Random field theory-based voxel-level fMRI positive predictive value calculation
In times of reproducibility crisis, there exists an increasing demand for adequately powered cognitive neuroimaging studies with high predictive value. For mass-univariate GLM-based fMRI studies, power analyses are complicated by the large number of simultaneous outcome measures (voxel data). A variety of specialized procedures has been developed that control different false positive rates of the ensuing multiple testing problem. In this talk, I will review an approach proposed by Hayasaka et al. (2007) that evaluates power and adequate sample sizes while incorporating the randomfield theory-based control of the family-wise false positive error rate. I will detail both the intuitive and technical foundations of power and sample size calculations and the random field approach for family-wise error control in fMRI. In addition, I will expand the approach by Hayasaka et al. (2007) to the evaluation of positive predictive value maps, an important concept that has recently given rise to a renewed interest in power calculations (e.g. Button et al., 2013; Szucs and Ioannidis, 2016).
Dec 05, 2016 | 04:00 PM