MRI supervised and unsupervised classification of Parkinson's disease and multiple system atrophy.

Abstract

Multimodal MRI approach is based on a combination of MRI parameters sensitive to different tissue characteristics (eg, volume atrophy, iron deposition, and microstructural damage). The main objective of the present study was to use a multimodal MRI approach to identify brain differences that could discriminate between matched groups of patients with multiple system atrophy, Parkinson's disease, and healthy controls. We assessed the 2 different MSA variants, namely, MSA-P, with predominant parkinsonism, and MSA-C, with more prominent cerebellar symptoms. Twenty-six PD patients, 29 MSA patients (16 MSA-P, 13 MSA-C), and 26 controls underwent 3-T MRI comprising T2*-weighted, T1-weighted, and diffusion tensor imaging scans. Using whole-brain voxel-based MRI, we combined gray-matter density, T2* relaxation rates, and diffusion tensor imaging scalars to compare and discriminate PD, MSA-P, MSA-C, and healthy controls. Our main results showed that this approach reveals multiparametric modifications within the cerebellum and putamen in both MSA-C and MSA-P patients, compared with PD patients. Furthermore, our findings revealed that specific single multimodal MRI markers were sufficient to discriminate MSA-P and MSA-C patients from PD patients. Moreover, the unsupervised analysis based on multimodal MRI data could regroup individuals according to their clinical diagnosis, in most cases. This study demonstrates that multimodal MRI is able to discriminate patients with PD from those with MSA with high accuracy. The combination of different MR biomarkers could be a great tool in early stage of disease to help diagnosis. © 2018 International Parkinson and Movement Disorder Society.