Developing a Dynamic Graph Network for Interpretable Analysis of Multi-Modal MRI Data in Parkinson's Disease...

Area of Science:

• Neurology
• Computer Science
• Artificial Intelligence

Background:

• Parkinson's disease (PD) presents a significant public health challenge due to population aging.
• Computer-aided diagnosis (CAD) methods are crucial for PD diagnosis and progression prediction.
• Graph Convolutional Networks (GCNs) are increasingly used for integrating multi-modal features and modeling subject correlations in deep learning.

Purpose of the Study:

• To develop an interpretable Graph-Learning Convolutional Network (iGLCN) for enhanced personalized Parkinson's disease diagnosis.
• To address limitations of existing GCNs, including fixed graph topologies and lack of interpretability.
• To improve the practical applicability and feasibility of AI in clinical PD diagnosis.

Main Methods:

• Proposed an interpretable Graph-Learning Convolutional Network (iGLCN) model.
• Implemented dynamic graph structure adjustment for GCNs by learning an optimal underlying latent graph.
• Incorporated interpretable feature learning to explain diagnosis outcomes.

Main Results:

• The iGLCN demonstrated enhanced flexibility and maintained high classification performance for PD diagnosis.
• The proposed method provides interpretable results, increasing clinical utility.
• Experimental results validated the effectiveness and interpretability of the iGLCN for PD diagnosis.

Conclusions:

• The iGLCN offers a promising approach for personalized and interpretable Parkinson's disease diagnosis.
• The dynamic graph learning capability enhances diagnostic accuracy and adaptability.
• The method's practicability, feasibility, and interpretability are expected to benefit clinical decision-making in PD.