Nonlinear spatiotemporal dynamics of functional MRI revealed by independent components analysis
Linear experimental designs have dominated the field of functional neuroimaging, but although successful at mapping regions of relative brain activation, the technique assumes that both cognition and brain activation are linear processes. To test these assumptions, we performed a continuous functional MRI experiment of finger opposition. Subjects performed a visually paced bimanual finger tapping task. The frequency of finger tapping was continuously varied between 1 Hz and 5 Hz, without any rest blocks. After continuous acquisition of fMRI images, the task-related brain regions were identified with Independent Components Analysis (ICA). When the timecourses of the task-related components were plotted against tapping frequency, nonlinear `dose-response' curves were obtained for most subjects. Nonlinearities appeared in both the static and dynamic sense, with hysteresis being prominent in several subjects. The ICA decomposition also demonstrated the spatial dynamics with different components active at different times. These results suggest that the brain response to tapping frequency does not scale linearly, and that it is history dependent even after accounting for the hemodynamic response function. This implies that finger tapping, as measured with fMRI, is a nonstationary process. When analyzed with a conventional General Linear Model, a strong correlation to tapping frequency was identified, but the spatiotemporal dynamics were not apparent.
