Hyperbolic Kernel Graph Neural Networks for Neurocognitive Decline Analysis From Multimodal Brain Imaging.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Multimodal neuroimages like diffusion tensor imaging (DTI) and resting-state functional MRI (fMRI) provide complementary insights into brain structure and function.
• Existing fusion methods often fail to capture the hierarchical organization of brain networks due to their Euclidean space implementation.

Purpose of the Study:

• To develop and validate a hyperbolic kernel graph fusion (HKGF) framework for enhanced neurocognitive decline analysis using multimodal neuroimages.
• To leverage hyperbolic geometry for a more effective representation of brain network hierarchies.

Main Methods:

• Constructing multimodal brain graphs from DTI and fMRI data.
• Employing hyperbolic kernel graph neural networks (HKGNNs) to encode brain graphs in hyperbolic space, preserving hierarchical structures.
• Implementing a cross-modality coupling module for effective data fusion and a hyperbolic neural network for prediction.

Main Results:

• The HKGF framework demonstrated superior performance compared to state-of-the-art methods in neurocognitive decline prediction tasks.
• Experiments on over 4,000 subjects confirmed the efficacy of hyperbolic space representation for capturing brain network complexities.

Conclusions:

• The proposed HKGF framework offers a powerful and generalizable approach for multimodal neuroimage analysis in the context of neurocognitive decline.
• HKGF facilitates objective quantification of brain connectivity changes associated with neurocognitive decline, paving the way for improved diagnostic tools.