Brain-behavior correlations in hemispatial neglect using CT and SPECT

Hemispatial neglect, a syndrome in which patients fail to orient or respond to contralesional space, is not completely understood. It has been elicited using various methods, and different explanations have been advanced,^1,2 ranging from sensory and attentional theories³ to the concept that neglect is due to abnormal internal representation of space.⁴ There are personal⁵ and extrapersonal forms of neglect, although the latter are more common. Extrapersonal neglect refers to awareness and interactions with "near" (within arm's length) and "far" extrapersonal space,⁵ and can be fractionated into sensory-attentional and motor-intentional systems⁶ that can differentially influence performance in neglect. The attentional system, responsible for sensory-input processing, mediates the ipsilesional spatial bias commonly seen in neglect patients. The intentional system, which is responsible for motor output but is believed to be functionally separable from the primary motor system, mediates overt actions in the desired spatial location. In our previous study that decoupled sensory and motor aspects of neglect, we suggested that a common internal spatial representation subserves both perception and action.⁷

Some investigators argue that both the parietal and frontal lobes interact and play complementary roles in attentional-intentional processing.⁸ A general theory that incorporates both systems relates neglect to dysfunction in an attentional-arousal system that is lateralized to the right hemisphere³ and arises from damage to an underlying anatomic network, including the frontal, parietal, and anterior cingulate cortices; basal ganglia; and thalamus. These regions have also been described as responsible for directed attention.^9,10 Each anatomic region is believed to contribute differentially to the attention-intention system, and damage in a specific region impairs its associated function.^9,10 Attentional neglect has been attributed to posterior damage(inferior parietal)³ and intentional neglect to anterior lesions (dorsolateral frontal).¹¹ However, motor neglect has also been shown after a discrete parietal lesion¹² and sensory neglect after frontal lobe damage.¹³

Hemispatial neglect could result from direct damage to anatomic nodes or to the white matter tracts connecting cortical-cortical and cortical-subcortical regions. Most previous localization studies of neglect¹⁴ were based on analysis of patients with structural damage on CT. Verification of neuroanatomic models can be better achieved, however, by combining structural (CT) and functional imaging such as SPECT in the same population. CT delineates direct physical damage to the cortex, subcortical nuclei, and white matter fiber tracts, whereas SPECT measures regional blood flow, which correlates with direct and remote effects of damage, that is, diaschisis¹⁵ or functional impairment of structurally intact regions at a remote distance from the primary lesion. In this study, lesion location and severity on CT and SPECT in right-hemisphere-damaged stroke patients were correlated with neurobehavioral tests of hemispatial neglect. Damage in the right-frontal, parietal, and anterior cingulate cortices; basal ganglia; and thalamus was hypothesized to be a predictor of left-hemispatial neglect, and all patients with neglect were predicted to have structural damage in at least one of the five key regions or their interconnections.

Methods. Population inclusion criteria. The study sample consisted of 120 patients selected from a prospectively studied stroke population admitted to the Acute Stroke Care Unit at Sunnybrook Health Science Centre. Study inclusion criteria specified that patients be right-handed, have at least 20/40 vision corrected, be able to undergo neglect testing, and have a single right-hemisphere, CT-confirmed lesion.

Sunnybrook Neglect Battery. All patients underwent assessment of neglect with the Sunnybrook Neglect Battery (SNB), which has been described elsewhere¹⁶ and comprises four subtests: spontaneous drawing and copying of a clock and daisy, line cancellation, line bisection, and shape cancellation.¹⁷ The battery was administered at a mean time of 12.6 days (standard deviation = 13.7 days) after stroke. Statistical tests were employed to examine the psychometric properties of the SNB.¹⁸ Briefly, all subtests significantly correlated with the total neglect score (r ∼ 0.8, p < 0.001) and with each other (r ∼ 0.6, p < 0.001), thus demonstrating internal consistency within the battery. Factor analysis showed that the subtests of the SNB were not redundant in that all four subtests contributed positively to a single factor(eigenvalue = 2.78), accounting for 69.4% of the variance. Last, the SNB was shown to have good external content validity when compared with another test of visuospatial neglect, the Visual Search Board task.¹⁹ Specifically, logistic regression of the four subtests against the Visual Search Board was highly significant (p < 0.001).

CT scan procedures. Stroke patients underwent CT scanning of the head on a General Electric 9800 scanner (GE Medical Systems, Milwaukee, WI), generally within 48 hours after stroke. When this study was conducted, a repeat scan was performed within 2 weeks for clinical purposes to document the extent of the lesion if no appropriate lesion was seen on the initial scan. The scan that maximally represented the lesion was used for localization analysis. The mean time after stroke for CT scan acquisition was 4.4 days (SD = 10.6 days). Scans were performed parallel to the orbitomeatal line and printed on radiographic film for further analysis. Lesions were also traced from film onto paper, the area corresponding to the lesion for each slice was digitized (Sigma Scan, Jandel Scientific, Corte Madera, CA), and the areas were summed to arrive at a lesion volume for that scan.

Anatomic localization was performed by reference to a stereotactic atlas.²⁰ To allow for differences in head size, the lesion on each of the transaxial slices was drawn on the best-matched template from the Talairach-Tournoux atlas,²⁰ and damage was recorded by region, Brodmann's area, and x-y-z coordinate. This technique allowed for dichotomous categorization, that is, a particular anatomic region was denoted as damaged or not, irrespective of its size or depth. To estimate regional damage severity, the number of slices on which a region of interest (ROI) appeared was first determined in a standard set of the atlas slices. The number of slices with visible damage to that region was divided by the number of slices on which that region appeared, and this ratio, which provided a measure of the vertical depth of the lesion, was used as an index of severity.

Twelve CT regions were used in analysis and included the anterior cingulate; the frontal, parietal, temporal, and lateral occipital cortices; the primary motor and sensory strips; and the basal ganglia and thalamus. The remaining three regions included white matter that was classified on transaxial slices of the Talairach-Tournoux atlas as anterior (matrix '4 4' to '6 7' transaxially, 'C' and 'D' coronally, 'b' sagittally), central (matrix '4 5' to '6 7,' 'E,' 'b'), and posterior (matrix'4 5' to '6 7,' 'F' and 'G,' 'b'). Further division into smaller lobular subregions (e.g., supramarginal gyrus, longitudinal fasciculi) within these regions was used in post-hoc analyses if the larger region showed significance.

SPECT scan procedures. Part way through this study SPECT became part of the clinical stroke assessment protocol, and scans were generally acquired within the first week of the stroke. Previously, SPECT scans were obtained in some stroke patients, often at longer intervals after stroke. Because neglect tends to diminish over time and can resolve quickly,²¹ different temporal inclusion criteria were allotted to patients with and without neglect. For patients without neglect, SNB administration and SPECT scanning had to be within 10 days of each other. A more liberal range (45 days) was allowed for patients with neglect to increase power²² for the localization analyses. The SNB was sometimes administered after the SPECT scan in patients with severe neglect who were often ill and unable to undergo testing for many weeks. If the patient had neglect at 45 days it was reasonable to infer that it was present initially. Eighty-eight patients met these criteria and were used in the SPECT and combined CT-SPECT analyses. In addition, SPECT scans were performed on 19 healthy volunteers as part of a research study.

After injection of 740 MBq of ^99MTc-HMPAO, on a GE 400 AT single-head gamma camera (GE Medical Systems, Milwaukee, WI), SPECT scans were acquired with patients in the supine position (imaging time, 30 minutes). Using step and shoot mode, 64 planar views were acquired over 360 degrees, with a 64 × 64 pixel acquisition frame per view (25 seconds per view) and a magnification factor of 1.33. After acquisition, each scan was reconstructed to correct for head tilt in the coronal and transaxial planes and to align each brain parallel to the orbitomeatal line in the transaxial plane, as delineated on the midsagittal slice. After reconstruction(ramp and Butterworth filter with a power factor of 15 and a cut-off frequency of 0.4 cm^-1, attenuation correction²³ μ = 0.12 cm^-1), a correction for HMPAO flow nonlinearity²⁴ was applied. Reconstructed image spatial resolution was approximately 1.2 cm full width at half maximum.

To improve the accuracy of anatomic localization across subjects, a linear-scaling technique was applied,²⁵ so that each brain was compressed or expanded into a predetermined volume with 12 slices. Specifically, all brains were scaled to be 54 pixels long and 48 pixels wide in maximum extent. SPECT-scan analysis involved semiquantitative measurement of cortical, subcortical, and medial cortical ROIs (figure 1). The cortical rim procedure adapted from Hellman et al.²⁶ involved finding the outer edge of the brain. An initial estimate of this edge was found using an arbitrarily defined threshold based on the median pixel value for the entire brain. In most cases, the edge obtained by this method was adequate. However, in scans with severe hypoperfusion, this edge was inappropriately placed by simple thresholding. To correct for this effect, an algorithm was developed to compare the left and right edges of the brain. In cases in which a large degree of asymmetry was detected, the edge on the abnormal side was replaced by the edge of the normal side, "mirrored" about the midline. Next, the algorithm moved in 11 pixels and divided the resulting rim into 24 equal annular segments, 12 per hemisphere, on six slices. An additional dorsal slice was divided into eight segments for a total of 152 segments calculated from the cortical rim procedure (seefigure 1a). The automated ROI procedure was used to measure counts in certain additional regions that could not be adequately captured by the cortical rim procedure. Preset ellipses were placed proportional to slice size over the basal ganglia, thalamus, anterior cingulate gyrus, and cerebellum (seefigure 1b). There were 18 ROI segments placed on each hemisphere for a total of 36 ROI segments. From both procedures, mean counts, SDs, and the number of pixels were obtained for each of the 94 segments per hemisphere.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Semiautomated semiquantitative SPECT procedures. Cortical rim procedure (a) and region-of-interest (ROI) procedure(b). AC = anterior cingulate; BG = basal ganglia; C = cerebellum; MO = medial occipital; PC = posterior cingulate; R= right hemisphere; TH = thalamus.

Because all SPECT scans were reconstructed parallel to the orbitomeatal line and linearly scaled, each scan was standardized such that when transaxial slices were viewed, each individual slice should have corresponded to similar brain anatomy across subjects. Using a stereotactic atlas²⁰ and a linearly scaled SPECT scan, it was possible to identify lobar anatomy on the scan for each of the 188 segments. To reduce the data, segments from similar anatomic regions were then grouped and averaged, correcting for segment size. In total, there were 10 averaged regions (eight cortical, two subcortical) for each hemisphere. In this report, only the right-hemisphere data were used, including the anterior cingulate, frontal, parietal, temporal, sensorimotor, occipital, parietal-temporal, and medial occipital regions; basal ganglia; and thalamus.

SPECT-scan standardization. Before SPECT-scan analysis, the data for each anatomic segment were standardized in a two-step procedure. Mean counts obtained in each segment were standardized by dividing by the mean counts in the cerebellum with the highest counts (i.e., ipsilesional cerebellum in 90% of patients). Each segment ratio (X_seg) was converted to a z score by subtracting from it a specified mean(x_norm) and dividing the result by the standard deviation of that mean (s_norm) using the following formula:Equation 1 The specified mean and SD values were calculated as the mean and SD of the homologous segment (same hemisphere) from the SPECT scans of the 19 healthy age-matched volunteers.

Multicollinearity in CT and SPECT data. Multiple linear regression (MLR) performs best when the independent variables are highly correlated with the outcome variable but have no or low intercorrelations among independent variables.²⁷ With highly correlated or multicollinear (r > 0.8) variables, MLR may not be able to find a reliable set of predictor variables, because highly correlated variables generally will not enter together into a regression equation. To investigate multicollinearity²⁸ between anatomic variables, Pearson bivariate correlation matrices and regression collinearity diagnostics were examined for the CT and SPECT regional variables.

Results. Group differences. Population demographics can be seen in table 1. There were no statistically significant differences for sex or education, but both lesion volume and age differed significantly; patients with neglect were older and had larger lesions. Thus, age and lesion volume were entered in all regression analyses as covariates. The CT data are graphically represented infigure 2. Six regions, the parietal cortex; anterior, central, and posterior white matter regions; and motor and sensory strips, had significantly more damage in the neglect group (seefigure 2). Overall, patients with neglect had lower mean count ratios; however, t-test analysis of the 10 SPECT regions showed that after correcting for multiple comparisons with p < 0.005 (i.e., Bonferroni correction for 10 ipsilateral regions for α = 0.05), the only significant difference was in the basal ganglia (t = 3.2, p < 0.001).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Mean percent of all possible slices damaged for 12 regions. AC = anterior cingulate; AWM = anterior white matter; BG = basal ganglia; CWM = central white matter; F = frontal; M = motor strip; O = occipital; P = parietal; PWM = posterior white matter; S = sensory strip; T = temporal; TH = thalamus. Error bars refer to SEM. *Significant on t-test between neglect groups (p < 0.004).

CT visual analysis. To address the hypothesis that all patients with neglect will have damage in at least one of the predicted anatomic regions, lesion localization data from the CT scans were further scrutinized(table 2). In the group with neglect, 69 (84%) of 82 patients had damage to at least one of the key theoretical regions. By comparison, 28 (74%) of 38 patients without neglect also had damage to at least one key region, and the difference between the groups was not statistically significant (X₍₁₎² = 1.84). However, the group of patients with neglect tended to have damage that included more than one of the key predicted regions. Specifically, in the group with neglect, 50(61%) of 82 patients had damage to two or more key theoretical regions compared with only 11 (29%) of 38 patients in the group without neglect, a significant finding (X₍₁₎² = 10.7, p < 0.002). Patients with neglect also had more overall damage to the predicted regions (12.6 ± 12.4% average damage to the five regions combined) than patients without neglect (6.6 ± 8.1% average damage, t = -3.3, p < 0.002).

Investigation of the 10 patients with neglect who did not have direct damage to the five key regions showed that all had damage to the association fibers in white matter fiber bundles that connect these key regions, either in the anterior or posterior white matter regions or both. The association fiber bundles affected most often in this population included the inferior(FLi, n = 6) and superior longitudinal fasciculi (n = 6 anteriorly and n = 5 posteriorly). Of the eleven patients without neglect in whom lesions also did not involve the five predicted regions, four patients also had damage to at least one of these association white matter fiber bundles (FLi, n = 3; pFLS, n = 2).

CT linear regression analysis. Before regression analyses, the issue of multicollinearity was addressed with respect to the CT data. Examination of Pearson bivariate correlations and collinearity diagnostics²⁸ produced with the regression analyses did not show any evidence of significant multicollinearity between CT regions. Two approaches were used used in building regression equations. The primary approach was hypothesis driven and included only the hypothesized regions. All five key regions (frontal, parietal, and anterior cingulate; basal ganglia; and thalamus) were forced in a regression model using neglect score as the dependent variable and the CT data as independent variables, along with age and lesion volume as covariates. Log transformation was performed on all neglect scores to reduce skewness, and arc sine transformation was appropriately applied²⁷ to the CT data before regression analysis. Statistical significance was set at p < 0.05. The second approach was exploratory and endeavored to determine if any additional brain regions would enter into a model predicting neglect score. Twelve anatomic regions, detailed earlier, were entered in a stepwise linear regression. Statistical significance was set at p< 0.004 for each region in the exploratory approach (Bonferroni correction for 12 regions for α = 0.05).

For the hypothesis-driven regression analysis, the model with all five predicted regions produced a significant model (F_(7,112) = 6.5, p < 0.001, R² = 0.29). The anterior cingulate and parietal regions were significant predictors (p < 0.05). Exploration of the parietal subdivisions found that the supramarginal gyrus was the most significant parietal subregion (p < 0.05). This was supported by the finding that 13% of patients with neglect had damage to the supramarginal gyrus, whereas none of the patients without neglect had damage to this region.

In the exploratory stepwise regression of all 12 regions, no variable entered into the regression equation after volume and age, which together accounted for 20% of the variance (F_(2,117) = 14.7, p< 0.001, R² = 0.20). Although no region significantly entered the equation, the posterior white matter region (p = 0.0068) and the lateral occipital regions (p = 0.0095) showed trends toward significance.

SPECT linear regression analysis. Pearson bivariate correlations and regression collinearity diagnostics were used to investigate multicollinearity in the SPECT data. Most correlations were positive, and the average correlation was 0.6 with the largest correlation (r = 0.9) between the temporal and parietotemporal regions. Examination of the regression collinearity diagnostics further confirmed the collinearity of the SPECT data. It was therefore inappropriate to use MLR in an exploratory analysis of the SPECT data. Instead, MLR was used only with the SPECT data to see if dysfunction in the right-frontal, parietal, and anterior cingulate; basal ganglia; and thalamus would predict neglect score. All five regions were forced, along with the covariates age and CT-lesion volume, into a regression model to predict score on the SNB. Linear regression showed that the five regions entered into a significant model (F_(7,80) = 5.8, p< 0.001, R² = 0.34), but only the right-parietal region showed a trend to significance (p = 0.08).

CT-SPECT linear regression analysis. To explore the effect of adding functional information, an additional regression was performed using both the CT and SPECT regions. First, all five predicted CT regions were forced, along with the covariates age and CT lesion volume, into a regression equation. Then stepwise regression was used to see if any additional SPECT variables entered (α = 0.05). The right parietal from SPECT entered and the resultant equation accounted for 40% of the variance (F_(7,80) = 5.5, p < 0.001), an 18% increase of the variance explained in the previous model.

Discussion. The structural and functional imaging results from this study support a role for the right-parietal and anterior cingulate cortical regions in hemispatial neglect. Greater structural damage on CT in these regions correlated with poorer performance on the SNB. When the imaging data were combined, the only significant functional predictor of SNB score was decreased perfusion in the right-parietal lobe on SPECT.

Although white matter fiber bundles have been implicated previously in hemispatial neglect,³ prior studies have based these findings either on small sample sizes or on CT-lesion overlays that did not specify the white matter regions involved.¹⁴ The current study provides strong empiric evidence that damage to posterior white matter fiber bundles, specifically involving the inferior and superior longitudinal fasciculi, is associated with neglect. This finding is congruent with a study¹⁴ of 110 right-hemisphere-damaged patients that implied that the white matter region beneath the temporal-parietal-occipital (TPO) junction on CT lesion overlaps of the lateral brain surface was commonly involved in patients with neglect. The posterior white matter fiber bundles mentioned above pass through this TPO white matter junction.

The fiber bundles deep to the TPO cortices are at a critical junction interconnecting the posterior lobar regions locally²⁹ and anterior-posteriorly.³⁰ Damage to this area has been shown to affect both nearby areas, such as the parietal lobe, and distant areas, such as the frontal lobe. In humans, the TPO junction includes connections with the visual, tactile, and auditory unimodal sensory association areas and is considered to be a polymodal sensory region.⁹ That more patients with neglect had damage to the white matter beneath the TPO region in our study compared with those without neglect (39% versus 20% of patients, p < 0.05) provides empiric support for the importance of the TPO region in neglect. Further, the combination of damage to the cortex and white matter fiber bundles may lead to a more severe and lasting neglect.¹⁸ In such cases, dysfunction may result not only from damage to the key cortical region but also be an effect of information not being forwarded as a result of damage to the connecting white matter fiber tracts. Damage in the posterior white matter region and lateral occipital showed a trend toward significance in the exploratory regression analyses of the CT data. This was further supported by the finding that patients with neglect were more likely to have two or more key anatomic regions damaged than patients without neglect.

Because patients with and without neglect had damage to at least one predicted region in approximately equal percentages, the results cannot be used as support for the prediction that at least one region in the neuroanatomic network for directed attention should be differentially affected in neglect. Nevertheless, the correlation of larger lesion size with neglect is consistent with the literature.³¹ Patients with neglect sustained a larger volume of damage overall compared with patients without neglect, and more key regions were damaged in patients with neglect. Specifically, 62% of patients with neglect had lesions involving two or more of the predicted key regions compared with only 29% of patients without neglect. Whether neglect occurs due to a larger volume of damage to the right hemisphere or as a result of damage to multiple regions in an underlying network subserving directed attention cannot be inferred from these descriptive results. To further understand the neuropathology of hemispatial neglect, evidence from the regression analyses was deployed because the influence of lesion size could then be partialled out.

Before the results from the regression analyses are discussed, it is important to mention two limitations that may have prevented statistically significant differences from emerging for some regions. Study size and low power limited the exploratory regression analyses of the CT data. Post-hoc power calculations for the analyses yielded power equal to 0.40; thus negative findings should not be emphasized too strongly. Although there were 120 patients in the study, making it one of the largest group studies ever conducted in neglect, traditional inferential statistics like linear regression are limited when there are many independent variables, such as in brain imaging data (recall that the Bonferroni correction of α = 0.05 for 12 regions became p < 0.004). Second, analysis of the SPECT data was limited due to the limited spatial resolution of the single-head SPECT camera and the collinearity of functional brain regions. In regression analysis, multicollinearity of the independent variables may prevent inclusion of additional significant regions.²⁷ This may be why the parietal region was the only functional brain region to reach significance. A statistical technique that is able to benefit from multicollinearity and is not hampered by study size, such as partial least squares,³² might be a more appropriate application for exploring the SPECT data.³³

The right-parietal cortex emerged in the regression analyses as significantly related to performance on the SNB, which conforms with the clinical literature. The parietal cortex has long been associated with hemispatial neglect since the earliest clinicopathologic studies.³⁴ In PET studies, the right-parietal lobe has been demonstrated to activate in relation to both left- and right-sided stimuli, although more so for contralateral stimuli.³⁵ In our study, patients with neglect had significantly more structural damage to the parietal lobe on CT compared with patients without neglect. In a post-hoc MLR analysis of the CT data, the supramarginal gyrus emerged as the most significant parietal subregion. This region within the inferior parietal lobe may be the human homologue of dorsolateral area PG in the macaque monkey.⁹

It seems likely from our data that the parietal region is a crucial region within the theoretical anatomic network for directed attention. However, it could be argued that the subtests from the battery used in this study preferentially tested extrapersonal visuospatial sensory-attentional neglect, and thus would be more sensitive to damage in the parietal lobe. Although we recognize this as a possible confound, note that 31% (25/82) of patients with neglect had damage restricted to the frontal cortex, basal ganglia, or thalamus, areas believed to be associated with the motor-intentional system, and only 38% (31/82) of neglect patients had structural damage that included the parietal cortex. Although investigators have reported methods to dissociate the two systems,^36,37 the techniques often require specific apparatus that are difficult and impractical to use in a clinical setting (i.e., at the bedside). This study sought to assess neglect at the bedside and not to dissociate these two systems; in fact, we suspect that the SNB probes both attentional and intentional neglect, and the frequency of anterior and posterior damage in our neglect population would tend to support this.

Although the parietal lobe has been recognized as an important neural component associated with neglect for over half a century, the evidence from single-case reports that neglect occurred with damage elsewhere and the theoretical network proposed in the 1980s^3,9,10 could be taken to suggest that all nodes are equally important. Our data reaffirm the primacy of parietal damage and are convergent with the only other large group localization study of neglect.¹⁴ Our results can therefore be taken as strong support for involvement of the right-parietal lobe, specifically the inferior parietal, in extrapersonal visuospatial neglect.

The other region that emerged was the right-anterior cingulate cortex. This region was anticipated to emerge as it has been shown to be activated during attention tasks on PET.³⁸ Although it was less frequently damaged, this region entered in the CT regression analysis as a significant predictor of SNB score. Whereas this region did not emerge significantly in any of the SPECT analyses, the anterior cingulate showed a trend toward significance in the combined CT-SPECT regression analyses. Thus, the data also support the involvement of the anterior cingulate in hemispatial neglect. In summary, in our combined multivariate and univariate analyses of the CT and SPECT data in this large consecutive right-hemisphere-damaged stroke population, the parietal, anterior cingulate and posterior white matter fiber tracts emerged as critical lesion sites for hemispatial neglect.

Although the frontal lobe, basal ganglia, and thalamus did not emerge from these analyses, this does not necessarily mean that they are not important areas involved in hemispatial neglect. The data for this study were derived from patients with brain damage whose functional network for directed attention was disrupted. That a region did not emerge in our analyses cannot be used as definitive evidence that those regions are not involved in a normal functioning brain. This can be more appropriately investigated by activation studies in healthy subjects.³⁵ Particularly relevant are recent studies providing supporting evidence for the postulated cortical network for directed attention in the normal-behaving adult human.^39,40 Using functional MRI in healthy subjects, the frontal, parietal, and cingulate cortices were activated in tasks requiring directed attention and spatial orientation. The results from the current study are based on lesion localization in patients with hemispatial neglect and inferences should only be drawn about regions that appear to be critical for the disruption of normal function.

Acknowledgments

The authors thank Drs. A.R. McIntosh, J.R. Wherrett, M-L Smith, G. Winocur, M-P McAndrews, and F.S. Prato for their comments on the MSc thesis on which this work was based. We also thank K. Barbour for helping with the CT data; K. Ma for helping with the SPECT data; N. Blair, J. Bondar, and D. Martin for collecting the behavioral data; our neurologic colleagues Drs. J.G. Edmeads and A. McLean for patient referrals; and the technologists of nuclear medicine for their assistance.