Mapping individual brains to guide restorative therapy after stroke: Rationale and pilot studies

Abstract Some treatments under development to improve motor outcome after stroke require information about organization of individual subject's brain. The current study aimed to characterize normal inter-subject differences in localization of motor functions, and to consider these findings in relation to a potential treatment of motor deficits after stroke. Functional MRI (fMRI) scanning in 14 subjects examined right index finger tapping, shoulder rotation, or facial movement. The largest activation cluster in left sensorimotor cortex was identified for each task, and its center expressed in Talairach stereotaxic coordinates. Across subjects, each task showed considerable variability in activation site coordinates. For example, during finger tapping, the range for center of activation was 7 mm in the x-axis, 19 mm in the y-axis, and 11 mm in the z-axis. The mean value for center of activation was significantly different for all three coordinates for all pairwise task comparisons. However, the distribution of activation site centers for the finger task overlapped with the other two tasks in the x- and y-axes, and with the shoulder task in the z-axis. On average, the center of activation for the three motor tasks were spatially separated and somatotopically distributed. However, across the population, there was considerable overlap in the center of activation site, especially for finger and shoulder movements. Restorative therapies that aim to target specific body segments, such as the hand, in the post-stroke motor system may need to map the individual brain rather than rely on population averages. Initial details are presented of a study using this approach to evaluate such a therapy.


INTRODUCTION
Stroke remains the leading cause of adult disability 1 . The most common impairment after stroke, and a major contributor to disability, is weakness 1,2 . Acute thrombolytic therapy can reduce long-term disability after stroke [3][4][5] , however, few patients in the US reach a medical facility early enough to be eligible for such interventions 6 . Therefore, increased attention has been focused on improving neurological status with therapies administered in the subacute or chronic phase of stroke.
Animal studies have suggested a number of potentially useful therapeutic strategies for patients who have passed the acute phase of stroke. One such approach involves focal cortical stimulation, with the intent being to improve behaviors arising from the stimulated cortical region. Studies supportive of this approach are reviewed elsewhere in this journal.
Animal studies use direct cortical stimulation to target the cortical area in which the function of interest is localized. This is generally not possible in human stroke patients. In such a situation, one might hope to stimulate cortex with respect to anatomical landmarks. However, the human brain shows substantial variability in the localization of functions such as language and movement. Thus, across subjects, sites of functional organization do not correspond precisely to features of brain anatomy [7][8][9][10][11] . Moreover, even within subject, functional localization can change in relation to short-term and long-term experience, disease, age, and other variables 1 1-1 7 .
The rst purpose of the current report is to characterize, at the individual level, inter-subject variability in the localization of hand motor function, and to contrast these ndings with localization of shoulder and face motor function. Results support the need to map individual subjects when localizing motor function, rather than rely on a single set of coordinates. The second purpose of this report is to provide initial details of a study that utilizes this approach in the treatment of patients with chronic stroke.

Subjects
Fourteen healthy subjects were studied. Each was righthanded 1 8 , free of neurological disease, and gave informed consent. This study was performed at the University of Washington, where it was approved by the human subjects committee.

Data acquisition
Each subject underwent functional magnetic resonance imaging (fMRI) that alternated rest with performance of one of the motor tasks. Detailed methods have been published previously 19 . In sum, subjects rehearsed each task prior to scanning. Imaging was performed at 1.5 Tesla. Scanning employed a gradient echo echoplanar pulse sequence with T2*-weighting for blood oxygenation level dependent contrast. Each of the three scans lasted 3:20 and alternated 20 sec rest with 20 sec of motor task performance. The rst scan alternated rest with 2 Hz right index nger tapping; then rest versus 1 Hz right shoulder rotation; then rest versus 1 Hz contraction of the right corner of the mouth. Scanning parameters included TR 2000, TE 50, in-plane resolution 3.75£3.75 mm, 14 contiguous 7 mm axial brain slices, 100 images/slice, plus four TRs to establish steady state.

Data analysis
Images were motion corrected and linear detrended, after which a voxelwise t-test contrasted active with rest state for each of the three tasks, with results expressed as a Z-map. Results were spatially smoothed with a 4 mm Gaussian lter. Each Z-map was then converted to stereotaxic space 20 by registering to the standard image supplied with MEDx 3.3 software (Sensor Systems, Sterling, VA, USA) using FLIRT (www.fmrib.ox.ac.uk/ fsl/). The activation cluster with the largest number of activated voxels in the area composed of left precentral plus postcentral gyri was identi ed and isolated in its entirety at the threshold of Zˆ4.2 (approximately). The center of this cluster was identi ed, and its Talairach x, y, and z coordinates were noted. In the current report, absolute values are used for all coordinates.
A two-tailed, nonpaired t-test was used to compare coordinate values across tasks, without correction for multiple comparisons.

RESULTS
For the 14 subjects, mean age ( § SD) was 51 § 19 years. There were eight males and six females. Multiple factors such as head motion reduced the number of available scans to 10 for nger movement, ve for shoulder movement, and ve for facial movement. Figure 1 shows scatterplots for the x-axis (higher numbers mean more lateral), y-axis (more negative numbers mean more posterior), and z-axis (higher numbers mean more dorsal) for each of the tasks. ANOVA testing identi ed a signi cant difference in the center of activation across the three tasks for each of the three coordinates (p < 0.001 for each). Comparing each pair of tasks found signi cant (p < 0.05) differences for x, y, and z coordinates. Table 1 presents the mean values across subjects for the x, y, and z coordinate for the center of activation of each task. The decreasing z-axis values describe a somatotopic distribution going ventrally down the central sulcus, from shoulder to hand to face. Consistent with human central sulcus anatomy, this ventral progression is accompanied by increasingly lateral and anterior progression in the mean subject site of center of Figure 1: The left primary sensorimotor cortex activation site center of activation is presented for each subject and for each motor task performed during fMRI brain mapping. Activation center is presented in Talairach x, y, and z stereotaxic coordinates activation. The range is also presented in Table 1.
The y-axis had the highest range for each of the three tasks.

DISCUSSION
Human brain mapping has provided many insights into changes in brain organization after stroke 21 . The current discussion, however, pertains to extending the utility of human brain mapping beyond this. Several authors have suggested that measuring features of cortical organization may be useful in the application of therapies that target patients with chronic stroke 2 2-24 . For therapies that aim to stimulate cortical regions in which a speci c function is localized, brain mapping may be valuable for identifying the target of interest. The current results describe a considerable range in the normal site of activation for each task (Figure 1 and Table 1). Results are concordant with prior studies. Pen eld and Boldrey 7 , reporting in 126 operations on patients with conditions such as epilepsy or tumor, described a broad distribution in the area in which unipoloar stimulation induced nger movements, extending 55 mm along the central sulcus. Hlustick et al. 25 described the mean ( § SEM) right primary motor cortex center of activation during fMRI of left nger movements by 11 controls. Results were presented for xaxis (34) and for y-axis (¡13), the latter value re ecting analysis of only activated voxels anterior to the central sulcus. Lotze et al. 26 performed fMRI in 30 subjects during right or left hand movements. Apart from average group maps, the authors presented box plots showing linear distance from brain vertex to site of maximal activation. Indovina and Sanes 2 7 described the range of Talairach coordinates in left motor cortex during 2 Hz right nger movements. The range was 26-42 (x-axis), 12-24 (y-axis), and 49-65 (z-axis). These ranges are generally higher than the values found in the current report ( Table 1), likely due to the use of a more liberal threshold (p < 10 ¡2 , versus 10 ¡5 in Table 1) to de ne signi cant activation.
These contrasts raise several caveats for efforts to noninvasively localize motor function. First, the threshold used to de ne activation in uences variability detected. Second, results are in uenced by whether analysis considers the entire activation cluster, or only those voxels on a speci c gyrus. In 37 of 42 control subjects, 2 Hz index nger tapping activated a cluster that extended from precentral to postcentral gyrus 28 . The current study examined the center of the entire activation cluster rather than a subregion, and therefore likely included regions related to sensory processing. Third, the extent of variability identi ed may be in uenced by the sample size studied. Finally, some of the differences between the above studies may relate to use of different activation tasks.
The current results show overall concordance with prior brain mapping studies and describe inter-subject variability. This method may be useful as a noninvasive assessment of cortical function for directing a stimulation-based therapy. This approach is currently being employed in a study to evaluate epidural stimulation in 18 patients with chronic motor de cits after stroke. Epidural stimulation to the motor cortex, applied to relieve chronic pain, has been described as improving Activated voxels with z > 4 have been colorized and superimposed upon in-plane anatomical images. The white arrow indicates the cortical activation cluster that contains the most strongly activated voxel in right (contralateral, strokeaffected) primary sensorimotor cortex. B: The patient's matching anatomical image. The voxel of interest has been colored white and is indicated by the arrowhead. It is located directly in the right central sulcus. In both images, the asterisk indicates the area of infarction chronic motor de cits after stroke, as reviewed by Brown et al. in this issue. Consequently, an industrysponsored study has been organized to assess the safety and ef cacy of targeted subthreshold epidural cortical stimulation, in association with occupational therapy, in 18 patients with stroke-induced hemiparesis affecting primarily the upper extremity. The primary endpoint of this study is safety, de ned as the proportion of patients who have any of the following outcomes between the time of enrollment and the time of epidural stimulation electrodes removal, which is approximately 23-28 days later: 1. death, 2. major medical morbidity, including myocardial infarction, pneumonia, wound infection, or deep venous thrombosis, 3. a generalized tonic clonic seizure, or 4. decrement in neurological status, de ned as a decrease in 20% on either the Fugl-Meyer scale or the hand function subscore of the Stroke Impact Scale.
Entry criteria include age 20-75 and a history of paresis-inducing stroke >4 months prior. De cits must be moderate-severe but nondevastating, as speci ed by an arm motor Fugl-Meyer score of 20-50 (normalˆ66), and active wrist extension of at least 5. Twelve of 18 patients will be randomized to cortical stimulation, and 6 to no stimulation, with all patients receiving the same occupational therapy regimen.
In this study, neurosurgical placement of the epidural stimulator is guided by brain mapping. Patients undergo fMRI during performance of a motor task by the paretic hand. The choice of task is determined by the patient's motor status, with the most able patients mapped during 0.25 Hz index nger tapping, and those with poorest motor status mapped mapped during 0.25 Hz wrist movements. The most strongly activated voxel in contralateral (stroke-affected) sensorimotor cortex is used to direct placement of the epidural cortical stimulation device. An example of a patient study is shown in Figure 2. This approach considers inter-subject variability in localization of motor function by using features of brain function to guide this potentially restorative treatment.

CONCLUSION
Normal subjects show variability in the site of primary sensorimotor cortex activation. Animal studies suggest that simulating key motor areas might be a useful therapy for improving motor outcome after stroke. Mapping an individual subject's brain may therefore help to achieve best clinical effect from therapies that aim to stimulate a cortical region underlying a speci c function.