A cluster mass permutation test with contextual enhancement for fMRI activation detection

Area of Science:

• Neuroimaging
• Statistical Analysis
• Brain Imaging

Background:

• Gaussian random field (GRF) methods are standard for neuroimaging statistical inference and family-wise error rate (FWE) control.
• GRF methods rely on assumptions about error field properties, including lattice approximation and differentiable spatial autocorrelation.
• Permutation tests offer data-driven FWE control without strict distributional assumptions.

Purpose of the Study:

• Introduce a novel analysis procedure, the cluster mass permutation test with contextual enhancement (CMPCE).
• Compare the performance of CMPCE against established GRF methods.
• Evaluate FWE rates, cluster detection probability, and delineation accuracy of both methods.

Main Methods:

• CMPCE involves pre-whitening data to remove temporal autocorrelations.
• Both CMPCE and GRF were tested using null datasets and simulated activation data.
• The methods were applied to functional MRI (fMRI) data from a tactile stimulation experiment.

Main Results:

• Both CMPCE and GRF exhibited inflated FWE rates on null data, suggesting partial failures in preprocessing steps.
• CMPCE demonstrated superior delineation accuracy, classifying fewer voxels incorrectly at equivalent cluster detection probabilities.
• GRF methods achieved higher maximal cluster detection probabilities, while both methods yielded similar qualitative results on fMRI data.

Conclusions:

• CMPCE is a promising new method for fMRI analysis, particularly when high delineation accuracy is critical.
• The findings highlight potential issues in preprocessing steps affecting FWE control in both methods.
• Further research may refine CMPCE for enhanced neuroimaging statistical inference.