Abstracts

Rationale: Neurosynth is a large-scale, open-source platform that uses text-mining, natural language processing, and meta-analysis to collate data from 14,371 published papers in the fMRI literature and provide a voxel-by-voxel statistical map of fMRI activations in MNI152 space. The voxel basis of localization in Neurosynth, however, precludes analysis of broader regional patterns of functional terms. We created a new web app, Parcelsynth, to reduce voxel-specific noise and hence more accurately localize regional function. Parcelsynth aggregates all voxel data that lie within the 1 cm parcels of the Yale Brain Atlas (YBA) to provide an anatomical, atlas-based representation of the Neurosynth database through an online user interface for education and research purposes._x000D_
Methods: Sourcing the Neurosynth dataset from its GitHub repository, we translated each MNI152 voxel activation into the corresponding YBA parcel and computed an average activation weighting for each parcel. We created a dataset with dimensions 696 YBA parcels by 1334 Neurosynth functional terms. Each element of this dataset reflects an averaged one-way ANOVA z-score of activations for a parcel. For additional descriptive content, we also created a dataset to reference the constituent voxel activations and paper details localized to each YBA parcel. We used R Shiny to create an online user interface for analysis and visualization of these datasets._x000D_
Results: Parcelsynth is a web application for exploring fMRI-based functional localizations collated from data in 14,371 published papers on an anatomical brain atlas. There are two core tools of Parcelsynth. The first is search-by-parcel (Figure 1). Search-by-parcel enables the user to click on a YBA parcel of interest to view relevant functional terms in a word cloud and bar plot organized by z-score, papers that report activations, and the voxels of reported activations. The second core tool is search-by-term (Figure 2). Search-by-term enables the user to search for a functional term and view a z-score map on YBA for that term. The z-score map colors the 696 parcels such that the color intensity reflects the magnitude of z-score. Left hemisphere, right hemisphere, and whole brain atlas views are available for all mesh plots featured on Parcelsynth._x000D_
Conclusions: We present a powerful tool for exploring fMRI-based functional localizations on an anatomical brain atlas. We translated the open-source fMRI data repository of Neurosynth into YBA space and incorporated new functionality such as the "search-by-parcel" tool, which enables users to view relevent functional terms associated with a selected region of interest. Parcelsynth can be used for education and research purposes where exploring large fMRI datasets through an anatomical database may be valuable._x000D_
Funding: This work was supported by grants from the National Institutes of Health (NIH NS109062), the CG Swebilius Trust, and two donors from Yale New Haven Hospital.