Introduction

Mild Cognitive Impairment (MCI) affects approximately 30% of individuals with Parkinson’s disease¹. However, no reliable imaging methodology currently exists to identify PD-MCI in PD. We hypothesized that multivariate analysis of sophisticated diffusion MRI (dMRI) voxelwise measures, conventional volumetric measures, and dMRI-derived graph-theoretical network measures will provide a set of imaging measures that are both sensitive and specific to PD-MCI.

Methods

Participants: We recruited 37 PD participants and conducted a comprehensive neuropsychological evaluation^1,2 to identify PD-MCI and PD-non-MCI (PD-nMCI) participants. For diagnostic accuracy, PD-MCI was classified after applying a threshold of 1.5 standard deviations^1,2 below appropriate norms on at least two neuropsychological tests following Movement Disorders Society (MDS) criteria¹. Based on clinical presentation and neuropsychological evaluations, a consensus diagnosis of PD-MCI was made by a practicing neurologist and a trained neuropsychologist. All participants were scanned with the following parameters on a 3T Siemens Skyra MRI scanner: 3D T1-weighted MRI acquisition: Isotropic spatial resolution=1mm³, inversion time (TI)=900ms, repetition time (TR)=2300ms, echo time (TE)=2.96ms. dMRI acquisition: Number of b-values=3, b-values=500s/mm², 1000s/mm², 2500s/mm² (multi-shell dMRI data was acquired in the same run), number of diffusion-encoding directions at each shell=71, number of non-diffusion weighted images (b0)=25, isotropic spatial resolution=1.5mm³, TR=5218ms, TE=100ms, multiband factor(MB)=3, acceleration factor (GRAPPA)=2, phase encoding direction= P>>A. We also acquired a b0 image with the same parameters but with opposite phase encoding direction (A>>P) for eddy-current distortion correction³. Total acquisition time=28 minutes. Pre-processing: Eddy-current distortion correction was performed using eddy tool in FSL 6.0, and translational head motion during the scan was computed for each participant. Volumetric analysis: FreeSurfer 6.0⁴ was utilized to generate cortical/subcortical volume, cortical thickness, and curvature measures for each participant. dMRI analysis: We computed single-tensor (ST) measures (fractional anisotropy (FA), radial diffusivity (RD), axial diffusivity (AxD), mean diffusivity (MD)), utilizing the lower shell dMRI data (b=500s/mm², 1000s/mm²) using dtifit toolbox in FSL 6. Additionally, in-house estimation of whole-brain free-water fraction (f_iso) was performed for each participant utilizing the lower shell dMRI data (b=500s/mm², 1000s/mm²), and f_iso-ST measures (f_iso-FA, f_iso-MD, f_iso-AxD, f_iso-RD) were estimated for all participants. Diffusion kurtosis imaging⁵ (DKI) measures (axonal kurtosis (AK), extra axonal diffusivity (EAD), radial kurtosis (RK), axonal water fraction (AWF), mean kurtosis (MK)), and neurite orientation dispersion and density imaging (NODDI)⁶ measures (KAPPA, orientation dispersion index (ODI), intracellular volume fraction (FICVF), isotropic volume fraction (FISO)) were also estimated for all participants using dMRI at all three-shells using appropriate toolboxes. All dMRI-derived measures for a representative participant are shown in Fig.2A. Network construction: AAL atlas⁷ was used to generate 90 nodes of the network in each subject’s native diffusion space after appropriate transformation⁸. Whole-brain tractography was performed using diffusion toolkit (http://www.trackvis.org/dtk/)⁹. Fibers smaller than 10mm¹⁰ or having FA<0.2 were removed from any further analysis. Each internode connection (edge) was weighted by the product of the number of fibres and average FA of the fibres connecting the two nodes. Graph-theoretical measures: Various global and local graph-theoretical measures were computed using GRETNA¹¹ for each participant. Statistical analysis: Nonparametric comparisons between the means of each dMRI measure/volumetric measure/cortical thickness measure/graph-theoretical measure between groups, and also correlations of these measures with clinical measures were performed using permutation analysis of linear models (PALM)¹² in FSL. In addition, network-based statistic (NBS)¹³ was used to statistically quantify differences in the weighted structural connectivity pattern between the groups. Significance was established at p_corr<0.05.

Results

22 PD-MCI and 15 PD-nMCI participants were identified in this study with no difference in demographics, global cognition, and head motion during the scan (Table.1). Significantly higher curvature in left-hemispheric precentral gyrus and lower cortical thickness in right-hemispheric insula was observed in PD-MCI (Fig.1). Furthermore, corpus-callosum (CC) volume was significantly associated with the dementia rating scale (DRS)-Memory score in PD-nMCI (Fig.1). All dMRI models suggested WM disorganization in CC and cingulate fibers in PD-MCI (Fig.2B). Similarly, the correlation between DRS-Conceptualization and almost all dMRI-derived measures were observed in both groups (Fig.3). AK, FICVF, and ODI showed a significantly different correlation between groups with DRS-Conceptualization (Fig.3). NBS revealed lower structural connectivity in PD-MCI encompassing posterior cingulate, cuneus, and Heschl gyrus (Fig.4A). Global network measures showed a significantly lower global hierarchy in PD-MCI, and a distinct correlation of path length, local efficiency, and global efficiency in PD-MCI (Fig.4B). Significantly higher nodal efficiency in right frontal inferior triangularis was observed in PD-MCI (Fig.4C). Path length of right caudate and left hippocampus showed a differential correlation with Brief Visuospatial Memory Test-Delayed Recall (BVMT-DR) and Montreal Cognitive Assessment (MoCA) score, respectively (Fig.4C). Global efficiency of right posterior cingulate also showed a differential correlation with DRS-Memory between groups (Fig.4C).

Discussion

Our preliminary analysis suggests that both dMRI-derived voxelwise measures and cortical volume/thickness of CC, insula, and limbic cortex are sensitive measure to identify PD-MCI and differentiate PD-MCI from cognitively normal PD. Furthermore, distinct topographical disorganization involving visual and parietal cortex was observed in PD-MCI.

Conclusion

Beyond ST dMRI-derived voxelwise measures from NODDI or DKI, in addition to conventional network measures and structural MRI measurements might provide both sensitive and specific measures for further multivariate analysis to identify imaging markers corresponding to PD-MCI. Analysis with longitudinal data and multimodal analyses utilizing resting-state functional MRI are currently underway.

Acknowledgements

This research project was supported by the NIH COBRE grant 5P20GM109025, Keep Memory Alive-Young Investigator Award, and philanthropic funds from Peter and Angela Dal Pezzo, Lynn and William Weidner, and Stacie and Chuck Matthewson.