An integrated cluster-wise significance measure for fMRI analysis

Area of Science:

• Neuroimaging
• Neuroscience
• Biostatistics

Background:

• Cluster-wise inference is a standard technique in functional magnetic resonance imaging (fMRI) analysis.
• Traditional methods using suprathreshold voxel counts for cluster-level statistics are suboptimal, potentially affecting statistical power and false-positive rates.
• These methods often overlook crucial information like voxel-wise significance levels and inter-voxel dependencies.

Purpose of the Study:

• To introduce a novel Integrated Cluster-wise significance Measure (ICM) for fMRI cluster-wise inference.
• To improve the power and control the family-wise error rate (FWE) in fMRI analyses.
• To provide a computationally efficient method for cluster-level significance determination.

Main Methods:

• Developed an Integrated Cluster-wise significance Measure (ICM) that combines cluster extent, voxel-level significance (p-values), and activation dependence.
• Employed a computationally efficient strategy for ICM based on probabilistic approximation theories.
• Validated the ICM method through extensive simulations and application to two real fMRI datasets.

Main Results:

• The proposed ICM method demonstrated improved statistical power compared to traditional methods.
• ICM effectively controlled the family-wise error rate (FWE).
• The computational load for ICM-based cluster-wise inference, including permutation tests, was found to be affordable.

Conclusions:

• The Integrated Cluster-wise significance Measure (ICM) offers a superior approach for cluster-level significance determination in fMRI.
• ICM enhances analytical power while maintaining robust control over FWE.
• This method provides a computationally feasible advancement for fMRI data analysis.