Evaluation of thresholding methods for activation likelihood estimation meta-analysis via large-scale simulations

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Statistical Analysis

Background:

• Threshold-free cluster enhancement (TFCE) is a popular statistical inference method in neuroimaging.
• TFCE is known for higher sensitivity and not requiring voxel-level thresholds compared to other methods.
• Activation Likelihood Estimation (ALE) is a common method for coordinate-based neuroimaging meta-analysis.

Purpose of the Study:

• To evaluate the applicability and performance of TFCE in ALE meta-analyses.
• To compare TFCE against voxel-level and cluster-level family-wise error (FWE) correction methods.
• To determine if TFCE offers advantages over existing correction methods in ALE.

Main Methods:

• Conducted large-scale simulations using over 200,000 artificial meta-analysis datasets.
• Varied the number of experiments and spatial convergence across datasets.
• Applied ALE to datasets and compared TFCE with voxel-level and cluster-level FWE correction.

Main Results:

• All three methods (TFCE, cFWE, voxel-level FWE) produced valid results with minimal spurious findings.
• TFCE sensitivity was comparable to cluster-level FWE (cFWE) but slightly worse in some scenarios.
• cFWE produced the largest clusters, followed by TFCE, while voxel-level FWE yielded the smallest, most specific clusters.

Conclusions:

• TFCE does not outperform standard cFWE correction in ALE meta-analyses.
• TFCE is computationally more expensive than cFWE.
• The general use of TFCE for ALE meta-analysis is not recommended due to lack of superior performance and increased computational cost.