Task-radMBNet: An Improved Task-Driven Dynamic Graph Sparsity Pattern Radiomics-Based Morphological Brain Network for...

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Diagnostics

Background:

• Alzheimer's disease (AD) analysis benefits from dynamic adaptive graph sparsity for improved focus and sensitivity.
• Radiomics-based morphological brain networks (radMBN) offer a framework for analyzing brain structure.

Purpose of the Study:

• To introduce a task-driven dynamic adaptive graph sparsity model (Task-radMBNet) for enhanced AD diagnosis.
• To integrate connectivity and radiomics features within a dual-channel graph convolutional network (GCN) framework.
• To improve the accuracy of early AD detection using advanced neuroimaging analysis.

Main Methods:

• Developed Task-radMBNet, incorporating a connectivity-GCN channel and a radiomics-GCN channel sharing a dynamic sparse brain network.
• The connectivity-GCN channel dynamically learns optimal sparse topology for the task.
• The radiomics-GCN channel integrates radiomics node features with dynamic topology for diagnostic enhancement.

Main Results:

• Task-radMBNet achieved superior classification accuracy: 87.8% and 86.0% for early AD diagnosis.
• Evaluated on 1273 subjects across AD Neuroimaging Initiative and European DTI Study on Dementia databases.
• Visualizations demonstrated topology heatmaps and important connectivity under varying sparsity settings.

Conclusions:

• Task-radMBNet shows significant promise for diagnosing neurological disorders, particularly Alzheimer's disease.
• The integration of Task-radMBNet with radMBN offers a powerful approach for neuroimaging analysis.
• Dynamic adaptive graph sparsity is a key factor in improving diagnostic sensitivity and accuracy.