Atlasing
The Atlasing project deals with the general problem of creating statistical maps of data from populations of subjects and creating a toolkit called the Morphometric Atlases and Statistics Toolkit (MAST) to address this problem.
Statistical maps may show summary statistics or may reflect testing of a statistical hypothesis. The types of maps that might be created are limited only by the imagination of the neuroscientist in designing a study and the types of statistical tests that have been implemented in the analysis software.
The types of maps that might be created are limited only by the imagination of the neuroscientist in designing a study and the types of statistical tests that have been implemented in the analysis software.
The Atlasing Project has four aims:
- To broaden the array of available statistical tests to include mixed models and multivariate models that are needed to address modern study designs. Creation of a statistical map involves testing a hypothesis at each of thousands of points in an image, which requires strategies for dealing with Type I error (false rejections of the null hypothesis).
- To implement and validate three types of strategies for dealing with this problem in the context of morphometric data: anisotropic random field theory, false discovery rate control, and permutation testing. The applicability and limitations of each of these methods will be explored and appropriate interfaces provided to validly address the imagewise multiple comparison problem for the various test statistics and models implemented in the first Aim. As an alternative to analyzing each anatomic location independently, data reduction strategies can be used to identify summary factors that characterize changes occurring at many locations simultaneously.
- To implement and validate three such strategies: principal components analysis, partial least squares and independent components analysis. Atlasing is concerned not only with the attributes to be displayed but also the locations at which they are displayed. In order to interpret a statistical map with confidence, it is important to know that the map is representative of the underlying anatomy that it portrays, that the points that are displayed have been sampled reasonably uniformly and that the statistical quantities being shown are all comparable in terms uniformity of relationships between adjacent data points.
- To provide both diagnostic maps that can be used to recognize non-uniformities in each of these domains and remediation strategies that can be used to correct non-uniformities when they are identified. In combination, these four aims will result in a toolkit that provides a unique and comprehensive strategy for handling the atlasing problem.
In combination, these four aims will result in a toolkit that provides a unique and comprehensive strategy for handling the atlasing problem.