MNC-Net: Multi-task graph structure learning based on node clustering for early Parkinson's disease diagnosis.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Early identification of Parkinson's disease (PD) is crucial for effective patient management and prognosis.
• Structural brain networks (SBNs) show promise for PD diagnosis, but analyzing their complex topology for abnormal patterns is challenging.
• Existing deep learning models for PD prediction often lack interpretability and fail to integrate clinical scores, limiting their practical application.

Purpose of the Study:

• To develop an interpretable graph learning framework for early Parkinson's disease classification using structural brain networks.
• To address the challenges of mining abnormal patterns from high-dimensional SBNs and improve the interpretability of deep learning models.
• To incorporate clinical scores into the diagnostic model, enhancing its practical applicability.

Main Methods:

• Proposed a multi-task graph structure learning framework based on node clustering (MNC-Net) for early PD diagnosis.
• Modeled SBNs into modular graphs to facilitate representation learning of abnormal patterns and reduce noise.
• Integrated a regression task for clinical scores and utilized multi-task learning to incorporate latent domain information.

Main Results:

• MNC-Net achieved 95.5% accuracy in distinguishing early-stage PD from healthy controls on the Parkinsons Progression Markers Initiative dataset.
• Graph structure learning effectively captured discriminatory features, as visualized by t-SNE plots.
• Identified salient brain regions (ROIs) associated with PD symptoms, demonstrating the model's interpretability and consistency with prior research.

Conclusions:

• The proposed MNC-Net framework offers an effective and interpretable approach for early Parkinson's disease diagnosis.
• Integrating structural brain networks and clinical scores through multi-task learning enhances diagnostic accuracy and clinical relevance.
• The identified brain regions provide insights into the neurobiological underpinnings of early PD symptoms.