HOI-brain: A novel multi-channel transformers framework for brain disorder diagnosis by accurately extracting signed higher-order interactions from fMRI data.
Authors
Affiliations (5)
Affiliations (5)
- School of Mathematics and Statistics, Shandong University, Weihai, 264209, Shandong, China. Electronic address: [email protected].
- School of Mathematics and Statistics, Shandong University, Weihai, 264209, Shandong, China; Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing, 100190, Beijing, China. Electronic address: [email protected].
- Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing, 100190, Beijing, China. Electronic address: [email protected].
- School of Computer Science and Technology, Shandong University, Qingdao, 266000, Shandong, China. Electronic address: [email protected].
- School of Mathematics and Statistics, Shandong University, Weihai, 264209, Shandong, China. Electronic address: [email protected].
Abstract
Accurately characterizing the higher-order interactions of brain regions and effectively extracting the interpretable higher-order organizational patterns from functional Magnetic Resonance Imaging (fMRI) data are crucial for the diagnosis of brain diseases. However, current graph models mainly focus on pairwise patterns, as well as triadic patterns within the brain while overlooking more higher-order patterns with signs, limiting an integrated understanding of brain-wide communication. To address these challenges, we propose HOI-Brain (Higher-Order Interaction in Brain Network), a novel computational framework that enables the utilization of signed higher-order interactions and signed organizational patterns in fMRI data for the diagnosis of brain diseases. Specifically, we present a new calculation of co-fluctuations based on Multiplication of Temporal Derivatives to detect higher-order interactions with adequate temporal resolution. Next, we further distinguish positively and negatively synergistic higher-order interactions and encode them in the monotonic weighted simplicial complexes, which can offer detailed insights into the communication within the brain. Moreover, we employ Persistent Homology in the monotonic weighted simplicial complexes of the brain to extract signed higher-dimensional neural organizations from a spatiotemporal perspective. Finally, a multi-channel transformers architecture is proposed to holistically integrate information from heterogeneous topological features. Comprehensive experiments across Alzheimer's disease (AD), Parkinson's disease (PD), and Autism Spectrum Disorder (ASD) datasets demonstrate the superiority, effectiveness, and interpretability of our framework. These key regions and higher-order patterns align with existing evidence: concordant HOIs decrease in AD but increase in PD and ASD, providing mechanistic clues to disorder-specific network changes.