Figure presents the lesion-density plots for neglect and control patients. The average lesion volume (Table ) was about three times greater in neglect than control patients [t(48) = 3.78, p < .001]. The region that was most frequently damaged in neglect patients involved the insula and underlying white matter, the posterior temporal and inferior parietal cortex (Figure ). However, as can be seen in Figure these regions were also frequently damaged in control patients. In order to identify regions that were more frequently damaged in neglect patients compared to the control group, we subtracted the superimposed lesions of the control group from the lesions of the neglect group (Figure ). This subtraction identified a region reaching from the posterior insular cortex into the white matter of the frontal lobe as predicting best the incidence of spatial neglect. However, the analysis shown in Figure was not only based on patients with cerebral ischemia, but also included a significant number of patients with haemorrhagic brain damage. Though the inclusion of patients with different vascular aetiologies renders our results comparable to previous studies, the fact that cerebral haemorrhage tends to be subcortical, does not always respect vascular territories and is difficult to delineate due to oedema surrounding the lesion might have biased the lesion overlap in favour of subcortical regions. We therefore performed the same analysis, but included only patients with ischemic brain damage (Figure and ). The subtraction map based on this comparison (Figure ) showed, that in comparison to patients without neglect brain damage in neglect patients was more frequent in a region surrounding the TPJ and involving the IPL, the posterior STG, and the insula (damage at least 40% more frequent in neglect compared to non-neglect patients). Note also that the region differentiating best between neglect and control patients was not confined to cerebral cortex, but reached far into the white matter beneath the TPJ and STG.
Figure 1 Lesion analysis of 28 patients with spatial neglect and 22 control patients without spatial neglect. The figure shows transverse sections of the MNI template brain with Talairach z-coordinates increasing from right (z = -16) to left (z = 48). A) Overlap (more ...)
Figure 2 Lesion analysis of 19 patients with spatial neglect and 17 control patients with ischemic brain damage. Same analysis as shown in Figure 2, but confined to patients with ischemic damage. A) Overlap map of neglect patients. B) Overlap map of control patients (more ...)
The advantage of the subtraction technique is that only regions for which frequency of involvement is different in the neglect compared to the control group are highlighted. However, the technique does not provide a statistical test of frequency of lesion involvement. In order to determine the frequency of STG and IPL damage in neglect and non-neglect patients, we determined the number of patients in each group that had damage to at least a portion of the STG or IPL. In order to demarcate the STG and IPL we used the criteria defined by Mort et al. [14
]. 20 patients with spatial neglect (71.4%) and 9 control patients (40.9%) had damage to the STG while 16 neglect patients (57.1%) and only 2 control patients (9.1%) had damage to the IPL. The incidence of STG [χ2
= 4.71, p < .05] and IPL damage [Fisher exact test, p < .001] was significantly higher in neglect compared to control patients. However, a limitation of this approach is that the STG has a greater anterior-posterior extension and is located more centrally within the vascular territory of the middle cerebral artery (MCA) compared to the IPL. Therefore, neglect patients may have damage to the STG simply because there is a high probability that MCA infarcts damage at least some portion of the STG. This hypothesis may be tested by computing the sensitivity (i.e. the probability that spatial neglect occurs as a consequence of damage to the region) and specificity (i.e. the probability that no neglect occurs when the region is spared) of STG and IPL damage as predictors of the occurrence of spatial neglect (Table ). Sensitivity was slightly, but not significantly higher for the STG compared to the IPL [χ2
= 1.24]. In contrast, specificity was significantly lower for the STG than the IPL [Fisher exact test, p < .02]. When patients with haemorrhages were excluded, sensitivity and specificity slightly increased (Table ), but there was still no statistical difference between the sensitivities of STG and IPL damage to neglect while IPL damage was highly predictive of spatial neglect in patients with infarction, and more so than STG damage [Fisher exact test, p < .05].
Measures of sensitivity and specificity
We further evaluated the incidence of damage to specific cortical and subcortical areas with a voxel-wise analysis. In this analysis, a χ2
-test is performed for every 'damaged' voxel testing the hypothesis whether it is involved significantly more often in the neglect group compared to the control group. In order to maximize the power of this analysis, only voxels that were damaged in at least 10 patients were examined, which in our sample resulted in 21 517 voxels tested. In view of the high number of statistical tests performed, we accepted only voxels for which the test reached a significance level of p < 0.01. This criterion was preferred to a Bonferroni-correction, which tends toe be over-conservative for very large numbers of comparisons [25
]. Figure displays only those voxels that differentiated with a χ2
of at least 6.63 (p < .01) between neglect and control patients. All voxels shown in the figure were damaged more often in the neglect than the control group. There were essentially two clusters of voxels for which the test was statistically significant, an anterior and a posterior cluster. The anterior cluster reached from the inferior to the superior insular cortex and further into the white matter of the frontal lobe anterior to the horn of the lateral ventricle. Within this cluster a group of voxels differentiating best between both groups [χ2
= 13.99, p < .0002] was situated slightly anterior to the head of the caudate nucleus (Talairach-coordinates: 21,22,14). The posterior cluster was located at the TPJ and reached into the supramarginal gyrus and the inferior postcentral gyrus. The STG was also involved, but only with its most posterior part neighbouring at the TPJ. Within the posterior cluster, a group of voxels in the white matter beneath the TPJ (Talairach-coordinates: 36,-45,24) differentiated best between neglect and control patients [χ2
= 12.41, p < .0005].
Figure 3 Voxel-wise lesion comparisons. A) Voxel-wise analysis evaluating anatomical differences between the neglect group and the control group (all patients included). B) Voxel-wise analysis including only patients with damage due to infarction. The colour codes (more ...)
A voxel-wise comparison including only patients with infarction showed a similar result (Figure ), the notable difference to the previous analysis being that the anterior cluster reached less far into the frontal white matter. However, note that because of the smaller sample size this analysis had less statistical power than the previous comparison.
Thus, our comparison between neglect and non-neglect patients revealed an anterior and a posterior cluster differentiating best between these two groups. Both, the IPL and the STG were damaged significantly more often in neglect than in control patients, although only the posterior part of the STG was involved.
Previous studies have suggested that patients with a strong bias on cancellation tasks may anatomically dissociate from patients with a large line bisection error. While frontal or subcortical injury has been reported in the former [19
], the latter are more likely to exhibit posterior brain lesions [26
]. These differences might explain some conflicting results regarding neglect anatomy reported in previous studies. In order to test this hypothesis, we identified in our neglect group 6 patients with a number of left omissions in the 'Bells test' that was higher than the average of the whole neglect group, but who showed bias on line bisection that was smaller than the average (group 'cancellation'). We compared these patients to 6 patients whose number of cancellation omissions was below average, but who had large biases on line bisection (group 'bisection'). The mean ipsilesional bisection bias and the number of left omissions in the cancellation test of these two groups are shown in Figure . The 'cancellation' group made significantly more left omissions in the 'Bells test' [t(10) = 7.88, p < .0001], whereas the 'bisection' group had a significantly larger ipsilesional bias on line bisection [t(10) = 3.38, p < .01].
Figure 4 Performance in cancellation and line bisection of subgroups of patients with neglect. The figure shows average performance of two groups of 6 neglect patients on the cancellation (number of omissions on the left of the sheet, max. = 15) and line bisection (more ...)
Figure and shows the lesion-density plots for the 'cancellation' and 'bisection' groups, respectively. In the 'cancellation' group five of six patients shared a common lesion, located in the white matter of the frontal lobe, anterior to the caudate nucleus. In the 'bisection' group, five of six patients shared a lesion centred on the insular cortex, comprising the white matter beneath the STG and the paraventricular white matter of the parietal lobe. The subtraction of lesions of the 'bisection' group from the 'cancellation' group (Figure ) confirmed that the 'cancellation' group had significantly more frequent damage in the white matter anterior to the caudate nucleus [red arrow; Fishers test: p < .05]. In contrast, damage of the insular cortex and the white matter beneath the TPJ and the posterior STG was more frequent in the 'bisection' than the 'cancellation' group [blue arrow; Fishers test: p < .05].
Figure 5 Anatomical bases of cancellation and line bisection performance. Lesion analysis of 6 patients with spatial neglect showing impaired cancellation, but relatively spared line bisection (A) and 6 patients with neglect showing the reverse pattern (B). The (more ...)