Structural MRI Volumetry Index for Differentiation of Progressive Supranuclear Palsy From Parkinson's Disease and Multiple System Atrophy by Automatic Segmentation: A Comparison With Magnetic Resonance Parkinsonism Index

BACKGROUND: Although Parkinson's disease (PD) and Parkinsonism-Plus syndromes (PPS) exhibit distinct brain atrophy, current clinical differentiation heavily relies on manual, plane-specific 2D metrics, which are labor-intensive and observer-dependent. To fill this gap, this study developed an automated, software-independent metric-the parkinsonism volumetric index (PDSVI)-calculated as [(pons volume × third ventricle volume)/(midbrain volume × thalamus volume) × 10]. PURPOSE: To evaluate the performance of PDSVI in precisely differentiating progressive supranuclear palsy (PSP) from PD, multiple system atrophy cerebellar (MSA-C) and parkinsonian (MSA-P) types. STUDY TYPE: Retrospective. POPULATION: 246 PD, 108 PSP, 63 MSA-C, 63 MSA-P (N = 480, 253 males and 227 females; mean age: 63.3 ± 9.2 years) and 252 healthy controls (HC) (134 males and 118 females; mean age: 64.9 ± 8.2 years). FIELD STRENGTH/SEQUENCE: 3 T, 3D inversion-prepared fast spoiled gradient echo sequence T1-weighted Imaging. ASSESSMENT: Manual magnetic resonance parkinsonism index (MRPI) was measured by two raters. Automated brain volumetry was performed using a pre-trained cascaded VB-Net. Patients were randomly assigned to training (n = 320) and internal validation (n = 160) cohorts via 2:1 stratified randomization. Diagnostic performance of manual MRPI, pons/midbrain volume ratio, and PDSVI were assessed. STATISTICAL TESTS: Diagnostic performance was assessed via receiver operating characteristic (ROC) curve analysis and DeLong's test; p < 0.05 was considered statistically significant. RESULTS: In internal validation cohort, ROC analysis showed that PDSVI distinguished PSP from non-PSP (AUC = 0.897), PD (0.873), MSA-P (0.950), MSA-C (0.939), and HC (0.909), comparable to manual MRPI (AUCs: 0.909, 0.886, 0.923, 0.984 and 0.862, p values: 0.760, 0.786, 0.492, 0.092, 0.287, respectively). DATA CONCLUSION: This study proposes a fully automated, 3D-T1-derived parameter that, while exhibiting lower sensitivity compared to the manual MRPI, provides comparable overall diagnostic performance among PSP, PD, MSA, and healthy controls. EVIDENCE LEVEL: 3. TECHNICAL EFFICACY: Stage 2.