Automated, high accuracy classification of Parkinsonian disorders: a pattern recognition approach

Area of Science:

• Neuroimaging
• Neurology
• Machine Learning

Background:

• Parkinsonian disorders, including progressive supranuclear palsy (PSP), multiple system atrophy (MSA), and idiopathic Parkinson's disease (IPD), present similar early symptoms, complicating clinical diagnosis.
• Distinct molecular pathologies underlie these conditions, necessitating objective biomarkers for accurate differentiation.
• Previous structural neuroimaging studies have shown incomplete separation of these disease groups.

Purpose of the Study:

• To evaluate the efficacy of multi-class pattern recognition using structural MRI for automated classification of PSP, MSA, and IPD.
• To identify specific anatomical patterns and brain regions that serve as reliable biomarkers for discriminating these disorders at the individual subject level.

Main Methods:

• Structural MRI scans were acquired from patients with PSP (n=17), IPD (n=14), MSA (n=19), and healthy controls (HCs, n=19).
• An advanced probabilistic pattern recognition approach was applied to analyze pre-defined anatomical patterns, including a subcortical motor network, its component regions, and the whole brain.
• The diagnostic value of these patterns for discriminating between the disease groups and HCs was assessed.

Main Results:

• Simultaneous high-accuracy discrimination of all disease groups was achieved using the subcortical motor network.
• The midbrain/brainstem emerged as the most accurate region for predicting diagnosis, differentiating all disease groups from each other and from HCs.
• The subcortical network outperformed whole-brain and individual region analyses in predictive accuracy.
• Specific patterns of atrophy in the midbrain/brainstem, cerebellum, and basal ganglia were associated with PSP and MSA, while IPD showed distinct midbrain/brainstem atrophy.

Conclusions:

• Automated analysis of structural MRI data can accurately predict diagnoses for individual patients with Parkinsonian disorders.
• The subcortical motor network, particularly the midbrain/brainstem region, provides valuable anatomical biomarkers for differentiating PSP, MSA, and IPD.
• This approach offers a promising tool for objective diagnosis and potentially earlier intervention in these neurodegenerative conditions.