False discovery rate revisited: FDR and topological inference using Gaussian random fields

Area of Science:

• Neuroimaging
• Statistical Inference
• Data Analysis

Background:

• The false discovery rate (FDR) is a common metric for controlling errors in multiple comparisons.
• Conventional voxel-wise FDR procedures are widely used in statistical parametric mapping (SPM).
• However, these methods may not be suitable for inferring topological features like activation regions in continuous imaging data.

Purpose of the Study:

• To evaluate the appropriateness of conventional voxel-wise FDR control for topological inference in SPM.
• To highlight the discrepancy between controlling voxel-wise FDR and controlling the FDR of activations.
• To propose a more suitable FDR control method for image-based statistical inference.

Main Methods:

• Revisiting existing literature on FDR control.
• Analyzing the characteristics of continuous signals in neuroimaging data.
• Developing and illustrating a novel FDR control strategy based on connected excursion sets.

Main Results:

• Controlling the FDR of individual voxels does not equate to controlling the FDR of actual activation clusters or regions.
• Inference based on conventional voxel-wise FDR is inadequate for identifying topological features in SPMs.
• A proposed solution involves controlling the FDR of connected excursion sets, characterized by their volume.

Conclusions:

• Standard voxel-wise FDR control is inappropriate for topological inference in statistical parametric mapping.
• A modified FDR approach, focusing on connected excursion sets, is necessary for accurate identification of activation patterns in neuroimaging.
• This revised approach ensures more reliable statistical inference on the spatial characteristics of brain activity.