White matter normalization – a novel analysis method with improved sensitivity can detect large decreases in cortical CBF in early Parkinson’s disease.
Borghammer P1, Ostergaard K2, Jonsdottir KY3, Gjedde A1,3.
1PET centre, Aarhus University Hospital, Denmark, 2Department of Neurology, Aarhus University Hospital, Denmark, 3Center for Functionally Integrative Neuroscience, Aarhus University Hospital, Denmark
Background: Several PET studies in Parkinson’s disease (PD) have found cortical decreases in cerebral blood flow (CBF) and glucose consumption, and increases in lentiform nucleus, thalamus, and several white matter (WM) areas including cerebellum, brainstem, and primary sensory-motor area. The data analysis of these studies is complicated by the very large inter-subject variability. The standard solution to this problem is to divide each voxel in a subject scan with the global mean (GM). It is an often overlooked fact, that subtle differences in global mean between the groups introduce bias into the analysis (i.e. if GM is decreased in one group, the normalization will inflate the voxel values in this group), even if such a global difference has not been detected by the standard comparison of global CBF in the two groups. We suggest, that the widespread practice of normalizing data to GM is flawed, in that it creates artificial hyperactivity in WM areas. We investigated the alternative of normalizing data to mean of central white matter (WM), assuming that WM is more stable and less likely to be systematically decreased in one group. This should theoretically improve analysis sensitivity and diminish the risk of creating artifacts.
Methods: Nine early-stage PD patients (H&Y I-II) and fifteen healthy age-matched controls underwent CBF PET scans while resting. We employed both region-of-interest (ROI) and voxel-based statistics. The analysis was carried out on non-normalized data and after normalization to GM and WM mean, respectively. The findings of these three analyses were compared.
Results: ROI: No significant differences were found in absolute values of CBF in whole brain or any of the four major lobes. After normalization to GM, PD patients had decreased CBF in parieto-occipital lobes, and in whole grey matter (all p<0.05). After WM normalization, we found significant differences in the same areas and, additionally, we detected a decrease in the frontal lobe (all p<0.05).
Voxel-based: GM normalization yielded only one single voxel-cluster of significant decrease in the medial occipital lobe and a cluster of significant increase in left primary motor area (mostly white matter). WM normalized data disclosed three much larger significant clusters of decreased CBF in occipital lobe and lateral parietal lobe bilaterally (p<0.05 for all clusters), but no increases. Voxel-based results are exhibited in figure 1.
Conclusions: WM normalization seems to be a considerably more sensitive method than common GM normalization. Using this method, we were able to demonstrate quite large decreases in cortical perfusion in a small sample of early PD patients. The GM/WM ratio was significantly decreased in early PD, indeed suggesting a decrease in global CBF. This has never been reported before.
Furthermore, WM normalization does not produce areas of debatable increases (e.g. putamen and white matter areas in cerebellum and primary sensory-motor cortex), which is consistently seen in GM normalized data. There is no basis for these increases in the PD animal model literature. Indeed, there is plentiful evidence that the putamen exhibits decreased, not increased, metabolism.
We conclude that most group comparison studies in PD have probably used an inadequate normalization procedure that blunts the analysis sensitivity and creates artifacts. Care should be taken to identify a reference region in which there is no evidence of systematic differences between the groups. Central white matter seems an attractive candidate.
Figure 1: Clusters of significantly different CBF voxels in the voxel-based analysis superimposed on an MRI atlas in Talairach space. Blue and red colors designate clusters where PD patients had decreased / increased CBF compared to healthy controls. The small inset images show coronal sections of the largest significant clusters. A. Analysis of data originally normalized to whole brain. A single cluster of decreased CBF in the V1 was found. A cluster of increased CBF was found in frontal lobe white matter expanding slightly into the SMA and anterior cingulate. B. When normalizing the data to white matter, we found several areas of decreased CBF in PD in V1, inferior occipital lobe bilaterally, lateral parieto-occipital cortices bilaterally, and right putamen / insula. There were no longer any significant clusters of increased CBF.