🤖 AI Summary
Functional MRI (fMRI) posterior inference suffers from cluster-size dependence, undermining reliability across spatial scales. Method: We systematically compare Notip and pARI—two permutation-based extensions of All-Resolutions Inference (ARI)—under varying cluster sizes, using a data-driven clustering and posterior proportion lower-bound estimation framework evaluated on the NeuroVault dataset. Results: Notip significantly outperforms pARI and baseline ARI for small clusters, yielding more robust and informative lower bounds that enable fine-grained “drill-down” analysis; conversely, pARI excels for large clusters. This reveals a fundamental trade-off between spatial precision and statistical sensitivity, refuting the assumption of universal superiority for any single method. Critically, we establish, for the first time, well-defined applicability boundaries for both approaches—providing an interpretable, context-aware basis for method selection in multi-scale fMRI inference.
📝 Abstract
All Resolutions Inference (ARI) is a post hoc inference method for functional Magnetic Resonance Imaging (fMRI) data analysis that provides valid lower bounds on the proportion of truly active voxels within any, possibly data-driven, cluster. As such, it addresses the paradox of spatial specificity encountered with more classical cluster-extent thresholding methods. It allows the cluster-forming threshold to be increased in order to locate the signal with greater spatial precision without overfitting, also known as the drill-down approach. Notip and pARI are two recent permutation-based extensions of ARI designed to increase statistical power by accounting for the strong dependence structure typical of fMRI data. A recent comparison between these papers based on large voxel clusters concluded that pARI outperforms Notip. We revisit this conclusion by conducting a systematic comparison of the two. Our reanalysis of the same fMRI data sets from the Neurovault database demonstrates the existence of complementary performance regimes: while pARI indeed achieves higher sensitivity for large clusters, Notip provides more informative and robust results for smaller clusters. In particular, while Notip supports informative ``drill-down''exploration into subregions of activation, pARI often yields non-informative bounds in such cases, and can even underperform the baseline ARI method.