ARIbrain - Valid circular inference for fMRI analysis #21
Description
Title
ARIbrain - Valid circular inference for fMRI analysis
Presentor and Affiliation
Wouter, Weeda, Leiden University (@wdweeda)
Xu Chen, Leiden University Medical Centre (@xuchen312)
Collaborators
Jelle Goeman, Leiden University Medical Centre
Jonathan Rosenblatt, University of the Negev
Livio Finos, University of Padua
Aldo Solari, University of Milan Bicocca
Github Link (if applicable)
https://github.com/wdweeda/ARIbrain
Abstract (max. 200 words):
Analysis of fMRI data is often constrained to a strict set of procedures to avoid statistical pitfalls like double-dipping or increased false-positive rates. In practice this means that detected regions of activation can often not be analysed afterwards.
One of the solutions to this problem is the recently developed method All-resolutions inference (ARI) [1]. ARI allows researchers to estimate the proportion of truly active voxels within freely chosen regions-of-interest (ROIs). These ROIs can be chosen based on a priori defined regions (e.g. anatomical regions or meta-analysis locations), or on ROIs derived from the same data (e.g. cluster-forming thresholds, or data from other experimental conditions). All analyses can be performed as many times as a researcher wants with full family-wise error (FWE) control.
This gives ARI almost unlimited freedom in analysing brain images. For example, researchers can use multiple thresholds within the same image or base analysis on activation data from the same experiment. In addition, researchers are allowed to ‘drill- down’ or zoom in within these ROIs, without losing FWE control.
In this demonstration, we present the ARIbrain package. We will highlight the flexibility of the ARI framework and show how the ARI method can be used in analysis pipelines.
Preferred Session
Oral sessions and demos - 3. Demo: New advances in open neuroimaging methods
Additional Context
[1] Rosenblatt et al. 2018. All-resolutions Inference. Neuroimage