Learning the Hierarchical Organization in Brain Network for Brain Disorder Diagnosis

Jingfeng Tang; Peng Cao; Guangqi Wen; Jinzhu Yang; Xiaoli Liu; Osmar R. Zaiane

cs.LGMar 10, 2026

Learning the Hierarchical Organization in Brain Network for Brain Disorder Diagnosis

Authors: Jingfeng Tang, Peng Cao, Guangqi Wen, Jinzhu Yang, Xiaoli Liu, Osmar R. Zaiane

Organizations: Computer Science and Engineering, Northeastern University, Shenyang, China · School of Computer Science and Artificial Intelligence, Shandong Normal University, Jinan, China · AiShiWeiLai AI Research, Beijing, China · Amii, University of Alberta, Edmonton, Alberta, Canada

Abstract

Brain network analysis based on functional Magnetic Resonance Imaging (fMRI) is pivotal for diagnosing brain disorders. Existing approaches typically rely on predefined functional sub-networks to construct sub-network associations. However, we identified many cross-network interaction patterns with high Pearson correlations that this strict, prior-based organization fails to capture. To overcome this limitation, we propose the Brain Hierarchical Organization Learning (BrainHO) to learn inherently hierarchical brain network dependencies based on their intrinsic features rather than predefined sub-network labels. Specifically, we design a hierarchical attention mechanism that allows the model to aggregate nodes into a hierarchical organization, effectively capturing intricate connectivity patterns at the subgraph level. To ensure diverse, complementary, and stable organizations, we incorporate an orthogonality constraint loss, alongside a hierarchical consistency constraint strategy, to refine node-level features using high-level graph semantics. Extensive experiments on the publicly available ABIDE and REST-meta-MDD datasets demonstrate that BrainHO not only achieves state-of-the-art classification performance but also uncovers interpretable, clinically significant biomarkers by precisely localizing disease-related sub-networks.

Figures & tables

Figure 1: (a) Average Pearson correlation between brain regions on the ABIDE dataset, revealing strong interactions of brain regions across predefined sub-networks. (b) Learned sub-networks on the ABIDE and REST-meta-MDD datasets. Derived from attention weights, these visualizations demonstrate that BrainHO uncovers disease-related sub-networks that span predefined sub-networks(identified via subgraph to graph attention weights) as well as functionally consistent sub-networks(determined by the proportion of functional sub-networks).

Figure 2: Framework of the proposed BrainHO. The model constructs the brain network in a bottom-up attention learning manner while enforcing top-down hierarchical consistency. The Hierarchical Attention module comprises three distinct stages: node-to-node, node-to-subgraph, and subgraph-to-graph.

Dataset	Methods	Input	ACC(%)	AUC(%)	SEN(%)	SPE(%)
REST- meta-MDD	BNT	$\mathcal{P}$	63.28 $\pm$ 1.48	69.08 $\pm$ 1.49	65.03 $\pm$ 4.83	61.41 $\pm$ 6.48
	Com-BrainTF	$\mathcal{P}$	63.24 $\pm$ 2.34	68.73 $\pm$ 1.76	60.93 $\pm$ 7.67	66.21 $\pm$ 6.90
	ALTER	$\mathcal{B},\mathcal{P}$	63.95 $\pm$ 0.72	68.91 $\pm$ 0.88	64.94 $\pm$ 6.28	62.75 $\pm$ 7.75
	DHGFormer	$\mathcal{B},\mathcal{P}$	61.60 $\pm$ 1.36	65.18 $\pm$ 1.79	64.73 $\pm$ 3.55	58.01 $\pm$ 3.84
	LHDFormer	$\mathcal{B},\mathcal{P}$	63.95 $\pm$ 1.88	69.96 $\pm$ 1.80	65.08 $\pm$ 5.25	62.79 $\pm$ 8.36
	BrainHO(Ours)	$\mathcal{P}$	64.71 $\pm$ 2.01	68.83 $\pm$ 1.34	67.43 $\pm$ 4.19	61.51 $\pm$ 3.65

Table 1: Performance comparison on ABIDE and REST-meta-MDD datasets. The Input column specifies the data type used by each method, where

\mathcal{B}

represents raw BOLD signals and

\mathcal{P}

denotes the static PCC connectivity matrix.

Figure 3: The interpretibility of the sub-networks identified by our model.

Explore similar work

Navigating Hierarchy: Hyperbolic Learning on Brain Graphs for Disorder Diagnosis
Jul 8, 2026Yapeng Li, Bo Jiang, Ziyan Zhang +2Medical Diagnosis Hierarchical Representation Learning
School of Computer Science and Technology, Anhui University, 111 Jiulong Road, Hefei, 230601, China
Exploring Subnetwork Interactions in Heterogeneous Brain Network via Prior-Informed Graph Learning
Mar 13, 2026Siyu Liu, Guangqi Wen, Peng Cao +4Brain Network Analysis Functional Connectivity
Computer Science and Engineering, Northeastern University, Shenyang, China · Key Laboratory of Intelligent Computing in Medical Image of Ministry of Education, Northeastern University, Shenyang, China · School of Computer Science and Artificial Intelligence, Shandong Normal University, Jinan, China +5
Learning Multi-Scale Hypergraph for High-Order Brain Connectivity Analysis
Jun 2, 2026Jaeyoon Sim, Soojin Hwang, Seunghun Baek +2Neuroimaging Hypergraph Learning
Pohang University of Science and Technology (POSTECH), Pohang, South Korea · University of North Carolina at Chapel Hill, Chapel Hill, USA

Jul 8, 2026J/K move · Enter open · S save

Navigating Hierarchy: Hyperbolic Learning on Brain Graphs for Disorder Diagnosis

Yapeng Li, Bo Jiang, Ziyan Zhang, Dongdong Chen+1

School of Computer Science and Technology, Anhui University, 111 Jiulong Road, Hefei, 230601, China

Medical Diagnosis Hierarchical Representation Learning Neuroimaging Hyperbolic Representation Learning Brain Network Analysis

Open paper AskPDF arXiv

Functional brain networks exhibit a hierarchical organization across ROI, community, and whole-brain levels, supporting local processing, inter-community coordination, and global integration. Recent studies have demonstrated that brain community-aware modeling is beneficial for both diagnosis and biomarker identification of brain networks. However, existing brain graph modeling methods often struggle to model ROI-community interactions, thereby failing to fully exploit the hierarchy across ROI, community, and whole-brain network levels. To address this issue, inspired by deep hyperbolic learning in modeling hierarchical structures, we propose a novel framework, termed Hyperbolic Learning on Brain Graphs (HLBG), for brain network analysis. The core idea of HLBG is to exploit the inherent hierarchical geometry of hyperbolic space to model the hierarchical relationships among ROIs, functional communities, and the whole-brain network, thereby learning hierarchy-aware and highly discriminative representations for brain network data. Specifically, HLBG first projects representations from ROIs, communities, and the whole-brain network into Lorentzian hyperbolic space. Then, the multi-level hierarchy is imposed via two geometric entailment constraints. In addition, we introduce a new Graph-aware Mamba (GaMamba) model, which incorporates topology-derived structural prompts into Mamba to capture long-range dependencies while preserving graph topological information. Experiments on ABIDE-I and REST-MDD demonstrate that HLBG outperforms state-of-the-art methods and identifies disorder-relevant functional biomarkers.

Abstract

Explore similar work

Navigating Hierarchy: Hyperbolic Learning on Brain Graphs for Disorder Diagnosis

Exploring Subnetwork Interactions in Heterogeneous Brain Network via Prior-Informed Graph Learning

Learning Multi-Scale Hypergraph for High-Order Brain Connectivity Analysis