Enhancing Neurodegenerative Disease Diagnosis Through Confidence-Driven Dynamic Spatio-Temporal Convolutional...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Dynamic brain networks offer superior characterization of functional connectivity compared to static networks for neurodegenerative disease diagnosis.
• Existing dynamic brain network classification methods face limitations due to spatio-temporal coupling and inadequate integration of complex topological features.

Purpose of the Study:

• To develop a novel method, Confidence-Driven Dynamic Spatio-Temporal Convolutional Network (CD-DSTCN), for improved classification of dynamic brain networks.
• To address the limitations of sliding window approaches in capturing complex spatio-temporal dependencies and topological features.

Main Methods:

• Utilized a spatio-temporal convolutional network with a temporal attention mechanism to extract features within each window.
• Implemented a confidence scoring system based on true class probability (TCP) to weight the importance of different time windows.
• Employed a multilayer perceptron (MLP) for final classification using confidence-weighted fused features.

Main Results:

• The proposed CD-DSTCN method demonstrated superior performance compared to state-of-the-art algorithms on Alzheimer's and Parkinson's datasets.
• The method effectively captures and integrates complex temporal and spatial dependencies across time windows.
• Identified valuable biomarkers for neurodegenerative disease diagnosis.

Conclusions:

• CD-DSTCN offers a promising advancement in the diagnosis of neurodegenerative diseases by leveraging dynamic brain network analysis.
• The confidence-driven weighting mechanism enhances the integration of spatio-temporal information for more accurate classification.
• The developed method provides a robust tool for identifying brain disease biomarkers.