Reliability-based voxel selection

• Published in: NeuroImage (Orlando, Fla.) Vol. 207; p. 116350
• Main Authors: Tarhan, Leyla, Konkle, Talia
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 15-02-2020; Elsevier Limited; Elsevier
• Subjects: Algorithms; Brain - physiology; Brain mapping; Brain Mapping - methods; Condition-rich designs; Data reliability; Datasets; fMRI; Functional magnetic resonance imaging; Human subjects; Humans; Hypotheses; Image Interpretation, Computer-Assisted - methods; Image Processing, Computer-Assisted - methods; Magnetic Resonance Imaging - methods; Neuroimaging; Pattern Recognition, Automated - methods; Reproducibility of Results; Researchers; Voxel selection; fMRI; Voxel selection; Data reliability; Condition-rich designs
• ISSN: 1053-8119; 1095-9572
• DOI: 10.1016/j.neuroimage.2019.116350

While functional magnetic resonance imaging (fMRI) studies typically measure responses across the whole brain, not all regions are likely to be informative for a given study. Which voxels should be considered? Here we propose a method for voxel selection based on the reliability of the data. This method isolates voxels that respond consistently across imaging runs while maximizing the reliability of multi-voxel patterns across the selected voxels. We estimate that it is suitable for designs with at least 15 conditions. In two example datasets, we found that this proposed method defines a smaller set of voxels than another common method, activity-based voxel selection. Broadly, this method eliminates the need to define regions or statistical thresholds a priori and puts the focus on data reliability as the first step in analyzing fMRI data.
•When predicting and mapping voxel responses, which cortex should be considered?•We introduce a method to isolate cortex that responds reliably across fMRI runs.•This method is suitable for condition-rich designs with at least 15 conditions.•Notably, it puts the focus on reliability as the first stage of fMRI data analysis.