Determining the number of states in dynamic functional connectivity using cluster validity indexes

Area of Science:

• Neuroscience
• Data Science
• Computational Biology

Background:

• Clustering analysis is crucial for identifying dynamic states in brain functional connectivity.
• Existing cluster validity indices (CVIs) lack clear guidance for dynamic functional connectivity (dFC) data.
• The suitability of various CVIs for dFC analysis remains undetermined.

Purpose of the Study:

• To comprehensively test and compare a wide range of CVIs for clustering dFC data.
• To identify the most effective CVIs for determining the optimal number of clusters in dFC analysis.
• To provide recommendations for robust dFC clustering in addiction research.

Main Methods:

• Evaluated twenty-four different cluster validity indices (CVIs).
• Applied CVIs to both simulated and real-world addiction dFC data.
• Compared the performance of established methods (e.g., Elbow-Criterion, Silhouette, GAP-Statistic) with less common ones.

Main Results:

• The Davies-Bouldin and Ray-Turi CVIs demonstrated superior performance in identifying the optimal number of clusters.
• These effective CVIs identify a local minimum critical point, amenable to automated computation.
• Performance comparison across twenty-four CVIs highlights Davies-Bouldin and Ray-Turi as most suitable for dFC.

Conclusions:

• Davies-Bouldin and Ray-Turi CVIs are recommended for determining cluster numbers in dFC analysis.
• These methods offer a more reliable approach compared to widely used but less effective indices.
• Automated identification of critical points enhances the practical application of these CVIs in dFC research.