Identification of activated brain regions while controlling false positives is a central problem in functional neuroimaging. As part of Phase I of the Human Brain Project (HBP), we propose to develop, evaluate, and implement nonparametric inference tools for neuroimaging. Requiring no assumptions or only weak distributional assumptions, nonparametric tests offer inferences with false positive rates exactly as specified. Further, preliminary work using our SnPM software has found that nonparametric methods can be substantially more powerful than parametric random field results, while requiring none of the random field smoothness and high-threshold assumptions. As part of our research plan, we will (1) evaluate existing nonparametric and parametric methods, (2) develop and evaluate new nonparametric methods, and (3) build, distribute and support nonparametric software tools. We will use nonparametric methods to validate parametric random field methods with real data, and will also evaluate and optimize existing nonparametric tools. We will create new nonparametric tools, which address shortcomings of existing methodology, for example, cluster size tests valid under heterogeneous smoothness, or variable threshold methods to focus or homogenize power. And most significantly, we will implement all of the proposed methods in our modality-independent software. We will make our software both scriptable and more user-friendly, and create web-based documentation. In addition, we will make our software interoperable with other widely used neuroimaging software tools, in the spirit of both the HBP and the NIMH & NINDS NIfTI initiative. Once prohibitively slow, nonparametric methods are now practical for researchers with even modest computing hardware. Our software will make nonparametric methods widely available and easy to use. Our software will facilitate methodological developments by allowing use of the arbitrary statistics, instead of just those with parametric results, and facilitate neuroscientific developments by giving researchers access to powerful nonparametric methods.
| Kang, Jian; Johnson, Timothy D; Nichols, Thomas E et al. (2011) Meta Analysis of Functional Neuroimaging Data via Bayesian Spatial Point Processes. J Am Stat Assoc 106:124-134 |
| Zhang, Hui; Nichols, Thomas E; Johnson, Timothy D (2009) Cluster mass inference via random field theory. Neuroimage 44:51-61 |
| Xu, Lei; Johnson, Timothy D; Nichols, Thomas E et al. (2009) Modeling inter-subject variability in FMRI activation location: a Bayesian hierarchical spatial model. Biometrics 65:1041-51 |
