Dr. Roberts is a Radiology Resident; Mr. Kelley and Mr. Carroll are Medical Students; Dr. George is a Distinguished Professor of Psychiatry, Neurology, and Radiology; and Dr. Haines is a Professor and the Chairman of the Department of Neurosurgery, Medical University of South Carolina, Charleston, SC.

Over the past 2 decades, detailed visualization of normal and pathologic brain anatomy has been achieved by magnetic resonance imaging (MRI) and computed tomography. With the development of new technologies--such as positron emission tomography, single-photon emission computed tomography, magneto-encephalography, and functional MRI (fMRI)­­functional brain information can be obtained along with anatomic detail. One neuroimaging method, blood-oxygen-level­dependent (BOLD) fMRI, provides high-resolution anatomic and functional information. Currently, BOLD fMRI has gained wide acceptance in research and has been used for neuroscience research throughout the world.

In the clinical arena, fMRI technology promises easily obtainable, noninvasive presurgical mapping for neurosurgical patients. The localization of eloquent cortex prior to surgery would allow neurosurgeons to plan surgical approaches and better discuss treatment options with patients before embarking on a surgical treatment plan. Already, many neurosurgeons are enthusiastically requesting fMRI cortical activation maps along with Wada testing. However, for such an important new tool, there has been surprisingly little research in the area of reproducibility of results. A few studies have shown "good reproducibility" and "high test­retest reliability" of fMRI activation; however, a great deal of inter-scan variability has been seen in even the most robust fMRI activation protocols. These studies raise questions about whether fMRI should be used in clinical practice before further validation studies have been performed.

Functional magnetic resonance imaging is derived from the BOLD effect. Following a period of stimulus-induced neuronal activation, an increase in local blood flow and volume leads to an increase in oxygenated blood. A T2* weighted signal that is proportional to the rise in oxyhemoglobin concentration can be detected. Images obtained while the subject performs a specific task are then subtracted from images obtained during a period of rest. The difference map is statistically weighted and thresholded. Areas of significant activation are mapped onto structural images. Functional MRI maps are generated relatively easily and 1.5 T MRI scanners with echoplanar capability are now readily available. Therefore, fMRI would be the functional neuroimaging modality of choice in many clinical situations. Functional MRI has been used for presurgical mapping, in neuropsychiatry, ^1,2 and in elucidating functional cortical networks. ^3-6 Yet fMRI has not been shown to reproduce functional activation with sufficient sensitivity for neurosurgeons to confidently rely on fMRI to replace intraoperative cortical mapping.

Various investigators have examined the reproducibility of BOLD fMRI activation patterns during motor, sensory, and visual tasks. ^7-16 Typically, reproducibility studies have been performed in healthy, cooperative volunteers. In patient populations, however, reproducibility may be even lower due to the presence of neuropathology and difficulty with patient participation during the exam. Lowered reproducibility rates may also be due to several uncontrolled experimental variables, including subject motion, equipment variables, data analysis methods, and physiological variables, such as the level of cortical excitability. Cortical excitability varies across individuals and within certain patient groups, such as epilepsy patients, who have been shown to have altered cortical excitability. ¹⁷ Cortical excitability also varies over time within a specific subject, which may contribute to intrasubject variability of fMRI activation patterns. Factors associated with changes in cortical excitability include sleep deprivation, levels of alertness and arousal, anxiety, and medications such as stimulants or caffeine.

Review of the literature on fMRI reproducibility indicates that the number of activated pixels and the location of activation tends to differ to some extent between subjects performing the same task, as well as within the same subject scanned on multiple occasions. For example, Yetkin et al ⁸ studied 4 subjects who repeated both sensory and motor fMRI tasks twice within one MRI session. Reproducibility was specified as a ratio of the number of active pixels on both the first and the second scan, divided by the number of pixels active on either the first or the second scan. The reproducibility ratio was found to be 0.57 (measured at a 0.5 correlation threshold) but could be increased to 0.81 if nearest neighbor pixel activation was included.

Rombouts et al ¹⁰ reported on the reproducibility of visual cortex fMRI activation. The area and location of activation during an fMRI visual task were measured for 18 volunteers who underwent 2 scanning sessions. Reproducibility was estimated using a ratio for the number of activated pixels (R _size ) and for activation overlap (R _overlap ), with both ratios equal to 1 with perfect reproducibility of the number of activated pixels and the location of activation. Size and location of activation were variable with mean intrasubject values of 0.83 ± 0.16 for R _size and 0.31 ± 0.11 for R _overlap . In 2 subjects, visual cortex activation was not seen at all.

Noll et al ⁹ examined the test­retest reliability of fMRI motor tasks in 7 subjects. For the 3 scans acquired within the same scanning session, an average true-activation detection rate was determined to be 0.7264. Three additional subjects underwent fMRI motor mapping over multiple scanning sessions. The average true-activation detection rate between intersession scans was determined to be 0.4741, significantly lower than the detection rate for within-session scans. Although a multistage procedure was performed for alignment of slices between sessions, residual misregistration of slices between sessions could have contributed to the lower detection rate.

Scholz et al ¹⁴ investigated the reproducibility of fMRI motor activation in motor cortex and subcortical nuclei. They found an intrasubject percent deviation in the number of active pixels to be 7.2% in motor cortex and even higher in supplementary motor area and basal ganglia (21.5% to 26.4%). Other fMRI reproducibility studies have shown similar results.

In addition to issues of reproducibility of fMRI activation, the basis of the BOLD effect has not been fully characterized, and functional maps have not been fully validated in comparison with the gold standards, such as intraoperative cortical stimulation, subdural grids, and Wada testing. Case reports and small series of combined fMRI and intraoperative direct electrical cortical stimulation (ECS) have generally shown good agreement between the two methods; however, concordance has not been 100% in larger series. A comparison of fMRI functional localization with a gold standard in a patient population large enough to demonstrate statistical significance is necessary before fMRI should be widely used in clinical situations.

Various smaller studies have been performed that compare fMRI with ECS. ^18-28 Puce et al ¹⁹ studied 4 patients in whom preoperative fMRI was performed and co-registered with a three-dimensional (3D) volume. The 3D MRI volume was used to create a
volume rendering of the brain's surface. The patients then underwent intraoperative cortical stimulation and somatosensory evoked potential (SSEP) recordings. Photographs of the surgical field were obtained and aligned with the surface renderings of the brain using anatomical landmarks. The authors stated that fMRI and electrophysiological studies have revealed good agreement on location of motor and sensory areas, but the spatial extent of activation, as determined by the two methods, differed. They proposed this difference could be due to anesthesia and the inherent variables in electrophysiological measurements.

Yetkin et al ²³ studied 28 patients who underwent preoperative fMRI with motor and language tasks. Intraoperatively, motor cortex and the location of speech arrest were identified by cortical stimulation. Numbered markers were placed on the brain's surface and a photograph of the surgical field was obtained. The photograph was superimposed onto the most superficial slice from the fMRI study by identification of sulci and venous structures. In all comparisons, fMRI activation and the area identified intraoperatively during the performance of a similar task were within 2 cm of each other; however, only 87% of the comparisons were within 1 cm of each other. The authors stated that the discrepancy could be due to the combined inaccuracies of each measurement. Repeated fMRI sessions generally showed an overlap of 50% to 80% of the areas of activation. Although intraoperative stimulation is considered the gold standard for functional mapping, its precision is approximately 1 cm due to the spread of current through the cortex. ²³

Fitzgerald et al ²¹ performed fMRI in 11 patients to identify language areas. Three-dimensional spoiled gradient-echo images were merged with an MR angiogram to create a surface rendering with vascular landmarks. Functional MRI language maps were then co-registered with this 3D volume, and foci of activation >= 1 cm were projected onto the cortical surface. Cortical stimulation was carried out intraoperatively, and language areas were marked with tags. A photograph of the surgical field was obtained, which was superimposed onto the surface rendering of the functional activation. The combined sensitivity for all language tasks performed across all patients was 81%, with sensitivity defined as the percentage of language tags placed intraoperatively that matched fMRI activation areas. Specificity was 54% and was defined as the percentage of nonlanguage tags not found by fMRI.

Fandino et al ²⁴ reported on 11 patients who underwent presurgical fMRI to identify areas of activation with a hand motor task. Intraoperative ECS was performed on all patients. The topographical relationship between tumor and primary motor cortices, as identified by ultrasonography, was used to compare fMRI activation with the results of ECS. The authors stated that in only 9 (82%) of the 11 patients, fMRI activation could be verified by intraoperative ECS.

Finally, the utility of fMRI in presurgical planning has not been investigated thoroughly. Before the use of fMRI becomes widespread, the information provided by fMRI scanning must be shown to be clinically useful. Lee et al, ²⁹ in an attempt to assess the clinical usefulness of fMRI, performed a retrospective study of 46 patients who had undergone preoperative fMRI. The authors reviewed patient medical records to document how often, and in which ways, the fMRI study had influenced clinical management. They concluded that preoperative fMRI was useful at three key points in the clinical decision-making process: 1) for assessing the feasibility of surgical resection, 2) to aid in surgical planning, and 3) for selecting patients for more invasive mapping procedures. Further discussions and studies investigating the clinical relevance of presurgical fMRI have yet to be conducted.

Conclusion

While the ultimate goal is to find the simplest and safest way to identify critical structures and plan surgical approaches, fMRI presurgical localization should not yet be used in place of standard techniques as it has not yet been demonstrated that fMRI functional mapping is at least as safe and effective as intraoperative ECS. Before fMRI can be used as a presurgical tool, reliable intrasubject activation with sufficient sensitivity must be demonstrated. Such a study could be performed as a multicenter trial, allowing the enlistment of a larger patient population, using a standardized protocol comparing fMRI activation with a designated gold standard.

Acknowledgment

Dr. Roberts would like to thank Dr. James Ravenel for his editorial assistance with the manuscript.