Novel applications of machine learning and computational neuroscience models to neuroimaging in Parkinson's disease and related disorders.

Purpose of review: Parkinsonian syndromes are a heterogeneous group of neurodegenerative diseases that pose challenges in early diagnosis, differentiation, and pathophysiological understanding. The objective of this review is to summarize recent contributions of computational models combined with neuroimaging data to the differential diagnosis of Parkinsonian syndromes, disease subtyping, and understanding of disease processes. Recent findings: Using machine learning algorithms trained with MRI features, diagnostic accuracies above 90% have been achieved for distinguishing patients with Parkinson's disease from healthy controls and for the differential diagnosis of Parkinsonian syndromes. Computational models, such as hierarchical cluster analysis and Subtype and Stage Inference (SuStaIn), have enabled the identification of distinct disease subtypes within Parkinson's disease based on imaging-derived brain features. Network models based on structural and functional connectomes have revealed that disease spread in Parkinson's disease is primarily driven by global connectivity. Additionally, local brain characteristics such as gene expression, cellular composition, and neuroreceptor profiles may contribute to selective vulnerabilities. Summary: Computational approaches enhance the diagnosis of Parkinsonian syndromes, particularly in the early stages, and refine the characterization of disease subtypes, benefiting clinicians, especially in non-expert centers. Such applications hold significant potential for enabling more personalized management and selecting appropriate candidates for clinical trials. Furthermore, a deeper understanding of pathophysiology supports the development of disease-specific therapies.