| Methods for the analysis of event-related blood oxygenation level dependent (BOLD) fMRI data typically assume that the signal of interest consists of a fixed, reproducible hemodynamic response to repeated stimuli in the presence of a constant noise background. This assumption is not appropriate if the underlying functional units contributing to this signal activate stochastically, especially if the spatial resolution is such that only a small number of functional units are present in each voxel. A method for analyzing trial by trial consistency of high resolution voxels, called specified resolution wavelet analysis (SRWA), was developed to assess this deterministic assumption. Monte Carlo methods were used to evaluate the sensitivity and specificity of this technique with both synthetic and real data. Despite the typically low contrast to noise ratio of BOLD fMRI data, SRWA was sufficient for the identification of voxelwise activation on the individual trial level.;Data were analyzed from the primary visual cortex (V1) of healthy human sujects performing a simple visual checkerboard task. The consistency of voxelwise activation from trial to trial was significantly less than assumed by a deterministic model at high spatial resolutions. The stochastic nature of this observed brain activity was not explained by fluctuations in behavior, physiology, movement, or any other of a number of common confounds typically found in fMRI experiments. Additionally, test insensitivity did not account for these low consistency rates. As spatial resolution is decreased, the observed activation became consistent with maps produced by traditional deterministic analyses, such as the t-test. Hence, the stochastic model of brain activity is compatible with traditional assumptions in the limit of spatiotemporal averaging. The decreasing activation consistency at increasing resolution suggests a stochastic selection of functional units from stimulus to stimulus. Although the brain responds to each event, the pattern of activation is not predetermined. Instead, identical stimuli can be processed by different collections of relevant functional units. |
