Patterns of iron deposition and classification in parkinsonian syndromes: a quantitative susceptibility mapping study.

Differential diagnosis of parkinsonian syndromes remains challenging because of overlapping clinical symptoms. We characterized brain iron deposition using whole-brain quantitative susceptibility mapping (QSM) and developed internally validated exploratory machine learning (ML) classifiers. This single-center retrospective study included 119 patients with Parkinson's disease (PD), 62 with multiple system atrophy-parkinsonian type (MSA-P), 25 with progressive supranuclear palsy (PSP), and 56 healthy controls (HC). Voxel-wise and region-of-interest (ROI) analyses were performed, and classifiers were built using whole-brain ROI QSM features for HC versus parkinsonian syndromes and combined QSM plus clinical variables for PD versus MSA-P. Compared with HC, all patient groups showed increased susceptibility in multiple regions, mainly involving the substantia nigra (SN) and basal ganglia. Compared with PD, MSA-P showed higher susceptibility in the SN, putamen, pallidum, parts of the frontal lobe, and dentate nucleus. ROI analyses showed that increased susceptibility in the substantia nigra pars compacta (SNpc) was common across patient groups; MSA-P showed more prominent putaminal, pallidal, and dentate alterations; and PSP was mainly characterized by increased susceptibility in the SNpc and pallidum. Internally validated classifiers achieved subject-level AUC values of 0.805 for HC versus parkinsonian syndromes and 0.800 for PD versus MSA-P. These findings suggest the potential value of QSM-derived features as exploratory biomarkers.