Automated Imaging Differentiation for Parkinsonism (AIDP) is a machine learning-based diagnostic system that uses 3-Tesla diffusion magnetic resonance imaging (MRI) combined with support vector machine (SVM) classification to differentiate between Parkinson disease (PD), multiple system atrophy (MSA), and progressive supranuclear palsy (PSP) during life. Published in JAMA Neurology (2025), this prospective multicenter study achieved near-pathology-level diagnostic accuracy, with 93.9% confirmation in autopsy cases. [@vaillancourt2025]
The AIDP system analyzes diffusion MRI metrics that capture the microstructural changes underlying neurodegeneration in parkinsonian syndromes:
- Free-water (FW) imaging: Measures the volume of free water in brain tissue, which increases with neurodegeneration as cells die and extracellular space expands
- Free-water-corrected fractional anisotropy (FAt): Assesses white matter integrity by removing the confounding effects of free water, revealing true axonal damage
- Regional analysis: 132 brain regions are analyzed across the entire brain to identify disease-specific patterns
Each parkinsonian syndrome shows characteristic patterns of regional neurodegeneration:
| Disease |
Primary Pattern |
Key Regions Affected |
| PD |
Limbic and olfactory pathway involvement |
Anterior olfactory nucleus, limbic regions |
| MSA |
Cerebellar and brainstem involvement |
Middle cerebellar peduncle, pontine basis, cerebellar hemispheres |
| PSP |
Midbrain and frontostriatal involvement |
Midbrain, superior cerebellar peduncle, frontal cortex, striatum |
In PD, diffusion MRI reveals:
- Free-water increases in the substantia nigra pars compacta reflecting dopaminergic neuron loss
- Relative preservation of white matter integrity in early disease
- Minimal cerebellar involvement (distinguishing from MSA)
In MSA, characteristic findings include:
- Marked free-water elevation in the middle cerebellar peduncle ("hot cross bun sign" precursor)
- Pontine basis degeneration seen as increased free-water
- Cerebellar hemisphere involvement (especially in MSA-C variant)
- Dorsal medulla involvement reflecting autonomic nucleus degeneration
In PSP, the imaging captures:
- Midbrain free-water elevation (underlying the "hummingbird sign" on conventional MRI)
- Superior cerebellar peduncle atrophy and free-water increase
- Frontostriatal network degeneration
- Global free-water elevations reflecting widespread neurodegeneration
The AIDP system utilizes standardized 3-Tesla diffusion MRI with:
- Diffusion-weighted imaging: Minimum 30 gradient directions
- b-values: At least 1000 s/mm² for adequate signal-to-noise
- Spatial resolution: ≤2mm isotropic for precise regional analysis
- Preprocessing: Eddy current correction, motion correction, brain extraction
The SVM classifier employs:
- Feature extraction: Free-water and FAt values from 132 brain regions
- Covariate adjustment: Age and sex as demographic covariates
- Model training: 5-fold cross-validation on 78% of data
- Validation: Independent test set (22% held-out data)
- Neuropathology confirmation: 49 autopsy cases for ground truth validation
Input Features (132 regions × 2 metrics + demographics)
↓
Support Vector Machine (RBF kernel)
↓
Classification Outputs:
- PD vs Atypical Parkinsonism
- PD vs MSA
- PD vs PSP
- MSA vs PSP
| Comparison |
AUROC |
95% CI |
PPV |
NPV |
| PD vs. Atypical Parkinsonism |
0.96 |
0.93–0.99 |
0.91 |
0.83 |
| PD vs. MSA |
0.98 |
0.96–1.00 |
0.97 |
0.97 |
| PD vs. PSP |
0.98 |
0.96–1.00 |
0.92 |
0.98 |
| MSA vs. PSP |
0.98 |
0.96–1.00 |
0.98 |
0.81 |
Of 49 autopsy-confirmed cases, AIDP predictions matched neuropathology in 46 cases (93.9%), providing unprecedented validation of in vivo imaging-based diagnosis. This represents the highest level of pathological confirmation for any imaging-based diagnostic method in parkinsonian syndromes.
- Prospective cohort: 249 patients (99 PD, 53 MSA, 97 PSP) from 21 US and Canada sites (July 2021–January 2024)
- Retrospective cohort: 396 additional patients for model refinement
- Independent test set: 145 patients for final validation
AIDP is recommended for integration into the standard diagnostic workup for patients with parkinsonism when:
- Clinical features are atypical or overlap between syndromes
- Disease duration is <5 years (when differentiation is most challenging)
- Patient enrollment in disease-modifying therapy trials requires accurate diagnosis
- Prognostic counseling requires syndrome-specific predictions
| AUROC Range |
Clinical Interpretation |
| >0.95 |
Excellent diagnostic accuracy, high confidence |
| 0.90–0.95 |
Good diagnostic accuracy, moderate confidence |
| 0.80–0.90 |
Moderate diagnostic accuracy, consider additional testing |
| <0.80 |
Poor diagnostic accuracy, clinical judgment prevails |
- Objective: Removes inter-rater variability in clinical assessment
- Quantitative: Provides reproducible, operator-independent metrics
- Non-invasive: Uses standard MRI sequences already available
- Pathology-validated: Highest correlation with autopsy findings
- Multiplex: Single test provides differential diagnosis for all three syndromes
- Requires specific diffusion MRI protocols (not all centers)
- Processing requires specialized software and expertise
- May not be available in community settings
- Cannot distinguish PD from Parkinson-plus syndromes with atypical features
- Still requires clinical correlation for final diagnosis
| Method |
Sensitivity |
Specificity |
Availability |
Pathology Validation |
| AIDP (MRI + ML) |
95% |
95% |
Limited |
93.9% |
| Clinical criteria (NINDS) |
70–80% |
70–80% |
Widely available |
Limited |
| DaT SPECT |
~80% |
~75% |
Widely available |
None |
| Tau PET (flortaucipir) |
~85% |
~80% |
Limited |
Partial |
| MRI visual assessment |
60–70% |
60–70% |
Widely available |
Limited |
- Hybrid diagnosis: Integration with clinical criteria for combined probability
- Extended syndromes: Inclusion of corticobasal syndrome (CBS) and DLB
- Automated pipelines: One-click preprocessing and analysis
- Cloud platforms: Democratized access without local MRI infrastructure
- Point-of-care MRI: Portable, low-field systems adapted for AIDP
- EHR integration: Automated results interpretation in electronic health records
- Real-time analysis: On-the-fly processing during MRI acquisition
- Longitudinal monitoring: Tracking disease progression and treatment response
- Multi-ethnic cohort validation
- Integration with emerging disease-modifying therapies
- Comparison with other ML approaches (deep learning, random forests)
AIDP represents a paradigm shift in the differential diagnosis of parkinsonian syndromes. By combining quantitative diffusion MRI with machine learning, clinicians can now achieve near-pathology-level accuracy during life. This technology addresses a critical unmet need in movement disorder neurology and has the potential to transform clinical trial design, prognostic counseling, and treatment selection for patients with PD, MSA, and PSP.
- Vaillancourt et al., Automated Imaging Differentiation for Parkinsonism (2025) — PMID: 40094699
- Pagano et al., AIDP Study Group - Machine Learning MRI Differentiation (2025) — JAMA Neurology
- Progressive Supranuclear Palsy - Diagnostic Methods (2026) — PMID: 40898879
- MRI Biomarkers for Atypical Parkinsonism (2024)