EMPIRICAL AND SUBSTANTIVE MODELS, THE BAYESIAN PARADIGM, AND METAANALYSIS IN FUNCTIONAL BRAIN IMAGING

Functional neuroimaging research is currently rediscovering and adapti ng established statistical methods for its use, including design of ex periments, the general linear model, contrasts, random field theory, l ongitudinal models, Fourier analysis, and general signal and image pro cessing methods. This brief paper gives an example of comparative perf ormance of five different statistical models applied to the same set o f data generated in an fMRI study of motor cortex. These methods inclu de a two-sample t-statistic, a Kolmogorov-Smirnov statistic, a princip al component/canonical variates approach, a pruned feedforward artific ial neural network with one hidden layer, and a frequency domain regre ssion convolution model allowing for spatially varying hemodynamic res ponses. Produced by essentially empirical statistical models, there ap pear to be more similarities than differences in these spatial activat ion patterns, yet all lack explicit incorporation of substantive prior information. Distinctions are drawn between exploratory models for hy pothesis generation and confirmatory models for hypothesis testing. Ln addition, the Bayesian paradigm helps to combine empirical and substa ntive models, and meta-analysis provides a rational means by which to combine information over a range of similar results affected minimally by publication bias. (C) 1997 Wiley-Liss, Inc.