Region of interest-wise multiple-comparison correction in structural MRI: Brief review and a simulation-based...

Region of interest (ROI)-based analyses are commonly employed in structural neuroimaging; however, practical guidance on defining the multiple-comparison test family and selecting appropriate error-control procedures remains limited. In this study, we combine a concise methodological review with a simulation-based twin-set framework using real MRI-derived ROI volumes to examine how the definition of the test family and the choice of the correction method jointly influence the minimum detectable effect size in ROI-wise analyses. T1-weighted MRI data from 110 participants were processed with FreeSurfer, and volumetric measures from 130 ROIs were grouped into anatomically defined frontal, temporal, parietal, and occipital families. Paired groups were constructed to minimize baseline differences, and controlled volume increases ranging from 1% to 20% were introduced into a single ROI, while all other regions remained unchanged. Statistical significance thresholds were then evaluated for global whole-brain versus lobe-wise test families using Bonferroni correction and false discovery rate (FDR) control with the Benjamini-Yekutieli procedure. On average, uncorrected significance was reached at a 5.7% volume increase. Global Bonferroni correction required approximately 13.0%, and lobe-wise Bonferroni correction required 11.6%. Under FDR-BY correction, the thresholds were 15.3% for the global family and 13.7% for lobe-wise families. Restricting the test family from the whole brain to anatomically constrained lobe-wise subsets reduced the minimum detectable effect size by approximately 1.5% under both correction frameworks, while preserving the relative ordering of ROI-level sensitivity. These findings suggest that in ROI-based structural MRI, statistical sensitivity may depend not only on the choice between family-wise error rate or FDR control but also on how the test family is defined. A pragmatic implication is that confirmatory analyses should predefine anatomically or functionally coherent ROI families and apply conservative multiplicity control within these families, whereas exploratory analyses may require broader families and more flexible inference with explicit acknowledgment of the associated error structure.