Spatially independent activity patterns in functional MRI data during the stroop color-naming task

Abstract

A method is given for determining the time course and spatial extent of consistently and transiently task-related activations from other physiological and artifactual components that contribute to functional MRI (fMRI) recordings. Independent component analysis (ICA) was used to analyze two fMRI data sets from a subject performing 6-min trials composed of alternating 40-sec Stroop color-naming and control task blocks. Each component consisted of a fixed three-dimensional spatial distribution of brain voxel values (a "map") and an associated time course of activation. For each trial, the algorithm detected, without a priori knowledge of their spatial or temporal structure, one consistently task-related component activated during each Stroop task block, plus several transiently task-related components activated at the onset of one or two of the Stroop task blocks only. Activation patterns occurring during only part of the fMRI trial are not observed with other techniques, because their time courses cannot easily be known in advance. Other ICA components were related to physiological pulsations, head movements, or machine noise. By using higher-order statistics to specify stricter criteria for spatial independence between component maps, ICA produced improved estimates of the temporal and spatial extent of task-related activation in our data compared with principal component analysis (PCA). ICA appears to be a promising tool for exploratory analysis of fMRI data, particularly when the time courses of activation are not known in advance.

Figure 1

Different classes of components detected by ICA decomposition of Stroop task fMRI data. (red, z ≥ 2.0; blue, z ≤ 2.0). Negative z values mean those voxels are activated opposite to the plotted time course. (a) Consistently task-related (CTR) component. (b) Transiently task-related (TTR) component. The dotted line shows the time course of the consistently task-related component for comparison. (c) Slowly varying, non-task-related component. The active region for this component was mostly localized to the ventricular system. The lower line shows the mean time course of the active voxels for this component. (d) Quasi-periodic component. This component was largely active in a single slice and had a dominant period of about 12 sec. The spatial distributions of such components were highly reproducible between trials. (e) Suspected abrupt head movement. Note abrupt change in time course, suggesting an abrupt head movement. (f) Component with a “ring-like” spatial structure is suggestive of a head movement.

Figure 2

Comparison of three linear models for analyzing fMRI data. PCA and two versions of ICA were used to linearly separate the data into partially spatially independent maps. The most consistently task-related component determined by each of the three methods from the first trial are shown, along with the correlation coefficient between the associated time courses and the reference function for the behavioral experiment. The ICA algorithm components resembled the task reference function much more strongly than the most highly correlated PCA components.

Figure 3

ICA separated one CTR component for each of the two trials. The right column superimposes the four control/task blocks in each trial on the linearly detrended CTR component. The mean of all eight component activations is shown at the bottom of the right column, superimposed on one cycle of the expected task reference function.

Figure 4

Examples of activations TTR to the task block design. The CTR component is shown with a dotted line. Bolded portions indicate time periods when TTR activations appear task-related.

Figure 5

A scatter plot comparing the map voxel values of a suspected artifact (head movement) component before and after removal of the CTR component. Map voxel values are scaled to z-scores. Note the reproducibility of map voxel values for both the highly active and less-active voxels in each map. Voxel values before CTR removal are shown on the abscissa.

Figure 6

(a) A marked effect on a TTR component of removing the CTR component. The CTR component for trial 1 (Upper) had areas of activation in predominately posterior regions. The same ICA decomposition revealed TTR activation during the second experimental block (blue rectangle) whose map voxel distribution was constrained by the algorithm to be independent to the CTR component (Lower Left). Application of ICA after the removal of the CTR component (by Eqs. 11–18) again revealed a TTR component with strong mesial frontal activation (Lower Right). (b) Scatter plot showing the effects of removal of the CTR component on a TTR component. Here, CTR-component removal had a significant effect on the map voxel values, in contrast to Fig. 5. Yellow voxels denote activation with the time course shown, and blue voxels denote suppression with the same time course.

Figure 7

Simulation depicting the effects of removal of an ICA component on subsequent ICA decomposition. (a) Three simulated activations with different spatial distributions were added to the data set from the first trial. Two of the simulated distributions were highly overlapping for all but the most highly active voxels in anterior and posterior regions (illustrated voxels). The simulated parietal distribution was independent from the other two distributions. (b) After spatial ICA decomposition, the spatial component voxel values for three of the ICA components correlated 0.92, 0.54, and 0.99, respectively, with the distributions of the simulated activations. (c) Removal of the CTR component and reanalysis by ICA resulted in a more accurate separation (r = 0.97) of the least-well-separated distribution (frontal). The separated parietal distribution was largely unaffected by CTR removal (r = 0.99).