Classification of MRCP based Task Parameters

Abstract

Human brain works through a host of electrical signals which are generated as a result of every thought and action. The subset of these generated signals associated with limb movement are called Movement Related Cortical Potential (MRCP). Efforts have been made to accurately detect and classify these signals in order to enable rehabilitation of patients suffering from neurological disorders. MRCP signals obtained via both invasive as well as non-invasive methods comprise a mixture of frequencies are highly prone to deterioration by noise. My focus of work is to develop a paradigm for pre-processing of MRCP signals associated with variable force and speed of stroke patients of a proprietary dataset, and to subsequently classify them. I have employed empirical Mode Decomposition (EMD) along with low pass filters to create a novel method for pre-processing of MRCP signals. I have subjected a host of extracted features to Principal Component Analysis (PCA) for dimensionality reduction and classified the signals through final selection of simple logical regression as a classifier. I have obtained a best accuracy of 77.2%. My work can be applied for neuro-rehabilitation of stroke patients with interface to bionics. The methodology has also been validated against a publically available dataset for our overall process validation. The outcome of this work can be used towards rehabilitation of stroke patients.