Home >  CCN talks >

Innovators in Cognitive Neuroscience: March 9, 2021

Stephanie Noble

Stephanie Noble

Postdoctoral Research Fellow

Department of Radiology and Biomedical Imaging

Yale School of Medicine

Noble webpage

Time: 12:00-1:00pm

Place: Zoom - https://dartmouth.zoom.us/j/95561161544?pwd=Y3FNdTdrTU1rcFRVdU9uVElnRWlwdz09

Charitable causes: Cientifico Latino and Clubes de Ciencia Perú

Leveling up: How broader levels of inference improve power in functional connectivity

Abstract: Inference in fMRI commonly occurs at the level of clusters of contiguous voxels, reflecting the observations that 1) neighboring voxels are dependent (i.e., share similar properties), and 2) the brain exhibits functional segregation between areas. This has been translated to the context of functional connectivity via the Network-Based Statistic (NBS), where inference occurs at clusters of contiguous edges. Yet while these approaches focus on local dependence, recent work has underscored the dependence between widespread areas spanning the brain. Here, we introduce nonparametric procedures for inference at the level of large-scale networks and the whole-brain, and empirically compare their power with edge- and cluster-level inference. Resampling 7 tasks and 3 group sizes from the Human Connectome Project revealed that broader levels of inference are better powered to detect effects. Differences were most striking at smaller sample sizes, with network-level inference achieving nearly double the power of edge- and cluster-level inference at n=40 (paired sample). All methods achieved valid FWER control, with particularly conservative control at the whole-brain level. Increasing effect size and power with more pooling highlights the broad, distributed nature of brain activity during tasks. In light of recent concerns about low power in fMRI, broader-level inference may present a needed avenue towards facilitating discovery in neuroimaging.