Decision Thresholds in Functional MR Image Analysis

In functional magnetic resonance imaging (fMRI) studies, statistical parametric maps (SPMs) plots in an anatomical background those voxels exceeding ap-value threshold. In this study, we explore some limitations of a sole use of p- values including the false discovering rate (FDR) control over experimental-wise error rates. As a comparison, we also include the receiver-operator characteristic (ROC) curve approach for finding the optimal decision threshold. In a real data example, we apply different methods to analyzing the same data set. The results show that both p-value and ROC approaches suggest similar findings when functional images are less contaminated by noise. The FDR control of experimental- wise errors is sensitive to the proportion of voxels being classified as active. We finally discuss the use of different approaches for analyzing fMRI data.