The selectivity, sensitivity, criterion validity, and test-retest reliability can be calculated for both PET and MRI studies. Establishing the test-retest reliability of most baseline PET measures is relatively straightforward (Ball, Fox, Herscovitch, & Raichle, 1988; Nyberg, Farde, & Halldin, 1996; Schmidt et al., 1996), although this literature remains surprisingly small considering the increasing use of these measures in clinical diagnosis. Establishing the test-retest reliability of activations caused by stimulation paradigms is a trickier issue. Reliability in these paradigms will always be task and region specific, making it impossible to make generalizable statements about reliability. Nevertheless, there are increasing attempts to define the test-retest reliability of the activations associated with specific cognitive and motor tasks (Fernandez et al., 2003; Kiehl & Liddle, 2003; Maitra, Roys, & Gullapalli, 2002; Specht, Willmes, Shah, & Jancke, 2003). This issue has proved particularly important when fMRI is used as part of presurgical planning for intractable epilepsy. Obviously, a neurosurgeon needs to know which measures (neuropsychological data, WADA procedure, etc.) provide the most valid and reliable information about functional localization of cognitive tasks (particularly language tasks) before choosing to remove part of a patient's cortex. However, determination of the psychometric properties of neuroimaging data is complicated by the fact that the data sets include information on magnitude of change (or the degree of temporal correlation) and location of activation. For instance, imagine performing a receptive language study on a patient on two occasions. In both cases the subject demonstrates activation in the left superior temporal gyrus, but the emerging foci, although within 5 mm of each other, do not overlap. Depending on one's criteria, this could be viewed as a replication or a failure to replicate. When viewed loosely (for instance, in terms of hemispheric asymmetries within the temporal or frontal lobe), such tasks have typically shown good reliability (Fernandez et al., 2003; Rutten, Ramsey, van Rijen, & van Veelen, 2002). In contrast, when viewed on a voxelwise basis, the overlap between activations across sessions tends to be much lower (Fernandez et al., 2003).
Attempts to use functional neuroimaging for diagnoses have also provided information regarding the sensitivity and selectivity of this information. This has received particularly strong attention in the diagnosis of early Alzheimer's disease (Petrella, Coleman, & Doraiswamy, 2003). Research along similar lines will clearly need to be performed if functional neuroimaging is to reach its full potential as an assessment tool, regardless of whether it is used in isolation or as part of a MTMM matrix.
Was this article helpful?