Using Diffusion Similarity Networks to Identify, Track, and Predict Movement Disorders

Characterizing individual structural brain alterations in movement disorders like Parkinsons disease (PD) and Essential Tremor (ET) remains a significant challenge in clinical neuroscience. While brain structural connectivity (SC) derived from Diffusion Tensor Imaging (DTI) tractography maps physical white matter pathways, it fails to capture localized microstructural remodelings within gray matter structures. In this study, we use Diffusion Similarity Networks (DSNs) as a novel framework to estimate microstructural connectivity from diffusion weighted imaging (DWI). First, we evaluate the ability of DSNs to differentiate individuals with PD from healthy controls (HC). We used Similarity Network Fusion (SNF) to visualize differences between the fused networks of both groups as well as the Brain Connectivity Toolbox (BCT) to quantify changes in network topology, including transitivity, clustering coefficient, global efficiency, and betweenness centrality. Second, to evaluate the capacity of DSNs to track disease progression over time, we utilized longitudinal data extracted from the Parkinsons Progression Markers Initiative (PPMI) database, mapping distinct functional subnetwork degeneration trajectories in PD patients. Finally, we developed a predictive Graph Neural Network (GNN) framework on a locally collected cohort to address the challenging task of differentiating PD from ET, utilizing DSNs as highly discriminant input features. Our results demonstrate that DSN metrics successfully capture distinct structural signatures that reliably separate clinical phenotypes, while longitudinal modeling reveals an accelerated, progressive decay in specific motor and basal ganglia subnetwork topologies over time. Furthermore, the GNN model achieved high classification performance (Balanced accuracy = 0.8684, AUC= 0.9420) in distinguishing PD from ET globally, with local explain- ability analysis successfully showing the specific brain Regions of Interest (ROIs) driving the differential diagnosis. Overall, these findings demonstrate that Diffusion Similarity Networks serve as a robust, non-invasive biomarker for movement disorders, offering deep insights into disease progression and an explainable AI tool for clinical decision support.