Year: 2018
Type: dissertation
Abstract: Steady State Visual Evoked Potentials (SSVEPs) have been the most commonly utilized Brain Computer Interface (BCI) modality due to their relatively high signal-to-noise ratio, high information transfer rates, and minimum training prerequisites. Up to date Canonical Correlation Analysis (CCA) and its extensions have been widely utilized for SSVEP target frequency identification. However, reliable and robust SSVEP identification performance is still a challenge, particularly for portable BCI systems operating in an Intensive Care Unit (ICU) department filled with various source of noise. As such, I propose an innovative partition-based feature extraction method that entails partitioning the score spaces of CCA and Power Spectral Density Analysis (PSDA) in three cases, extract efficient descriptors from each partition, then concatenate the extracted measures to generate more discriminative fusion spaces. Moreover, I investigate transforming the fusion spaces to lower dimensions utilizing Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). Finally, to validate the proposed method, I compare the performance of the partition-based feature extraction and score space fusion method to a well-established SSVEP identification method based on Multivariate Linear Regression (MLR). The experimental results of this investigation report that the proposed method enhances the identification performance of the CCA-based BCI system from 63% to 78%. The identification performance is further improved to 98% after the discriminative transformation with LDA outperforming MLR, which achieved an average overall 86% identification accuracy. As such, the proposed method is a promising approach to implement and operate BCI systems in the ICU.
Author: Muhamed Farooq
Institution: University of Michigan–Dearborn
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W2801892837
