Purpose:To investigate the reliability of functional magnetic resonance imaging (fMRI), an approach for mapping and quantifying reliably activated voxels was developed.
Materials and Methods:First, a SPM99 analysis was performed, and the resulting statistical maps were taken as the basis for subsequent analyses of reliability. Several approaches were demonstrated using 1) a voxel-wise intraclass correlation coefficient (ICC); 2) an analysis of scatter plots, calculating the correlation of contrast t-values for pairs of activation maps; and 3) the ratio of overlapping volumes as suggested in the literature. The methods were applied to an fMRI study in which subjects were asked to vary their attentional effort during watching a flickering checkerboard pattern with varying letters in the center. The subjects had to ignore or attend to the presentation, or they had to detect a target letter within the checkerboard.
Results:The imaging data showed good reliability in terms of ICC for regions of visual processing, as well as for frontal areas, especially in the letter detection task. Furthermore, the size of reliable clusters depended on the presumed attentional effort of the subjects.
Conclusion:Application of the method demonstrated that the activation due to visual stimulation could also be detected very consistently during a no-attend condition, but the reliability of the activations were best during the attended tasks. THIS PAPER DESCRIBES statistical techniques employed for functional magnetic resonance imaging (fMRI) studies that allow assessment of the reliability of activations. The detected signal change in the time course caused by the BOLD effect (blood oxygenation level dependence) is an indirect measurement of neural activity. The BOLD effect can be attenuated because of several underlying technical, physiological, and psychological phenomena. The signal change of about 2%-5% is not much higher than the general noise components in the fMRI time series. The amount of activation can also be affected by remaining motion after realignment, as noted by Casey et al (1) in a comparison of fMRI results across four institutes, whereas the activation maps are in general similar between institutes, modalities, and analysis methods (2).However, use of the fMRI technique for clinical purposes requires reliable results within single subjects measured at different times, for comparisons between subjects or patients, as well as for diagnostic examinations. Aguirre and coworkers (3) compared the variability of BOLD responses within and between subjects and found a higher variability in the shape of responses between subjects than within. McGonigle et al (4) investigated the same subject in several repeated measurements and concluded that single-subject results could lead to erroneous interpretations. This should be taken into account in preoperative planning for which fMRI is becoming more and more important (5). The relationship between neural activity and the signal change is also still a matter of discussion. Arthurs ...