The New Jersey Institute of Technology's
Electronic Theses & Dissertations Project

Title:	Predictive decoding of neural data
Author:	Halchenko, Yaroslav O.
View Online:	njit-etd2009-045 (xxvii, 197 pages ~ 12.9 MB pdf)
Department:	Department of Computer Science
Degree:	Doctor of Philosophy
Program:	Computer Science
Document Type:	Dissertation
Advisory Committee:	Wang, Jason T. L. (Committee co-chair) Hanson, Stephen Jose (Committee member) Michalopoulou, Eliza Zoi-Heleni (Committee member) Nassimi, David (Committee member) Liu, Chengjun (Committee member) Pearlmutter, Barak A. (Committee member)
Date:	2009-05
Keywords:	Brain imaging FMRI Machine learning Multimodal analysis EEG PyMVPA
Availability:	Unrestricted
Abstract:	In the last five decades the number of techniques available for non-invasive functional imaging has increased dramatically. Researchers today can choose from a variety of imaging modalities that include EEG, MEG, PET, SPECT, MRI, and fMRI. This doctoral dissertation offers a methodology for the reliable analysis of neural data at different levels of investigation. By using statistical learning algorithms the proposed approach allows single-trial analysis of various neural data by decoding them into variables of interest. Unbiased testing of the decoder on new samples of the data provides a generalization assessment of decoding performance reliability. Through consecutive analysis of the constructed decoder's sensitivity it is possible to identify neural signal components relevant to the task of interest. The proposed methodology accounts for covariance and causality structures present in the signal. This feature makes it more powerful than conventional univariate methods which currently dominate the neuroscience field. Chapter 2 describes the generic approach toward the analysis of neural data using statistical learning algorithms. Chapter 3 presents an analysis of results from four neural data modalities: extracellular recordings, EEG, MEG, and fMRI. These examples demonstrate the ability of the approach to reveal neural data components which cannot be uncovered with conventional methods. A further extension of the methodology, Chapter 4 is used to analyze data from multiple neural data modalities: EEG and fMRI. The reliable mapping of data from one modality into the other provides a better understanding of the underlying neural processes. By allowing the spatial-temporal exploration of neural signals under loose modeling assumptions, it removes potential bias in the analysis of neural data due to otherwise possible forward model misspecification. The proposed methodology has been formalized into a free and open source Python framework for statistical learning based data analysis. This framework, PyMVPA, is described in Chapter 5.