Identification of essential tremor and dystonic tremor using Graph Convolutional Networks with multiple connectivity...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Imaging

Background:

• Essential Tremor (ET) and Dystonic Tremor (DT) are distinct movement disorders.
• Understanding their underlying neuropathological mechanisms is crucial for diagnosis and treatment.
• Graph convolutional networks (GCNs) offer a powerful approach for analyzing complex brain data.

Purpose of the Study:

• To identify salient brain regions and connectivity patterns associated with ET and DT using a multi-connection GCN (MCGCN) model.
• To explore the neuropathological mechanisms underlying ET and DT.
• To differentiate between ET, DT, and healthy controls (HCs) based on brain connectivity.

Main Methods:

• Collected resting-state functional MRI (Rs-fMRI) data from 55 ET patients, 51 DT patients, and 52 HCs.
• Constructed functional connectivity (FC) matrices using three distinct modes and input them into four GCN architectures for classification.
• Utilized Grad-CAM for identifying discriminative brain regions and employed graph theory and correlation analyses for validation.

Main Results:

• GCN models achieved high classification accuracies: 91.36% (DT vs. HCs), 85.91% (ET vs. HCs), and 86.64% (ET vs. DT).
• Discriminative brain regions identified were primarily in the basal ganglia, cerebello-thalamo-cortical motor circuitry, and non-motor cortical areas.
• Nodal efficiency in these salient regions showed a negative correlation with clinical characteristics.

Conclusions:

• The study implicates the classic tremor network in the pathogenesis of ET and DT.
• Findings enhance the understanding of FC-based pathophysiological mechanisms in these tremor disorders.
• MCGCN provides a robust method for identifying neuroimaging biomarkers for ET and DT.