Pathway Anchored Multimodal Clustering Reveals Circuit Level Signatures in Parkinson's Disease
Vinod, A., Eliendula, A.S., Bhardwaj, S., Dev, A., Dominic, A., Bajaj, C.
View full paper →
Multimodal biomarker patterns map onto six clinically relevant brain pathways, highlighting disease heterogeneity across motor, cognitive, limbic, and vascular domains.
Key findings
- A pathway-anchored co-clustering framework (SRVCC) integrates structural MRI, DTI, and DaT-SPECT within six predefined neuroanatomical circuits.
- Multimodal Pathway Integrity Scores (MPIS) reveal circuit-level imaging heterogeneity linked to motor severity and cognitive function.
- Five imaging-driven patient clusters emerge with distinct nigrostriatal, frontostriatal, and sensory-visuospatial profiles.
- Lower nigrostriatal and frontostriatal pathway integrity is associated with higher motor burden; sensory-visuospatial pathway shows the strongest association with cognitive function.
Incoming patients are placed within a population-derived subgroup landscape. The nearest cohort and key clinical features are surfaced to the treating neurologist.
Six neuroanatomical pathways — nigrostriatal, frontostriatal, sensory-visuospatial, limbic, microvascular, and balance/cerebellar — mapped to their clinical burden domains.
Cohort and methods
The study draws from the PPMI cohort: 185 Parkinson's disease patients, 72 healthy controls, and 37 SWEDD participants, all with baseline T1 MRI, DTI, and DaT-SPECT imaging.
The SRVCC (Scalable Robust Variational Compositional Co-clustering) framework learns joint patient and feature clusters. Multimodal Pathway Integrity Scores aggregate circuit-specific imaging features (fractional anisotropy, mean diffusivity, regional volume, and specific binding ratios) with equal modality weighting across six predefined PD-relevant circuits.