Influence of analytic techniques on comparing DTI-derived measurements in early stage Parkinson's disease

Abstract

Diffusion tensor imaging (DTI) studies in early Parkinson's disease (PD) to understand pathologic changes in white matter (WM) organization are variable in their findings. Evaluation of different analytic techniques frequently employed to understand the DTI-derived change in WM organization in a multisite, well-characterized, early stage PD cohort should aid the identification of the most robust analytic techniques to be used to investigate WM pathology in this disease, an important unmet need in the field. Thus, region of interest (ROI)-based analysis, voxel-based morphometry (VBM) analysis with varying spatial smoothing, and the two most widely used skeletonwise approaches (tract-based spatial statistics, TBSS, and tensor-based registration, DTI-TK) were evaluated in a DTI dataset of early PD and Healthy Controls (HC) from the Parkinson's Progression Markers Initiative (PPMI) cohort. Statistical tests on the DTI-derived metrics were conducted using a nonparametric approach from this cohort of early PD, after rigorously controlling for motion and signal artifacts during DTI scan which are frequent confounds in this disease population. Both TBSS and DTI-TK revealed a significantly negative correlation of fractional anisotropy (FA) with disease duration. However, only DTI-TK revealed radial diffusivity (RD) to be driving this FA correlation with disease duration. HC had a significantly positive correlation of MD with cumulative DaT score in the right middle-frontal cortex after a minimum smoothing level (at least 13mm) was attained. The present study found that scalar DTI-derived measures such as FA, MD, and RD should be used as imaging biomarkers with caution in early PD as the conclusions derived from them are heavily dependent on the choice of the analysis used. This study further demonstrated DTI-TK may be used to understand changes in DTI-derived measures with disease progression as it was found to be more accurate than TBSS. In addition, no singular region was identified that could explain both disease duration and severity in early PD. The results of this study should help standardize the utilization of DTI-derived measures in PD in an effort to improve comparability across studies and time, and to minimize variability in reported results due to variation in techniques.

Keywords: Neuroscience.

Fig. 1

Skeletonwise results of WM organization in PD. The location of the cluster showing a significantly (p_corr<0.05) negative relationship between FA and disease duration using both TBSS (a) and DTI-TK (b). The top and bottom panel of (b) shows the location of the cluster showing a significantly (p_corr<0.05) negative and positive relationship between FA (top panel) and RD (bottom panel), and disease duration using DTI-TK. R and L represent the right and left hemispheres respectively. Color bar represents the range of p-values in the overlaid cluster.

Fig. 2

VBM-based results of the effect of smoothing before statistical analysis. (a) Top panel: Location of the cluster, involving right middle frontal gyrus, where MD in HC was significantly (p_corr<0.05) correlated with DaT score. Bottom panel: Scatterplot of the extent of the cluster and p-values as a function of spatial smoothing (left panel), along with scatterplot of the extent of the cluster and effect size as a function of spatial smoothing (right panel) is shown. R and L represent the right and left hemispheres respectively.