| The rapid growth and wide acceptance of functional magnetic resonance imaging (fMRI) as a noninvasive method for brain function studies have stimulated the development of new methods for analyzing and visualizing brain imaging data. The focus of this study is to introduce a modified neural gas (NGMD) algorithm that incorporates spatial information for the analysis of fMRI data.; In this research, we have made three main contributions to the area of fMRI data analysis. First, we developed the NGMD algorithm. The results on simulated and experimental data showed that the method was robust for detecting areas of brain activation, especially under difficult conditions such as low contrast-to-noise ratio, complex response and presence of phase delay. The method was also able to identify unknown hemodynamic responses that traditional model-driven methods were not able to find out. Second, we developed an algorithm for cluster merging after clusters of activation had been detected by data-driven methods. This technique allowed automatic selection and merging of the clusters identified by model-free methods; and at the same time it ensured that the homogeneity of the merged cluster was acceptable at a level of significance. This technique was demonstrated on both simulated and experimental fMRI data. Third, we combined the NGMD algorithm with techniques for the segmentation and reconstruction of the cortical surface of the brain. We used the reconstruction of the cortical surface to limit the detection of functional activation only to those voxels of a functional data set that lie within the cortex. Since the NGMD algorithm took into account the true distance information between various brain regions in flattened surface data, a more accurate estimation of spatial layout of the brain activations was expected. We validated this technique on fMRI data collected in an actual muscle fatigue experiment. |
