Data Science MS Thesis Presentation By Naveen Pothayath: Human Brain Activity Classification with EEG Data Surface Reconstruction

Tuesday, April 17, 2018
11:00 am to 12:00 pm
Floor/Room #: 

Thesis Advisor: Fatemeh Emdad, Ph.D. 
Co-advisor: Elke A. Rundensteiner, Ph.D. 
Reader: Xiangnan Kong, Ph.D.

EEG has been used to explore the electrical activity of the brain for many decades. During that time, different components of the EEG signal have been isolated, characterized, and associated with a variety of brain activities. However, no widely accepted model describing the spatio-temporal structure of the full-brain EEG signal exists to date. Modeling the spatio-temporal nature of the EEG signal is a daunting task. The spatial component of EEG is defined by the locations of recording electrodes (ranging between 2 to 256 in number) placed on the scalp, while its temporal component is defined by the electrical potentials the electrodes detect. The EEG signal is generated by the composite electrical activity of large neuron assemblies in the brain. These neuronal units often perform independent tasks, giving the EEG signal a highly dynamic and non-linear character. These characteristics make the raw EEG signal challenging with which to work. Thus, most research focuses on extracting and isolating targeted spatial and temporal components of interest. While component isolation strategies like independent component analysis are useful, it is limited by noise contamination and poor reproducibility. These drawbacks to feature extraction could be improved significantly if they were informed by a global spatio-temporal model of EEG data.
This thesis aims to introduce a new data-surface reconstruction (DSR) technique for EEG, which can model the integrated spatio- temporal structure of EEG data. To produce physically intuitive results, we utilize a hyper-coordinate transformation, which integrates both spatial and temporal information of the EEG signal into a unified coordinate system. We then apply a non-uniform rational B spline (NURBS) fitting technique that minimizes the point distance from the computed surface to each element of the transformed data. To validate the effectiveness of this proposed method, we conduct an evaluation using a 5-state classification problem with one baseline and four meditation states comparing classification accuracies using the raw EEG data versus the surface reconstructed data in the broadband range and the alpha, beta, delta, gamma and higher gamma frequencies. Results demonstrate that the fitted data consistently outperforms the raw data in the broadband spectrum and all frequency spectrums.