) where the classification system takes over only after the presence of MRCP signal has been detected. To achieve this, the MRCP signal was first spatially (Laplacian) and later band pass filtered to improve the signal to noise ratio, then a matched filter was applied to detect the signal. This was obtained with a detection latency of −458 ± 97 ms before the movement execution. Then six temporal features were extracted from 400 ms data after the point of detection to be classified by a standard linear support vector machine. The overall accuracy of 73 % was achieved for the online detection and classification for four different types of movements which is very close to the base line accuracy of 74 % using the offline system. The whole system was tested on Matlab and verified on a Raspberry Pi II as a portable device. The results show that the online implementation of such a system is feasible and can be adapted for stroke patient rehabilitation.