In this talk I will discuss our improvements to the current use of Random Field Theory (RFT) for performing multiple testing in neuroimaging. RFT has a long established history of being used for multiple testing (Worsley 1996). However recent work (Eklund 2016) showed that, using large resting state based analyses, many of the standard assumptions of RFT (i.e. smoothness, stationary and Gaussianity) do not hold in fMRI. As a result traditional means of applying this framework cannot be relied upon to control false positive rates. I will show how it is possible get around each of the 3 standard assumptions. Moreover I will demonstrate that this results in a fast and powerful framework for performing multiple testing inference in fMRI which correctly controls the voxelwise false positive rate. I will present resting state validations based on on 7000 subjects from the UK BioBank in order to demonstrate that the error rate is correctly controlled in practice.
