Power Analysis for High Dimensional Neuroimaging Studies

This article considers the problem of sample size determination for power analysis of neuroimaging studies in-volving multiple comparisons. High dimensional neuroimaging data arise naturally in the context of developing biomarkers for patients with neurologi-cal disorders. The standard methodol-ogy for power analysis cannot be used or extended in such cases because mul-tiple comparisons involving high di-mensional data are more complex and require more parametric information, rather than simply providing the effect size that often plays the vital role in a traditional power analysis problem. This article introduces concepts such as varying effect size and discrimina-tion of null and non-null distributions, while exploring the complexity of rel-evant data. It systematically develops a sample size determination approach that has a simple interpretation and satisfies some desirable intuitive prop-erties (eg, larger effect size requires a smaller sample, larger sample provides better power). Results are illustrated with a real data set.