Can Neuroimaging Predict Dementia in Parkinson's Disease?

Loss of connections between specialized brain regions (e.g. occipito-temporal) and between cerebral hemispheres (inter-hemispheric) are commonly associated with PD-MCI. Conflicting results with respect to global graph theory metrics (i.e. CC and GE) are likely due to methodological differences in connectome construction.
AUC = area under the curve; BG = basal ganglia; CC = clustering coefficient; fMRI = functional MRI; GE = global efficiency; GT = graph theory; HC = healthy controls; LE = local efficiency; MNI = Montreal Neurological Institute; Mod = modularity; NBS = network-based statistics; NS = not significant; PD-non-MCI = Parkinson's disease without mild cognitive impairment or dementia; PL = path length.
^*+ΔSymbols in 'Analysis' column indicate which comparisons are referred to in the 'Regional changes' column.

Authors and Disclosures

Juliette H. Lanskey^1,2,*, Peter McColgan^3,*, Anette E. Schrag⁴, Julio Acosta-Cabronero⁵, Geraint Rees^5,6, Huw R. Morris^4,7 and Rimona S. Weil<sup>5,8

1</sup> Institute of Neurology, UCL, Queen Square, London, UK
² Department of Clinical Neurosciences, University of Cambridge, Cambridge, UK
³ Huntington's Disease Centre, UCL, Queen Square, London, UK
⁴ Department of Clinical Neurosciences, Royal Free Campus UCL Institute of Neurology, UK
⁵ Wellcome Centre for Human Neuroimaging, UCL, Queen Square, London, UK
⁶ Institute of Cognitive Neuroscience, UCL, Queen Square, London, UK
⁷ Department of Movement Disorders, UCL, Queen Square, London, UK
₈ UCL Dementia Research Centre, Queen Square, London, UK
*These authors contributed equally to this work.

Correspondence to
Dr Rimona Weil UCL Dementia Research Centre, 8–11 Queen Square, London WC1N 3BG, UK E-mail: r.weil@ucl.ac.uk

Competing interests
A.E.S. reports personal fees from Medtronic and AstraZeneca. H.R.M. reports personal fees from Teva, AbbVie, Boehringer Ingelheim, and GSK. R.S.W. reports personal fees from GE Healthcare

Sidebar

Glossary of Graph Theory Metrics

Area under the curve: Graph theory metrics are calculated across a range of sparsity thresholds and the area under the curve is calculated in order to avoid choosing an arbitrary threshold value.

Betweenness centrality: The fraction of all shortest paths in the network that pass through a given node.

Clustering coefficient: Fraction of a node's neighbours that are connected to each other.

Degree: Represents the number of binary connections a brain region has.

Edge: Connections between brain regions defined using structural or functional brain imaging.

Eigen vector centrality: A measure that assigns increasing importance to a node if both they and their neighbours are highly connected to hub regions.

Global efficiency (GE): Inverse of shortest average path length.

Modularity: Subdivision of the network such that a module is defined as a group of regions that are highly connected to each other with minimal connections outside the group.

Motif: Patterns of interconnections occurring in a brain network at a significantly higher frequency than those occurring in randomized networks.

Network-based statistics (NBS): This is a statistical method that controls the family-wise error rate when mass univariate tests of all connections in a connectome are undertaken.

Node: A brain region usually defined using a brain atlas

Path length (PL): Minimum number of connections that must be traversed to go from one node to another.

Small worldness: The optimal balance between local clustering (clustering coefficient) and shortest path length.

Strength: A weighted version of degree that represents the sum of connection weights for a given brain region.

Thresholding: Applying a threshold below which connections are removed from the connectivity matrix. This is used to remove spurious connections and create sparsity in the matrix. Many methods are available; however, there is currently no consensus with respect to the optimal approach.

Weighting: Brain networks can be binary, where connections or either absent or present, or weighted based on diffusion metrics, such as fractional anisotropy, or the magnitude of temporal correlation between time series in the case of rs-fMRI.