multi-view-o-information-brain-networks

name	multi-view-o-information-brain-networks
description	Higher-order brain interaction analysis using O-information and Multi-View Information Bottleneck for fMRI-based psychiatric diagnosis. Decomposes multivariate neural interactions into redundant and synergistic components across multiple brain views for improved diagnostic classification. (arXiv:2604.17713, April 2026)
tags	["O-information","higher-order interactions","brain networks","fMRI","psychiatric diagnosis","information bottleneck","multi-view learning","redundancy","synergy"]

Multi-View O-Information for Higher-Order Brain Network Interactions

arXiv: 2604.17713 (April 20, 2026) Categories: cs.LG

Summary

A multi-view information-theoretic framework for analyzing higher-order brain interactions in fMRI data for psychiatric diagnosis. Uses O-information to decompose multivariate neural dependencies into redundancy-dominated and synergy-dominated interactions, combined with a Multi-View Information Bottleneck (MVIB) approach that integrates multiple brain views for improved diagnostic classification of psychiatric conditions.

Key Methodology

O-Information Framework

1. Higher-Order Interactions: O-information quantifies whether a system of N variables exhibits redundancy-dominated (shared information) or synergy-dominated (emergent information) interactions
2. Decomposition: O-information decomposes total correlation into redundant and synergistic components
3. Neural Application: Applied to fMRI BOLD time series to characterize brain region interactions beyond pairwise connectivity

Multi-View Information Bottleneck

1. Multi-View Formulation: Treat different brain network views (functional networks, anatomical regions, frequency bands) as separate views
2. Compression-Extraction: Information bottleneck principle compresses each view while preserving diagnostic-relevant information
3. Fusion Strategy: Learned representations from multiple views are fused for downstream classification
4. Diagnostic Task: Applied to psychiatric disorder classification (schizophrenia, depression, ADHD)

Technical Pipeline

1. fMRI Preprocessing: Standard pipeline (motion correction, normalization, smoothing)
2. Brain Parcellation: ROI extraction using established atlases (AAL, Schaefer, etc.)
3. O-Information Computation: Calculate O-information for brain region subsets
4. Feature Extraction: Redundancy/synergy features from O-information decomposition
5. Multi-View Fusion: MVIB combines features from different brain views
6. Classification: Diagnostic prediction with cross-validation

Key Findings

• Higher-order interactions capture diagnostic information missed by pairwise connectivity
• Redundancy-dominated interactions distinguish psychiatric subtypes
• Multi-view fusion significantly improves classification accuracy over single-view approaches
• O-information provides interpretable biomarkers for clinical understanding

Practical Applications

When to Use This Approach

• Psychiatric disorder classification from fMRI data
• Analyzing higher-order brain interactions beyond pairwise connectivity
• Multi-modal brain data fusion for clinical diagnosis
• Understanding redundancy vs synergy in brain networks

Implementation Steps

1. Extract ROI time series from preprocessed fMRI
2. Compute O-information for region triplets and higher-order subsets
3. Classify interactions as redundancy- or synergy-dominated
4. Define brain views (network-level, hemisphere, frequency band)
5. Train MVIB model with view-specific encoders
6. Fuse compressed representations for classification
7. Evaluate with cross-validation and interpret via O-information maps

Limitations & Considerations

• Computational Cost: O-information scales combinatorially with number of regions
• Sample Size: Requires substantial fMRI datasets for reliable estimation
• Parcellation Sensitivity: Results depend on brain parcellation choice
• Clinical Validation: Further validation needed for clinical deployment

Related Skills

• brain-higher-order-structures — Higher-order brain network analysis
• multimodal-brain-connectivity-gnn — Multimodal brain connectivity
• brain-network-controllability — Brain network control theory