This item is under embargo and not available online per the author's request. For access information, please visit

Date of Award

Spring 5-15-2016

Author's School

School of Engineering & Applied Science

Author's Department

Biomedical Engineering

Degree Name

Doctor of Philosophy (PhD)

Degree Type



Electrocorticography (ECoG) has gained a lot of momentum and has become a serious contender as a recording modality for the implementation of Brain-Computer Interface (BCI) systems in the last few years. ECoG signals provide the right balance between minimal invasiveness and robust spectral information to accomplish a BCI task. However, all the BCI studies until now have used signals recorded from a large number of implanted electrodes and a larger number of spectral features. The recording and processing of these signals uses a lot of electrical power and thus hinders its use outside the research setting. To translate this research to the clinic as a chronic recording modality for neural prosthesis, minimizing the number of features and thus, the power consumption to record and process them, is of prime importance. This thesis develops and investigates two different techniques to minimize the feature space required to obtain a robust BCI control in a virtual environment setting. ECoG electrodes embedded in thin-film polyimide or Silastic were implanted in the epidural space over pre-motor, primary motor and parietal cortical areas in non-human primates. Subjects tested this thesis had had their electrode arrays implanted at least 1-2 years before the beginning of these experiments. Monkeys were trained to perform a classic 2D center out task using the recorded signals and one of two new BCI decoding algorithms developed in this thesis. Both the algorithms used for BCI control updated the decoding model using data from the previous trials. The parameters of the decoding algorithms were varied every 1-2 weeks to gradually reduce the number of features being used for control. A robust BCI control was obtained using only 30-40% of the available feature set. Post hoc analysis of the reduced feature set revealed a significant presence of mid-gamma (75-115Hz) band followed by the beta band (15-30 Hz). A novel, 1D Up-Down BCI task was used to study the modulation frequency of these two bands and the differences between them. It was observed that though subjects gradually increased the frequency of modulation in both the bands over a few weeks, they were able to modulate the mid-gamma band at a faster rate. Finally, as a proof concept, two previously trained subjects were used to perform the 2D center-out task with features recorded from only 4 ECoG electrodes. The laboratory recording system and a low power recording system were used in different sessions of experiments, and a robust control was obtained in both the cases. The overall observations and results of these studies provide with a strong basis for ECoG as a low power recording modality that can be chronically used for neural prosthesis.


English (en)


Daniel W. Moran

Committee Members

Dennis L. Barbour, Baranidharan Raman, Lawrence H. Snyder, Kilin Q. Weinberger,


Permanent URL:

Available for download on Monday, July 09, 2018