Topological Cluster Statistic (TCS): fMRI cluster enhancement using anatomically-guided inference
Date:
Poster presentation on Topological Cluster Statistic (TCS)
Abstract:
Task-evoked changes in brain activity are known to be distributed over widespread anatomically coordinated neural populations. Nevertheless, conventional functional magnetic resonance imaging (fMRI) studies widely use cluster-based statistical inference approaches to detect patterns of local change in brain activity. These approaches are based on grouping spatially contiguous regions of significant effect and thus neglect the complex anatomical topology that links spatially disjoint brain regions together. Furthermore, insufficient power and disproportionate false positive rates reportedly hinder the potential capability of these methods for optimal inference. Here, we propose a structural-connectivity-guided clustering framework, called Topological Cluster Statistic (TCS), that enhances the sensitivity and interpretability of cluster-based statistical inference by leveraging white matter anatomical connectivity information. Our method harnesses multimodal information from both diffusion tractography and functional imaging to better cluster task fMRI activation maps. In contrast to conventional approaches, TCS provides a consistent improvement in statistical power over a wide range of effects. This improvement can range up to a 10%-50% sensitivity increase in the detection of local effects for various fMRI tasks and sample sizes with the greatest gains for medium-sized effects. TCS additionally enables inspection of the underlying anatomical networks implicated in an observed effect and can thus uncover knowledge regarding the anatomical underpinnings of brain activation. This novel approach is made available in FSL PALM software to facilitate future usability for a wide range of study designs. Given the increasing recognition that activation does not occur in isolation but instead reflects widespread, coordinated processes, TCS provides a solution to integrate the known anatomical structure underlying widespread activations into neuroimaging analyses moving forward.