Analysis of motor imagery fnirssignal for braincomputer interface

Abstract

The Brain-Computer Interface (BCI) technology is a promising approach that enables the individuals affected by neuromuscular disorders to control an external assistive device without any activation of the muscles. The BCI translates the user s intent from the neural activity acquired from the brain and transform it to commands to control an external device thus providing a direct communication pathway between the human brain and the device. This thesis is directed towards the path to fulfil an utmost need to utilize a safe, non-invasive and practical technique to enable the movement impaired individuals to interact with the external environment purely based on their thought processes alone. Functional near-infrared spectroscopy (fNIRS) technique is a noninvasive optical modality that measures the cerebral blood oxygenation changes associated with the neuronal activation. fNIRS maps regional brain activity by detecting activation-induced concentration changes in deoxygenated haemoglobin (HbR) and oxygenated haemoglobin (HbO). In this thesis, the fNIRS technique, which is safe and non-invasive, is explored as an alternative to well-established techniques such as Electroencephalography (EEG) in the field of BCI. For an effective BCI implementation, it is pivotal to address any issues arising from the signal acquisition, processing of signals, feature extraction to choosing the appropriate feature translation algorithm. Hence, the first contribution of this work is to investigate different preprocessing methods for the removal of noise in the fNIRS signal. Here, six different noise reduction approaches such as Band-pass filtering, Correlation-basedsignal improvement (CBSI), Median filtering, Savitzky-Golay filtering, Wavelet filtering and Independent Component Analysis (ICA) have been analysed and compared with each other for their effectiveness to reduce motion artefacts and physiological noise present in the fNIRS signal. The performance of the various methods is compared using metrics such as contrast to noise rat