The use of Random Field Theory (RFT) to control the false positive rate in neuroimaging is a well established testing framework. In a recent manuscript, Bansal and Peterson (2018) found that in simple Gaussian simulations RFT incorrectly estimates the number of clusters of a Gaussian field that lie above a threshold. However, these results contradict the existing literature and our own previous findings, and exhibited highly counter-intuitive features. Unfortunately this implies that a number of their conclusions regarding the performance of RFT are invalid.
The purpose of this repository is to provide freely available and reproducible code the evaluations considered by Bansal and Peterson (2018); showing simulation and theory align in line with the standard literature. Our results show that in these simulations the expected Euler characteristic closely matches the average number of clusters as predicted by theory. Unfortunately it appears that Bansal and Peterson (2018) have miscalculated the Euler characteristics and have simulated their random fields incorrectly. Unfortunately this implies that their conclusions regarding the performance of RFT are invalid. See our response article HERE for full details.
To run the code in this repo you will need to have MATLAB and to install the RFTtoolbox - A MATLAB toolbox designed for generation and analysis of random fields both continuously sampled and on a lattice. In addition, SPM12 must be installed in order for the 3D LKC estimation (performed using SPM) to work.
These figures compare out results with those obtained in Bansal and Peterson (2018) (which is denoted as BP in the plots). The figures demonstrate that in these simulation settings RFT provides a close estimate of the expected number of clusters (at all applied smoothness levels) contradicting their findings.
