Ordinal Pattern Tree: A New Representation Method for Brain Network Analysis.

Area of Science:

• Neuroscience
• Graph Theory
• Machine Learning

Background:

• Brain network analysis commonly uses graph theory but often overlooks edge weight information.
• Existing methods for representing brain networks have limitations in capturing complex relationships.

Purpose of the Study:

• To propose a new brain network representation method, the ordinal pattern tree (OPT), that utilizes edge weights and hierarchical structures.
• To develop an optimal transport-based ordinal pattern tree (OT-OPT) kernel for measuring brain network similarity.
• To evaluate the effectiveness of the OT-OPT method on benchmark datasets for classification and regression tasks.

Main Methods:

• Representing brain networks using ordinal pattern trees (OPTs) with ordinal edges derived from weighted edge relationships.
• Developing the optimal transport (OT) based ordinal pattern tree (OT-OPT) kernel to compute similarity between OPTs.
• Utilizing optimal transport distances to calculate node transport costs within the OT-OPT kernel, proving its positive definiteness.

Main Results:

• The proposed ordinal pattern tree (OPT) method effectively leverages weighted information and hierarchical node relationships in brain networks.
• The OT-OPT kernel demonstrates superior performance in measuring brain network similarity compared to existing graph-based methods.
• Experimental results on ADHD-200, ABIDE, and ADNI datasets show significant improvements in classification and regression tasks.

Conclusions:

• The ordinal pattern tree (OPT) offers a powerful new approach for brain network representation and analysis.
• The OT-OPT kernel provides a robust and effective measure of brain network similarity.
• The proposed method advances the state-of-the-art in neuroimaging analysis for clinical applications.