Use of Empirical Mode Decomposition for Classification of MRCP Based Task Parameters

Accurate detection and classification of force and speed intention in Movement Related Cortical Potentials (MRCPs) over a single trial offer a great potential for brain computer interface (BCI) based rehabilitation protocols. The MRCP is a non-stationary and dynamic signal comprising a mixture of frequencies with high noise susceptibility. The aim of this study was to develop efficient preprocessing methods for denoising and classification of MRCPs for variable speed and force. A proprietary dataset was cleaned using a novel application of Empirical Mode Decomposition (EMD). A combination of temporal, frequency and time-frequency techniques was applied on data for feature extraction and classification. Feature set was analyzed for dimensionality reduction using Principal Component Analysis (PCA). Classification was performed using simple logistic regression. A best overall classification accuracy of 77.2% was achieved using this approach. Results provide evidence that BCI can be potentially used in tandem with bionics for neuro-rehabilitation.