IN 1981, Phelps et al. 1published a positron emission tomography study showing that cerebral glucose utilization increases in the human visual cortex when subjects are scanned with their eyes open versus when they are scanned with their eyes closed. Not a dramatic finding by today’s standards, but this landmark study helped found modern functional brain imaging because it demonstrated that those brain regions that are relatively more active have a regionally specific increase in metabolism, which can be noninvasively measured in vivo in the living human brain. Twenty-three years later, we can now image regional changes in human brain activity without the risk of radioactivity exposure that positron emission tomography imaging requires by using the widely available functional magnetic resonance imaging (fMRI) technique with blood oxygenation level–dependent (BOLD) contrast.2,3This technique exploits the fact that changes in the oxygen state of hemoglobin lead to measurable changes in the local magnetic field susceptibility of the blood and surrounding tissue. By tracking these susceptibility changes over time during a physiologic manipulation (i.e. , a cognitive or sensorimotor challenge), one can measure localized changes in cerebral blood flow as a correlate of brain activity at finer spatial and temporal resolution than that afforded by existing positron emission tomography techniques. Application of fMRI with BOLD contrast to pain4and anesthesia research5are still in their infancy, and a multitude of questions regarding the proper application of this technique remain to be determined. In this issue of Anesthesiology, Ibinson et al. 6move the field forward by providing answers to three parameters that are relevant to the conduct of fMRI studies of pain.
The first parameter addressed by Ibinson et al. concerns the question of signal attenuation. In an fMRI pain study where a pain stimulus is presented to subjects in an on-and-off manner (a so-called boxcar design), does the fMRI BOLD signal get smaller from one boxcar to the next as the session continues? Some studies suggest that it fades or attenuates from one block to the next,7,8whereas other studies have not reported this effect.9,10Ibinson et al. found the BOLD signal in the fourth “on” block to be significantly lower than that seen in the first “on” block in the three brain regions of the anterior cingulate cortex, the somatosensory cortex, and the cerebellum, with the greatest attenuation (i.e. , as much as 50%) occurring in the somatosensory cortex.
The second parameter addressed by Ibinson et al. 6concerns the period of time over which the BOLD signal might remain suppressed, given within-session signal attenuation. Ibinson et al. determined whether the attenuation from one session would continue into another session that starts 4 min after the first. If so, this might be problematic for any study in which the experimental conditions are changed between sessions. If there is an attenuation carryover effect, the second imaging session would start with a reduced BOLD signal and, if an investigator did not counterbalance the active and control scan sessions, a systematic error might be falsely attributed to an experimental manipulation. Fortunately, for future studies, a signal attenuation carryover effect is not evident when a 4-min between-session break is used.
The third parameter addressed by Ibinson et al. concerns the most effective echo time to use in a pain study. These investigators determined whether a 40-ms or a 60-ms TE time would provide the most robust BOLD signal for imaging pain, given a repetition time of 3 s on a 1.5-T scanner. Their results do not reveal much difference between an echo time of 40 or 60 ms, and either seems appropriate.
The source of the within-session BOLD signal attenuation is not known, and many postulates have been offered. Activation of descending pain modulation pathways is one postulate, but this seems unlikely because the psychophysical perception of pain does not fade over a 30-s block of time,7,11and it may actually increase near the end of that time window.9Pain-induced global changes in cerebral blood flow have also been postulated,6,8,12and this idea deserves further study. Another possibility may be stimulus-induced motion that directly varies with the BOLD signal. This is an important additional parameter for future fMRI studies to consider.
The studies showing BOLD signal attenuation have also noted some difficulty with pain-induced subject movement. Indeed, 31% of the scan sessions in the study of Ibinson et al. were eliminated from analyses after applying an exclusion threshold limiting head motion to less than 1 mm of movement from one scan to the next and excluding sessions showing significant task-correlated motion. In the pain study of Kurata et al. ,8pain “on” stimulation periods were limited to 15 s, as opposed to 30 s for visual and motor tasks, to reduce pain-induced movement artifact and suggested that “more sophisticated techniques to correct head-motion artifacts . . . might permit longer duration of pain stimulus.” All of these studies applied the motion correction technique of realigning all of the session scans to the first scan in the series, but is this sufficient to correct for stimulus-induced motion artifact in the fMRI time series? As stated by Friston et al. 13in reference to their retrospective motion correction procedure, “This [motion correction] approach is predicated on the observation that movement-related effects are extant even after perfect realignment.”
The studies that did not report a within-session pain-induced attenuation of the BOLD response used slightly different techniques that may have minimized motion-related effects. Pain-induced motion artifact may have been reduced in the study by Apkarian et al. 9because they used a surface head coil and they had their subjects move their own hands onto the pain stimulus, which allowed their subjects to anticipate and prepare for the pain. In a recent study by our group, we used the approach of Friston et al. 13for motion correction. With this approach, the within-session time series movement parameters that were used to realign the time series are entered into the design matrix of the study such that the linear translations and rotations of the image volume are treated as confounds within the statistical model. This procedure effectively eliminates the signal covariance that is directly due to subject movement and is particularly effective at minimizing or removing stimulus-induced movement effects. Friston et al. 13claim that, using their motion correction technique, their own “empirical analyses suggest that (in extreme situations) over 90% of fMRI signal can be attributed to movement, and that this artifactual component can be successfully removed.”
Motion-induced artifact in the fMRI time series will continue to pose a problem for pain researchers well into the future. This is especially true for studies that attempt to precisely quantify changes in the evoked hemodynamics caused by various pain stimuli. Perhaps recent developments in prospective motion correction techniques will help to minimize concern related to stimulus-evoked motion.14
The potential for functional imaging to help elucidate mechanisms of pain and elucidate anesthesia is enormous. The work of Ibinson et al. reminds us that that any field of research moves forward sometimes with giant steps and sometimes by simply understanding the basic parameters that produce the experimental signals of interest.