ROC analysis of statistical methods used in functional MRI: Individual subjects

The complicated structure of fMRI signals and associated noise sources make it difficult to assess the validity of various steps involved in the stati stical analysis of brain activation. Most methods used for fMRI analysis as sume that observations are independent and that the noise can be treated as white gaussian noise. These assumptions are usually not true but it is dif ficult to assess how severely these assumptions are violated and what are t heir practical consequences. In this study a direct comparison is made betw een the power of various analytical methods used to detect activations, wit hout reference to estimates of statistical significance. The statistics use d in MRI are treated as metrics designed to detect activations and are not interpreted probabilistically. The receiver operator characteristic (ROC) m ethod is used to compare the efficacy of various steps in calculating an ac tivation map in the study of a single subject based on optimizing the ratio of the number of detected activations to the number of false-positive find ings. The main findings are as follows: Preprocessing. The removal of inten sity drifts and high-pass filtering applied on the voxel time-course level is beneficial to the efficacy of analysis. Temporal normalization of the gl obal image intensity, smoothing in the temporal domain, and lowpass filteri ng do not improve power of analysis. Choices of statistics, the cross-corre lation coefficient and t-statistic, as well as nonparametric Mann-Whitney s tatistics, prove to be the most effective and are similar in performance, b y our criterion. Task design. the proper design of task protocols is shown to be crucial. In an alternating block design the optimal block length is b e approximately 18 s. Spatial clustering. an initial spatial smoothing of i mages is more efficient than cluster filtering of the statistical parametri c activation maps, (C) 1999 Academic Press.