Voxel-based and ROI techniques are powerful methods to detect group differences in brain structure and function. However, the underlying assumption of both of these techniques is that the group differences are localized to the same brain region in at least the majority of patients. It is yet unknown if WM pathology occurs in the same location in each patient with schizophrenia. In fact, the currently DTI literature has considerable variability in the location of WM findings (Kanaan et al., 2005
; White et al., 2008
). If there are individual differences in the location of the WM abnormalities, both ROI and voxel based approaches may lack the power to identify these differences. In addition, it is possible that variability in the location of WM abnormalities may contribute to the clinical heterogeneity of schizophrenia.
To test for spatially heterogeneous abnormalities in FA, we developed an algorithm to detect contiguous voxels that fall below their voxel-based mean. This method is able to detect clusters between individuals that differ in brain location, but are localized within the WM. These we labeled WM potholes. We found that patients with EOS demonstrated a significantly greater number of WM potholes compared to matched controls. Since this is a novel approach, we tested the method across a number of different z-scores and volume thresholds (see ). The patient group had significantly greater number of ‘potholes’ at multiple thresholds and cluster sizes. These clusters are most prominent at approximately 2 standard deviations below the mean and have sizes that exceed 1,000 voxels (1 cc), which were found in the corpus callosum. The control group did not have a significantly greater number of ‘potholes’ than patients at any threshold or volume.
Interestingly, the spatial location of these potholes did emerge in regions that have been identified commonly in studies of schizophrenia. For example, numerous studies have identified aberrant WM regions in the corpus callosum, cingulate bundle, and frontal WM tracts (Kyriakopoulos et al., 2008
; White et al., 2008
). Utilizing the standard TBSS approach, we did not find any regions which were different between patients and controls. However, with a less conservative threshold for multiple testing, we found significant differences in the corpus callosum and frontal WM regions, regions which overlap with the pothole approach. Thus, it is possible that certain regions that have a convergence of WM pathways, such as the corpus callosum, are more likely have an overlap of WM abnormalities, which are then detected using voxel-based techniques.
Considering the variability of WM pathology in disorders such as multiple sclerosis (Gilmore et al., 2008
), it is possible that similar heterogeneous patterns may also exist in the WM pathology found in schizophrenia. While there may be involvement in multiple regions of the brain, certain areas, such as the cingulate bundle, corpus callosum, or frontal WM tracts, may be more susceptible to alterations. This would account for the overrepresentation of positive findings in these regions using voxel based and ROI studies. However, other individuals or subgroups may have different areas of susceptibility that are based on genetic or environmental influences. One possible etiology would involve disruptions in the distribution or number of oligodendrocytes. Postmortem studies of schizophrenia often focus on specific regions and thus do not provide a gross distribution of abnormalities. However, postmortem studies have identified abnormal numbers of oligodendrocytes in the prefrontal cortex (Uranova et al., 2004
; Uranova et al., 2007
) and cingulum (Stark et al., 2004
While it was not a goal of this paper to assess developmental differences in the number of potholes, we did find an inverse correlation between potholes and age; i.e., younger children tended to have a greater number of smaller potholes. Since studies using both postmortem and DTI methods have demonstrated an increase in FA that is associated with development, this technique may be beneficial to identify regions associated with neurodevelopmental processes such as myelination. Future work using this technique with larger populations of typically developing children will be necessary to confirm this finding. Finally, we found a significant positive correlation between the duration of illness and the right inferior fronto-occipital fasciculus. Since these EOS patients are in the early stages of their illness (mean duration of illness was 2.3 years), a greater range of illness duration or a longitudinal design would be best suited to evaluate the relationship between illness duration and the number of potholes. This is important since there is evidence that patients with chronic schizophrenia have regions of lower FA has compared to first-episode patients (Friedman et al., 2008
This is a preliminary investigation of a novel method to assess WM microstructure in clinical populations and we note that there are several limitations to the study. We have a relatively small sample size, although the detection of differences in a small sample actually reflects large effect sizes. It would be beneficial to apply the method to a larger dataset with both a first-episode and chronic patient groups, since a progression of WM abnormalities has been demonstrated between first-episode and chronic groups (Friedman et al., 2008
). Finally, we used the Johns Hopkins WM atlas to mask the region in which we searched for contiguous voxels falling below a set threshold. While we attempted to control for multiple tests, some of the potholes may be false positives. However, it would be expected that there would be an equal number of false positives present in both the EOS and control groups. Yet, we demonstrated a greater number of potholes in the patient group across different thresholds and cluster sizes. At any threshold or minimum cluster size, the control group never had a greater number of potholes compared to the patient group.
In summary, we describe a novel approach to detect WM abnormalities that may not be detected using ROI or voxel-based approaches. This algorithm was applied to a group of patients with EOS and the patient group had a significantly greater number of WM potholes compared to the control group. Variations of this algorithm could be used to assess for local minima within z-transformed FA maps rather than our approach of thresholding the WM maps. In addition, reductions of the variance may be obtained by regressing variables such as age into the model that have been shown to have an effect on FA.