Speeding up Permutation Testing in Neuroimaging.

Area of Science:

• Neuroimaging
• Statistical analysis
• Computational neuroscience

Background:

• Multiple hypothesis testing is a pervasive challenge in neuroimaging, necessitating accurate Family-Wise Error Rate (FWER) estimation.
• Traditional Bonferroni correction is overly conservative, while standard permutation testing is computationally intensive.
• Existing methods struggle to balance statistical rigor with computational efficiency in neuroimaging data analysis.

Purpose of the Study:

• To develop a computationally efficient permutation testing method for estimating FWER in neuroimaging.
• To address the limitations of existing FWER correction techniques, particularly their conservativeness or high computational cost.
• To enable more powerful and feasible statistical inference in neuroimaging studies.

Main Methods:

• Proposed a novel permutation testing methodology by analyzing the low-rank plus low-variance structure of the permutation matrix.
• Utilized matrix completion techniques on a highly sub-sampled matrix (0.5%) to approximate the full permutation matrix.
• Evaluated the method's performance on four diverse neuroimaging datasets.

Main Results:

• Achieved a computational speedup factor of approximately 50× compared to standard permutation testing.
• Demonstrated high accuracy in recovering the Family-Wise Error Rate (FWER) distribution.
• Showcased stable and faithful recovery of the estimated alpha-threshold.

Conclusions:

• The proposed sub-sampled matrix completion approach offers a significant speedup for permutation testing in neuroimaging.
• This novel methodology maintains the fidelity of FWER estimation, enhancing statistical power.
• The findings suggest a more computationally tractable approach to robust statistical inference in neuroimaging research.