ON THE USE OF TEMPORAL CORRELATION-COEFFICIENTS FOR MAGNETIC-RESONANCE MAPPING OF FUNCTIONAL BRAIN ACTIVATION - INDIVIDUALIZED THRESHOLDS AND SPATIAL RESPONSE DELINEATION

Functional activation of the human visual and motor system was studied by magnetic resonance imaging (MRI) at 2.0 T using dynamic series of oxygenation-sensitive gradient-echo images at high spatial resolution. Activation maps were computed by correlating signal intensity time co urses with a reference waveform on a pixel-by-pixel basis. Although th is strategy readily demonstrates stimulus-related functional cooperati vity of activated regions in thresholded maps of correlation coefficie nts, intertrial variability in the underlying distributions of correla tion coefficients precludes the use of correlation coefficients as dir ect thresholds for defining activation. Because stimulus-related effec ts emerge as positive deviations from an otherwise symmetric distribut ion of correlation coefficients, invariance against intertrial differe nces as well as adequate visualization of activated areas may be achie ved by the following procedure. First, a symmetrized noise distributio n is reconstructed from the actual activation map that allows rescalin g of correlation coefficients into percentile ranks with respect to th e integral of the noise distribution. Second, a high percentile rank ( or correspondingly low error probability) can be used as threshold to define primary sites of activation with high specificity. And third, t he spatial extent of activation may be delineated by adding directly n eighboring pixels with lower values provided their correlation coeffic ients are high enough to contribute to the positive deviation from the noise distribution. The outlined approach yields robust activation ma ps but still awaits a more thorough statistical treatment of activatio n in MRI correlational mapping. (C) 1995 John Wiley & Sons, Inc.