الفهرس 
Introduction
Real dataOnly 14 pages are availabe for public view
 |  | from | 247 |  |  |
| | | from | 247 | | |
Abstract
Activation detection methods of fMRI data are required to find the response waveforms and the associated activated regions. Generally, these methods can be divided into two categories depending on whether or not they require prior knowledge about activation patterns: (i) model-based and (ii) model-free techniques. Both model-based and model-free techniques treat the fMRI data as linear mixtures of activation sources; which may impose limitations on their accuracy, especially for complex data sets. The present work investigates thoroughly the performance of a number of the existing techniques commonly used for the processing and activation detection of fMRI data. Moreover, it attempts to develop a new technique based on the Hilbert-Huang transform (HHT) which enables the analysis of both the nonlinear and nonstationary signals. Two types of fMRI databases are used: synthetic and real data sets. Because the truth is not known in a real experiment, synthetic fMRI data representing the brain function under both resting and activated states were generated. The synthetic fMRI data are volumetric data divided into two types: (i) based on boxcar stimulation, and (ii) event-related stimulation. The real database was obtained from the website of the Institute of Neurology at University College London. The comparative study of performance has shown that denoising the fMRI data using the time-series-based (TSB) techniques provides better results than using image-based (IB) techniques. Moreover, comparison of time-series based techniques (time domain and transformed domain) has shown that the Wiener filter implemented with the wavelet packet provides the best results at all noise levels of the synthetic data. As for the activation detection, the results obtained from synthetic data and human data show that the HHT is effective in noise and nonstationarity removal. It provides the best results for activation detection using boxcar as well as event-related paradigms. Moreover, using the HHT technique makes it possible to overcome the nonlinearity of the signal and to detect small levels of stimulations.