MV²Net: Multi-Variate Multi-View Brain Network Comparison over Uncertain Data.

Area of Science:

• Neuroscience
• Data Visualization
• Computer Science

Background:

• Identifying effective biomarkers from human brain networks is challenging.
• Existing methods focus on individual connectivity features and assume perfect data fidelity.
• Real-world comparisons require handling data uncertainty and multi-variate analysis.

Purpose of the Study:

• To present MV^2Net, a visual analytics system for comparing brain networks.
• To integrate multi-variate, multi-view visualization with interactive data wrangling for uncertainty management.
• To improve biomarker detection and comparison of brain networks between subject groups.

Main Methods:

• Integrating multiple extraction methods for diffusion and geometric connectivity features.
• Employing an anomaly detection algorithm for data quality assessment.
• Utilizing single- and multi-connection feature selection for biomarker detection.
• Developing novel visualization designs for level-of-detail comparisons (juxtaposed, explicit-coding, composite, fiber tract views).

Main Results:

• MV^2Net successfully integrates analysis and visualization for brain network comparison.
• The system handles data uncertainty through interactive wrangling.
• Novel visualization designs facilitate detailed comparisons across subject groups and features.
• Expert studies and case analyses validate the system's effectiveness and superiority.

Conclusions:

• MV^2Net offers a superior approach to brain network comparison and biomarker discovery.
• The system effectively addresses limitations of existing methods in handling data uncertainty.
• The integrated visual analytics framework advances the study of neurological diseases through brain network analysis.