Mixture of Spectral Wavelets on Simplicial Complex: Analysis of Brain Connectome with Neurodegeneration

Yechan Hwang, Minjae Jeong, Jaeyoon Sim, Guorong Wu, Won Hwa Kim

Keywords: neurodegenerative disease, spectral graph analysis, simplicial complex

Abstract:

Understanding how pathological changes disrupt communication patterns in the brain requires models that examine interactions across multiple structural levels of a network. Many existing graph-learning methods emphasize node representations while providing limited treatment of signals defined on edges, and their use of predetermined spectral filters can restrict sensitivity to heterogeneous frequency behavior.
To overcome these issues, we introduce a framework that couples a simplicial wavelet–based representation—capable of handling signals on both vertices and connections—with an adaptive filtering module that selects informative spectral components in a data-dependent manner. This combination enables flexible multi-scale analysis and highlights structural patterns relevant to neurodegenerative conditions. Evaluations on widely used brain graph benchmarks show consistent gains in predictive performance as well as clearer interpretation of disease-related network alterations. The implementation of this work will be released upon publication.

Primary Area: applications to neuroscience & cognitive science

Submission Number: 11658