Strategies for block-design fMRI experiments during task-related motion of structures of the oral cavity

Abstract

Functional MRI (fMRI) studies of jaw motion, speech, and swallowing disorders have been hampered by motion artifacts. Tissue motion perturbs the static magnetic field, creating geometric distortions in echo-planar images that lead to many false positives in activation maps. These problems have restricted blood oxygenation level-dependent (BOLD) fMRI studies involving orofacial muscles to event-related designs, which offer weak contrast-to-noise ratios when compared to block designs. Two new approaches are described that greatly reduce false positives in the activation maps created by the distortions in block-design fMRI studies involving jaw and tongue motion during chewing. First, an appropriate task duration of 10−14 s was found to maximize functional contrast while minimizing motion artifacts. Second, three motion-sensitive postprocessing methods were applied successively to examine the temporal and spatial characteristics of responses and identify and remove false positives caused by motion artifacts. These techniques are shown to allow the use of block design in an fMRI study of a jaw motion task. Extension to speech and swallowing tasks is discussed.

Introduction

Diagnosis of jaw motion, speech, and swallowing disorders can be enhanced using neuroimaging methods. Part of the motivation for the present experiment was to establish the use of functional MRI (fMRI) to probe the neurophysiological basis of temporomandibular joint (TMJ) disorder. Even if a chewing task does not elicit perceptual pain, below-perception activation may occur that would allow an early diagnosis of the disease. It is noted that rectal-distension-induced activity in cortical areas associated with pain perception was observed in an fMRI study for stimulation that was below the perception level (Kern et al., 2001b).

Problems exist, however, when using fMRI to acquire blood oxygenation level-dependent (BOLD) activation maps with tasks involving motion of the head and muscles of the oral cavity. First, interscan head motion can have a significant effect on study sensitivity (Zeffiro, 1996). Task-related head motion can also create false positive activation along high-contrast boundaries. Volume registration routines (Cox and Jesmanowicz, 1997) have been developed for rigid-body alignment of brain images to correct for slight movements of the head that occur despite head restraints. Second, tissue motion can create distortions in the image. Differences in the magnetic susceptibilities at tissue interfaces create magnetic field inhomogeneities that require shimming to produce a more uniform static magnetic field, B₀ (Haacke et al., 1999). Magnetic field inhomogeneities create geometric distortions in echo-planar images, principally along the phase-encoding direction (Jezzard and Balaban, 1995). Motion of tissue during echo-planar imaging (EPI) alters the static magnetic field in a dynamic fashion. Even if the moving tissue lies outside the imaging field-of-view (FOV), these static field changes can still create dynamic geometric distortions or signal dropouts in the image (Yetkin et al., 1996). These distortions or signal dropouts can generate false positive activation or obscure true activation during fMRI studies.

Motion-induced signal changes were shown to equal or exceed BOLD activation-induced signal changes (Yetkin et al., 1996). The motion artifact responses (MARs) were generally found to have faster rise times than BOLD responses and to be both in-phase and out-of-phase with BOLD responses. The MARs with the strongest responses were identified at the superior and inferior edges of the phantom and at the high-contrast boundaries of the head. The largest magnetic field perturbations caused by speech occur in the inferior regions of the brain, suggesting that increased amounts of distortion occur closer to the tissue in motion (Birn et al., 1998). Similarly, the largest reductions in image signal-to-noise ratio (SNR) during speech occur in the inferior axial slices (Barch et al., 1999). The cortical representation of jaw and tongue motion is located in the inferior region of the primary motor cortex (Kandel et al., 2000). Thus, localization of jaw and tongue activation is especially susceptible to motion artifacts.

Event-related (ER) experimental designs, which are affected somewhat less by motion artifacts, have been employed for tasks involving orofacial motion. In an event-related design, the task occurs briefly. Images acquired during the brief period of motion can be ignored and BOLD activation maps can be found by identifying the peaks of the BOLD responses that are naturally delayed by 5−6 s (Birn et al., 1999). ER designs have been used to detect artifact-free activation for speaking (Birn et al., 1999, Huang et al., 2001, Palmer et al., 2001), swallowing (Birn et al., 1999, Kern et al., 2001a), jaw clenching (Birn et al., 1999), tongue movement (Birn et al., 1999), and smiling (Gosain et al., 2001). In a block-design study, Barch et al. (1999) found that discarding the images acquired during an overt task was more effective in both revealing true activation and removing motion artifacts when applied to group analysis. This is because motion artifact responses are less likely to be repeated in similar locations across subjects. The disadvantage of ER designs is that the short task leads to a relatively small BOLD signal. For example, the contrast-to-noise ratio (CNR) of an ER experiment using a 2-s task was found to be 35% less when compared to a block-design experiment using a 20-s task (Bandettini and Cox, 2000). Low CNR weakens the detection of BOLD responses and may cause activation to go undetected in fMRI experiments.

Alternatively, it may be possible to use block designs in fMRI studies involving muscles of the oral cavity if proper task durations are used. Birn et al. (2004) found that a block duration of 10 s during a speech task offset the MAR from the BOLD response by a quarter cycle, allowing good separation between the phases of the two types of responses. One of the aims of the current study was to investigate the optimum task duration that would maximize the CNR while minimizing the number of false positives caused by MARs.

For speech studies, many researchers have investigated eliminating motion altogether by replacing overt speech tasks with covert speech tasks (Barch et al., 1999, Huang et al., 2001, Palmer et al., 2001). These studies assume that covert speech tasks will stimulate the same language areas of the brain as overt speech. However, different patterns of brain activation in Broca's area have been reported when comparing overt and covert speech (Barch et al., 1999, Huang et al., 2001). Thus, overt speech cannot be represented simply by brain activation produced by covert speech plus primary motor and auditory cortex activation. In addition, covert speech has been shown to produce BOLD responses of a lower magnitude than overt responses in motor and non-motor cortices (Palmer et al., 2001, Barch et al., 1999). Thus, fMRI acquisitions with covert speech tasks may fail to detect activation that could be detected with overt speech tasks.

Dynamic geometric distortion can be reduced to some degree using special pulse sequences. An EPI sequence containing a real-time autoshimming sequence was developed to provide first-order correction for motion-induced changes in the magnetic field (Ward et al., 2002). However, this technique only employed a linear correction, which may fail to correct distortions near regions containing sharp changes in magnetic susceptibility.

Another solution would be to acquire field maps with each successive image so that the magnetic field inhomogeneities can be mapped and each image unwarped. Attempts have been made to acquire two successive images with different echo times (TE) for each slice in an EPI sequence (Birn et al., 1998, Hutton et al., 2002). The phase images of these two images were employed to generate a field map, which was then used to unwarp the geometric distortions specific to each acquired image (Jezzard and Balaban, 1995). Dynamic field map estimation has also been achieved via the use of a spiral-in/spiral-out sequence (Sutton et al., 2004). The spiral sequence approach was effectively used to track magnetic field drift and respiration-induced phase oscillations. The problem with these methods as they have been applied is that the time delay between collection of the two successive images allows motion of the head, adding noise to the phase images. This noise leads to inaccuracies in the field map, reducing the effectiveness of the geometric correction. Hutton et al. (2002) suggested that although the use of unique field maps for each successive image provided improvement in the registration of functional images to anatomical images, errors in the field map resulted in an increase in the variance of voxel time-series. A pulse sequence that shortens the time delay between collection of successive images could alleviate this problem.

Several papers describe mathematical methods to correct geometric distortions in EPI images (Andersson et al., 2001, Hutton et al., 2004, Ernst et al., 1999). Instead of acquiring field maps, echo-planar images of the brain were processed via affine transformations (rotation, translation, scaling, and shearing) to register the surface of the functional brain with the surface of a high-resolution anatomical image (Ernst et al., 1999). The authors of this study point out that this method suffers from an inability to correct the extreme geometric distortions and signal loss caused by high-order magnetic field inhomogeneities. Alternatively, a map of magnetic field inhomogeneities was estimated by comparing subsequent images from EPI data (Andersson et al., 2001). The effect of geometric distortions caused by task-related motion of muscles in the head on these correction techniques is not known, however.

Motion artifact responses are likely to possess temporal characteristics different from true BOLD responses, suggesting that postprocessing methods can be used to separate false positives from true positives in data analysis. The rise of the MAR is immediate with the start of the motion-related task, whereas the rise of the BOLD response is delayed by 2−3 s and the peak is delayed by 5−6 s. Huang et al. (2001) used a complex cross-correlation analysis (Lee et al., 1995) to compute the phase of the correlation for each voxel in an ER study. They excluded activated voxels that had a phase corresponding to a motion response. However, they only applied their method to ER fMRI. In the present study, complex cross-correlation analysis is applied to block-design fMRI. A threshold was created to identify and remove false positives that have a temporal phase corresponding to motion responses. MARs also have a wide range of signal intensity changes, whereas BOLD responses tend to be on the order of 5−10% of the baseline, depending on voxel size. Furthermore, the BOLD response is likely to repeat itself identically over several epochs of a block-design run, whereas the MAR is not. A second threshold was created to remove false positives for which the variances of values in the averaged response are excessively high. Lastly, a cluster analysis threshold was employed to remove isolated activated voxels remaining after the previous thresholds, since these are likely to be false positives.

In brief, there were two main aims of this experiment. The first aim was to find the optimum task duration that would maximize the CNR while minimizing the number of false positives caused by MARs. The second aim was to optimize postprocessing methods that examine temporal and spatial aspects of the responses to remove false positives caused by motion artifacts. The combined use of these strategies is shown to allow the analysis of a block-design functional MRI experiment involving task-related jaw motion.