The Incremental Cluster Threshold-Free Cluster Enhancement Algorithm for Functional Connectivity Analysis

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Statistical Inference

Background:

• Threshold-free cluster enhancement (TFCE) is a widely used neuroimaging statistical method.
• Current TFCE implementations are computationally intensive, limiting their application with increasing data complexity and sample sizes.
• The quadratic growth of functional connectivity (FC) edges with regions of interest (ROIs) makes standard TFCE computationally infeasible for fine parcellations.

Purpose of the Study:

• To develop a computationally efficient algorithm for TFCE that produces numerically equivalent results to standard TFCE.
• To decouple the runtime of TFCE from discretization precision.
• To enable TFCE analysis on large-scale neuroimaging datasets with fine parcellations.

Main Methods:

• Introduced Incremental Cluster TFCE (IC-TFCE), an algorithm that builds clusters incrementally, avoiding recomputation at each threshold step.
• Stored TFCE results on a region of interest (ROI)-based structure instead of an FC edge structure for enhanced speed.
• Developed and validated a novel graph transformation for applying IC-TFCE to voxel data.

Main Results:

• IC-TFCE achieves a 3-93x speedup for FC TFCE, depending on the precision parameter $dh$.
• The algorithm makes TFCE analyses with over 1000 ROIs computationally tractable.
• Mathematical proofs and numerical comparisons validate the correctness of IC-TFCE.
• An empirical power analysis guided parameter selection for practitioners.

Conclusions:

• IC-TFCE offers a computationally efficient and scalable alternative to standard TFCE for neuroimaging statistical inference.
• This method significantly enhances the feasibility of TFCE for large-scale, high-resolution neuroimaging studies.
• IC-TFCE facilitates more robust statistical analyses and aids in parameter selection for researchers.