Comparing statistically pooled brain maps in FMRI studies using parametric and non-parametric methods
MetadataShow full item record
In a typical functional brain imaging experiment, scientists aim to map the specifi c areas of the brain that are activated while subjects perform a designated task (which may be cognitive, motor, or other). For comparison purposes (comparing patients vs. controls, females vs. males, as some examples), combining the brain maps from the subjects in an efficient way becomes imperative so that we can get an overall picture of activity for each group. We use statistical tests that have been developed historically for combining independent sources of information to create maps for each group of subjects in a neuroimaging study. These statistical tests follow two basic approaches - combining p-values and meta-analysis. Through these methods we aim to draw conclusions about the behavioral pattern of two or more groups with respect to each other. We also want to compare the performance of the diff erent methods. Group comparisons have been done in the past using 'group maps" for each population through a fi xed e ffects model or a random eff ects model. This dissertation explores some pre-existing statistical combination techniques used in combining and inter- preting 'group maps". We will use parametric and non-parametric approaches to compare between two or more populations. While combining and comparing brains, there are two aspects that arise - spatial and statistical. We will only focus on the latter aspect. We will assume that the voxels of the brain are independent of each other. However, as we conduct various tests to compare group maps at each voxel, to minimize false positives i.e. voxels declared active when they are not, we will threshold at each voxel. In this dissertation we will explore thresholding through false discovery rate, permutation tests and bootstrapping and we compare these methods to draw a conclusion about which one would be apt to use.