Model based estimation and detection of hemodynamic response in event-related fMRI
Magnetic Resonance Imaging (MRI) has rapidly emerged as a powerful, noninvasive diagnostic tool. Research in functional MRI (fMRI) has made accelerated progress on both experimental and analysis fronts. This work is directed at enhancing fMRI analysis through effective incorporation of physiologic knowledge in the detection and estimation procedures. Two projects that have advanced our ability to accurately characterize fMRI activity are described. In project I, the robustness of the phase-encode mapping (PEM) technique is evaluated. This is the first work to find the approximate statistical significance (i.e., p-values) of observed maps through the assessment of type I and II error rates caused by non-stationarity of response latency. Results indicate that latency variance has a greater effect on type II error rate, while size of the cortical area under examination controlled type I error rate. The goal of project II is to develop a novel framework for event-related fMRI analysis, that will effectively incorporate prior knowledge of the spatio-temporal characteristics of brain activation into the detection and estimation of the hemodynamic response (HDR). The proposed procedure is a segmentation algorithm where parameters are estimated by fitting a chosen HDR model to the data. These parameters are used to generate for each voxel, a reference waveform which serves as the basis for clustering. The algorithm yields clusters with corresponding representative responses. Activity is obtained at a regional level, rather than individual voxel level. The method is also implemented on synthetic data and results are compared with those of a standard t-test.
Talavage, Purdue University.
Electrical engineering|Biomedical research
Off-Campus Purdue Users:
To access this dissertation, please log in to our