Hierarchical cluster analysis of multimodal imaging data identifies brain atrophy and cognitive patterns in Parkinson’s disease

Abstract

Background: Parkinson's disease (PD) is a heterogeneous condition. Cluster analysis based on cortical thickness has been used to define distinct patterns of brain atrophy in PD. However, the potential of other neuroimaging modalities, such as white matter (WM) fractional anisotropy (FA), which has also been demonstrated to be altered in PD, has not been investigated. Objective: We aim to characterize PD subtypes using a multimodal clustering approach based on cortical and subcortical gray matter (GM) volumes and FA measures. Methods: We included T1-weighted and diffusion-weighted MRI data from 62 PD patients and 33 healthy controls. We extracted mean GM volumes from 48 cortical and 17 subcortical regions using FSL-VBM, and the mean FA from 20 WM tracts using Tract-Based Spatial Statistics (TBSS). Hierarchical cluster analysis was performed with the PD sample using Ward's linkage method. Whole-brain voxel-wise intergroup comparisons of VBM and TBSS data were also performed using FSL. Neuropsychological and demographic statistical analyses were conducted using IBM SPSS Statistics 25.0. Results: We identified three PD subtypes, with prominent differences in GM patterns and little WM involvement. One group (n = 15) with widespread cortical and subcortical GM volume and WM FA reductions and pronounced cognitive deficits; a second group (n = 21) with only cortical atrophy limited to frontal and temporal regions and more specific neuropsychological impairment, and a third group (n = 26) without detectable atrophy or cognition impairment. Conclusion: Multimodal MRI data allows classifying PD patients into groups according to GM and WM patterns, which in turn are associated with the cognitive profile.

White Matter Abnormalities in Patients With Parkinson's Disease: A Meta-Analysis of Diffusion Tensor Imaging Using Tract-Based Spatial Statistics

“…A total of 23 whole-brain TBSS studies (Hattori et al, 2012;Kamagata et al, 2013;Kim et al, 2013;Melzer et al, 2013;Agosta et al, 2014;Carriere et al, 2014;Worker et al, 2014;Diez-Cirarda et al, 2015;Ji et al, 2015;Vercruysse et al, 2015;Vervoort et al, 2016;Wen et al, 2016;Acosta-Cabronero et al, 2017;Chen B. et al, 2017;Georgiopoulos et al, 2017;Luo et al, 2017;Firbank et al, 2018;Li et al, 2018;Minett et al, 2018;Guan et al, 2019;Quattrone et al, 2019;Inguanzo et al, 2020;Pelizzari et al, 2020) were identified using our search protocol. Six of them compared separate independent patient subgroups with the same HC groups, with details present in Table 1.…”

Impaired Structural Connectivity in Parkinson's Disease Patients with Mild Cognitive Impairment: A Study Based on Probabilistic Tractography

“…Conversely, beyond the acceptance of MCI definition (Litvan et al 2012 ) as useful clinical criteria to identify patients with worse cognitive profiles and dementia risk, recent evidence suggested the existence of a more complex picture, identifying PD subtypes based on neuropsychological, clinical, and MRI data (Dujardin et al 2013 ; Uribe et al 2016 ; Fereshtehnejad et al 2017 ; Inguanzo et al 2021 ). In light of our results, it could be suggested that the study of structural connectivity in PD subtypes might facilitate the study of different patterns of cognitive deterioration and shed light on their anatomical basis/substrates.…”

“…For example, when both GM and WM changes are considered in the same sample, WM appears to be explaining just a small part of the degenerative pattern. In Inguanzo et al (2021), we used GM and WM measures to perform a hierarchical cluster analysis, and we found three subgroups, of which only one presented WM alterations. Accordingly, cognitive performance in PD has been consistently seen to correlate with GM structural parameters (Garcia-Diaz et al 2018;Mak et al 2014), and with functional connectivity (Baggio et al 2015).…”

Comprehensive subtyping of Parkinson’s disease patients with similarity fusion: a case study with BioFIND data

“…In this context, researchers have turned to data-driven perspectives, where patient subtyping is transformed into a typical clustering problem 11 – 16 . These works have focused on several different types of data collected from patients besides just clinical assessments, which include neuroimaging data 17 – 19 , genomic data 20 , and neurophysiological assessment data 21 . Without any prior assumption patients are grouped into clusters, each of which corresponds to a specific subtype, such that patients within the same subtype manifest similar PD characteristics while those from different subtypes are distinct.…”