Using large-scale neuroimaging data to overcome historic cerebellar neglect

Brisbane Convention & Exhibition Centre 

Coordinated-based meta-analyses such as activation likelihood estimation (ALE) have contributed greatly to the understanding that the cerebellum plays important roles across diverse motor, cognitive, and emotional behaviors. However, cerebellar imaging is faced with unique challenges. One is that it has historically been forgotten about in literature, especially in non-motor domains. This leads to stark differences in reporting rates of superior and inferior (underreported) activations. In turn, this renders ALE’s null-hypothesis invalid and calls into question the accuracy of previous cerebellar meta-analyses. Here, we introduce cerebellum-specific ALE (C-SALE), that overcomes differences in reporting rates across the cerebellum. It does so by updating its null model to represent voxel-wise probabilities of finding activation foci. The null model was created by combining all studies in the extensive, manually indexed BrainMap database. We first compare the classic ALE method to our new C-SALE method, showing much improved specificity of the latter. C-SALE finds convergence in twelve of forty task datasets, including aspects of action, memory, social cognition, emotion, and perception. Having access to many moderate-to-large-sized tasks, we were able to extensively characterize C-SALE stability using repeated subsampling. We show that mapping stability increases across task domains quite similarly as subsampling proportions increase. However, the Action domain stands out, with high subsample consistency leading to convergence across most subsamples. We next compare C-SALE maps to existing cerebellar parcellations, mappings, and gradients, often finding significant spatial correspondence between behaviorally related C-SALE and target maps. Lastly, we apply the same unbiased framework to reveal brain-wide coactivation networks across action, working memory, and vision task sets. Together, our findings provide important refinement of the cerebellar behavioral topography. Our method can be flexibly extended to any volumetric brain region-of-interest. Full code to do this (github.com/NevMagi/cerebellum_specific_ALE) including a fast GPU implementation of ALE and C-SALE (speed-ups ~100x; github.com/amnsbr/nimare-gpu) are openly shared. Concluding, our study furthers understanding of cerebellar and brain subregional functions, suggesting regions suitable for study of basic and clinical applications.