Diagnostic potential of automated subcortical volume segmentation in atypical parkinsonism

Abstract

Objective: To determine whether automated and observer-independent volumetric MRI analysis is able to discriminate among patients with Parkinson disease (PD), multiple system atrophy (MSA), and progressive supranuclear palsy (PSP) in early to moderately advanced stages of disease. Methods: T1-weighted volumetric MRI from patients with clinically probable PD (n 40), MSA (n 40), and PSP (n 30) and a mean disease duration of 2.8 ± 1.7 y were examined using automated volume measures of 22 subcortical regions. The clinical follow-up period was 2.5 ± 1.2 years. The data were split into a training (n 72) and a test set (n 38). The training set was used to build a C4.5 decision tree model in order to classify patients as MSA, PSP, or PD. The classification algorithm was examined by the test set using the final clinical diagnosis at last follow-up as diagnostic gold standard. Results: The midbrain and putaminal volume as well as the cerebellar gray matter compartment were identified as the most significant brain regions to construct a prediction model. The diagnostic accuracy for PD vs MSA or PSP was 97.4%. In contrast, diagnostic accuracy based on validated clinical consensus criteria at the time of MRI acquisition was 62.9%. Conclusions: Volume segmentation of subcortical brain areas differentiates PD from MSA and PSP and improves diagnostic accuracy in patients presenting with early to moderately advanced stage parkinsonism. Classification of evidence: This study provides Class III evidence that automated MRI analysis accurately discriminates among early-stage PD, MSA, and PSP.