Thresholding in correlation analyses of magnetic resonance functional neuroimaging

The definition of objective and effective thresholds in MRI of human brain function is a crucial step in the analysis of paradigm-related activations. This paper introduces a user-independent and robust procedure that calculates statistical parametric maps based on correlation coefficients. Thresholds are introduced as p values and defined with respect to the physiologic noise distribution of the individual maps. Experimental examples from the human visual and motor system rely on dynamic acquisitions of gradient-echo echo-planar images (2.0 T, TR = 2000 ms, 96 x 128 matrix) with blood oxygenation level-dependent contrast. The results demonstrate the disadvantages of thresholding with fixed correlation coefficients. In contrast, taking the individual noise into account allows for a derivation of p values and a reliable identification of highly significant activation centers. An adequate delineation of the spatial extent of activation may be achieved by adding directly neighboring pixels provided their correlation coefficients comply with a second lower p value threshold. (C) 2003 Elsevier Inc. All rights reserved.