Multi-Level Graph Neural Network With Sparsity Pooling for Recognizing Parkinson's Disease.

Area of Science:

• Neuroscience
• Machine Learning
• Medical Imaging

Background:

• Parkinson's disease (PD) is a neurodegenerative disorder characterized by motor symptoms.
• Machine learning (ML), particularly deep learning, shows promise in assisting PD diagnosis.
• Current ML approaches for PD using MRI data face challenges in graph construction efficiency and overfitting on limited datasets.

Purpose of the Study:

• To propose a novel multi-layer graph neural network (GNN) model for enhanced Parkinson's disease (PD) prediction using magnetic resonance imaging (MRI) data.
• To address the limitations of existing GNN models regarding graph construction efficiency and overfitting on small datasets.

Main Methods:

• Development of a novel multi-layer GNN model incorporating a fast graph construction technique.
• Integration of a sparsity-based pooling layer with an attention mechanism.
• Incorporation of graph structure sparsity as prior knowledge to reduce model overfitting during training.

Main Results:

• The proposed GNN model demonstrated effectiveness in PD prediction using real-world MRI datasets.
• Experimental results indicated the model's superiority compared to established baseline methods.
• The novel approach successfully addressed challenges in graph construction and overfitting.

Conclusions:

• The developed multi-layer GNN model offers an effective and efficient solution for PD prediction from MRI data.
• The proposed methods for fast graph construction and sparsity-based pooling significantly improve model performance and generalization.
• This research advances the application of GNNs in neurodegenerative disease diagnosis.