Task-correlated Head Movement in fMR Imaging: False Activations Can Contaminate Results Despite Motion Correction

In recent years, functional magnetic resonance (fMR) imaging has greatly expanded our capacity to investigate the neuronal substrates of human cognitive processes. This methodology, which relies on blood oxygenation level–dependent (BOLD) contrast, has proven to be a valuable tool for addressing not only questions regarding the basic nature of human cognitive function, but also questions concerning how aging and disease can alter this function.

As in all scientific endeavors, neuroimaging is susceptible to errors. A great deal of attention has been devoted to avoiding statistically false-positive results. That is, given that independent statistical tests are conducted on thousands of voxels, and that each test has a small probability of falsely concluding that there is a significant activation, the sheer number of tests results in a large number of brain regions that spuriously appear to be activated.

False-positive results can also occur when head motion is correlated with the task design, and it is this type of artifact that Field and colleagues address in their article in this issue of the AJNR (page 1388). As an example, if a study investigates brain regions underlying movement of the fingers, and employs an experimental protocol in which periods of finger movement are alternated with periods of rest, it is possible that the subject's finger movements will translate subtle motion to the head during the finger movement blocks. Such motion manifests as translation along or rotation about the x, y, or z axis, and can produce regional differences in signal magnitude between two contrasting conditions (movement vs rest in this example) that reach statistical significance. In contrast to statistically false-positive results (ie, type I errors), it is difficult to assess the probability of motion-induced false-positive results. Most researchers strive to minimize this probability either “on-line” by reducing the opportunity for head motion in the first place by using bite-bars or other head restraint devices, “off-line” by using motion correction postprocessing algorithms to realign all the brain volumes to a reference volume, or by a combination of these methods.

Although previous investigators (1) have demonstrated that relatively large movements (approximately 3-mm translation) can result in spurious activations that are reduced by motion correction algorithms, the investigation by Field et al is unique in three ways. First, a phantom approximating the size and shape of the human brain was constructed, along with an apparatus for introducing controlled in-plane translations and rotations. Thus, simulations of fMR imaging experiments with alternating blocks of two different trial types could be performed and, in contrast to studies using human volunteers, task-correlated motion could be guaranteed to be present while task-correlated neuronal activation was guaranteed to be absent. Second, the effects of subtle movements (< 1 mm) with varying degrees of task-correlated motion were investigated to simulate realistic experimental conditions. Finally, false-positive results due to motion were assessed after employing sophisticated postprocessing algorithms, including motion correction, removal of low-frequency components, and corrections for multiple comparisons using spatial extent. These analytical methods are commonly employed in fMR imaging investigations.

Field et al observed that, despite the subtlety of movement and the inclusion of accepted postprocessing procedures, false activation appeared when movement correlated with the task at r > 0.52, and appeared on every experiment with r > 0.67. The authors argue that “the degree of correlation between stimulus and motion may be more important than the magnitude of motion in creating these artifacts.” Although the investigation of Field et al has methodological limitations (eg, the phantom has a different structural and chemical composition than the human brain, which could result in relatively greater sensitivity to motion-related artifacts), their results should nevertheless raise concerns within the neuroimaging community about the degree to which motion contributes to fMR imaging activation maps.

Although Field et al have increased awareness that the potential for motion-related false-positive results may be present even when motion has been “prevented” or “corrected,” their results raise a number of questions:

• How should investigators modify their procedures to reduce the probability of motion-related false-positive results? At a minimum, it seems reasonable to suggest that investigators monitor the magnitude of correlation between the task and motion for each subject. The most conservative approach would be to discard subjects with unacceptably high correlations, but other corrective measures may be possible and deserve further attention.

• What is the effect of motion on false-negative results? That is, how often does subtle motion eliminate or reduce genuine neuronally derived activation?

• Is through-plane movement more or less likely to produce artifacts than in-plane movement? Field et al investigated only in-plane movement.

• Are event-related fMR imaging investigations less susceptible to motion-related artifacts than block designs? Field et al simulated a block design with six alternating “on” and “off” epochs of 30 seconds each.

Further investigation of these matters will likely improve the quality of functional neuroimaging data, and will increase confidence that results reflect genuine activation rather than motion.