TAFNet: Trusted Multiview Associative Fusion Neural Networks for Analyzing Dynamic Brain Networks

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Dynamic functional connectivity (DFC) is vital for understanding brain network temporal dynamics.
• Existing DFC methods often overlook data heterogeneity in clinical settings, assuming uniform quality across time windows.
• This limitation hinders accurate analysis of complex brain states.

Purpose of the Study:

• To propose a novel framework, TAFNet, for robust dynamic brain network analysis.
• To address data heterogeneity and improve the reliability of DFC analysis in clinical environments.
• To enhance diagnostic accuracy for neurological disorders using advanced machine learning.

Main Methods:

• Developed a trusted multiview associative fusion neural network (TAFNet) framework.
• Employed a local-global convolutional filtering module for evidence extraction from independent temporal window views.
• Integrated a multiview associative fusion mechanism using mutual information and a top-k view selection strategy.
• Incorporated a dynamic trust assessment (DTA) module and Dempster combination rule for reliable fusion and confidence quantification.

Main Results:

• TAFNet demonstrated superior performance compared to state-of-the-art methods on three schizophrenia datasets.
• The framework achieved a diagnostic accuracy improvement of at least 2.15%.
• The dynamic trust assessment module effectively quantified predictive confidence by aligning evidential beliefs with probabilistic predictions.

Conclusions:

• The proposed TAFNet framework offers a significant advancement in analyzing dynamic brain networks, particularly in heterogeneous clinical data.
• TAFNet enhances diagnostic accuracy and provides reliable confidence estimations, crucial for clinical applications.
• This approach paves the way for more accurate and trustworthy neuroimaging-based diagnostics.